AKANIL

NEW PERSPECTIVES FOR ANALYZING THE
BREAKUP, ENVIRONMENT, EVOLUTION,

COLLISION RISK AND REENTRY OF
SPACE DEBRIS OBJECTS

A Thesis
Submitted for the Degree of
Doctor of Philosophy
in the Faculty of Engineering

By
A. K. ANILKUMAR
VIKRAM SARABHAI SPACE CENTRE, TRIVANDRUM

Department of Aerospace Engineering
Indian Institute of Science
Bangalore -560 012, INDIA
February 2004

D
DDE
EEC
CCL
LLA
AAR
RRA
AAT
TTI
IIO
OON
NN

I hereby declare that the entire work recorded in this thesis has been
carried out by me under the supervision of Prof. M. R. Ananthasayanam,
Department of Aerospace Engineering, Indian Institute of Science, Bangalore
and Dr. P. V. Subba Rao, Head, Applied Mathematics Division, Vikram
Sarabhai Space Centre, Trivandrum and no part of it has been submitted for
any degree or diploma of any institution previously.

Bangalore, A K Anilkumar
February 2004.
Table of contents
i

Contents Page
No.

Acknowledgements
Abstract
List of Publications
List of Figures
List of Tables
List of Symbols and Abbreviations

Chapter 1: An Overview of the Space Debris Environment

1.1 Introduction
1.2 Sources for the Creation of Space Debris
1.3 Sink Mechanisms for the Decay of Space Debris
1.4 Measurement of Space Debris

1.4.1 Tracked Debris Population
1.4.2 Untracked Debris Population

1.5 Evolution of Debris Clouds
1.6 Risks due to Orbital Debris
1.7 Design and Operational Measures to Control Debris

xi
xiii
xvii
xix
xxix
xxxv

1
4
6
8

11
13
16

TABLE OF CONTENTS
New perspectives for analyzing the breakup, environment, evolution, collision risk and reentry of space debris
objects
ii
Contents Page
No.

1.8 NASA Guidelines for Limiting Orbital Debris
1.9 Models of the Space Debris Scenario
1.10 Future Scenario of the Debris
1.11 Importance of Space Debris Studies
1.12 Contributions in the Present Study

Chapter 2: Review of Literature on the Major Issues in Space Debris
Studies

2.1 Introduction
2.2 Earlier Breakup Models of Space Objects

2.2.1 Breakup Model of Johnson and McKnight
2.2.2 Breakup Model of Jehn
2.2.3 NASAs Breakup Model

2.3 Models of the Space Debris Scenario

2.3.1 MASTER (Meteoroid and Space Debris
Terrestrial Environment Reference) Models
2.3.2 ORDEM (Orbital Debris Engineering Model)
2.3.3 Other Debris Models
2.3.4 Uncertainties in Various Models
2.3.5 Comparison of the Models

2.4 Models for the Evolution of Space Debris Scenario

18
21
22
24
25

29
30

33

51

Table of contents
iii
Contents Page
No.

2.5 Collision Probability Assessment
2.6 Reentry Prediction of Risk Objects
2.7 Conclusions

Chapter 3: A Proposed Model of On Orbit Breakup ASSEMBLE
and Simulation of Some Historical Breakups

3.1 Introduction
3.2 Conceptual Basis for the Present Model
3.3 Earlier Approaches to Modeling the Orbital Breakups
3.4 Breakup of the Indian PSLV-TES Mission Spent Upper stage

3.4.1 Delta Velocity Imparted to the Fragments
3.4.2 Components of the Delta Velocities
3.4.3 Distribution of the Mass of the Fragments
3.4.4 Determination of the Area to Mass Ratio
3.4.5 Mass of the Debris Fragments and their
Imparted Delta Velocities

3.5 The Approach in the Present ASSEMBLE Model
3.6 The Proposed ASSEMBLE Model
3.7 Steps for Simulating a Typical Breakup
3.8 Simulation of Some Historical Breakups Using the Proposed
ASSEMBLE Model

3.8.1 Case Study of PSLV-TES Spent Upper Stage

53
55
56

57
57
59
61

66
67
68
70

objects
iv
Contents Page
No.

3.8.2 Case Study of STEP II R/B
3.8.3 Case Study of CBERS-I/ SACI -1 R/B
3.8.4 Case Study of COSMOS 1813
3.8.5 Case Study of SPOT 1 R/B

3.9 Comparison with the Existing Models
3.10 Conclusions

Chapter 4: A New Stochastic Impressionistic Low Earth (SIMPLE)
Engineering Model of the Space Debris Scenario

4.1 Introduction
4.2 Genesis of Space Debris Environment Model
4.3 Differences between the Nature and Environment of Space
Debris and Gas Molecules
4.4 Present Probabilistic Approach for Modeling the Debris
Environment
4.5 Database Utilized for the SIMPLE Engineering Model
4.6 Choice of Suitable Probability Distributions in the Present
SIMPLE Model

4.6.1 Distribution of the Number Density of Debris
with Altitude: f (n, h, t)
4.6.2 Distributions of the Number Density of Debris
with Perigee Height: f(n, h
p
, t)

73
73

89
91
92

94

95
97

Table of contents
v
Contents Page
No.

4.6.3 Distributions of the Eccentricity of the Debris:
f (n, e, t)
4.6.4 Distributions of the Ballistic Coefficients of the
Debris: f (n, B, t)

4.7 Statistical Correlation among the Fragment Characteristics

4.7.1 Correlation between Altitude and Eccentricity
4.7.2 Correlation between Perigee Height and
Eccentricity
4.7.3 Correlation between Eccentricity and B

4.8 The Proposed SIMPLE Model for the Number Density with
respect to Altitude, Perigee Height, the Eccentricity and the
Ballistic Coefficient
4.9 Analysis of the Data for Different Inclination Bands

4.9.1 Parameters of the Distributions for the Number
of Debris with Altitude in Various Inclination
Bands
4.9.2 Parameters of the Distributions for the Number
of Debris with Perigee Heights in Various
Inclination Bands
4.9.3 Distribution of the Eccentricity of the Debris in
the Different Inclination Bands

106

107

111

objects
vi
Contents Page
No.

4.9.4 Distribution of the Ballistic Coefficient of the
Debris in the Different Inclination Bands

4.10 SIMPLE Model for Different Inclination Bands
4.11 Further Analysis on the Distribution of the Eccentricity and
Ballistic Coefficients

4.11.1 Distribution of the Eccentricity and Ballistic
Coefficient in Various Altitude Bins
4.11.2 Distribution of the Eccentricity and Ballistic
Coefficient in Various Perigee Height Bins
4.11.3 Characterization with respect to Inclination
Bands

4.12 Application of SIMPLE Nature of Debris Environment
4.13 Conclusions

Chapter 5: Evolution of the Orbital Debris Scenario based on Kalman
Filter Approach (Fast Number Density Simulator and
Equivalent Ballistic Coefficient Estimator)

5.0 Prologue
5.1 Introduction
5.2 Present Model

5.2.1 Concept of Binning

117
119

132
133

143
144
146

Table of contents
vii
Contents Page
No.

5.2.2 Concept of Equivalent Fragment and its
Properties in Terms of the Individual Fragment
Characteristics
5.2.2 Evolution and Update of the EQF
Characteristics

5.3 Essence of the Concept of Using Constant Kalman Gain
5.4 Implementation of the Constant Kalman Gain Approach
5.5 The Real World and the Filter World Scenario
5.6 Propagation of the Orbital Characteristics of the
Fragments in the Bins and the EQF
5.7 Kalman Filter Approach (to Assimilate the Measurement
Information and Expand the Scenario to Estimate the Ballistic
Coefficient of the EQFs in the Various Bins)

5.7.1 Filter World State Equations
5.7.2 Filter World Measurement Update Equations

5.8 Case Studies and the Analysis of the Results
5.9 Evolution of Debris Objects Generated due to Explosions

5.9.1 Evolution of a Single Breakup
5.9.2 Variation of the Constant Kalman Gain for the
Number of Fragments
5.9.3 Evolution of a Single Breakup Followed by
Another Breakup

150
151
152
155

156

159
160

objects
viii
Contents Page
No.

More than One Breakups
More than One Breakups and Some Launch
Activities

5.10 Inference of Fragment Properties from the EQF
Characteristics
5.11 Conclusions

Chapter 6: A Reference Collision Probability Estimation Model for
Target Risk Assessment due to Orbital Debris

6.1 Introduction
6.2 Definition of Useful Terms
6.3 Different Approaches for Estimating Collision Probability

6.3.1 Collision Probability as a Poisson Process
6.3.2 Monte Carlo Analysis to Obtain the
Distribution of Closest Approach
6.3.3 A New Approach for the Collision Probability
Assessment
6.3.4 Long Term Evolution of the Debris Using
Sophisticated Environment Models
6.3.5 Collision Avoidance (COLA)

165

166

185
186
187

Table of contents
ix
Contents Page
No.

6.4 Risk Analysis
6.5 Approach Based on Monte Carlo Technique and Extreme
Value Distribution
6.5.1 Collision Probabilities and Time Between
Collisions
6.6 Application of the Proposed Reference Collision Probability
Model
6.7 Comparison of the Present Results with Other Models
6.8 Conclusions

Chapter 7: Prediction of Reentry of Debris Objects: Constant Gain
Kalman Filter Approach

7.1 Introduction
7.2 Filter World Scenario: State Equations
7.3 Filter World Scenario: Measurement Equations
7.4 Adaptive Extended Kalman Filtering Approach with Time
Varying Gains

7.4.1 Adaptive Procedure of Myers and Tapley for
Estimating Q and R

7.5 Constant Kalman Gain Approach
7.6 Genetic Algorithm
7.7 Results from the Adaptive Kalman Filtering Approach
7.8 Results and Discussions: Constant Kalman Gain Approach

196
197

198

203
204

209
212
212
213

216
217
217
218

objects
x
Contents Page
No.

7.8.1 Simulated Case Studies
7.8.2 Case Study with US Sat No. 25947, SOYUZ
7.8.3 Case Study with SROSS-C2 Satellite
7.8.4 Case Study with COSMOS 1043 Object
7.8.5 Case Study with COSMOS 389 Satellite

7.9 Conclusions

Chapter 8: Conclusions and Suggestions for Further Work

Appendices

1 Two Line Elements (TLE) Format
2 Number Mass Relationships in On Orbit Breakup
3 Basics of Orbital Motion
4 Statistical Distributions: Laplace, Log Normal, Weibull
Distributions
5 Delta Velocity Relationships
6 Delta Velocity Components Equations
7 Genetic Algorithms
8 A Brief Introduction to Kalman Filter and its Implementation
using Constant Gains
References

224

235

237
247
253
259

263
265
267
269

283

Acknowledgements
xi

A Ac ck kn no ow wl le ed dg ge em me en nt ts s

I am deeply indebted to Prof. M. R. Ananthasayanam, Department of Aerospace
Engineering, Indian Institute of Science, Bangalore, for his inspiring and invaluable
guidance, continuous support and encouragement. I place on record my gratitude towards
him for the suggestions, comments and fruitful discussions, which had a great influence on
the work that has been carried out.

I acknowledge my special gratitude towards Dr. P. V. Subba Rao, Head, APMD, Vikram
Sarabhai Space Centre (VSSC) who supported and encouraged me throughout the entire
period of this work by technical discussions, suggestions and critical evaluation.

I sincerely thank Sri. G. Madhavan Nair, Chairman, Indian Space Research Organization for
permitting me to pursue the research programme at Indian Institute of Science, Bangalore.

I would like to thank Dr. B. N. Suresh, Director, VSSC and Sri. R. V. Perumal, Director
(Projects), VSSC for providing me the opportunity to carry out this work. I express my
thanks to Dr. V. Adimurthy, Deputy Director, VSSC, the first person who discussed with me
about the Space Debris aspects and to Sri. Madan Lal, Group Director, AFDG, VSSC for
providing me the necessary supports. I would like to thank Dr. A. R. Acharya, Visiting
Professor, IISc (formerly Deputy Director, VSSC) and Sri. K. L. Prasad, TCS, Trivandrum
(formerly Scientist, VSSC) for encouraging me to pursue the studies. I thank my colleagues
in APMD, specifically Dr. R. K. Sharma and Sri. N. S. Madhavan for their support.

I would like to express my deepest appreciation to my wife Smt. Lekha and to our two small
kids Athul Krishna and Ashwin Krishna for their daily sacrifices and their love and affection
showered on me. I am happy to acknowledge my parents, Sri. Krishnankutty Nair and Smt.
objects
xii
Anandavally Amma, for their silent recognition of my efforts and continued support and my
mother in law, Smt. Bhavani Amma, for her constant encouragement. I will place my thanks
to in-laws, brothers and sisters for the encouragement given to me in doing the research.

I am grateful to Sri. Manmatha Kumar for the secretariatal assistance provided in preparing
the thesis.

A K Anilkumar
Abstract

xiii

In the space surrounding the earth there are two major regions where orbital debris causes
concern. They are the Low Earth Orbits (LEO) up to about 2000 km, and Geosynchronous
Orbits (GEO) at an altitude of around 36000 km. The impact of the debris accumulations are
in principle the same in the two regions; nevertheless they require different approaches and
solutions, due to the fact that the perturbations in the orbital decay due to atmospheric drag
effects predominates in LEO, gravitational forces including earths oblateness and luni solar
effects dominating in GEO are different in these two regions. In LEO it is generally known
that the debris population dominates even the natural meteoroid population for object sizes 1
mm and larger. This thesis focuses the study mainly in the LEO region.

Since the first satellite breakup in 1961 up to 01 January 2003 more than 180 spacecraft and
rocket bodies have been known to fragment in orbit. The resulting debris fragments
constitute nearly 40% of the 9000 or more of the presently tracked and catalogued objects by
USSPACECOM. The catalogued fragment count does not include the much more numerous
fragments, which are too small to be detected from ground. Hence in order to describe the
trackable orbital debris environment, it is important to develop mathematical models to
simulate the trackable fragments and later expand it to untrackable objects. Apart from the
need to better characterize the orbital debris environment down to sub millimeter particles,
there is also a pressing necessity of simulation tools able to model in a realistic way the long
term evolution of space debris, to highlight areas, which require further investigations, and to
study the actual mitigation effects of space policy measures.

The present thesis has provided newer perspectives for five major issues in space debris
modeling studies. The issues are (i) breakup modeling, (ii) environment modeling, (iii)
evolution of the debris environment, (iv) collision probability analysis and (v) reentry
prediction.
ABSTRACT
objects
xiv
The Chapter 1 briefly describes an overview of space debris environment and the issues
associated with the growing space debris populations. A literature survey of important earlier
work carried out regarding the above mentioned five issues are provided in the Chapter 2.

The new contributions of the thesis commence from Chapter 3. The Chapter 3 proposes a
new breakup model to simulate the creation of debris objects by explosion in LEO named A
Semi Stochastic Environment Modeling for Breakup in LEO (ASSEMBLE). This model is
based on a study of the characteristics of the fragments from on orbit breakups as provided in
the TLE sets for the INDIAN PSLV-TES mission spent upper stage breakup. It turned out
that based on the physical mechanisms in the breakup process the apogee, perigee heights
(limited by the breakup altitude) closely fit suitable Laplace distributions and the eccentricity
follows a lognormal distribution. The location parameters of these depend on the orbit of the
parent body at the time of breakup and their scale parameters on the intensity of explosion.
The distribution of the ballistic coefficient in the catalogue was also found to follow a
lognormal distribution. These observations were used to arrive at the proper physical,
aerodynamic, and orbital characteristics of the fragments. Subsequently it has been applied as
an inverse problem to simulate and further validate it based on some more typical well
known historical on orbit fragmentation events. All the simulated results compare quite well
with the observations both at the time of breakup and at a later epoch. This model is called
semi stochastic in nature since the size and mass characteristics have to be obtained from
empirical relations and is capable of simulating the complete scenario of the breakup.

A new stochastic environment model of the debris scenario in LEO that is simple and
impressionistic in nature named SIMPLE is proposed in Chapter 4. Firstly among the
orbital debris, the distribution of the orbital elements namely altitude, perigee height,
eccentricity and the ballistic coefficient values for TLE sets of data in each of the years were
analyzed to arrive at their characteristic probability distributions. It is observed that the
altitude distribution for the number of fragments exhibits peaks and it turned out that such a
feature can be best modeled with a tertiary mixture of Laplace distributions with eight
parameters. It was noticed that no statistically significant variations could be observed for the
parameters across the years. Hence it is concluded that the probability density function of the
Abstract

xv
altitude distribution of the debris objects has some kind of equilibrium and it follows a three
component mixture of Laplace distributions. For the eccentricity e and the ballistic
parameter B values the present analysis showed that they could be acceptably quite well
fitted by Lognormal distributions with two parameters. In the case of eccentricity also the
describing parameter values do not vary much across the years. But for the parameters of the
B distribution there is some trend across the years which perhaps may be attributed to causes
such as decay effect, miniaturization of space systems and even the uncertainty in the
measurement data of B. However in the absence of definitive cause that can be attributed for
the variation across the years, it turns out to be best to have the most recent value as the
model value. Lastly the same kind of analysis has also been carried out with respect to the
various inclination bands. Here the orbital parameters are analyzed with respect to the
inclination bands as is done in ORDEM (Kessler et al 1997, Liou et al 2001) for near circular
orbits in LEO. The five inclination bands considered here are 0-36 deg (in ORDEM they
consider 19-36 deg, and did not consider 0-19 deg), 36-61 deg, 61-73 deg, 73-91 deg and 91-
180 deg, and corresponding to each band, the altitude, eccentricity and B values were
modeled. It is found that the third band shows the models with single Laplace distribution for
altitude and Lognormal for eccentricity and B fit quite well. The altitude of other bands is
modeled using tertiary mixture of Laplace distributions, with the e and B following once
again a Lognormal distribution. The number of parameter values in SIMPLE is, in general,
just 8 for each description of altitude or perigee distributions whereas in ORDEM96 it is
more. The present SIMPLE model captures closely all the peak densities without losing the
accuracy at other altitudes.

The Chapter 5 treats the evolution of the debris objects generated by on orbit breakup. A
novel innovative approach based on the propagation of an equivalent fragment in a three
dimensional bin of semi major axis, eccentricity, and the ballistic coefficient (a, e, B)
together with a constant gain Kalman filter technique is described in this chapter. This new
approach propagates the number density in a bin of a and e rapidly and accurately without
propagating each and every of the space debris objects in the above bin. It is able to
assimilate the information from other breakups as well with the passage of time. Further this
objects
xvi
approach expands the scenario to provide suitable equivalent ballistic coefficient values for
the conglomeration of the fragments in the various bins. The heart of the technique is to use a
constant Kalman gain filter, which is optimal to track the dynamically evolving fragment
scenario and further expand the scenario to provide time varying equivalent ballistic
coefficients for the various bins.

In the next chapter 6 a new approach for the collision probability assessment utilizing the
closed form solution of Wiesel (1989) by the way of a three dimensional look up table, which
takes only air drag effect and an exponential model of the atmosphere, is presented. This
approach can serve as a reference collision probability assessment tool for LEO debris cloud
environment. This approach takes into account the dynamical behavior of the debris objects
propagation and the model utilizes a simple propagation for quick assessment of collision
probability. This chapter also brings out a comparison of presently available collision
probability assessment algorithms based on their complexities, application areas and sample
space on which they operate. Further the quantitative assessment of the collision probability
estimates between different presently available methods is carried out and the obtained
collision probabilities are match qualitatively.

The Chapter 7 utilizes once again the efficient and robust constant Kalman gain filter
approach that is able to handle the many uncertain, variable, and complex features existing in
the scenario to predict the reentry time of the risk objects. The constant gain obtained by
using only a simple orbit propagator by considering drag alone is capable of handling the
other modeling errors in a real life situation. A detailed validation of the approach was
carried out based on a few recently reentered objects and comparison of the results with the
predictions of other agencies during IADC reentry campaigns are also presented.

The final Chapter 8 provides the conclusions based on the present work carried together with
suggestions for future efforts needed in the study of space debris. Also the application of the
techniques evolved in the present work to other areas such as atmospheric data assimilation
and forecasting have also been suggested.

list of publications
xvii

The following reports and papers were prepared based on the work carried out for this thesis

1. Anilkumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., " Collision Probability
Analysis of Debris Clouds", Fluid Mechanics Report, 2004 FM 1, Dept. of Aerospace
Engineering, Indian Institute of Science, Bangalore, February 2004.
2. Ananthasayanam, M. R., Anilkumar, A. K., and Subba Rao, P. V., " Evolution of the
Orbital Debris Scenario based on Kalman Filter Approach", Fluid Mechanics Report,
2004 FM 2, Dept. of Aerospace Engineering, Indian Institute of Science, Bangalore,
February 2004.
3. Anilkumar, A. K., Ananthasayanam, M. R., Subba Rao, P. V., and V. Adimurthy,
Onorbit Collision Probability in LEO Using SIMPLE Model, IAC-03-IAA5.2.09,
Presented in IAF Conference, Bremen, October 2003 and also to be published in the
Proceedings of IAF conference, Science and Technology Series, 2003.
4. Ananthasayanam, M. R., Anilkumar, A. K., Subba Rao, P. V., and V. Adimurthy,
Characterisation of Eccentricity and Ballistic Coefficients of Space Debris in Altitude
and Perigee Bins, IAC-03-IAA5.p.04, Presented in IAF Conference, Bremen, October
2003 and also to be published in the Proceedings of IAF conference, Science and
Technology Series, 2003.
5. Anilkumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., Prediction of Reentry
of Space Debris Objects : Constant Gain Kalman Filter Approach, AIAA 2003-5393,
Proceedings of AIAA AFM Conference, TEXAS, August 2003.
6. Sharma, R. K., and Anilkumar, A., K., An integrated approach for risk object reentry
prediction: Development and validation of KSGEN software, VSSC Technical Report,
VSSC/APMD/TR/073/2003, 2003.
7. Sharma, R. K., Anilkumar, A., K., and Xavier James Raj, M., IADC Reentry prediction
campaign 2003 #1: Reentry of COSMOS 389 satellite, VSSC Technical report,
VSSC/APMD/SP/082/2003, 2003.
LIST OF PUBLICATIONS
objects
xviii
8. Anilkumar, A. K., Initiatives and future challenges in Space Debris studies,
Presentation in one day colloquium Aero Vision 2003, VSSC, Trivandrum, 2003.
9. Anilkumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., Simulation of Some
Historical Onorbit Breakups Using ASSEMBLE Model, AIAA-2003-5760, Proceedings
of 41
st
Aerospace Science Meeting & Exhibit (Atmospheric Environment), Reno, January
2003 and also Fluid Mechanics Report, 2002 FM 7, Dept. of Aerospace Engineering,
Indian Institute of Science, Bangalore, April 2002.
10. Ananthasayanam, M. R., Anilkumar, A. K., and Subba Rao, P. V., "A Proposed
Stochastic IMPressionistic Low Earth (SIMPLE) Engineering Model of the Space Debris
Scenario", Fluid Mechanics Report, 2002 FM 5, Dept. of Aerospace Engineering, Indian
Institute of Science, Bangalore, April 2002.
11. Anilkumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., " Simple Model:
Characterization of Eccentricity and Ballistic Coefficient of Space Debris in Altitude and
Perigee Bins", Fluid Mechanics Report, 2002 FM 6, Dept. of Aerospace Engineering,
12. Anilkumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., Analysis and
Modeling of PSLV-TES Mission Spent Upper Stage Breakup in LEO, Fluid Mechanics
Report 2002 FM 4R, Department of Aerospace Engineering, Indian Institute of Science,
Bangalore, February 2002 and also presented in Inhouse Symposium, Department of
aerospace Engineering, Indian Institute of Science, Bangalore, March 2002.
13. Anil Kumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., A New Modeling
Approach for Orbital Breakup in Space, COSPAR 02-A-01843, 2002.
14. Ananthasayanam, M. R., Anil Kumar, A. K., and Subba Rao, P. V., A New Stochastic
Impressionistic Low Earth (SIMPLE) Model of the Space Debris Scenario, COSPAR
02-A-01772, 2002.
15. Anil Kumar, A. K., and Subba Rao, P. V., Reentry Prediction Accuracy Improvement
Using Genetic Algorithm, COSPAR 02-A-01372, 2002.
16. Suresh, I. And Anilkumar, A., K., A note on assessment of NASA guidelines for debris
in the context of PSLV-C3 mission, VSSC / APMD/ TM-PSLV/029/2001, 2001

List of figures
xix

Figures Page
No.


Figure 1.1 An Overview of Space Objects
Figure 1.2 Categorization of Space Debris Population
Figure 1.3 Space Debris Sources and Sinks
Figure 1.4 Distributions of Tracked Objects
Figure 1.5 Spatial Densities of the Various Kinds of Tracked Objects
in LEO
Figure 1.6 Percentages of Various Types of Tracked Objects
Figure 1.7 Evolution of a Debris Cloud
Figure 1.8 Cumulative Numbers of Major Collisions: Future
Predictions

Studies

Figure 2.1 A Satellite Fragmentation Model
Figure 2.2 Comparisons of ORDEM96 and ORDEM 2000
Figure 2.2 Spatial Density Comparisons between Different
Models

3
4
5
10
10

11
12
22

31
43
51

LIST OF FIGURES
objects
xx
Figures Page
No.


Figure 3.1 Three Aspects of Orbital Breakup Modeling and Analysis.
Figure 3.2 The Gabbard Diagram from the TLEs of PSLV-TES Spent
Upper Stage Breakup Fragments after 31 Days of Breakup
Figure 3.3 The Distribution of Apogee, Inclination, Eccentricity and B*
from the TLEs of PSLV-TES Spent Upper Stage Breakup
Fragments after 31 Days of Breakup
Figure 3.4 Flow Chart of the ASSEMBLE Model.
Figure 3.5 PSLV-TES Spent Upper Stage Debris Cloud at the Time of
Breakup Simulated Using ASSEMBLE Model
Figure 3.6 PSLV-TES Spent Upper Stage Debris Cloud after 31 Days of
Breakup Simulated Using ASSEMBLE Model
Figure 3.7 DeltaV Values of PSLV-TES Spent Upper Stage Debris
Fragments from the Observations
Figure 3.8 DeltaV Values of PSLV-TES Spent Upper Stage Debris
Fragments Simulated by ASSEMBLE Model
Figure 3.9 Step II R/B Debris Cloud of 713 Fragments at the Time of
Breakup Simulated using ASSEMBLE Model
Figure 3.10 Step II R/B Debris Cloud of 713 Fragments as of August 29,
1996 (89 Days after the Breakup) as Reconstructed from US
SN Database
Figure 3.11 Step II R/B Debris Cloud of 713 Fragments after 89 Days
of Breakup Simulated using ASSEMBLE Model

59
76

77

78
79

79

80

80

81

81

82

List of figures
xxi
Figures Page
No.

Figure3.12 The Deltav, Semi Major Axis, Inclination and Eccentricity
Distribution of the Fragments of Step II R/B Simulated
Breakup Using ASSEMBLE Model
Figure 3.13 Component Velocity DeltaVr and DelatVt of the Fragments of
Step II R/B Simulated Breakup using ASSEMBLE Model
Figure 3.14 Component Velocity DeltaVt and DelatVn of the Fragments of
Step II R/B Simulated Breakup using ASSEMBLE Model
Figure 3.15 Component Velocity DeltaVn and DelatVr of the Fragments
of Step II R/B Simulated Breakup using ASSEMBLE Model
Figure 3.16 Area to Mass Ratio and Component Velocities of the
Fragments of Step II R/B Simulated Breakup using
ASSEMBLE Model
Figure 3.17 CBERS-1 /SACI-1 Long March 4 Third Stage Debris Cloud
of 280 with One Day of the Event as Reconstructed from US
SN Database
Figure 3.18 CBERS-1 /SACI-1 Long March 4 Third Stage Debris Cloud
of 280 at the Time of Breakup Simulated using ASSEMBLE
Model
Figure 3.19 Cosmos 1813 Debris Cloud of 846 Fragments as
Reconstructed from PARCS Radar Observation Taken about
10 Hours after the Breakup.
Figure 3.20 Cosmos 1813 Debris Cloud after 10 Hours of Breakup
Simulated using ASSEMBLE Model
Figure 3.21 Cosmos 1813 Debris Cloud after 10 Hours of Breakup
Simulated Using the Model Proposed by Jehn 1996.

82

83

83

84

84

85

85

86

86

87

objects
xxii
Figures Page
No.

Figure 3.22 Spot I R/B Debris Cloud of 463 Fragments Three Months
after the Event as Reconstructed from US SN Database
Figure 3.23 Spot I R/B Debris Cloud of 463 Fragments Three Months
after the Event as Simulated by ASSEMBLE Model
Figure 3.24 DelatV and Eccentricity Distributions from the Breakup
Approach of Johnson and McKnight

Chapter 4: A New Stochastic Impressionistic Low Earth
(SIMPLE) Engineering Model of the Space Debris
Scenario

Figure 4.1a Number Density Distributions in Altitude and Perigee
Height bins for the Year 2002
Figure 4.1b Eccentricity and B Distributions of the Debris Objects for
the Year 2002
Figure 4.2 Comparison of Model with Year Wise and Distribution Fit of
Number Density in Altitude
Figure 4.3a Comparison of the Number Density of Debris from
Measurements and Model Distributions for Inclination Band
(0, 36]
Figure 4.3b Comparison of the Number Density of Debris from
(36,61]
Figure 4.3c Comparison of the Number Density of Debris from
(61, 73]

87

88

88

134

135

136

136

137

137

List of figures
xxiii
Figures Page
No.

Figure 4.3d Comparison of Number Density of Debris from
Measurements and Model Distributions in Inclination Band
(73, 91]]
Figure 4.3e Comparison of Number Density of Debris from
Measurements and Model Distributions in Inclination Band
(91,180]
Figure 4.4 Probability Plots of Eccentricity and Ballistic Coefficient
in Altitude Bins for the Year 2002
Figure 4.5 Variation of Location Parameter for Eccentricity in
Individual Altitude Bins
Figure 4.6 Variation of Location Parameter for Eccentricity in Moving
Window Altitude Bins
Figure 4.7 Comparison of Model Values of Location Parameter for
Eccentricity in Moving Window Altitude Bins and
Individual Altitude Bins
Figure 4.8 Probability Plots of Eccentricity and Ballistic Coefficient in
Perigee Height Bins for the Year 2002
Figure 4.9 Lognormal Nature of Eccentricity and B with respect to
Inclination Bands in Altitude Bin (700, 800] Km for the
Year 2002
Inclination Bands in Altitude Bin (800, 900] Km for the
Year 2002

138

138

139

140

140

141

141

142

142

objects
xxiv
Figures Page
No.

Chapter 5: Evolution of the Orbital Debris Scenario Based on Kalman
Filter Approach (Fast Number Density Simulator and
Equivalent Ballistic Coefficient Estimator)

Figure 5.1 Graphical Representation of the Orbital Propagation
Algorithm in the Eccentricity e Vs. Semi Major Axis a
Space
Figure 5.2 Updating the Ballistic Coefficients Only Using Constant
Gains for B Bin (0.0301, 0.0818)
Gains for B Bin (0.0818, 0.2226)
Gains for B Bin (0.6056, 1.6476)
Figure 5.5 Updating the Ballistic Coefficients and Number of Objects
Using Constant Gains for B Bin (0.0301, 0.0818)
Figure 5.8 Kalman Gains for Number of Objects
Figure 5.9 Kalman Gains for Ballistic Coefficients
Figure 5.10 The Estimated Ballistic Coefficients
Figure 5.11 One Source of Explosion Included. Updating the Ballistic
Coefficients and Number of Objects Using Constant Gains
for B Bin (0.0301, 0.0818)

168

169

170

171

172

173

174

175
176
177
178

List of figures
xxv
Figures Page
No.

for B Bin (0.0818, 0.2226)
for B Bin (0.6056, 1.6476)
Figure 5.14 Comparison between Estimated and Observed Number of
Objects: Two Sources of Explosions Included. Updating the
Ballistic Coefficients and Number of Objects Using
Constant Gains for B Bin (0.0301, 0.0818)
Objects: Two Sources of Explosions Included. Updating the
Ballistic Coefficients and Number of Objects Using
Constant Gains for B Bin (0.6056, 1.6476)
Objects: Two Sources of Explosions and Some New
Launches Included. Updating the Ballistic Coefficients and
Number of Objects Using Constant Gains for B Bin
(0.0301, 0.0818)
Objects: Two Sources of Explosions and Some New
Launches Included. Updating the Ballistic Coefficients and
Number of Objects Using Constant Gains for B Bin
(0.6056, 1.6476)

179

180

181

182

183

184

objects
xxvi
Figures Page
No.

Target Risk Assessment due to Orbital Debris

Figure 6.1 Convergence of Scale and Shape Parameters in Weibull
Distribution Fit for Target Altitude 500 km and Inclination 0
deg.
Figure 6.2 Convergence of Scale and Shape Parameter in Weibull
Distribution Fit for Target Altitude 500 km and Inclination
28.5 deg.
Figure 6.3 PDF and Distribution Function of Weibull Distribution Fit
Figure 6.4 Goodness of fit of the Weibull Distribution for Closest
Approach Distances
Figure 6.5 Ballistic Coefficients of the 3D Table (H, H
0
, B)
Figure 6.6 Number of Objects Remaining Out of 6000 Objects
Simulated Breakup at 650 Km Altitude
Figure 6.7 Relative Frequencies of Objects Sweeping through
Different Altitudes Due to the Simulated Debris Cloud of
6000 Objects
Figure 6.8 Collision Probabilities at Different Altitudes from
Catalogued Objects


Figure 7.1 Simulation Case 1: Ballistic Coefficient, Cost and Errors in
Apogee and Perigee Heights Using Adaptive Filtering
Technique

205

205

206
206

207
207

208

208

228

List of figures
xxvii
Figures Page
No.

Figure 7.2 Simulation Case 1: Measurement Noise Estimates in
Apogee and Perigee Heights Using Adaptive Filtering
Technique
Figure 7.3 Simulation Case 1: Process Noise Estimates in Semi Major
Axis, Eccentricity and Ballistic Coefficient Using Adaptive
Filtering Technique
Figure 7.4 Simulation Case 1: Measurement and Estimates in Apogee
and Perigee Heights Using Adaptive Filtering Technique
Figure 7. 5 Sensitivities of the Reentry Prediction with Different B
0

Values (Sat. No. 25947)
Figure 7.6 Estimate of Time Varying B with Different B
0
Values
(Sat. No. 25947)
Figure 7.7 Reentry Predictions with Different Sets of Kalman Gains
(Sat. No. 25947)
Figure 7.8 Final Prediction of Reentry Time with Near Optimal
Kalman Gains (Sat. No. 25947)
Figure 7.9 Estimated Ballistic Coefficients and the COST J
Figure 7.10 Sat25947 Reentry Predictions: Measurement Noise
Estimates in Apogee and Perigee Heights from the
Adaptive Filtering Technique
Figure 7.11 Sat25947 Reentry Predictions: Process Noise Estimates in
a, e and B from the Adaptive Filtering Technique
Figure 7.12 Sat25947 Reentry Predictions: Observed Measurements
and Estimated Values of Apogee and Perigee Heights
Figure 7.13 The US reconstruction of the Soviet ELINT Spacecraft
Figure 7.14 Comparisons of COSMOS 389 Satellite Reentry
Predictions by Different Agencies

228

229

229

230

230

231

231

232
232

233

233

234
234

objects
xxviii
Figures Page
No.

Appendices

3. Basics of Orbital Motion

Figure 1 A Graphical Representation of all the Orbital Elements of a
Satellite
Figure 2 A Graphical Representation of a, e and <

8. A Brief Introduction to Kalman Filter and its Implementation using
Constant Kalman Gains

Figure 1 Variation of the Error for Different Constant Kalman Gains
Figure 2 Measured Data for Constant Signal Followed by Sinusoidal
Variation
Figure 3 Residual Errors with Different Kalman Gains
Figure 4 Zoomed Residual Errors with Different Kalman Gains
Figure 5 Actual Signals and the State Prediction before Update

253

254

280
280

281
281
282
282

List of tables
xxix

Tables Page
No.


Table 1.1 Estimated Number, Size, Measurement and
the Effect of Spacecraft Mission
Table 1.2 Orbital Box Score
Table 1.3 Debris Assessment Issues and Corresponding Guideline
Descriptions

Studies

Table 2.1 Some Well Known Debris Environment Models


Table 3.1 Details of the Analyzed Breakup Events
Table 3.2 Location and Scale Parameters of the Distributions Used
in ASSEMBLE Model for Various Breakups Based on
Data and Gabbard diagram

2

8
20

36

75
76

LIST OF TABLES
objects

xxx
Tables Page
No.

Chapter 4: A New Stochastic Impressionistic Low Earth
(SIMPLE) Engineering Model of the Space Debris
Scenario

Table 4.1 Differences between Space Debris and Gas Molecules
Table 4.2 Summary of the Table of Contents of the Results
for the Various Distributions
Table 4.3 Parameters of the Tertiary Mixture of Laplace Distributions
for the Number Density of Catalogued Debris Objects with
Altitude.
Table 4.4 Parameters of the Tertiary Mixture of Laplace Distributions
for the Number Density of the Catalogued Debris Objects for
Perigee Height.
Table 4.5 Parameters of the Lognormal Distribution for the
Eccentricity of Catalogued Debris Objects
Table 4.6 Parameters of the Distributions for B Values of
Catalogued Debris Objects.
Table 4.7 Model Values for the Frequency of the Number Density of
the Catalogued Debris Objects with Altitude
Table 4.8 Model Values for the Frequency of the Number Density of
the Catalogued Debris Objects with Perigee Height.
Table 4.9 Number of Debris with Time in the Different Inclination
bands.
Table 4.10 The Location and Scale Parameters of the Tertiary Mixture of
Laplace Distributions for the Number Density of Debris in
Altitude for Various Inclination Bands

93
100

101

103

104

106

109

110

112

113

List of tables
xxxi
Tables Page
No.

Table 4.11 The Location and Scale Parameters of the Tertiary Mixture of
Laplace Distributions for the Number Density of Debris in
Perigee Height for Various Inclination Bands
Table 4.12 The Location and Scale Parameters of the Lognormal
Distributions for the Eccentricity in the Different
Inclination Bands.
Table 4.13 The Scale and Location Parameters of the Lognormal
Distributions for the Ballistic Coefficient in the Different
Inclination Bands.
Table 4.14 Parameters of the SIMPLE Model for the Number Density
with Altitude Distribution for Different Inclination Bands.
Table 4.15 Parameters of the SIMPLE Model for Perigee Distribution
for Different Inclination Bands.
Table 4.16 Parameters of the Eccentricity Distribution Model for
Different Inclination Bands.
Table 4.17 Location Parameters of the Lognormal Fit for Eccentricity
in Altitude Bins of Size 100 Km.
Table 4.18 Scale Parameters of the Lognormal Fit for Eccentricity
Table 4.19 The Coefficients of Model Polynomial Fit for Eccentricity
in Altitude Bins.
Table 4.20 Coefficients of Polynomial Fits for Location and Scale of B
in Altitude Bins across the Years.
Table 4.21 Location Parameters of the Lognormal Fit for Eccentricity in
Perigee Height Bins of Size 100 Km.

114

115

116

117

118

118

120

121

123

124

126

objects

xxxii
Tables Page
No.

Table 4.22 Scale Parameters of the Lognormal Fit for Eccentricity in
Perigee Height Bins of Size 100 Km.
Table 4.23 The Coefficients of Polynomial Fit for Eccentricity in
Perigee Height Bins
Table 4.24 The Coefficients of Model Polynomial Fit for B in
Perigee Height Bins
Table 4.25 The Coefficients of the Best 5
th
Degree Model Fit with
respect to Altitude Bins and Inclination Bands for
Eccentricity.
rd
Degree Model Fit with
respect to Perigee Height Bins and Inclination Bands for
Eccentricity

Chapter 5: Evolution of the Orbital Debris Scenario Based on
Kalman Filter Approach (Fast Number Density Simulator
and Equivalent Ballistic Coefficient Estimator)

Table 5.1 Real World Scenario without Binning
Table 5.2 Filter World Scenario with Binning
Table 5.4 Parameters Utilized in the Implementation of GA

Target Risk Assessment Due to Orbital Debris

Table 6.1 Summary of the Broad Nature of Various Approaches for
Collision Risk Assessment

127

128

128

130

131

153
153
159

188

List of tables
xxxiii
Tables Page
No.

Table 6.2 Ballistic Coefficient Values from the Three Way Table for
(H
o
, H, B) for One Year Duration
Table 6.3 Relative Number of Objects Sweeping Through the Target
in a Year
Table 6.4 Number of Debris Objects Swept through Different Target
Altitudes during the First 6 Years
Table 6.5 Spatial Densities and Collision Probabilities at Different
Target Altitudes from the Catalogued Objects
Table 6.6 Comparisons of Collision Probabilities at
Different Target Altitudes by Various Approaches


Table 7.1 Parameters Utilized in the Implementation of GA for the
Reentry Problem
Table 7.2 Details of the Various Case Studies
Table 7.3 Results of the Various Case Studies
Table 7.4 Comparisons of the Last Predictions Made by Various
Agencies : COSMOS 389 Satellite Reentry Predictions

Appendices

1. Two Line Elements (TLE) Format

Table 1 Two-Line Element Set Format Definition, Line 1

195

200

201

202

203

217

225
226
227

238

objects

xxxiv
Tables Page
No.

Table 2 Two-Line Element Set Format Definition, Line 2

3. Basics of Orbital Motion

Table 1 Orbital Perturbations of Close Earth satellites

242

255

List of Symbols and abbreviations
xxxv

a : Semi major axis
A
eff
: Effective area of cross section
A/m : Area to mass ratio
B : Ballistic coefficient
B* : Term in TLEsSet related to B
C
d
: Drag coefficient
D : Spatial density
d : Size of the fragment (m)
e : Eccentricity
f : Functional form of PDF
F : Distribution function
h : Altitude above the earth surface
H : Measurement sensitivity matrix
H
t
: Scale height
h
a
:Apogee height above the earth surface
h
p
: Perigee height above the earth surface
i : Inclination
J : Cost function in Kalman filter
J
s
: Modified Bessel function of order s used in orbital decay equations
K : Kalman gain matrix
k : Number of events with respect to collision probability
m : Mass of the fragment
M : Mass of the parent object
N : Number of debris fragments also number of measurements
N
m
: Cumulative number of fragments with mass m or greater
f
m
: The ratio of the reference mass of 1000 kg to the parent mass M

LIST OF SYMBOLS
objects

xxxvi
P
0
: Initial state error covariance
P
k
: Probability of occurrence of k events
p : Weight parameter for the mixture of Laplace distribution
Q : Process noise covariance
Roc : Rate of occurrence parameter
R : Measurement noise covariance
r : Position of the satellite
S : Volume of the toroid or size of the object as the case may be
t : Time
T
m
: Expected time to collision
Vr : Relative collision velocity
v : velocity of the satellite
X : State variable vector
V : Incremental velocity imparted to the fragment with respect to the parent
Vt, Vn, Vr: Incremental velocity in the transverse, normal and radial directions
respectively with respect to the orbital plane (m/s)
P : Deviation of the orbital period of the fragments from the period of the
parent body (min)
i : Deviation of the inclination of the fragments from the inclination of the
parent body (deg)
: True anomaly (deg)
: Azimuth (deg)
: Elevation (deg)
: Location parameter of the distribution
: Scale parameter of the distribution
: Shape parameter of the distribution
:
Atmospheric density
0
: Density at a reference altitude h
0
(m/s)
2 2 2
to equivqlent is which
r
v
n
v
t
v + +
List of Symbols and abbreviations
xxxvii
i
: i
th
Location parameter of tertiary mixture of Laplace distribution
i
: i
th
Scale parameter of tertiary mixture of Laplace distribution
: Shape parameter in the distribution
r
c
: Minimum distance of closeness assumed as collision.
: Innovation
: State transition matrix
: Right ascension of the ascending node
:Argument of perigee
w : Process noise
v : Measurement noise

SUPERSCRIPTS
~ : Value before update
^ : Value after update
T : Transpose of a matrix
-1 : Inverse of a matrix

SUBSCRIPTS
0 : Initial condition and values related to parent body
k : Discrete time index

AIAA American Institute of Aeronautics and Astronautics
ASSEMBLE A Semi Stochastic Environment Modeling for Breakup in LEo
BU Breakup
CDF Cumulative Distribution Function

ABBREVIATIONS
objects

xxxviii
DAS Debris Assessment Software
ESA European Space Agency
EQF Equivalent Fragment
GA Genetic Algorithm
GEO Geosynchronous Orbits
IAA Inter National Academy of Astronautics
IADC Inter Agency Debris Co-ordination Committee
IDES Integrated Debris Evolution Suite
KF Kalman Filter
LEO Low Earth Orbits
LEGEND Leo to Geo Environment Debris Model
MASTER Meteoroid and Space Debris Terrestrial Environment Reference
MC Monte Carlo
NASA National Aeronautics and Space Administration
NORAD North American Aerospace Defense Command
ORDEM Orbital Debris Engineering Model
PDF Probability Density Function
PSLV Polar Satellite Launch Vehicle
RSA Russian Space Agency
SIMPLE Stochastic Impressionistic Low Earth Model
SDM Semi Deterministic model
SDPA Space Debris Prediction and Analysis
TES Technology Experiment Satellite
TLE Two Line Element
USSPACECOM US Space Command Network
USSN US Space Surveillance Network
UTC Universal Time

Chapter 1 An overview of the space debris environment
1
CHAPTER 1

AN OVERVIEW OF THE
SPACE DEBRIS ENVIRONMENT

1.1 INTRODUCTION

The conquest of space by mankind can only be achieved by a proper understanding and
characterization of the space environment surrounding the near earth and beyond. The
various kind of environment are mainly (Tribble 1995)

(i) Vacuum
(ii) Neutral atmosphere
(iii) Plasma
(iv) Radiation
(v) Cosmic particles and
(vi) Micro meteoroids

All these are in general hostile for space exploration and cause for several spacecraft
anomalies ranging from minor, temporary, or permanent damage. However if these are
accounted for during the design process they do not lead to major issues. Moreover in
addition to the above existing for time immemorial, the very activity of putting artificial
satellites into an orbit around the earth going on for just less than about 50 years has created
the so called orbital debris. The dawn of the space age was on 4
th
October 1957 when the
Sputnik 1 was launched by USSR. So far nearly 25,000 satellites have been launched mainly
by USSR and USA (http://www.orbitaldebris.jsc.nasa.gov/library/EducationPackage.pdf).
The orbital debris is a serious threat, in particular for near earth satellites. Modeling and
objects

2
analyzing the various major issues arising out of space debris is the subject matter of the
present study.

Orbital debris generally refers to material that is in orbit as the result of space initiatives, but
is no longer serving any useful function. The Figure 1.1 provides an overview of the space
objects from the report Orbital Debris: A Technical Assessment (1995). The orbital debris
consists of mainly the spent upper stages, passive payloads, debris fragments due to on orbit
breakup of spent stages and payloads, objects generated by the mission related activities,
paint flakes due to material degradation and many other small sized objects due to various
reasons.

The Figure 1.2 shows a categorization of the orbital debris population (Orbiting Debris : A
Space Environmental Problem 1990). Table 1.1 gives a view of the estimated number of
debris objects with the size ranges and its impact on spacecraft missions. As is clear from this
table and figures, there are more than 1,00,000 debris objects with size above 1 cm and the
debris particles with smaller sizes runs into millions.

Table 1.1 Estimated Number, Size, Measurement and the Effect of Spacecraft Mission

Size (cm) Number of Objects Measurements Effect
> 10 > 9000 Radar Can breakup a satellite
1-10 > 1 E 5 Optical Telescope Penetrate the satellite walls
0.1-1 > 3 E 7 Statistical Estimate Surface or Component damage
< 0.1 10 E 9 10 E 14 LDEF Sensor interference

The present work concentrate mainly on five important aspects among others of the space
debris studies namely, breakup modeling, environment modeling, evolution of the debris
scenario, collision risk assessment and the reentry prediction of risk objects.

3

Figure 1.1 An Overview of Space Objects
( From Orbital Debris: A Technical Assessment 1995)
objects

4

Figure 1.2 Categorization of Space Debris Population
(From Orbiting Debris: A Space Environmental Problem 1990)

1.2 SOURCES FOR THE CREATION OF SPACE DEBRIS

There are many sources for the creation of the debris environment. Figure 1.3 provides some
of the sources and sinks of space debris. A major contributor to the criterion of orbital debris
background scenario has been the breakup of space objects. As mentioned earlier more than
180 breakups have been identified (Anz Meador 2001), and more are believed to have
occurred. It is mainly the explosions and a few collisions have contributed to the breakups.
Explosions can occur when propellant and oxidizer inadvertently mix in spent stages,
residual propellant becomes over pressurized due to heating and so on. Some satellites have
5
been deliberately detonated. Explosions can also be indirectly triggered by collisions with
debris.

Another major source is discarded hardware of many launch vehicle upper stages left out in
the orbit. Many satellites are also abandoned at the end of useful life. Another source of
debris are the objects released during spacecraft and mission operations, such as deployments
and separations. These typically involve items such as separation bolts, lens caps, momentum
flywheels, nuclear reactor cores, clamp bands, auxiliary motors, launch vehicle fairings, and
adapter shrouds (www.aero.org/cords/orbdebris.html, IAA Position Paper on Orbital Debris
2000).
.
Figure 1.3 Space Debris Sources and Sinks
(From Orbital Debris: A Technical Assessment 1995)

Material degradation due to atomic oxygen, solar heating, and solar radiation has resulted in
the production of particulates such as paint flakes and bits of multilayer insulation. Solid
objects

6
rocket motors used to boost satellite orbits have produced various debris items, including
motor casings, aluminum oxide exhaust particles, nozzle slag, motor-liner residuals, solid
fuel fragments, and exhaust cone bits resulting from erosion during the burn.

Three collisions are known to have occurred since the beginning of the space age. In
addition, the debris research community has concluded that at least one additional breakup
was caused by collision. The cause of approximately 22 percent of observed breakups is
unknown (Anz Meador 2001, IAA Position Paper on Orbital Debris 2000).

Approximately 70,000 objects estimated to be 2 cm in size have been observed in the 850-
1,000 km altitude band. NASA has hypothesized that these objects are frozen bits of nuclear
reactor coolant that are leaking from a number of Russian RORSATs
(http://www.aero.org/cords/).

1.3 SINK MECHANISMS FOR THE DECAY OF SPACE DEBRIS

Natural processes provide some relief to the debris problem and help to reduce the risks
posed by it. Orbit decay is one of the natural sink mechanisms for the removal of debris
objects. Once in orbit, a debris fragment is affected by perturbing forces that can change its
trajectory and even remove it eventually from the orbit reenter the altitude upto 120 km to be
equally burnt out completely. The primary forces acting on a space object in lower orbits
(below about 800 km) are atmospheric drag gravity and its anomalies. These gravitational
perturbations, however, although affecting some of the orbital parameters, do not generally
strongly affect the orbital lifetime. For space objects in higher orbits, luni solar gravitational
perturbations and solar radiation pressure become more important factors. Small debris are
mainly affected by solar radiation pressure, with minor affecting plasma drag, and
electrodynamic forces.

The rate at which a space object loses altitude is a function of its mass, its average projected
cross-sectional area, aerodynamic drag coefficient and the atmospheric density. Although the
7
earth's atmosphere technically extends to great heights, its retarding effect on space objects
falls off rapidly with increasing altitude. Atmospheric density at a given altitude, however, is
not constant and can vary significantly due to the 11-year solar cycle (MSIS 1986, Roes
1989). This natural phenomenon has the effect of accelerating the orbital decay of debris
during periods of solar maximum (increased sunspot activity and energy emissions). After
the last two peaks in the solar cycle, the total cataloged space object population actually
declined, because the rate of orbital decay exceeded the rate of space object generation
through new launches and fragmentations.

While the actual effect of the atmospheric drag is dependent up on the satellites mass and
cross sectional area (ballistic coefficient), virtually all satellites below 500 km will reenter
the Earths atmosphere within a span of usually about fifteen years.

A second sink mechanism, primarily of importance to elliptical orbits, is luni solar
perturbations. The gravitational effects of the Sun and the Moon are often the dominant
factor in the behavior of highly elliptical satellite orbits, particularly those at low inclinations.
These forces change the perigee height to either rise or fall. By taking advantage of these
natural forces, spent geosynchronous transfer stages can be made to reenter the Earths
atmosphere in a short period of time (http://www.aero.org/cords/).

A third although less prominent, sink effect is the solar pressure on a satellite. The effect of
photons striking satellite payloads and rocket bodies is usually negligible. However for very
small particles, particularly those of few grams or less, solar pressure can induce rapid orbital
decay. Solar pressure appears to be a dominant force in quickly removing solid rocket motor
effluents from the space environment.

The combination of all of these forces has caused approximately 16,000 cataloged objects to
reenter the atmosphere since the beginning of the space era. In recent years, an average of
two to three space objects large enough to be cataloged (as well as numerous smaller debris
particles) reenter the Earth's atmosphere each day.
objects

8
1.4 MEASUREMENT OF SPACE DEBRIS

Currently, the US Space Surveillance Network (USSN) is using ground based radar, optical
and infrared sensors to track large objects. The minimum size of a trackable object is about
10 cm for a low Earth orbit, and about 1 m for the geostationary orbit. The objects, which are
regularly tracked and monitored, constitute objects larger than 10-20 cm size in low Earth
orbits, or about a meter in geosynchronous altitudes. There are about 9000 such objects that
are catalogued with information regarding them being provided as two line elements (TLE).
The Table 1.2 provides the number of catalogued objects as on December 31, 2003 from the
Orbital Debris Quarterly News (Vol. 8, Issue 1, January 2004).

Table 1.2 Orbital Box Score
(From Orbital Debris Quarterly News Vol. 8, Issue 1, January 2004)

The TLEs consists of the orbital details such as eccentricity, inclination, argument of perigee,
mean anomaly, ascending node, mean motion and its derivatives together with a term
9
denoted by B*, an adjusted value of the ballistic coefficient B using the reference value of
atmospheric density,
o
.

B* = B
o
/2 and B* has units of (earth radii)
-1
.

The ballistic coefficient B represents how susceptible an object is to air drag and is the
product of its coefficient of drag C
D
and its cross-sectional area A divided by its mass m.

B = C
D
A/m

TLEs also provide, apart from the above parameters, USSPACECOM/ NORAD catalogue
number, international designation and the evolution number. Over 40% are fragments
generated due to the breakup of disintegrated satellites and upper stages of rockets.
Operational payloads are only less than 10 % of the catalogued objects. If one considers
smaller objects of 1 cm size, which of course cannot be tracked by usual methods, the
number of debris far exceeds 100,000. Appendix 1 provides a description of the TLE format
together with observed elements for a satellite namely, COSMOS 389 which has reentered
the earths atmosphere recently.

1.4.1 TRACKED DEBRIS POPULATION

USSPACECOM is presently tracking more than 9000 objects larger than about 10 cm in
diameter in the low earth orbit of the space. These originate from explosions of spent stages
or anti-satellite tests. Over the last 10 years this population has grown at a rate of about 10%
per year as shown in Figure 1.4 which shows the increase in the number of tracked objects
from 1961 to 2001. The average number density of objects varies with altitude for the
observed (tracked) population of objects. Figure 1.5 provides the visualization of the
concentrations of the debris objects in LEO in terms of spatial density (Allahdadi, 1993). The
most prominent features in this figure are the spikes located around 800 km and 1400 km.
These correspond to the Iridium and Globalstar commercial communication spacecraft
objects

10
constellations (Anz Meador, 2000). Other peaks that can be observed in this figure are
occurring at about 600 km and 950 km altitudes and are mainly consisting of payloads and
rocket bodies. These density peaks are due to very low earth satellite launches at 600 km, sun
synchronous orbits around 950 km and also anti satellite tests (Johnson and McKnight, 1987
http://www.aero.org/cords).

Figure 1.4 Distributions of Tracked Objects
(From IAA Position Paper on Orbital Debris 2000)

Figure 1.5 Spatial Densities of the Various Kinds of Tracked Objects in LEO
(From Anz Meador 2001)
11

Figure 1.6 Percentages of Various Types of Tracked Objects

The Figure 1.6 shows that about 40 % of the tracked objects are due to on orbit breakup, 30
% are payloads, 18 % are spent rocket bodies and about 12% are due to operational debris.

1.4.2 UNTRACKED DEBRIS POPULATION

The untracked population of objects is known to exist because of a variety of reasons. The
vast majority of manmade debris comprises objects smaller than 10 cm, which are not
tracked during routine operations. Around 180 explosions have occurred in space, producing
undoubtedly many objects too small to be tracked. These events have been the main source
that has generated a population of objects larger than 1cm on the order of 70000 to 150000.
For example objects of 4 cm in size not detectable at higher altitudes can be detected when
their orbits decay to about 400 km, before reentry.

1.5 EVOLUTION OF DEBRIS CLOUDS

Debris clouds are formed whenever debris fragments are generated by a source. For example,
discarded upper stages generally are surrounded by a cloud of particulates that are released
over time by degradation of various materials such as paint and multilayer insulation. Any
objects

12
concentration of debris particles or fragments in a well defined region of space is referred to
as a debris cloud.

Also whenever an orbital breakup occurs, a debris cloud is instantly formed. Such debris
clouds first take on the form of an expanding three dimensional ellipsoid. The debris cloud
gradually spread in a spiral pattern. As time passes, the debris cloud eventually could
envelops other satellites in the nearby vicinity. The Figure 1.7 shows the short term
evolution of the clouds from Orbital Debris: A Technical Assessment (1995).

Figure 1.7 Evolution of a Debris Cloud
13

Due to the physical processes involved in an explosion or collision and due to the dynamical
laws of orbital motion, fragments are not spread uniformly throughout a debris cloud. At
some locations, spatial density of fragments is much greater than at other locations.
Whenever the spatial density of the orbital fragments is high, the collision risk posed to
satellites that fly through the cloud is also naturally increased.

Certain regions of the debris cloud are constricted to nearly one or two dimensions. There are
three types of debris cloud constrictions: pinch points, pinch lines, and pinch sheets. Spatial
fragment density is very high at these constrictions
(http://www.aero.org/cords/deb_clouds.html).

Pinch points and pinch lines are particularly important to satellite constellations. Neither of
these constrictions moves with the debris cloud around its orbit. They remain fixed in inertial
space while the debris cloud repeatedly circulates through them. In many satellite
constellations there are multiple satellites in each orbital ring. If one of these satellites breaks
up, the remaining satellites in the ring will all repeatedly fly through the pinch point and
pinch line. If many fragments are produced by the breakup, the risk of damaging another
satellite in the ring may be significant. If satellites from two orbital rings collide, two debris
clouds will be formed, one in each ring. The constrictions of each cloud then pose a hazard to
the remaining satellites in both rings (http://www.aero.org/cords/deb_clouds.html).

1.6 RISK DUE TO ORBITAL DEBRIS

Space debris presents a real threat to present and future space programs. Concern about the
orbital debris hazard has grown for the last couple of decades. A number of events, including
the breakup of several rocket upper stages and the replacement of Shuttle windows after
impacts by small particles, helped to increase awareness of the problem, as did the need for
space debris considerations into the design of International Space Station Freedom. Impacts
with the much more numerous debris particles that are a millimeter or less in diameter can
objects

14
damage optics, degrade surface coatings, or even crack windows. At present spacecraft, its
mission design, as well as the operations of the Space Shuttle and the International Space
Station or satellite launches, are already influenced by debris threat considerations, and
countermeasures, like shielding and collision avoidance schemes and maneuvers, being
devised.

The first confirmed accidental collision between two catalogued objects occurred in July
1996 (http://www.orbitaldebris.jsc.nasa.gov/library/EducationPackage.pdf). In this the
gravity gradient attitude control boom of the French Satellite CERISE was damaged due to
the collision with a fragment from an Ariane rocket body. Also the very large number of
small sized particles can be a source of a continuous nuisance value. The small particle
impacts on the Space Shuttle lead to replacing on average one of the eight main windows
after each flight (http://www.orbitaldebris.jsc.nasa.gov).

If appropriate measures are not taken now, the future space programs will be severely
affected by the damage due to space debris. Recognizing this, all the space faring nations in
the world have taken up systematic scientific investigations on various aspects of the space
debris problem. The important aspects are

(a) the causes and growth of space debris
(b) their measurement and modeling and
(c) various measures available to mitigate the impact on present and future space
endeavors.

The relative velocity between the operational satellites and the orbital debris fragments are
very high. In low Earth orbit (altitudes lower than 2,000 km) the average relative velocity at
impact is 10 km/sec (36,000 km/hr or 21,600 mph). At this velocity, even small particles
possess significant amount of momentum and kinetic energy. For example, NASA frequently
replaces space shuttle orbiter windows because they are significantly damaged by objects as
small as a flake of paint. An aluminum sphere 1.3 mm in diameter has damage potential
15
similar to that of a .22-caliber long rifle bullet. A fragment 10 cm long is roughly comparable
to 25 sticks of dynamite (www.aero.org/cords/debrisks.html).

Debris fragments between 1 and 10 cm in size will penetrate and damage most spacecraft. If
the spacecraft bus is impacted, satellite function will be terminated and, at the same time, a
significant amount of small debris will be created. In large satellite constellations, this can
lead to amplification of the local smaller debris population and its associated erosional effect.

If a 10-cm debris fragment weighing 1 kg collides with a typical 1,200-kg spacecraft bus,
over one million fragments 1 mm in size and larger can be created. This collision results in
the formation of a debris cloud, which poses a magnified impact risk to any other spacecraft
in the orbital vicinity such as other members in the constellation.

Debris particulates smaller than 1 mm in size do not generally pose a hazard to spacecraft
functionality. However, they can erode sensitive surfaces such as payload optics. Hence,
while the spacecraft may survive, payload degradation can still result in mission loss. Debris
fragments from 1 mm to 1 cm in size may or may not penetrate a spacecraft, depending on
material selection and the shielding if any. Penetration through a critical component, such as
the flight computer or propellant tank, can result in the loss of spacecraft.

At geosynchronous altitude, average relative velocity at impact is much lower than in low
Earth orbit, about 200 m/sec (720 km/hr or 432 mph). This is because most objects in the
geosynchronous ring move slowly along their orbits. Nevertheless, fragments at this velocity
can still cause considerable damage upon impact. A 10-cm fragment in geosynchronous orbit
has roughly the same damage potential as a 1-cm fragment in low earth orbit. Similarly a 1-
cm geosynchronous fragment is roughly equivalent to a 1-mm low earth orbit fragment.
While larger pieces of debris could cause serious damage to a spacecraft, small debris,
particles usually less than 1mm in diameter, will not usually cause catastrophic damage, even
when travelling at speeds of 10km/s. Such small particles can however lead to the gradual
decay of solar panels, camera lenses, telescope reflectors, or other highly machined surfaces.
objects

16
Large pieces of debris can generate damage exceeding that of explosive charges, and can end
the operations of a previously functioning satellite, or if it collides with a rocket stage or
payload could create millions of small particles. Over 1 million fragments 1mm in size or
larger could be created in a collision of a typical 1,200kg spacecraft with a 1kg, 10cm debris
fragment. These may in turn pose more threat to operational satellites. In cases of
unexplained damage or loss of spacecraft, debris has been brought forward as a possible
cause (Kozdron, 1999).

In addition to presenting a collision hazard to space operations, orbital debris can also have
other detrimental effects (Orbital Debris: A Technical Assessment 1995). For example,
debris can affect astronomical observations by leaving light trails on long-exposure
photographs with wide fields of view. In addition, debris reentering the atmosphere can
potentially harm people and property on the ground. In the past, this has been a minor hazard,
since most reentering debris objects burn up completely in the atmosphere. However, there
have been some exceptions, for example, Kosmos 954, Skylab, and Salyut-7/Kosmos 1686
(Orbital Debris : A Technical Assessment 1995), and the exact number of objects surviving
reentry is unknown. Reports by the American Institute of Aeronautics and Astronautics
(AIAA), the European Space Agency (ESA), the U.S. Interagency Group on Space, the
International Academy of Astronautics (IAA), and the Japan Society for Aeronautical and
Space Sciences also helped to define the problem better and offer some suggestions for
mitigation.

1.7 DESIGN AND OPERATIONAL MEASURES TO CONTROL
DEBRIS

Limiting the creation of debris through mitigation measures best controls the risk due to
them. However, debris mitigation usually increases the cost of the mission. But the debris
mitigation procedures will have minimal impact on the mission cost if they are specified
early during the development phase. For example, deployment procedures can be designed to
prevent ejection of objects. Tethered lens caps and bolt catchers for explosive bolts are
17
examples of preventive design (Orbital Debris: A Technical Assessment 1995, McKnight
1990).

In order to prevent explosions, satellite components that store energy can be passivated at the
end of useful life. For instance either venting or burning up to depletion can eliminate the
residual propellant in upper stages and satellites. Batteries can be designed to reduce the risk
of explosion.

Further to prevent debris accumulation in preferred mission orbits due to collisions, satellites
and other objects must be removed from the mission orbit at the end of life. The NASA's
guidelines (NASA standard 1740.14 1995 and DAS 2001) for limiting orbital debris
recommend that an object should not remain in its mission orbit for more than 25 years.

The satellites, upper stages of the launch vehicles, and deployed objects in low earth orbit
can take advantage of the earth's atmosphere to reduce the time spent on orbit. At sufficiently
low altitudes, atmospheric drag on the object will cause the object's orbit to decay and result
in reentry within 25 years. Vehicles at higher altitudes can perform post mission maneuvers
to drop the perigee further down into the atmosphere. Propellant must be reserved for this
maneuver. Hence, the penalty to satellites is reduced mission life, and to upper stages it is
reduced performance. If atmospheric decay is exploited to remove an object from orbit, then
the risk posed to the ground by reentry of the object must also be considered.

At altitudes above 2,000 km, it is not feasible to force reentry within 25 years using current
space technology. At this point of time, it is generally recommended to place the objects in
disposal orbits called "graveyard" orbits. Many spacecraft in geo synchronous orbit are
already boosted into a higher disposal orbit at end of mission life (Position paper on Orbital
Debris 2000).

For some missions, it may be necessary to perform collision avoidance. The space shuttle
orbiter has maneuvered to avoid collisions with other objects on several occasions
objects

18
(http://www.aero.org/cords/cntl_risk.html). Regarding satellite constellations, if a potential
collision can lead to the creation of a debris cloud that may result in damage to other
constellation members, it may be worthwhile to perform a collision avoidance maneuver
(Protecting the Space Shuttle from Meteoroids and Orbital Debris 1997).

In the more distant future, it may be called upon to completely remove all nonfunctional
satellites and upper stages from orbit. This removal will not be feasible until new technology
is developed.

Current debris control practices aim at reducing the growth rate. The total amount and mass
of debris is still rising, since rate of removal of debris by natural mechanisms, due to air drag,
is exceeded by the rate of deposition in space of man made objects. These practices, if
unchanged, will sooner or later lead to a major increase in the amount of debris as a
consequence of collisions between earth orbiting objects. In order to avoid the excessive and
uncontrolled growth of space debris, ultimately more efficient and more costly mitigation
measures may have to be undertaken. These will likely include the removal of rocket bodies
and spacecraft from space upon completion of their missions (Flury 1996). Such measures
will have far reaching implications for the design and cost of space systems.

Periodically over the peak of the solar sunspot eleven-year cycle, the earth's atmospheric
density rises significantly. During such periods, the drag effects cause particles to migrate
faster to the lower altitudes. Even such a process is inadequate to deal with past or present
rates of debris generation and to decrease the population of debris objects to low values.

1.8 NASA GUIDELINES FOR LIMITING ORBITAL DEBRIS

NASA has put forward a set of guidelines for debris assessment to control the debris
accumulation in the space in the document NASA Standard 1740.14 "Guidelines and
Assessment Procedures for Limiting Orbital Debris in year 1995. As per these guidelines
the debris assessment must address the potential for orbital debris generation that results
19
from normal operations and malfunction conditions, on orbit collisions and the post mission
disposal strategies for the spent stages. Malfunction conditions refer to those credible failure
scenarios or conditions that can result in the direct generation of orbital debris or that can
disable the spacecraft to preclude post mission disposal. Examples of orbital debris generated
during normal operations include items such as lens covers, shrouds, and staging components
that are released into the environment. An on orbit explosion is an example of debris
generation by malfunction. Examples of debris generation by collisions include debris
generation by collisions with large objects and by loss of control of a spacecraft or payload as
a result of impact with small debris during mission operations (NASA Safety Standard
1740.14 (1995).

The Table 1.3 provides a summary of debris assessment issues and corresponding guideline
descriptions as given in NASA Safety Standard 1740.14 (1995). The Debris Assessment
Software (DAS 2001) has been developed to assist NASA programs in carrying out an
assessment of the orbital debris as described in NASA Safety Standard 1740.14, "Guidelines
and Assessment Procedures for Limiting Orbital Debris." The software follows the structure
of the standard and provides the user with tools to ensure compliance with the guidelines or
to assess debris mitigation options to bring a program to fall within guidelines.

In order to satisfy the guidelines in the above standard the program or project manager may
need to plan for such things like

1. Depleting onboard energy sources after completion of mission,
2. Limiting orbit lifetime after mission completion to 25 years or maneuvering to a
disposal orbit,
3. Limiting the generation of debris associated with normal space operations,
4. Limiting the consequences of impact with existing orbital debris or meteoroids, and
5. Limiting the risk from space system components surviving reentry as a result of post
mission disposal.

objects

20
Table 1.3 Debris Assessment Issues and Corresponding Guideline Descriptions
(From NASA Safety Standard 1740.14, 1995)

Debris
Assessment areas
Guideline Description Comments
Release of debris
during normal
mission operations
Limit number, size, and orbit
lifetime of debris larger than 1 mm
Limit lifetime of objects passing
through GEO

Includes staging components,
deployment hardware or other objects
larger than 1 mm that are known to be
released during normal operations
tethers or tether fragments left in orbit
are considered operational debris
Accidental
explosions
Limit probability of accidental
explosion during mission
operations
Deplete on-board stored energy at
end of mission life
Includes systems and components
such as range safety systems,
pressurized volumes, bipropellant
fuels, and batteries
Intentional
breakups
Limit number, size, and orbit
lifetime of debris larger than 1mm
Assess risk to other programs for
times immediately after a test when
the debris cloud contains regions of
high debris density
No assessment of orbital hazard
for breakups occurring below
altitude 90 km
Intentional breakups include tests
involving collisions or explosions of
flight systems and intentional breakup
during space system reentry to reduce
the amount of debris reaching the
ground

Collisions with
large objects
during mission
operations

Assess probability of collision with
intact space systems or large debris
Collisions with intact space systems
or large debris will create a large
number of debris fragments that pose
a risk to other operating spacecraft. A
significant probability of collision
may necessitate design or operational
changes
Collisions with
small debris
during mission
operations

Assess and limit the probability of
damage to critical components as a
result of impact with small debris

Damage by small debris can result in
both mission failure and failure to
perform post mission disposal. A
significant probability of damage may
necessitate shielding, use of
redundant systems, or other design or
operational modifications
Post mission
disposal
Remove spacecraft and upper
stages from high value regions of
space so they will not threaten
future space operations
Options are to transfer to a disposal
orbit or transfer to an orbit where the
space system will reenter within 25
years. Disposal orbits are defined
away from LEO, GEO, and semi
synchronous (12 hour) circular orbit
Debris surviving
reentry and
impacting in
populated areas
Limit number and size of debris
fragments that survive uncontrolled
reentry
This guideline limits human casualty
expectation

21
As of now many satellite launch vehicle programmes are checked against the NASA
guidelines in order to maintain a reduced number of debris objects generated out of the
programme. Such a study carried out for INDIAN PSLV Mission is reported in Suresh et al
(2001).

1.9 MODELS OF THE SPACE DEBRIS SCENARIO

The generation of the simulation tools to characterize the space debris environment as well as
the on orbit breakup was necessitated by the pressing requirement of providing better
estimates for the collision probability of the space debris objects with the launch activities
and hence to take adequate counter measures and decide on mitigation options. The
inadequacy or the limitation of the available database, mainly the non availability of the
characteristics of the small debris objects other than the bigger catalogued ones compels
people working in the area of orbital debris to model the on orbit breakups and the time
evolution of the debris environment. Space debris models are expected to provide a
mathematical description of the distribution of objects in space, their orbital characteristics,
the flux of objects and the useful physical characteristics of objects such as their size, mass,
density, reflection properties. These models can be deterministic in nature wherein each
object is described individually by its orbital parameters and physical characteristics, or
statistical in nature based on a probabilistic characterization or perhaps even the combination
of the above two types. These models can be applied profitably to risk and damage
assessments, prediction of debris detection rates for ground based sensors, prediction of
avoidance maneuvers of operational spacecraft and long term analysis of the effectiveness of
debris mitigation measures. At present some models, like ORDEM, MASTER and
EVOLVE, are available for the characterization of the space debris environment. These
models are limited by the sparse amount of data available, in particular related to the physical
characteristics, to validate the derived relationships. The models rely mostly on historical
records of satellite characteristics, launch activities, and on orbit breakups. In addition, there
are only limited data on spacecraft material response to impact and exposure to the orbital
environment.
objects

22
1.10 FUTURE SCENARIO OF THE DEBRIS

The amount of debris on orbit in future will depend upon whether the creation or removal
rate dominates. Currently, the only major mechanism for debris removal is orbital decay
through atmospheric drag, which ultimately leads to reentry. This mechanism is only
effective in a restricted range of low earth orbits. At higher orbits, it takes hundreds to
thousands of years for objects to reenter, and so there is no effective removal mechanism.
Historically, the creation rate of debris has outpaced the removal rate, leading to a net growth
in the debris population in low earth orbit at an average rate of approximately 5 percent per
year. The Figure 1.8 provides a typical prediction of the growth of the number of on orbit
collisions in future. The prediction is based on the assumption for the growth rate of launch
activities.

Figure 1.8 Cumulative Number of Major Collisions: Future Predictions

23
A major contributor to the current debris population has been fragment generation via
explosions. As the debris mitigation measure of passivation becomes more commonly
practiced, it is expected that explosions will decrease in frequency. It may take a few decades
for the practice to become implemented widely enough to reduce the explosion rate, which
currently stands at about four per year.

In the near future, there will be an increase in the number of individual intact objects
launched by man. The trend in space technology is moving towards smaller satellites
operating cooperatively in coordinated constellations. This trend is mainly due to the
revolution in commercial satellite based communications systems, and satellite micro
technology. As satellites become smaller, they will become more vulnerable to the smaller
debris population. Hundreds of such satellites have been proposed for operation in low Earth
orbits.

While functioning satellites in individual constellations will be controlled to maintain
constellation geometry and hence preclude collisions among themselves, traffic will increase
through each constellation due to associated operations and maintenance of their neighbors.
In addition, satellite failures will eventually occur, resulting in uncontrolled satellites that
drift through operating constellations. The close proximity of satellite orbits may result in
increased collision risk as well as increased risk of multiple losses if a satellite breakup
occurs and forms a debris cloud with constrictions.

Because of the increased number of larger objects, lack of sufficient debris mitigation efforts
such as collision avoidance could eventually result in collision driven population growth.

If the past trend continues, the number of tracked objects in space is predicted to increase by
a factor of 2 to 8 (depending on the rate of future explosions) within the next 20 years,
significantly increasing the probabilities of collision. Furthermore, there may be sufficient
objects in space so that collisions could be expected between non operational objects in low
earth orbit. Just one of these collisions could produce about 10
4
particles larger than 1 cm,
objects

24
and about 10
6
particles larger than 1 mm. A 1-cm object in low earth orbit could penetrate a
5-cm thick solid aluminum. The collision frequency between a 50-m diameter spacecraft and
the 1-cm population in the year 1995 could be about 0.033/yr at 500 km and 0.25/yr at 850
km. The increased number of small fragments could significantly reduce the reliability of
even small satellites.

Long term predictions, of stochastic processes like the growth of (i) economy, (ii)
atmospheric constituents and (ii) satellite orbits, all suffer from the fact that the underlying
physical processes can be changed by human intervention. Thus only reasonable short term
predictions with updates from measurements are the best to follow the evolution. In
particular, future traffic growth and the application of mitigation measures will have a major
influence on the outcome of model predictions. Space debris models must be continually
updated, since they are stochastic corrective process in the sense, they are to assimilate
observational and experimental data sets.

1.11 IMPORTANCE OF SPACE DEBRIS STUDIES

Despite the apparent infinite vastness of space, debris is a problem that cannot be ignored.
Unless a more reasonable approach is developed for space exploration, the space
environment in particular around earth orbit will become increasingly inhospitable. In
general, reclamation or disposal of space objects is not perceived by the user community as
being economical or even important, quoting The Ethics Concerning Space Debris(1999)
from the website http://cseserv.engr.scu.edu/nquinn/
ENGR019_299Fall2000/StudentWebSites/Leopold/ResearchPaper.htm. There is a tendency
among them to view the collision problem only as it impacts the economic loss of a specific
satellite or in terms of the legal liability of the responsible agency. The issue of space debris
control needs to be dealt with not in such a manner, but as a common problem shared by all
space users and the human community.

25
1.12 CONTRIBUTIONS OF THE PRESENT STUDY

In near earth space there are two major regions where orbital debris causes concern. These
are in Low Earth Orbits (LEO), up to about 2000 km, and in Geosynchronous Orbits (GEO),
at an altitude of around 36000 km. The impacts of these debris accumulations are in principle
the same in the two regions; nevertheless they require different approaches and solutions, due
to the fact that the perturbations in the orbital decay due to (i) atmospheric effects, (ii)
gravitational forces including earths oblateness and (iii) luni solar effects are different in
these two regions. In LEO it is generally known that the debris population dominates even
the natural meteoroid population for object sizes 1 mm and larger (Johnson and McKnight
1987). In this thesis the study will be focused mainly on this LEO region.

Since the first satellite breakup in 1961 up to 01 January 2003 more than 180 spacecraft and
rocket bodies have been known to fragment in orbit (Anz Meador 2001). The resulting debris
fragments constitute nearly 40% of the 9000 or more of the catalogued objects by
USSPACECOM. The catalogued fragment count does not include the much more numerous
fragments, which are too small to be detected from ground. Hence in order to describe the
trackable orbital debris environment, it is important to develop mathematical models to
simulate the trackable fragments and later expand it to untrackable objects.

Apart from the need to better characterize the orbital debris environment down to sub-
millimeter particles, there is also a pressing necessity of simulation tools able to model in a
realistic way the long term evolution of space debris, to highlight areas, which require further
investigations, and to study the actual mitigation effects of space policy measures.

The contribution in the present thesis concentrates on

(i) a new modeling of the on orbit breakup by a semi stochastic approach,
(ii) a simple characterization of the debris environment scenario in LEO by a
stochastic impressionistic model,
objects

26
(iii) an estimation theoretic approach utilizing the Kalman filter with tuned
constant gains to describe and follow the evolution of the space debris
fragment scenario,
(iv) a probabilistic approach to estimate the collision probability for a target object
with space debris and
(v) a very efficient and effective constant Kalman gain approach to predict the
reentry time of risk objects reentering the earths atmosphere.

The above five aspects comprise the most of the major issues of interest in the area of space
debris studies.

This Chapter 1 briefly brought out an overview of space debris environment and the issues
associated with the growing space debris populations. A literature survey of important earlier
work carried out regarding the above five issues are provided in the next Chapter 2.

The Chapter 3 proposes a new breakup model to simulate the creation of debris objects by
explosion in LEO named A Semi Stochastic Environment Modeling for Breakup in Leo
(ASSEMBLE). This model is based on a study of the characteristics of the fragments
generated out of an on orbit breakup as provided in the TLE sets, of the INDIAN PSLV-TES
mission spent upper stage. These observations were used to arrive at the proper physical,
aerodynamic, and orbital characteristics of the fragments. Subsequently it has been applied as
an inverse problem to simulate and further validate it based on some more typical well
known historical on orbit fragmentation events. All the simulated results compare quite well
with the observations both at the time of breakup and at a later epoch.

A new stochastic environment model of the debris scenario in LEO that is simple and
impressionistic in nature named SIMPLE is proposed in Chapter 4. The many advantages of
this engineering model over some other well known models is also provided in this chapter.

27
The Chapter 5 treats the evolution of the debris objects generated by on orbit breakup. A
novel innovative approach based on the propagation of an equivalent fragment object in a
three dimensional bin of semi major axis, eccentricity, and the ballistic coefficient (a, e, B)
together with a constant gain Kalman filter technique is described in this chapter. This new
approach propagates the equivalent fragment in each of the a and e bins rapidly and
accurately without propagating each and every of the space debris objects in the above bins.
Further this approach expands the scenario to provide suitable ballistic coefficient values for
the conglomeration of the fragments in the various bins. Further it is able to assimilate the
information from other breakups as well with the passage of time. The heart of the technique
is to use a constant Kalman gain which is optimal to track the dynamically evolving fragment
scenario and further expand the scenario to provide appropriate average time varying ballistic
coefficients for the various bins. The present concept has been validated through extensive
simulated studies.

The next Chapter 6 brings out the application of the present SIMPLE model for the
prediction of collision probability and closest approaches in a statistical sense. This chapter
provides a comparison of presently available collision probability assessment algorithms as
well. The details of collision probability assessment based on Poisson distribution and
extreme value distribution integrated with Monte-Carlo approach are also presented in this
chapter. A new approach for the collision probability assessment utilizing the closed form
solution of Wiesel (1989) by the way of three dimensional look up table, which takes only air
drag effect and an exponential model of the atmosphere, is also provided in this chapter.

The Chapter 7 proposes once again the application of the efficient and robust constant gain
Kalman approach that is able to accommodate the many uncertain, variable, and complex
features existing in the scenario and still able to predict the reentry time of the risk objects.
The constant Kalman gain based on using only a simple orbit propagator by considering drag
alone is capable of handling and accounting for the other modeling errors in a real life
situation. A detailed validation of the approach was carried out based on a few recently
objects

28
reentered objects and comparisons of the results with the predictions of other agencies are
presented.

The Chapter 8 provides the conclusions based on the present work carried together with
suggestions for future efforts needed in the study of space debris.
Chapter 2 Review of literature on the major issues in space debris studies

29
CHAPTER 2

REVIEW OF LITERATURE
ON THE MAJOR ISSUES IN
SPACE DEBRIS STUDIES

2.1 INTRODUCTION

There are four broad topics in the study of space debris. These are (Ananthasayanam 2001)

(i) Modeling,
(ii) Measurements,
(iii) Mitigation, and
(iv) Protection.

This thesis deals with the first topic namely modeling and in particular, the major issues
associated with it. Among many the five major issues involving modeling and analysis in
orbital space debris studies are

(i) The breakup modeling of space objects,
(ii) The environment model to characterize the space debris scenario,
(iii) The evolution of the space debris environment,
(iv) The estimation of the collision probability of debris with a target object, and
(v) The estimation of the reentry time for space objects.

The modeling of any aspect or issue or problem in any subject strives to progress the field.
These are broadly based on three approaches namely intuitive beginning, theoretical
framework, and experimental measurements. During the development of the subject the
information from these approaches flow back and forth to arrive at a best possible model at
objects
30
any given point of time. Enormous amount of work has been carried out during the past
many decades at various institutions and laboratories, and agencies dealing with the space
debris problem. In this chapter the earlier work carried out with regard to the above five
issues are reviewed. Finally, the newer contributions made in this thesis to further analyze the
modeling of space debris are described.

2.2 EARLIER BREAKUP MODELS OF SPACE OBJECTS

There are three aspects that involve modeling of the breakups of space objects by any of the
approaches in the literature. For the breakup fragments, these are (Johnson and McKnight
1987, Jehn 1996)

(i) The number and their velocity distribution,
(ii) The orbital parameters at the time of breakup, and
(iii) The physical and aerodynamic characteristics like mass, size, and ballistic
coefficient.

These help to properly predict their further evolution in the scenario, the collision risk they
pose to target objects and eventually their lifetime. A study of the earlier approaches shows
that the differences among the approaches are in the way in which the incremental velocities
(v) of the fragments are specified. The orbital parameters of the fragments are related by the
dynamical equations to v, and the physical and aerodynamic characteristics are generally
related through empirical relations derived from laboratory experiments.

2.2.1 BREAKUP MODEL OF JOHNSON AND MCKNIGHT

This model due to Johnson and McKnight (1987) has been used by many for the simulation
of the fragments for example by Ganeshan and Ananthasayanam (1997). A flow chart of this

31

Figure 2.1 A Satellite Fragmentation Model
from Johnson and McKnight (1987).
INPUT
Parameters of the
Delta Velocity
Distributions
Generate by Monte
Carlo Simulation
Delta velocities
Using assumed
distributions

B Values assigned
via hypothetical
distributions
INPUT

Integration Parameters
INPUT

Parent Orbital Parameters
a, e, i, ,
CONVERT
Orbital Parameters to state
vector
ADD DELTA
VELOCITIES
to original state
vector
INTEGRATE
Debris objects
SUMMARIZE DATA
PLOT DATA
CONVERT
New state
vector back
to orbital
elements
I= 1, N I= 1, N
I= N
objects
32
approach is given in Figure 2.1. Here the orbital parameters of the parent satellite are initially
converted into inertial coordinates and the breakup is then simulated by replacing the parent
body with a large number of fragments.

The velocity vector of each of the fragment is then formed by adding a delta velocity
component to each the three orthogonal velocity components of the original velocity vector.
The magnitude and distribution of delta velocities are assigned through a Monte Carlo
simulation with an assumed probability distribution for the incremental velocities. These new
velocity vectors can be converted next into the orbital parameters of these debris fragments
together with their further derived ballistic coefficient which can be utilized to study their
orbital evolution and collision probability assessment with a target object.

2.2.2 BREAKUP MODEL OF JEHN

The model suggested by Jehn (1996) basically utilizes the number versus the mass
relationships in accordance with the intensity of explosion. The Appendix 2 provides some
typical (among many) number to mass relationships derived by various studies in the field.
Jehn (1996) utilizes the relation by Su and Kessler (1985) in his first model and relations
based on the data from the Battle Institute (Fucke 1993) for the low intensity explosion and
the classical exponential relation for the first model and the modern power law as used in
MASTER model for the second model (Sdunnus and Klinkrad, 1993) with respect to the high
intensity explosion. For breakup due to collision the relation utilized is the one by Kessler
and Cour-Palais (1978). After obtaining the mass of the fragments, the effective area is
derived by utilizing a modified relation of Badhwar and Anz-Meador (1989) and Kessler and
Cour-Palais (1978). In order to take into account the scattering around the expected mean
value, the effective area is multiplied by a random variable from a lognormal distribution
with mean zero and standard deviation of 0.8. Making use of the assumption that the
deterministic delta velocity imparted to a fragment is a function of its size, the model by
Reynolds (1990) is utilized to obtain v from the effective area. Further in order to provide
random dispersions in v, the values are multiplied by a random variable from a triangular

33
distribution. Once the values of v for the fragments are thus available, this model proceeds
to derive the orbital parameters from the parent body orbital parameters and the variation in
orbital parameters due to v variations.

2.2.3 NASAS BREAKUP MODEL

The breakup models implemented in EVOLVE and other long term debris environment
models was known to be inadequate in two major areas (Johnson et al 2000). First it treated
all fragmental debris as spheres of a density, which is derived as a function of fragment
diameter, which is directly related to mass. However, it underestimated the generation of
fragments smaller than 10 cm in the majority of explosions. Hence a new breakup model was
proposed (Johnson et al 2000, Wegner et al 2000) which takes into account the additional
data and their analyses, results of three additional ground tests, better on orbit size and mass
estimation techniques, more regular optical tracking and reporting, additional radar sources
dedicated to the observation of small objects, and a longer time period with which to observe
the debris and its decay. The recent NASAs breakup model (Johnson et al 2000)

having a
bi-modal normal distribution to derive area to mass from diameter and a Lognormal
distribution depending on A/m is applied to spread the incremental velocities of the
fragments (Wegner et al 2000). It was noticed that as a consequence the clouds remain
centered much around the orbit of the parent body.

2.3 MODELS OF THE SPACE DEBRIS SCENARIO

Space debris models provide a mathematical description of the distribution of objects in
space, the movement and flux of objects and the physical characteristics of objects (namely
size, mass, density, reflection properties and intrinsic motion). These models can be
deterministic in nature which means that each object is described individually by its orbital
parameters and physical characteristics, or statistical in nature meaning the characterization
of an ensemble or a combination of both types. These models can be applied to risk and
damage assessments, prediction of debris detection rates for ground based sensors, prediction
objects
34
of avoidance maneuvers of operational spacecraft and long term analysis of the effectiveness
of debris mitigation measures. At present, some models are available for the characterization
of the space debris environment. These models are limited by the available data to validate
the derived relationships like the number density distributions and relationships between area
to mass ratio and fragmental velocity increments. These models rely on the historical records
of satellite characteristics, launch activity and on orbit breakups; in addition, there are only
limited data on spacecraft material response to impact and exposure to the orbital
environment. Furthermore, major assumptions are made in applying these models to predict
the future environment. In particular, future traffic scenarios and the application of mitigation
measures will have a major influence on the outcome of model predictions. Space debris
models must be continually updated, turning it to be stochastic corrective in sense, and
validated to reflect improvements in the detail and size of observational and experimental
data sets.

There are three types of models in debris environment scenario (Technical Report on Space
Debris 1999, Kessler et al 1997 and Jehn et al 1997). These are called as

(i) Support models
(ii) Evolutionary models and
(iii) Engineering models

Support models help in the analysis of the space debris environment and cover specific
topics, for example

(i) satellite breakup models,
(ii) flux models for evaluating collision risks,
(iii) traffic growth models,
(iv) lifetime estimation, and
(v) reentry studies,


35
Evolutionary models consist of a set of support models combined together to predict the
orbital scenario. They are based on historical population, which has been created by
simulating the known on orbit fragmentation the examples being EVOLVE and IDES.

Engineering models combine the outputs of other orbital debris models with measurements
of the environment to produce an environment definition model. Examples of engineering
models are ORDEM 96 and the recent ORDEM 2000.

Environment models may take two forms: as discrete models, which represent the debris
population in a detailed format, or as an engineering approximation. Furthermore, these
models can be short term in nature (considering timeframes of up to 10 years) or long term
(considering timeframes of over 10 years). In the preparation of all these models, the initial
debris population is represented at a particular starting epoch and propagated forward in time
in a stepwise manner, taking account of source and sink mechanisms and relevant orbit
perturbations. Neither the short-term nor the long-term models account for the periodic
concentrations of debris that exist for hours and months following a breakup. Such "very
short term" models are occasionally used to assess the hazard to specific space systems only.
The available databases for the space debris models are

(v) United States Space Command Satellite catalogue and the Russian Space
Surveillance catalogue for deterministic data on decimeter sized and larger
objects,
(vi) Dedicated radar campaigns in LEO for data on centimeter sized objects,
(vii) Statistical data from the analysis of retrieved surfaces and from in situ impact
sensors encountering sub millimeter debris populations,
(viii) Ground based telescopes for statistical data on decimeter and larger objects in
LEO,
(ix) Ground based simulations of explosive fragmentations, and
(x) Ground based simulations of hypervelocity collisions with satellite and rocket
bodies.
objects
36

The availability of models to estimate the impact risk from the man made space debris and
the natural meteoroid environment is essential for both, manned and unmanned satellite
missions. Various independent tools for modeling the space debris environment based on
different approaches have been developed during the past many years. Due to an increased
knowledge of the debris environment and its sources, for example, from improved
measurement capabilities, these models could be updated regularly, providing more detailed
and more reliable simulations.

Table 2.1 Some Well Known Debris Environment Models

Model name Source
Evolutionary
period
Engineering model
available
Minimum
size
Orbital
regime
CHAIN NASA Long term No 1 cm LEO
CHAINEE ESA Long term No 1 cm LEO
EVOLVE NASA
Short and long
term
No 1 mm LEO
IDES DERA
Short and long
term
No 0.01 mm LEO
LUCA TUBS Long term No 1 mm LEO/MEO
MASTER ESA Short term Yes 0.1 mm LEO/GEO
NAZARENKO RSA
Short and long
term
No 0.6 mm LEO
ORDEM 96 NASA Short term Yes 1 mm LEO
SDM/STAT ESA/CNUCE
Short and long
term
No LEO/GEO


37
The pertinent characteristics of the presently available debris environment models are
compared (Technical Report on Space Debris 1999) in Table 2.1. Among the models listed in
Table 2.1 two important models are ORDEM and MASTER. The MASTER model was
first released in 1997 and further an improved version was released in 1999. The ORDEM is
a semi deterministic engineering model with its first version released in 1996 known as
ORDEM 96 and a later version known as ORDEM 2000 was released in the year 2000.
Apart from MASTER and ORDEM we highlight some other known models also in this
chapter.

In the next section we discuss the basic characteristics and the utilities of some of the
presently available space debris environment models.

2.3.1 MASTER (METEOROID AND SPACE DEBRIS TERRESTRIAL
ENVIRONMENT REFERENCE) MODELS

The purpose of MASTER (Klinkrad et al 1997, Sdunnus et al 2002) is the characterization of
the natural and the man made particulate environment of the earth, and the consequent fast
and simple evaluation of the resulting effects on space missions. The model is based on a
semi-deterministic analysis of a reference population derived from the simulation of all major
space debris source terms. Also the micrometeoroids, as the natural component of the Earth
particulate environment, are also accounted for in MASTER.

MASTER consists of the Standard Application for a quick assessment of debris flux
characteristics, and the Analyst Application with high resolution flux results and additional
analytical capabilities, as for example flux on oriented surfaces, future constellation traffic
simulation, or meteoroid flux determination. All applications are made available together
with an easy to use Graphical User Interface (GUI) and online documentation on a CDROM
(Sdunnus et al 2002).

objects
38
The model can be installed and executed on a broad variety of platforms such as Sun Solaris,
IBM AIX, HP-UX, Linux, Windows 9x/NT and Apple Mac-OS.

For each simulated source, a corresponding debris generation model in terms of
mass/diameter distribution, additional velocities, and directional spreading has been
developed. A comprehensive perturbation model was used to propagate all objects to the
reference epoch.

Currently, apart from spent payloads and upper stages (TLE background), fragmentation
from on orbit collisions and explosions, dust and slag from Solid Rocket Motor (SRM)
firings, sodium-potassium (NaK) coolant droplets from RORSAT satellites, surface
degradation particles (paint flakes), and ejecta are also considered.

In order to describe the steady state natural meteoroid environment the Divine Staubach
meteoroid model (1993) was implemented into MASTER. This model considers core and
asteroidal meteoroids and is available as high resolution flux model via the MASTER
Analyst Application. Further the time correlated, seasonal meteoroid stream events are also
included based on the optional models of Cour-Palais (1969), Jenniskens (1994), or McBride
(1998).

The MASTER model offers a full three dimensional description of the terrestrial debris
distribution extending from LEO in altitude beyond 6564 km to the GEO region up to the
altitude 45164 km both from the center of the earth. Flux results relative to an orbiting target
or to an inertial volume can be resolved into source terms, impactor characteristics and orbit,
as well as impact velocity and direction. Also three dimensional flux analyses with respect to
an orbiting target or to a fixed volume in inertial space is possible.

MASTER is provided in a CDROM containing the stored, binned data together with the
processing software. A simplified version, called the MASTER engineering model, is
available as well. This version is applicable for target orbits with small eccentricities.

39
MASTER 97 MODEL

MASTER 97 (Klinkrad et al 1997 and Sdunnus 1995) uses a semi deterministic approach,
which represents the debris environment by modeling its history from the beginning of space
flight upto to the present time. It uses three source terms for the debris population: launches,
explosions and collisions and follows their orbital evolutions. The space debris population at
an epoch of 31 March 1996 is derived from 132 low and high intensity on orbit breakups.

A linear growth rate of 2 percent (uniform for all sizes) is applied for other years. Solar cycle
effects on cleansing rate are included as for the NASA engineering model (Anderson 1994)
given in terms of the mass (or diameter) dependent binning is used to assign weights to a
condensed population of about 240000 objects larger than 0.1 mm. An analyst application
allows to interrogate the spatial debris distribution to determine collision fluxes for an
arbitrary target orbit passing through the control volume. Flux results can be analyzed with
respect to the magnitude of the collision velocity, its direction (azimuth and elevation), the
orbit location, and the three dimensional position where the flux was encountered.

MASTER 99 MODEL

Compared to the previous MASTER 97 version, the features that have been significantly
updated or added in the MASTER 99 release are basically due to the source terms.

Apart from the TLE catalogue objects, the MASTER 97 version considered only
fragmentations as source term for man made space debris. The MASTER 99 release adds
Al
2
O
3
dust and slag objects from solid rocket motor (SRM) firings, sodium potassium (NaK)
coolant releases from RORSAT reactors, particles from surface degradation (paint flakes),
and ejecta from minor impacts. The processing of space debris generation mechanisms such
as SRM firings, fragmentations, NaK release events and so on was carried forward up to the
new reference epoch of August 1
st
1999.

objects
40
A TCL/TK based Graphical User Interface (GUI) was developed to allow for an intuitive and
efficient installation and execution of the MASTER tools. The GUI comes as a self
contained package and does not require the installation of additional interpreters.

The MASTER Standard Application is the successor of the former Engineering Application
MASTER 97. It is based on an approach developed by Divine (1993) and allows to compute
spatial densities and fluxes from user defined populations, which are defined by means of
orbital element distributions. The MASTER Standard Application has proven to be a fast
and efficient storage tool.

Meteoroid streams based on Cour-Palais (1969) and Jenniskens (1994) have been
implemented to account for the dynamic part of the meteoroid population. Directional effects
may be studied with this option. The user is allowed to edit all parameters of the considered
streams, and may even create new streams.

While MASTER 97 always generated flux on a sphere, MASTER 99 can furthermore
compute flux on oriented surfaces, a randomly tumbling plate, or on a set of up to 10 user-
defined surfaces, which can be defined in an Earth oriented, Sun oriented, or inertially fixed
coordinate system.

Setting up a user editable file containing the basic constellation characteristics can simulate
the effect of planned large constellations on the flux contributions from TLE background
objects.

MASTER 2001 MODEL

The ESA MASTER (Wegner et al 2000) model offers a full three dimensional description of
the terrestrial debris distribution reaching from LEO up to the GEO region. Flux results
relative to an orbiting target or to an inertial volume can be resolved into source terms,
impactor characteristics and orbit, as well as impact velocity and direction. All relevant

41
debris source terms are considered by the MASTER model. For each simulated source, a
corresponding debris generation model in terms of mass/diameter distribution, additional
velocities, and directional spreading has been developed. A comprehensive perturbation
model was used to propagate all objects to a reference epoch.

2.3.2 ORDEM (ORBITAL DEBRIS ENGINEERING MODEL)

The NASA Orbital Debris Program Office at Johnson Space Center has developed computer-
based orbital debris engineering models ORDEM 96 and ORDEM 2000 (Kessler et al 1997,
Liou et al 2001) which describe the orbital debris environment in the low Earth orbit region
between 200 and 2000 km altitude.

ORDEM 96 MODEL

The NASA engineering model (Kessler et al 1997) ORDEM 96, analytically represents
similar sources to MASTER plus additional non fragmentation source distributions like paint
flakes, Aluminium Oxide particles from solid rocket motor burns and Sodium/Potassium
droplets. These non fragmentation sources mainly contribute to the small size population.
The model is also derived from recent radar measurements and impact data from retrieved
surfaces. The debris environment distribution is approximated with six different inclination
bands. Each band has a unique distribution of semi major axis, for circular orbits, and a
unique perigee distribution for highly elliptical orbits. In addition, each inclination band has a
unique, analytically defined size distribution, which depend on the source of debris. Collision
probability equations are used to relate the orbital element distributions of the population to
the flux measured on a spacecraft or to the flux measured through the field of view of a
ground sensor. This model also provides directional information of the impacting flux.

Impact velocities can range from 0 to about 15.5 km/s with an average velocity of 10 km/s
for low inclination and of 12 km/s for high inclination orbits. According to this model, the
average density of particles larger than 0.62 cm in diameter is assumed to be 2.8x10
-0.74

objects
42
g/cm3. The average density of smaller space debris particles is thought to be in the range 2.8
- 4.7 x10
-0.74
g/cm3. The ORDEM 96 space debris model is applicable for particle sizes
larger than 1 micron and for circular and elliptical Earth orbits up to altitudes of 2000 km.

Nominal and worst case debris model parameters and mass densities utilized in ORDEM 96
are solar activity, index F10.7 between 70 and 140 and debris density between 4.0 - 8.0
(g/cm3).

For design purposes the nominal parameter values shall be used. A spherical shape shall be
assumed to convert particle masses and diameters. The default growth rates for the respective
models shall be used. For the assessment of impact effects the complete velocity distribution
of space debris should be used. For a preliminary analysis a constant debris impact velocity
of 10 km/s shall be used. Uncertainty factors for space debris fluxes shall not be applied.

ORDEM 2000 MODEL

The NASA Orbital Debris Program Office at Johnson Space Center has developed a new
computer based orbital debris engineering model ORDEM 2000, which describes the orbital
debris environment in the low Earth orbit region between 200 and 2000 km altitude (Liou et
al 2001) e. It is assumed that the model is appropriate for those engineering solutions
requiring knowledge and estimates of the orbital debris environment consisting of debris
spatial density, flux, and so on. ORDEM 2000 can also be used as a benchmark for ground
based debris measurements and observations. The model utilizes a large set of observational
data, covering the object size range from 10 mm to 10 m. The observational data is
incorporated into the ORDEM 2000 debris database, utilizing a maximum likelihood
estimator to convert observations into debris population probability distribution functions.
These functions then form the basis of debris populations. ORDEM 2000 contains a finite
element model to process the debris populations to form the debris environment. It is pointed
out by the developers that an improved input and output structure and a user friendly GUI is
also implemented in the model. ORDEM 2000 has been subjected to a significant

43
verification and validation effort. This model ORDEM 2000 supersedes the previous
ORDEM 96 model. The availability of new sensor and in situ data, as well as new analytical
techniques, has enabled the construction of this new model.

DIFFERENCES BETWEEN ORDEM 2000 AND ORDEM 96

Once a debris population is characterized in terms of size and orbital parameters, one can
build a LEO debris environment model from the population to describe the debris spatial
density and velocity distribution in space.

Figure 2.2 Comparison of ORDEM 96 and ORDEM 2000
(From Liou et al 2001)

LEO debris
population

derived from
observations
or in situ
measurements
6 inclination
bands
. 2 eccentricity
families
. Random ,
. Size distribution
. Altitude
dependence
Model debris
environment
(equations)
. Impact flux (s/c)
. Observed flux
ORDEM96
Model debris
environment
(data files)
. Impact flux (s/c)
. Observed flux
ORDEM 2000
objects
44
The Figure 2.2, from The New NASA Orbital Debris Engineering Model ORDEM 2000
(2001), shows the differences between ORDEM 96 and ORDEM 2000 in describing the flux
densities of LEO orbital debris environment.

The ORDEM 96 simplifies the population into six inclination bands and 2 eccentricity
families. The six bands are represented by the inclinations 7
0
, 28
0
, 51
0
, 65
0
, 82
0
and 98
0
.
Objects within each inclination band are assumed to have the same inclination rather than a
distribution of inclinations. Objects with eccentricities less than 0.2 are considered to be in
circular orbits, implying zero eccentricity, while the rest are considered to be in elliptical
orbits with eccentricities between 0.5 and 0.6. In addition, it also assumes that the longitudes
of the ascending node and arguments of perigee of objects are randomly distributed in space.
With additional approximations regarding the size distribution and the orbital perigee
distribution, a set of equations can be derived to represent the LEO debris environment. One
can then use the equations and calculate the impact flux on an orbiting spacecraft or the
debris flux expected to be observed by a ground based telescope or radar.

2.3.3 OTHER DEBRIS MODELS

Several other space debris models are in use or under development. These models can be
used in special cases for impact analyses or risk assessments if the user is familiar with the
features and range of validity of the models. Some of these models are described below.

IDES (INTEGRATED DEBRIS EVOLUTION SUITE)

The Integrated Debris Evolution Suite (IDES) space debris model (Walker et al 1996) uses a
deterministic representation of individual objects greater than 10 cm and a statistical
representation of the populations between 10 m and 10 cm by fragment orbit discretization
in dimensions of perigee, radius, eccentricity, inclination and mass. The debris source terms
considered include launch related objects, high and low intensity explosions, collisions, paint
flakes and secondary ejecta. These sources are modeled and evolved from the beginning of

45
the space age to the present day and extend into the long term future. For impact risk
assessments the directional and velocity distribution of the encountered flux is included.
Circular and elliptical target orbits may be analyzed. IDES is applicable to LEO altitudes and
for particle sizes larger than 10 m. The tool also allows studying the long term evolution of
the debris population and the long term impact risk to target orbits under different traffic
scenarios, and the effectiveness of different mitigation measures.

NASA 90 MODEL

From about 1990 until 1996 the NASA space debris engineering model (Anderson 1991) has
been most widely used for design applications. This model is given in terms of simple
analytical expressions. It is relatively easy to use and widely distributed. However, it has
some known shortcomings of which the assumption of spherical objects for all debris
particles is probably the most severe. The model can still be used for some preliminary risk
assessments but however the user should be aware of its shortcomings.

NAZERENKO MODEL

Nazarenko has developed a model (1993) at the Centre for Programme Studies (CPS) of
RSA, a semi analytic, stochastic model for both short term and long term prediction of the
LEO debris environment, providing spatial density, velocity distributions and particle fluxes.
The model takes into account, in average form, the debris sources (except for the cascading
effect) and of atmospheric drag. It has been adjusted on the basis of Russian and American
catalogued data and published measurements of somewhat smaller fragments (more than 1
mm) while also accounting for any a priori information.

EVOLVE MODEL

The EVOLVE model has been developed by NASA (Reynolds 1991). It is a semi-
deterministic model (SDM), wherein debris objects are described by a set of parameters. In
objects
46
addition to being capable of modeling the present debris environment, it can be used to
investigate future evolutionary characteristics under various mitigation practices using Monte
Carlo techniques. EVOLVE has been the primary NASA orbital debris environment model
for the past decade. EVOLVE is NASA's One Dimensional Orbital Debris Evolutionary
Model.

It is used for modeling long term (a few decades to a few centuries) evolution of the orbital
debris environment. EVOLVE is a 1-D simulation model, in that it places launched objects,
explosion or collision breakup fragments, calculates how these orbits will change in time,
and calculates from these orbits debris environment characteristics such as orbital debris
spatial density as a function of time, altitude, and debris size. EVOLVE places breakup
fragments in the environment according to the size and velocity distribution of these
fragments in the NASA breakup models. Debris mitigation measures are modeled in
EVOLVE through a scenario definition file. Scenarios have been run controlling future
launch rates, accidental explosions, and post mission orbit lifetime for payloads and upper
stages. Launched objects are placed in orbit in accordance with historical data, or for
environment projections with mission model data that specifies the launch date, orbit,
payload, and upper stage data for future launches.

The typical EVOLVE environment projection (web site http://sn-
callisto.jsc.nasa.gov/model/modeling.html) is started by calculating the current environment,
using the historical record of launches and breakup events, where fragments from these
events are placed in orbit using the breakup model. This current environment is then used as
the initial environment for debris environment projections. Because EVOLVE calculates the
current environment, comparisons between this environment and measurements of the
current environment can be used to provide confidence that historical sources of orbital
debris are being handled properly and that similar sources are being handled correctly in the
environment projections. EVOLVE results are periodically compared with the US Space
Command satellite catalog and the Haystack/Haystack Auxiliary (HAX) radar data. Because
EVOLVE explicitly simulates processes contributing to the evolution of the orbital debris

47
environment, it has proven to be very flexible in its ability to study new types of debris
mitigation, to model different scenarios on the future use of space, and to analyze the
sensitivity of environment projections to many of the multitude of factors on which the
projections depend.

LEGEND (LEO to GEO Environment Debris) MODEL

LEGEND is NASA's Three-Dimensional Orbital Debris Evolutionary Model (web site
http://sn-callisto.jsc.nasa.gov/model/modeling.html). In order to continue to improve our
understanding of the orbital debris environment, the NASA Orbital Debris Program Office
initiated an effort in 2001 to develop a new model to describe the near earth debris
environment. The LEGEND, a LEO to GEO Environment Debris model, is a full-scale three-
dimensional debris evolutionary model. It covers the near Earth space between 200 and
50,000 km altitude, including Low Earth Orbit (LEO), Medium Earth Orbit (MEO), and
Geosynchronous Earth Orbit (GEO) regions. The model is capable of providing debris
characteristics (number, type, size distribution, spatial density distribution, velocity
distribution, flux, and so on as functions of time, altitude, longitude, and latitude. LEGEND
includes a historical simulation component (1957 to present) and a future projection
component. Populations included in the model are active and spent satellites, rocket bodies,
breakup fragments, operational debris, and Sodium-Potassium (NaK) droplets. The minimum
size (diameter) threshold in the model is 1 mm.

The main function of the historical component of LEGEND is to reproduce the historical
debris environment (1957 to present) to validate the techniques used for the future projection
component of the model. The model utilizes a recently updated historical satellite launch
database (DBS database), two efficient state of the art propagators (PROP3D and
GEOPROP), and the NASA Standard Breakup Model. The overall program structure and the
input and output options are optimized to allow reasonable execution time and manageable
file storage and memory space. The program is capable of providing multi dimensional
descriptions of the debris environment. Additional modules have been developed to better
objects
48
analyze and visualize the model outputs. The main function of the LEGEND future
projection component is to allow the user to understand how the orbital debris environment
evolves in the future. It is also a reliable tool to examine how various mitigation practices
may help to protect the environment. A key element in the LEGEND future projection
component is a three dimensional collision probability evaluation model. It provides a fast
and accurate way to estimate future on orbit collisions from LEO to GEO. Since no
assumptions regarding the right ascensions of the ascending node and arguments of perigee
of objects involved are required, this model captures the collision characteristics in real three
dimensional physical space. It is a critical component of a true three dimensional debris
evolutionary model.

The typical projection period in LEGEND is 100 years. Due to uncertainties involved in the
process (namely, future launch traffic, solar activity, explosions, collisions), conclusions are
usually drawn based on averaged results from 30 Monte Carlo simulations. The main
program and subroutines of LEGEND are written in standard FORTRAN 90 language.
Additional analysis tools are written in FORTRAN 90 and IDL.

SDPA 2000 (SPACE DEBRIS PREDICTION AND ANALYSIS) MODEL

The Russian Space Debris Prediction and Analysis (SDPA) model (Nazerenko et al 2001,
Sdunnus 2002) is the semi analytical stochastic tool for medium and long term forecast of the
man made debris environment (with size larger than 1 mm), for construction of spatial
density and velocity distribution in LEO and GEO as well as for risk evaluation. The last
version of SDPA 2000 consists of ten individual modules related to the aforementioned tasks.
The total characteristics of space debris of the different sizes are considered (without
partition of these characteristics into specific sources). The current space debris environment
is characterized by

(a) the spatial density dependence on the altitude and latitude of a point, as well as on
size of objects, and

49
(b) a statistical distribution of the magnitude and direction of space objects velocities

in an inertial geocentric coordinate system. These characteristics are constructed on the basis
of the complex application of the accessible measuring information and series of a priori
data.

2.3.4 UNCERTAINTIES IN THE VARIOUS MODELS

The meteoroid and space debris environment flux models given above contain several known
approximations and other uncertainties.

One of the known uncertainties are in the meteoroid models and mainly resulting from
uncertainties in particle densities and masses. Fluxes for meteoroids larger than 10
-6
g are
well defined, but the associated masses are quite uncertain. The mass density of meteoroids
spans a wide range, from about 0.15 g/cm
3
to values as large as 8 g/cm
3
. At a set mass this
implies an uncertainty in the flux of a factor 0.1 to 10. For meteoroids smaller than 10
-6
g
flux uncertainties at a given mass are estimated to be a factor of 0.33 to 3. The space debris
flux models were developed as a best estimate rather than a conservative one. The factor of
uncertainties for debris fluxes in different size regimes giving the 90 % confidence level are
provided in the literature and that the real debris flux is within a bandwidth defined by the
model flux multiplied by the uncertainty factors.

A third uncertainty factor is regarding the particle sizes. It is reported that for objects with
size greater than 10 cm , the uncertainty factor is from 0.5 to 1.5, for sizes between 0.05 cm
to 10 cm the factor is from 0.33 to 3.0 and for other sizes less than 0.05 cm, the uncertainty
level is from 0.5 to 2.0. Other uncertainties of the debris model include the debris density,
the debris shape, the annual growth rates, the altitude dependence, especially for altitudes
above 800 km, the velocity distribution, especially for low impact velocities and the
dependence on solar activity.

objects
50
2.3.5 COMPARISON OF THE MODELS

Due to the assumptions and limitations of the various models described above, whose
descriptions not or cannot fully agree with each other the results of the long term debris
models do not agree quantitatively because of differences in assumptions and initial
conditions. However, the basic trends and tendencies obtained by the models agree
qualitatively.

The broad features in the long term debris model trends can be summarized as follows.

(a) The debris population may grow in an accelerated manner in the future if the
space flight activities are as frequent in the past. Given the increased objects
more collisions will occur in orbit.
(b) Currently fragments from explosions are the main source of space debris.
Without some mitigation efforts incorporated in the design and operations,
collision fragments may become one of the main debris sources in the twenty-
first century.
(c) Collisional fragments may contribute to further subsequent collisions. This
means that, at some point in the future, which is difficult to predict, the
population could grow enormously. Only by limiting the accumulation of
mass in LEO this can be prevented.
(d) Suppressing explosions can reduce the increase in the number of objects in
orbit, but cannot prevent collisions, which are driven by the total mass in orbit
and the number of large objects.

The Figure 2.3 from the Technical Report on Space Debris (1999) provides a picture on the
variations of the spatial density predictions between various models, with respect to the size
of the space debris objects.


51
Figure 2.3 Spatial Density Comparisons between Different Models
(From Technical Report on Space Debris 1999)

2.4 MODELS FOR THE EVOLUTION OF SPACE DEBRIS
SCENARIO

Most of the environment models contain evolution models as a orbital propagator in it. These
models consider the influence of a force model to simulate the evolution and dispersion of
the debris objects. These force models include drag forces, luni solar perturbations, earths
oblateness in terms of J
2
gravitational term and other perturbing forces like solar radiation
pressure. But to obtain the long term evolution of millions of objects, utilization of such
propagation models is impossible. In order to handle such situations, different evolutionary
objects
52
approaches are available in the literature. Jehn (1996) presented a fast orbital propagator
based on a look up table generated utilizing the (a, e) bins and simple orbital decay equations
due to drag effect alone (King-Hele 1987). The basic equations for such a propagation are
provided in Appendix 3.

Another model STAT was generated by Rossi et al (1997). The (STochastic Analog Tool)
STAT by Rossi et al (1997) code simulates the time evolution of the debris environment by
considering the number of objects contained in a set of discrete bins of semimajor axis,
eccentricity and mass. The evolution algorithm is based upon a set of finite difference
equations, which take into account launches, retrievals, explosions, collisions and orbital
propagation under the influence of atmospheric drag.

The bins are provided by a three fold subdivision,

(i) in semimajor axis (from 6378 to 46378 km),
(ii) eccentricity (from 0 to 1), and
(iii) mass (from 1 mg to 10,000 kg).

The state variables are initialized at the actual population, which is taken from the CNUCE
1994.0 Orbital Debris Reference Model (1994). Once the variables have been initialized, the
calculation and summation of different contributions at each time step is performed. An
important feature of STAT is that it propagates the orbit of an individual representative
object for each bin and therefore the computing time is almost independent of the dimension
of the population.

This approach yields results with respect to the propagation of a single point per bin, without
being as expensive in terms of computing time as the propagation of a statistically significant
set of representative points per bin.


53
2.5 COLLISION PROBABILITY ASSESSMENT

There are different approaches available in literature for the assessment of collision
probability of space debris objects with a target object. One method utilizes the approach
based on Poisson distributions (Johnson and McKnight 1987) and the principles of the
Kinetic theory of gas, another one utilizes the deterministic calculation of the closest
approaches for designing the collision avoidances and a third approach is based on an MC
technique (Vedder et al 1991) to obtain the closest approaches and thus obtain the
distribution of the closest approaches by an extreme value distribution.

In the third approach the closest distances are calculated based on either propagation the
debris objects with respect to a target spacecraft at different time intervals during an orbital
period of the target body. Identifying an accurate propagator and propagating all debris
objects to an epoch considered for the spacecraft are tedious and time consuming. It is
observed that an MC approach based on randomly varying mean anomalies of debris objects
provide nearly the same results as in the former case (Johnson and Mc Knight 1987, Vedder
et al 1991 and Chobotov et al 1997). The basic concept of this approach is to observe the
frequency of debris approach distances to a target on a larger, observable scale, and then to
use statistical methods to extrapolate to frequencies of the virtually unobservable collision
proximity distances. In order to observe the observable data we create a Monte Carlo
simulation of the orbital dynamics of the catalogued objects and an arbitrarily chosen target
body. Using this simulation the closest distance between any of the debris objects with the
target at a specified epoch, mean anomaly, in an orbit of the target body are calculated. These
closest distances at different mean anomalies are then approximated by a statistical
distribution.

Theoretically, this set of minimum distances should be characterized by a particular
asymptotical distribution function. The extreme value distribution, namely the Weibull
distribution (Mclauglin 1999) is the best distribution for fitting the minimum of the observed
data, to extrapolate the required information on unobservable collision probabilities. Once
objects
54
the parameters of the distribution have been determined from the data, the distribution of the
closest approach is fully described. Quantitative estimates of the fraction of an extended time
interval during which the target is within a specified distance of a debris object can be
directly determined from this distribution. In order to estimate the number of encounters
during this interval, it is necessary to define the maximum proximity distance that entails a
collision and to derive the expected time for encounter E(Tc), Tc being the encounter time,
and it enables estimating the number of collisions in a given time period for a specified
target.

The collision probability of earth orbiting satellites may be calculated through the use of
kinetic theory of gases concepts (Johnson and McKnight 1987). Here a number of
assumptions are made about the arrangement of satellites in earth orbits to approximate the
spatial density of the objects. The space is divided into toroidal sectors, )S, formed by the
intersection of different altitudes and planes of different latitudes. The rate of occurrence
parameter, R is obtained from the above theory. It is defined as the number of collisions
during a given length of time between any molecules of gas with others in a confined
volume, and is given by the product of the molecule size, density of gas, relative speed and
the length of the time considered.

By using the tracked debris objects in low earth orbit, the average velocity may be calculated
through numerical simulation of possible encounters or through direct averaging concepts.
Both of these methods utilize an average relative velocity as 10km/s.

By applying the above arguments to a confined volume in space, the collision rate between a
specified object and all other objects in a given volume can be found for an interval of time.
Hence R would be the product of the target objects crosses sectional area (A), average
spatial density of all other object (S), the average relative velocity (V) between the target
object and other objects under consideration and the time (T). Hence the probability P of
collision for time duration T is (Johnson and MCKnight 1987, Meshishnek 1995)

55

Where A is the cross sectional area is the projected area of the object along the trajectory.
Since objects could actually be tumbling through space, and hence changing the projected
cross sectional area, using the largest possible cross sectional area will provide the most
conservative collision estimate. Here the assumption is made that all objects in the population
are small compared to the object under consideration. Thus, the collision cross section can be
substituted by the cross sectional area of the object for a reasonable first order
approximation. Relative velocity is the function of the relative angle of encounter of the
object with another object and the velocity of the colliding object. The time at risk is the
length of time the object poses a risk to the population. If the object is to remain in orbit for a
predetermined length of time, the time at risk is this value. If an object is traveling in an
ascending motion the time of trajectory to reach the object is the time at risk, which may be
very short. If an object is an orbit, which is subjected to decay by atmospheric drag, the
lifetime of the object must the time at risk.

2.6 REENTRY PREDICTION OF RISK OBJECTS

The accurate estimation of the orbital life time of decaying objects is of considerable
importance for the prediction of risk object reentry time and proper planning of mitigation
strategies and hazard assessment. Space debris studies demand assessment of orbital life
times of a large number of objects resulting from on orbit fragmentation or collision.

Generally the orbital propagators as mentioned earlier are utilized for the prediction of
reentry time or lifetime of debris objects (Sharma and Anilkumar 2003, Anilkumar and
Subba Rao 2002). Since the ballistic coefficient of the debris objects are not known precisely,
some models are developed to take into account this uncertainty. Such method utilizes a
maximum likelihood estimation based on the orbital data measurements to estimate the
ballistic coefficients for predicting the lifetime. Some other methods utilize optimization
) * * * (
) * * * (
1 T V S A
T V S A
e P
=
objects
56
techniques such as genetic algorithm to estimate the reentry time by minimizing the variation
in lifetime obtained by TLEs at different epoch. These methods are successfully utilized in
the prediction of the reentry time of risk objects as part of the IADC reentry campaigns
(Sharma et al 2003).

It can be stated that in the literature, there appears to be no one single approach to the
assimilation of measurement data with the predictions to obtain the updates. The Kalman
filter and in particular with a constant Kalman gain provides just about the best approach to
assimilate and also expand the scenario to update the ballistic coefficient as explained in the
later chapters.

2.7 CONCLUSIONS

This chapter provided a description of the available previous work on the five major issues in
space debris modeling. The availability of models to estimate the impact risk from the man
made space debris and the natural meteoroid environment is essential for both, manned and
unmanned satellite missions. Various independent tools based on different approaches have
been developed in the past many years. Due to an increased knowledge of the debris
environment and its sources, for example, from improved measurement capabilities, these
models could be updated regularly, providing more detailed and more reliable simulations.
This chapter discussed some of the available on orbit breakup models and also the
environment models such as ORDEM and MASTER and their later updates. This chapter
also describes some of the existing evolutionary models to describe the debris scenario at
later times. Further the many collision probability estimation approaches have been received.
Also the resistance of many reentry time prediction techniques is noted.

However it should be maintained that in a large majority of cases, the details are not provided
explicitly in literature, particularly about the evolution and reentry time predictions. This
precludes a more detailed description and comment on the work.
Chapter3 A proposed model of on orbit breakup ASSEMBLE and simulation of some historical breakups

57
CHAPTER 3

A PROPOSED MODEL OF
ON ORBIT BREAKUP
ASSEMBLE AND SIMULATION OF SOME
HISTORICAL BREAKUPS

3.1 INTRODUCTION

Since the first on orbit breakup of TRANSIT 4A rocket body in 1961 till date there are about
more than 180 spacecraft and rocket bodies that have fragmented in orbit as reported by Anz
Meador (2001) and in the Orbital Debris Quarterly News of January 2002. The resulting
debris fragments as mentioned earlier constitute nearly 40% of the 9000 or more catalogued
objects by USSPACECOM as on 31
st
December 2003 (Orbital Debris Quarterly News,
January 2004). All of the fragments generated by the breakup form a cloud of debris and
move in the environment later merging into the background of debris scenario. There is the
probability of collision risk for a target body due to these debris objects. The catalogued
fragment count does not include the numerous fragments, which are too small to be detected
from ground. Hence in order to describe the trackable orbital debris environment, it is
necessary to develop mathematical models to simulate the trackable fragments formed by
explosion or collision and later expand it to untrackable objects. Subsequently a complete
risk assessment can be carried out due to all the orbiting debris.

3.2 CONCEPTUAL BASIS FOR THE PRESENT MODEL

Any of the modeling process, considered as a physical theory obviously has three
approaches, namely, intuitive beginning, mathematical modeling and consistency with
objects

58
experiments. These are similar to the various approaches to probability theory (Papoulis
1991, Feller 1950), and have increasing amount of information. The mathematical modeling
of breakups (BU) of objects in space is quite a difficult task due to the randomly varying
ways in which the combination of the debris velocity, mass, size, and shape can occur. Hence
it is better to be based on the measurements of the debris fragments, which could help in a
better mathematical modeling of such processes.

This chapter deals with a new approach to model the on orbit breakup of an object in space.
The present model is based on an analysis of the available catalogued USSPACECOM data
of the fragments soon after the breakup. Any breakup creates many fragments and each have
their orbital characteristics such as the semi major axis, eccentricity and inclination, and
other parameters like the ballistic coefficient, size, shape and many more. The ballistic
coefficient is provided in the USSPACECOM catalogue or can be derived from the available
experimentally measured orbital data of these fragments at various times. This is because the
orbital decay is sensitive to air drag through the ballistic coefficient. However the size and
shape cannot be easily (or sufficiently accurately) estimated based on orbital decay
characteristics and so will have to be obtained from empirical relations. Such relations are
based on some laboratory experiments, which provide sample relations only and not that of a
population. But one thing is that all these variables characterizing the fragments are in some
way connected due to the underlying physical process at the time of explosion. What this
implies is that one cannot assign arbitrary random values for their various properties. The
question is from which of the properties one can commence with and later derive the others?
This is similar to obtaining the height, weight and age of a person. Where do we start for best
possible estimates? It turns out that there is a hierarchy among the variables of a system. The
hierarchy in terms of whatever it implies, dictates in the present case that it is better to
commence from the measurable independent variables such as the semi major axis and
eccentricity to arrive at the velocity change in a breakup (BU). This is in contrast to earlier
approaches wherein one assumes the velocity of the fragments at the time of breakup.


59
3.3 EARLIER APPROACHES TO MODELING THE ORBITAL
BREAKUPS

The Figure 3.1 below shows the three aspects of any modeling effort for the orbital debris
breakup. The Block 1 shows the USSPACECOM catalogued orbital characteristics as well as
the ballistic coefficients. The Block 2 contains the distribution of the velocity of the
fragments and is the focus of all the breakup modeling approaches by the way that this aspect
is handled. The Block 3 relates the number of fragments, their individual mass, and size
through empirical relations obtained from laboratory experiments.

Figure 3.1 Three Aspects of Orbital Breakup Modeling and Analysis.

In Chapter 2, the earlier approaches for the modeling of orbital breakup were outlined. The
approaches in the framework of the above three aspects and its implications are presently
discussed. The approach of Johnson and McKnight (1987) was followed by Ganeshan and
Ananthasayanam (1997). In this one commences from Block 2 and proceeds to Block 1 and
Block 3 as follows. In Block 2 the magnitude of the velocity of the fragments is assumed to
obey a normal distribution and their directions in space are assumed to be uniformly
distributed in azimuth and elevation .

In the first instance when there is lack of information, this would be natural and justified.
However there is the difficulty to choose the parameters of the normal distribution. In general
the fragment velocities would obviously depend on the location of the breakup in the orbit
Breakup altitude, mass and orbital
parameters of the parent body at
the time of BU
Orbital Parameters of the
individual fragments, namely
a, e, i, and B*
V, ,
or
V
n
, V
t
, V
r
Empirical relations
Connecting
Number, Mass, Size,
Effective Area and
Ballistic Coefficient

Block - 1
Block - 2
Block - 3
objects

60
specified by true anomaly , the intensity and also the directionality of the explosion. Later
the characteristics in Block 1 are derived from the information of Block 2.

The approach of Jehn (1996) utilized in MASTER 99 model (2000) commences from Block
3, namely the relations firstly connecting the number of debris fragments and their mass and
secondly the mass to the size. A host of relationships are available in the literature relating
number and mass of the fragments based on the intensity of explosion and collision. The
Appendix 2 provides a list of some of relationships. Jehn next derives the Block 2 properties
from Block 3 information based on the size s versus velocity V relationship. However
such a relation cannot be a deterministic one to one type. The velocity V derived from the
above relation is subsequently randomly perturbed based on a triangular distribution between
0.1V and 1.3V. This triangular velocity distribution has been proposed by Reynolds (1991).
Later utilizing the information from Block 2 one carries out the Block 1 simulation. Due to
the many intermediate empirical relations in the procedure of Jehn (1996) one can expect
uncertainties or inaccuracies in the evolution of the orbital parameters of the debris based on
in particular their ballistic coefficient. Such a procedure appears compelling if one does not
know the severity, directionality, and other useful details of the many breakups that have
occurred in the past.

The recent NASAs breakup model of Johnson (2000)

incorporated in MASTER 2001 differ
from MASTER 99 in having a bimodal normal distribution to derive area to mass from
diameter and a Lognormal distribution depending on A/m is applied to spread the
incremental velocities of the fragments. It was noticed that as a consequence the clouds
remain centered much around the parent bodys orbit.

It may also be noted that both the above approaches do not appear to utilize the available
information of the breakup such as the Gabbard diagram or the USSPACECOM data, the
observed dispersion of the orbital period and the inclination of the catalogued objects, thus
these are unable to account for the intensity of explosion.

61
The basic structure of the present model has been obtained based on an analysis of the
characteristics of the parameters of the fragments of the PSLV-TES Mission spent upper
stage breakup on 19 December 2001 as given by the TLE set of the USSPACECOM
catalogue (Anilkumar et al 2002). If one is able to utilize as is done presently the orbital
characteristics soon after the breakup the velocity distribution of the fragments can be
obtained more accurately and later derive their physical properties from empirical relations
which are anyway unavoidable. The details of the analysis of this breakup are provided in
the next section.

3.4 BREAKUP OF THE INDIAN PSLV-TES MISSION SPENT
UPPERSTAGE

The INDIAN PSLV-TES Mission was launched on 22 October 2001, from the then
Sriharikota Range, presently the Satish Dhawan Space Centre (SDSC) of the Indian Space
Research Organization. After the successful completion of the satellite mission, the spent
upper stage of PSLV was inserted in an orbit of nearly 675 km x 550 km at an inclination of
97.9 deg. This rocket body exploded on 19 December 2001 after which the fragments were
tracked and catalogued in USSPACECOM TLE sets. The details

of this breakup event are
(Orbital Debris Quarterly News 2002)

International Designation : 2001049D
USSPACECOM Number : 26960
Name : PSLV-TES R/B
Mass : ~ 885 Kg
Breakup epoch and Location : ~ 1140 UTC, 19 December 2001,~ 25
0
S, 340
0
E
Altitude : ~ 670 km
Pre-event Orbit of PS4 : ~ 550 km x 675 km, 97.9 deg (Incl.)
(a = 6990.5 Km, e = 0.0089)
Number of Large : > 300
Catalogued Debris
objects

62

The USSPACECOM catalogued TLE elements for the 301 fragments of this explosion are
available as on 20
th
January 2002. The TLE data provides the information on the best
possible data with the least amount of uncertainty of the quantities, namely, the mean motion,
e the eccentricity, i the inclination, B* the ballistic coefficient related parameter (equivalent
to (B (
0
/2)) and the argument of the perigee. The orbital parameters, say, a the semi major
axis, h
a
the apogee height, h
p
the perigee height and its period can be derived from the
above quantities.

Based on the TLE data it is possible to infer all the kinematic orbital characteristics of the
catalogued fragments. However the other desirable fragmentation properties such as the mass
and the characteristic area cannot be obtained from the TLE data. The orbital decay of the
debris fragments depends strongly on the ballistic coefficient. The damage from a collision
depends on the mass m of the fragments. The observable radar cross section of a fragment
depends on its characteristic area A
eff
or the diameter d. These are strongly related to the
actual physical mechanism and the intensity of the explosion or collision. The detailed BU
being unknown and random in order to estimate the above, it is compelling to use one of the
available empirical relations in the literature. Also the effect of tumbling can change the
value of the drag coefficient C
D
and A
eff
in the ballistic coefficient as also the gas surface
interaction.

The Figure 3.2 provides the Gabbard diagram of the 301 fragments after nearly 30 days of
breakup. The slight coalescence in the left arm of the Gabbard diagram is due to the decay of
the orbits of the fragments during the 30 days. Utilizing suitable probability or cumulative
distribution functions, the Figure 3.3 shows the Laplace distribution fit (McLaughlin 1999)
for apogee height, and lognormal fits for eccentricity and the ballistic parameter B values and
Laplace fit for the inclination of the orbital fragments. This figure clearly brings out the
validity of the basic structure of the above quantities proposed in evolving the
ASSEMBLE model. The Appendix 4 provides briefly the description of the Laplace and
the lognormal distributions.

63

3.4.1 DELTA VELOCITY IMPARTED TO THE FRAGMENTS

The delta velocities imparted to the fragments at the time of explosion can be obtained from
the relations available in literature (Culp and McKnight 1985), which is provided in
Appendix 5. The delta velocities estimated for each of these fragments range from 0.2 m/s to
400 m/s with an average of 52.5 m/s.

3.4.2 COMPONENTS OF THE DELTA VELOCITIES

The calculated components of the delta velocities, namely the radial Vr, the transversal Vt,
and the normal Vn, components imparted to the fragments both in plane and out of planes are
useful to obtain further knowledge about the process of explosion. The relations for the
above component velocities are derived assuming impulsive fragmentation using the
equations developed from Lagranges planetary equations (Jehn 1996, Meirovitch 1970,
Johnson and McKnight 1987) as given in Appendix 6.

3.4.3 DISTRIBUTION OF THE MASS OF THE FRAGMENTS

Since there is no data or even otherwise it is very difficult to infer the mass of the debris
fragments, it is compelling to depend on some empirical relations to arrive at the distribution
of the mass of the fragments. The literature provides a large number of relations, which can
be utilized for explosions (high intensity or low intensity) and collisions based on the
explosion or collision experiments as listed in Appendix 2. Obviously the number and the
mass distribution depends on the density and shape of the parent mass, the directionality, and
the intensity of explosion or collision. Further the laboratory tests can provide only sample
measurement characteristics among many possible breakups. Hence there is always a certain
amount of uncertainty both in the structure and the parameters of the fitted empirical
relations. In the present study we have utilized the first four relations listed in Appendix 2
objects

64
with suitably tuned parameters. Here we provide the results utilizing the rescaled relations of
Su and Kessler quoted by Reynolds (1990).

<
>
=
m m
m m
f g m mf
f g m mf
N
/ 1936 for ), 05756 . 0 exp( 869
/ 1936 for ), 02056 . 0 exp( 171

where
m = the mass of the fragments in grams,
M
t
= the mass of the breakup object in grams,
N = cumulative number of fragments with mass m or greater,
and f
m
= the ratio of the reference mass of 1000 kg to the satellite mass M
t
.

This above expression estimates the number of fragments created by the present explosion as
868 with the mass ranging from nearly 56 kg to 2 grams, which accounts for a mass of about
840 kg. This means that for the largest 301 objects the cumulative sum of mass is 840 kg.
These mass values have been scaled up to conserve the total initial object mass of 885 kg.
This feature might appear arbitrary but considering the uncertainty in the above relations, this
should be acceptable.

3.4.4 DETERMINATION OF THE AREA TO MASS RATIO

For further analysis and extrapolation of the data to the untrackable objects, a relation
connecting the area to the mass has to be chosen. Many different formulae are available in
the literature (Jehn 1996, McKnight 1991). A simple dimensional analysis would indicate the
mass versus area relationship as
m ~ A
eff
1.5

and among different materials with varying densities the constant of proportionality and the
index would vary. Jehn (1996) provides the relation

65

where the coefficient and the exponent are related to the density of the breakup body
material.

In the present study we have utilized the values of m / A
eff
obtained from the B* (C
D
is
assumed to be 2.2) and estimated the coefficient and the exponent in the relation

m / A
eff
= A
eff

in a least square sense. Hence the modified relation is found to be of the form

22 . 1
62 . 46
eff
A m = .

3.4.5 MASS OF THE DEBRIS FRAGMENTS AND THEIR IMPARTED DELTA
VELOCITIES

Associating mass to the debris objects is a difficult task. It is generally accepted that the mass
of the object will have some kind of inverse relation to the size of the object, assuming
uniform density of the material of the parent body. The debris sizes are related to the
imparted delta velocities. A detailed literature survey brings out different forms involving
diameter and delta velocities. It is reported in the literature that the delta velocities imparted
to the fragment follow a triangular distribution (Jehn 1996, Reynolds 1991). In this report
the delta velocities are associated with the size of the debris (and hence to the mass of the
debris directly) utilizing the Reynoldss relation (1991)

514 . 1 ) (log 804 . 0 ) (log 0676 . 0 ) log(
2
= d d V
42 . 1
67 . 52
eff
A m =
objects

66
where d is the size of the fragment in meters and V, the delta velocity in km/sec, which
describes the peak in the velocity distribution curve. In order to provide dispersion in V,
the velocity is picked randomly from a triangular distribution whose minimum is 0.1 times
and maximum is 1.3 times respectively of the peak V.

3.5 THE APPROACH IN THE PRESENT ASSEMBLE MODEL

One can expect the direction in the inertial space, and the intensity of explosion to control the
distribution of the orbital characteristics of the fragments. Hence we have utilized the best
possible orbital parameter data as available from the USSPACECOM catalogue as the
starting point, as is provided in the Gabbard diagram in Figure 3.2. It turns out very
interestingly that based on the physical mechanisms in the breakup process both the apogee,
perigee heights (limited by the breakup altitude) closely fit suitable Laplace distributions and
the eccentricity follows a Lognormal distribution. The location parameters of these depend
on the orbit of the parent body and their scale parameters on the intensity of explosion
(Anilkumar et al 2002). The distribution of the ballistic coefficient in the catalogue was also
found to follow a Lognormal distribution. Thus all the necessary information in Block 1 is
available and one can proceed from here to derive the details of Block 2 and Block 3.

In dealing with the inverse problem of simulating typical breakups using the ASSEMBLE
model one commences with the information on orbital parameters from Block 1 and proceed
to Block 2 to derive the incremental velocities of the various fragments. Then finally Block 3
is handled independently and the mass and size are assigned to the simulated fragments based
on the magnitude of V and later changing it randomly by utilizing the triangular distribution
nature of V. Hence the proposed procedure utilizes probabilistic distributions for orbital
characteristics of the fragments and empirical relations for the physical characteristics and
relates them through the magnitude of V.


67
Though all the characteristics are in general random the word semi stochastic is used since
some properties of like orbital characteristics follow one to one deterministically from the
random delta velocities, but the characteristics such as mass and size are randomly related to
their velocities. This model is named as ASSEMBLE (A Semi Stochastic Environment
Modeling for Breakup in LEO).

For the purpose of simulating many earlier breakups generally one has the information of the
Gabbard diagram together with the number of fragments, the maximum P and i (Anz
Meador 2001), and other details about the parent body at the time or prior to the breakup
event. However the above does not provide all the quantitative orbital and physical
characteristics of all the fragments that have been generated. It is demonstrated that
ASSEMBLE is able to simulate the debris clouds in a statistical way by generating the
fragmentation events by a suitable choice of the parameters involved in the procedure. Also it
is shown further the characterization of the distribution of the ballistic coefficient of the
fragments helps to provide an improved description of the evolution of the debris with time.
The present approach being based on the available information on breakup, the observations
on Lognormal nature of incremental velocities and clustering of fragments around the parent
body follow as consequences.

3.6 THE PROPOSED ASSEMBLE MODEL

Figure 3.4 provides a detailed flow chart of the ASSEMBLE model. This model
commences from the distribution of the orbital elements, namely perigee height, eccentricity
and inclination, of the fragments. Then from the above elements it derives the incremental
velocities and subsequently the mass, size and other characteristics are generated utilizing the
empirical relations (Jehn 1996, Su and Kessler 1985) for low intensity or high intensity
explosions in accordance with the fragmentation event to be simulated. If the velocities of the
fragments are assumed to be normal, or derived otherwise from the size distribution it turned
out that in particular the eccentricity and the ballistic coefficients do not follow the
Lognormal distribution and hence inconsistent with the USSPACECOM measured data.
objects

68

In order to simulate a typical breakup using the ASSEMBLE model the information about
the breakup altitude or breakup true anomaly, apogee and perigee height, inclination at the
time of breakup and also the mass of the parent body are necessary. Apart from these one
needs the maximum delta velocity available for explosion or range of apogee or perigee
height and maximum P and maximum I. For simulating the historical breakups, the
maximum P, the maximum i and the ranges of apogee and perigee heights for the
catalogued objects are available in Anz Meador (2001).

The last mentioned ones are very crucial and provide the dispersions on the apogee or
perigee height, eccentricity and inclination. It is well known that the apogee and the
eccentricity are related to radial and transversal components of the velocity and the
inclination to the normal component with respect to the orbital plane. Usually along with the
Gabbard diagram one has the information about P, and i, the former can also be inferred
from the maximum change in the apogee and perigee heights. However there is the necessity
of some subjective tuning of the scale parameters of the distributions based on the qualitative
information in the Gabbard diagram. Generally it is assumed that the observed spread in the
Gabbard diagram corresponds to 3 standard deviation of the random variable.

3.7 STEPS FOR SIMULATING A TYPICAL BREAKUP

Having provided the conceptual basis for the ASSEMBLE model, this section deals with
the inverse problem of simulating a typical breakup. The following are the various steps in
simulating the breakup fragments and their other characteristics.

1. The location parameters for apogee or perigee and inclination distributions,
both of which follow generally a Laplace distribution, and the eccentricity
which obeys a Lognormal distribution can be obtained from the breakup point
details of the parent body as the perigee height, eccentricity and inclination at
the time of breakup. Next based on an assumed intensity of explosion (namely

69
the additional velocity and its directionality or Maximum P and i provided
for explosion), one can obtain shape parameters of the corresponding
distributions. The number of fragments generated namely N, and the mass of
individual fragments, can be obtained by using the Number versus Mass
relationship, such as Su Kessler (1985) or any other suitable relation for an
explosion.

2. Generate N random sets of the apogee or perigee height, inclination and
eccentricity combinations from the respective Laplace and Lognormal
distributions (McLaughlin 1999) by restricting the apogee height or perigee
height bounded by the breakup altitude.

3. Using the above sampled values of the apogee or perigee height and
eccentricity, calculate the true anomalies of each of the fragments at the
breakup point.

4. Obtain the velocity imparted to the each of the fragments and assign these
delta velocities to the masses based on triangular distribution constraint (Jehn
1996) or any other method that is found suitable. The relationships utilized to
calculate delta velocities are given in Appendix 5.

5. Depending on the intensity of explosion use a suitable empirical relation
connecting the number, mass and size to obtain the ballistic coefficients of the
fragments.

It is to be pointed out that these orbital parameters in the above Step 2 could in general
follow a joint distribution with appropriate marginal distributions. Hence while simulating
these, the cross coupling between the parameters can be taken into account. But the multiple
correlations between these parameters are mostly unknown. However, the assumption of zero
correlation if assumed, and each of the parameters are independently simulated from the
objects

70
marginal distributions and randomly coupled, does not significantly alter the fit
characteristics.

3.8 SIMULATION OF SOME HISTORICAL BREAKUPS USING THE
PROPOSED ASSEMBLE MODEL

In this section we describe the results based on the simulation of some of the well known
historical on orbit fragmentation events. This helps to further validate and strengthen the
present ASSEMBLE model. The fragmentation details have been obtained from the
History of On Orbit Fragmentation by Anz Meador (2001). Table 3.1 provides the important
details of the analyzed breakup events and Table 3.2 provides the details of the parameters
utilized in the ASSEMBLE model with their location and scale parameter values for
simulating and analyzing the four fragmentation events.

3.8.1 CASE STUDY OF PSLV-TES SPENT UPPER STAGE

Figure 3.5 provides the Gabbard diagram of the debris clouds at the instant of breakup
simulated by ASSEMBLE model. Figure 3.6 provides the Gabbard diagram of the 300
fragments obtained after propagating for 31 days after the breakup. These figures can be
compared with the one given in Figure 3.2. The differences in the left arm of the apogee
height between Figures 3.2 and Figure 3.5 with the latter first at the BU can be attributed to
the decay effect of 31 days in its orbits. However the Figures 3.2 and 3.6 both are after 31
days decay compares quite well. In order to further demonstrate the goodness of fit, the delta
velocity imparted by the observed fragments and the simulated fragments are estimated and
provided in Figures 3.7 and 3.8 respectively. It may be noted that the figures compare well.
The goodness of fit test based on the
2
distribution provides a probability of goodness nearly
0.85. Based on an estimation of the orbital decay and the lifetime the above goodness of fit
at 85 % confidence level appears to be good and acceptable.


71
3.8.2 CASE STUDY OF STEP II R/B

The Figure 3.9 provides the Gabbard diagram of the debris clouds at the instant of breakup of
Step II R/B simulated by ASSEMBLE model. Figure 3.10 provides the Gabbard diagram
of the 713 fragments as of August 29, which is nearly 90 days after the breakup (Anz Meador
2001, Johnson 1996). The differences in the left arm of the apogee height between Figures
3.9 and 3.10 can be attributed to the decay effect of 90 days in circularizing the debris orbits
and apart from this, in a statistical sense, the match is quite good. In order to demonstrate this
fact, the simulated fragments were propagated for 90 days and the Gabbard diagram is shown
in Figure 3.11. Now it can be seen that Figure 3.10 and Figure 3.11 are matching well, the
collapse of the left arm of the apogee being due to the effect of atmospheric drag. The
propagation routine used is the one given by KingHele (1987) which considers only drag
effect of the atmosphere and the ballistic coefficients corresponding to the fragments are the
ones generated by ASSEMBLE utilizing the empirical relations connecting number-mass-
size of the fragments (Jehn 1996, Anilkumar et al 2002). The orbital decay equations due to
drag effect is provided in Appendix 3.

Figures 3.12 to 3.16 provide extensively the various other characteristics of the fragments as
simulated by the ASSEMBLE Model. It is noted that the features shown in these figures
are matching well with those provided by the analysis of Anz Meador (2000) in Orbital
Debris Quarterly News about the STEP II rocket body (1994-029B) breakup. As expected it
can be seen that the derived quantities like V and the ballistic coefficient B* show their
nature in following a Lognormal distribution. Thus ASSEMBLE model provides all the
required information regarding the debris clouds.

3.8.3 CASE STUDY OF CBERS-I/ SACI-1 R/B

Figure 3.17 provides the Gabbard diagram of the above breakup as reconstructed from
USSSN database (Anz Meador 2001) just after a day of the breakup. This was the second
Long March 4 to breakup in four missions. The first breakup occurred on 4
th
October 1990
objects

72
one month after the launch (Zang and Liao 1993). The Gabbard diagram of the fragments
simulated by using the present ASSEMBLE model just at the time of breakup is provided
in Figure 3.18. The acceptably good statistical match between the simulation and
measurements can be seen clearly from these figures.

3.8.4 CASE STUDY OF COSMOS 1813

Figure 3.19 provide the Gabbard diagram of Cosmos 1813 debris cloud for 846 fragments
(Kling and Dowdy 1987, Johnson 1996) as reconstructed from PARCS radar observation
taken about 10 hours after the breakup as extracted from Anz Meador (2001). Figure 3.20
shows the simulated Gabbard diagram after 10 hours of evolution of the debris clouds using
ASSEMBLE model. As the breakup occurred at a very low altitude, the evolution of debris
clouds for 10 hours brings a change in the left arms of Gabbard diagram. Figure 3.21
provides a Gabbard diagram of the fragments 10 hours after the event as simulated using the
model proposed by Jehn (1996) and is extracted from it. The comparison between the
Figures 3.19, 3.20 and 3.21 clearly brings out the ability of the ASSEMBLE model better
than following Jehns (1996) approach in simulating the debris cloud.

3.8.5 CASE STUDY OF SPOT 1 R/B

Figure 3.22 provides the Gabbard diagram of the debris cloud of SPOT I R/B after three
months of evolution as reconstructed from US SSN database (Anz Meador 2001, Loftus
1989, Johnson 1987) and Figure 3.23 provides the Gabbard diagram corresponding to the
fragments after three months of propagation simulated by the ASSEMBLE Model. It is
clear that ASSEMBLE simulates the debris clouds to obtain a good match with the
observations.


73
3.9 COMPARISON WITH THE EXISTING MODELS

The results of Ganeshan and Ananthasayanam (1997) do not provide obtained by following
the approach of Johnson and McKnight (1987) the distribution characteristics of apogee
height, perigee height and eccentricity. It is seen that the Laplace distribution property of the
apogee height or perigee height and the lognormality of the eccentricity are not properly
simulated in these approaches of assuming normal distribution for the incremental velocity of
the fragments. The Figure 3.24 provides a case study of this approach. The DeltaV is
assumed to follow normal distribution and the fragments properties are simulated. The
second graph in this figure clearly shows that the simulated eccentricities do not follow the
lognormal distribution.

As already discussed the approach laid down by Jehn (1996) and utilized in the MASTER 99
model (2000), begins with the breakup model simulation from the number - size - mass
relations and generates the incremental velocity from the empirical relation connecting the
size and the incremental velocity V together with the subsequent modification based on a
triangular distribution. This model proceeds to obtain the changes in the orbital elements
from the relations connecting the components of V values and the changes in the state
vectors and in turn the changes in the semi major axis, eccentricity and inclination. Jehn
(1996) demonstrated that his approach provide some match with the measurements of the
breakup. As shown by Figures 3.20 and 3.21 for the COSMOS debris cloud, comparing the
Jehns approach and the present approach, it can be seen that ASSEMBLE provides a
better match for the breakup simulation.

3.10 CONCLUSIONS

The present ASSEMBLE model has been proposed based on an analysis of the properties
of the catalogued fragments generated from the breakup of the INDIAN PSLV-TES mission
spent upper stage. Subsequently it has been applied as an inverse problem to simulate some
objects

74
typical historical on orbit fragmentation events. All the simulated results with some known
major breakups are comparable quite well with the observations both at the time of breakup
and at a later epoch. Such an improved feature has been possible since ASSEMBLE
utilizes the observational orbital parameter data rather than assuming or obtaining
incremental velocities of the fragments based on intuitive empirical relations or distributions.
It is thus demonstrated that ASSEMBLE simulates the ensemble of fragments quite well
including the distribution of the ballistic coefficients which help in obtaining an improved
evolution of the debris cloud.

75

Table 3.1 Details of the Analyzed Breakup Events

TITLE

STEP II R/B

CBERS-1/SACI-1
R/B

SPOT 1 R/B

COSMOS
1813

INTERNATIONAL
DESIGNATION
USSPACECOM
CATALOGUE NUMBER

1994-029B

23106

1999-57C

25942

2001-049D

26960

1987-004A

17297
SATELLITE DATA
TYPE
OWNER
LAUNCH DATE
DRY MASS (KG)

Pegasus HAPS
USA
19 May 1994
97

Long March4 3
rd
stage
PRC
14 Oct 1999
1000

Ariane 1 3
rd
stage
ESA
22 Feb. 1986
1400

Payload
CIS
15 Jan.1987
6300
EVENT DATA
DATE
LOCATION
TIME
ALTITUDE (km)

3 Jun 1996
67 S, 56 E (asc)
1518 GMT
625

11 March 2000
51.2 S, 48.5 W
1304 UTC
741

13 Nov 1986
7 N, 42 E
1940 GMT
805

29 Jan 1987
73 N, 122 E
0555 GMT
390
PRE-EVENT ELEMENTS
MEAN ANOMALY (deg)
RIGHT ASCENSION (deg)
MEAN MOTION
INCLINATION (deg)
ECCENTRICITY
ARG. OF PERIGEE (deg)

108.3711
197.8565
14.56780581
81.9749
0.0165742
249.9583

43.0989
145.5131
14.46866365
98.5373
0.0012467
316.9224

300.1947
18.0087
14.2216
98.6973
0.0021203
60.1312

178.1696
256.7724
15.60427
72.8163
0.0043147
182.01
DEBRIS CLOUD DATA
MAXIMUM P (min)
MAXIMUM I (deg)
NO. OF CATALOGUED
FRAGMENTS

27.9
2.4
~ 713

10.985
0.99
~ 300

6.2
1.2
~ 488

9.1
0.1
~ 846
objects

76
Table 3.2 Location and Scale Parameters of the Distributions Used in ASSEMBLE
Model for Various Breakups Based on Data and Gabbard Diagram
LPD - Laplace distribution; LND - Lognormal distribution

Fragmentation event Orbital
parameters at
BU
Parameters
of the
distribution
SPOT 1
R/B
COSMOS
1813
STEP II
R/B
CBERS-1/
SACI-1
R/B
Location 805 360 500 726 Perigee height ~
LPD (Km) Scale 100 100 85 100
Location -6.2 -5.5 -4.1 -6.6 Eccentricity ~
LND Scale 1.0 1.1 0.67 1.35
Location 98.7 72.82 82 98.5 Inclination ~
LPD
(Deg.)
Scale 0.4 0.03 0.6 0.55

Figure 3.2 The Gabbard Diagram from the TLEs of PSLV-TES Spent Upper Stage
Breakup Fragments after 31 Days of Breakup

77

Figure 3.3 The Distribution of Apogee, Inclination, Eccentricity and B* from the TLEs
of PSLV-TES Spent Upper Stage Breakup Fragments after 31 Days of Breakup
objects

78

Figure 3.4 Flow Chart of the ASSEMBLE Model.
INPUTS
1. BU Point Parameters : a, e, i,
2. An estimate of intensity of explosion
(or range of periods, range of apogee /
perigee heights of the fragments
3. The number of fragments
Simulate individual fragments a, e, i using the probability distributions
Location parameters - Same as BU point values
Scale parameters derived from intensity of explosion or range of periods and
range of apogee height/ perigee height
apogee/perigee ~ Laplace (Location, Scale)
e ~ Lognormal (Location, Scale)
i ~ Laplace (Location, Scale)

Calculate the
individual V
for each of the
simulated
fragments
Generation of the Mass
of the individual
fragments using number
versus mass relations
Association of (a, e,
i) to mass of the
fragments using
triangular
distribution relation
between V and
size of the
fragments
Calculate of A
eff
using
Mass Size
relationship and hence
obtain the Ballistic
coefficient of the
fragments
DEBRIS
BREAKUP
SCENARIO

Orbital
parameters,
mass, size and
the ballistic
coefficient of
the fragments

79
94 96 98 100 102 104 106
200
400
600
800
1000
1200
1400
Period (min)
A
p
o
g
e
e
/
P
e
r
i
g
e
e

A
l
t
i
t
u
d
e

(
k
m
)
Apogee
Perigee

Figure 3.5 PSLV-TES Spent Upper Stage Debris Cloud at the Time of Breakup
Simulated Using ASSEMBLE Model

Figure 3.6 PSLV-TES Spent Upper Stage Debris Cloud After 31 Days of Breakup
Simulated Using ASSEMBLE Model.
objects

80

Figure 3.7 DeltaV values of PSLV-TES Spent Upper Stage Debris Fragments from the
Observations.

Figure 3.8 DeltaV values of PSLV-TES Spent Upper Stage Debris Fragments
Simulated by ASSEMBLE Model.

81

Figure 3.9 Step II R/B Debris Cloud of 713 Fragments at the Time of Breakup
Figure 3.10 Step II R/B Debris Cloud of 713 Fragments as of August 29, 1996 (89 Days
after the Breakup) as Reconstructed from USSN Database
(Anz Meador, 2001 Johnson 1996)
objects

82

Figure 3.11 Step II R/B Debris Cloud of 713 Fragments after 89 Days of Breakup

Figure 3.12 The DeltaV, Semi Major Axis, Inclination and Eccentricity Distribution of
the Fragments of Step II R/B Simulated Breakup Using ASSEMBLE Model

83

Figure 3.13 Component Velocities DeltaVr and DeltaVt of the Fragments of Step II R/B
Simulated Breakup Using ASSEMBLE Model

Figure 3.14 Component Velocities DeltaVt and DeltaVn of the Fragments of Step II R/B
objects

84

Figure 3.15 Component Velocities DeltaVn and DeltaVr of the Fragments of Step II R/B

Figure 3.16 Area to Mass Ratio and the Component Velocities of the Fragments of
Step II R/B Simulated Breakup Using ASSEMBLE Model

85
Figure 3.17 CBERS-1 /SACI-1 Long March 4 Third Stage Debris Cloud of 280 with one
Day of the Event as Reconstructed from US SN Database (From Anz Meador 2001)

Figure 3.18 CBERS-1 /SACI-1 Long March 4 Third Stage Debris Cloud of 280 at the
Time of Breakup Simulated Using ASSEMBLE Model
objects

86
Figure 3.19 Cosmos 1813 Debris Cloud 846 Fragments as Reconstructed from PARCS
Radar Observation Taken about 10 Hours after the Breakup
(Anz Meador 2001, Jehn 1996)

Figure 3.20 Cosmos 1813 Debris Cloud after 10 Hours of Breakup Simulated Using
ASSEMBLE Model

87
Figure 3.21 Cosmos 1813 Debris Cloud after 10 Hours of Breakup Simulated Using the
Model Proposed by Jehn 1996
Figure 3.22 Spot I R/B Debris Cloud of 463 Fragments Three Months after the Event as
Reconstructed from US SN Database (Anz Meador 2001, Johnson 1987)
objects

88

Figure 3.23 Spot I R/B Debris Cloud of 463 Fragments Three Months after the Event as
Simulated by ASSEMBLE Model

100 120 140 160 180 200 220 240 260 280 300
0
20
40
60
80
DelatV of fragments (m/s)
N
u
m
b
e
r

o
f

O
b
j
e
c
t
s
-11 -10 -9 -8 -7 -6 -5 -4 -3 -2
0.001
0.003
0.01
0.02
0.05
0.10
0.25
0.50
0.75
0.90
0.95
0.98
0.99
0.997
0.999
log(Eccentricty)
P
r
o
b
a
b
i
l
i
t
y

Figure 3.24 DeltaV and Eccentricity Distributions based on the Approach of
Johnson and McKnight (1987)

Chapter 4 A new stochastic impressionistic low earth (SIMPLE) engineering model of the space debris scenario
89

CHAPTER 4

A NEW STOCHASTIC
IMPRESSIONISTIC LOW
EARTH (SIMPLE) ENGINEERING MODEL OF
THE SPACE DEBRIS SCENARIO

4.1 INTRODUCTION

The purpose of any model in science and technology is to condense the available information
in the simplest possible way to characterize some important features or properties and be
useful for further applications to predict or control as is necessary. In the context of Space
Debris analysis a variety of tasks or applications (Ananthasayanam 2001) exist for the model.
These are

(i) Definition of the environment,
(ii) Risk analysis and impact location,
(iii) Collision avoidance schemes,
(iv) Selection of orbit for mission design, and
(v) Prediction of future debris scenario.

There has been considerable interest for many years now in the formulation of suitable
models of the space debris environment for use in the design and operation of spacecraft. The
approach and the philosophy that has been adopted for the evolving the present model is
similar to that utilized for evolving the International Standard or other various Reference
Atmosphere as spelt out by Jacchia and Champion in CIRA (1972) as well as Narasimha and
objects

90
Ananthasayanam (1980, 1981). It is fruitful to quote the philosophy with regard to the
International Standard Atmosphere (ISA) as stated by Toussaint in 1919 (from Gregg 1922).

in order to define the standard atmosphere what is needed is not exact representation
of that curve (nonlinear variation of temperature with altitude) but merely a law that
can be conveniently applied and which is sufficiently in accordance with the means
adhered to.

The World Meteorological Organization (WMO) has defined the standard atmosphere as

a hypothetical vertical distribution of atmospheric temperature, pressure and density
which, by international agreement, is roughly representative of year-round mid-
latitude conditions.

It is useful to recapitulate the way the standard and reference atmosphere models were and
are being constructed based on the statistics of the distributions of the atmospheric
temperature, pressure and density (Narasimha and Ananthasayanam 1980). In the present
case also one can presume the space debris environment manifests through its statistical
properties.

The development of the standard or reference atmosphere have utilized the statistics of the
multidimensional joint probability distributions followed by temperature, pressure and other
properties. For the present modeling of the orbital debris scenario the appropriate marginal
distributions derived from the joint distributions have been used to characterize the orbital
elements of the fragments namely apogee and eccentricity and their other properties such as
the ballistic coefficient.

Further the present model the emphasis is on a hypothetical nature of the distributions of
orbital debris environment for low earth orbit, in terms of altitude, perigee and eccentricity.
Firstly a global model, with respect to the above parameters, which is a roughly a
91

representative of the total characteristics over all the inclinations is given. This is followed by
similar models for each of the five inclination conditions, namely 28
o
, 51
o
, 65
o
, 82
o
and 98
o
as in ORDEM 96 (Kessler et al 1997), representing the five inclination bands (0,36], (36,61],
(61, 73], (73, 91] and (91, 180].

At present we call this as an impressionistic model since we have not accounted for the effect
of solar flux and growth rate in describing the distributions. To assimilate the newer data and
adaptively change the model parameters, the present approach has a suitable structure that
can provide such flexibility. Later under such a situation the present impressionistic model
would become stochastic improved model of the low earth orbiting debris!

4.2 GENESIS OF SPACE DEBRIS ENVIRONMENT MODEL

Modeling of the space environment for many characteristics such as the neutral atmosphere,
constituent species, ionosphere, magnetosphere, meteoroids, which are natural and existing
from time immemorial and presently the space debris, which is man made and of very recent
origin as from around 1960, is a highly useful task. The modeling of the space debris
environment helps in proper design and operation of a spacecraft such as estimating the
collision risk, and launch window determination and for many other purposes.

The present work deals with a new approach of modeling the low earth near circular orbital
debris environment. This approach is based on the following observations about the nature
of the orbital debris and philosophies to construct the model. These are

1. The space debris environment is a randomly evolving dynamical process.
2. The multidimensional probability distribution function and the statistics of the
various characteristics based on the observed data at various times are used to
qualitatively and quantitatively model the debris environment.
3. The description should be close to the mean over appropriate space and time
and within the ranges of the variations of the data.
objects

92
4. The model should have a balance between accuracy and simplicity.
5. The model should posses easy mathematical tractability for further
simulations, analysis of the environment and the estimation of collision
probabilities.
6. To possess the smallest possible number of parameters in describing the
environmental scenario.
7. To be able to assimilate newer data and adaptively change the model
parameters with the passage of time.

4.3 DIFFERENCES BETWEEN THE NATURE AND
ENVIRONMENT OF SPACE DEBRIS AND GAS MOLECULES

It is useful to note the differences between the gas molecules in the atmosphere and the space
debris in order to appreciate the difficulties in evolving a suitable model for the debris
scenario. The following Table 4.1 lists the major differences between the gas molecules and
the fragments of space debris. Mainly it may be noted that there are wide differences in the
shape and size among the debris, which is not so in the constituents of the gas molecules,
which are few and known from experiments. In contrast the debris have random variations
among themselves and there are no well known theories or experiments based on which the
shape and size distributions can be determined.

Further even the environment in which the debris evolves is random and not easily specified.
These are just some of the reasons that make it compulsive and useful to attempt to
characterize the debris scenario in terms of a suitable multidimensional joint probability
distribution function as discussed in the subsequent section.
93

Table 4.1 Differences Between Space Debris and Gas Molecules

No.

PARAMETERS

GAS MOLECULES

SPACE DEBRIS
1 Flow regimes

Free molecular to
Continuum

Free molecular
2 Particles

Equal size, shape and
mass

Unequal size, shape and
mass
3 Evolution

Binary collision among
themselves and body

Mainly Convection and
at times collision

4

Environment

Similar molecules
and body
Uncertain Atmosphere,
Earth's gravitational
Harmonics , Solar Radiation
Pressure, Luni-Solar
Perturbations,
5 Distribution
function
f (r, v, t)

f (r, v, m, S, t)
6 Source terms Generally not present

Influx is Present
7 Periodic
measurements

Not essential (theory is
adequate)

Essential
8 End objective

Forces acting on body,
momentum and heat
transfer.

Risk analysis, orbit
selection among many other
uses
9 Present Status of
Subject
Well developed Theory
With experimental
Supplement

Not fully developed.
Difficult for theory and
experiment

Note: r = location in space; v = velocity;
m = mass; S = cross sectional area
objects

94

4.4 PRESENT PROBABILISTIC APPROACH FOR MODELING THE
DEBRIS ENVIRONMENT

It is well known that the trackable objects of size greater than 10 cm are in thousands and the
untrackable object runs into millions. Since the space debris scenario contains a very large
number of objects of varying size, mass, effective area and orbital parameters (semi major
axis, eccentricity, inclination and so on), there is generally an advantage in such models to be
characterized by the probability density functions. For a single orbiting fragment around the
earth the orbital parameters are

a the semimajor axis,
e the eccentricity,
i the inclination,
the longitude of the ascending node,
the argument of perigee and
the true anomaly.

In general due to perturbations, such as gravitational anomaly, atmospheric drag, luni solar
perturbations, solar radiation pressure all the above parameters change with time. For LEO
the atmospheric drag is the major perturbation factor and determines the lifetime of debris.

Among the orbital debris the ones that are tracked and cataloged provide the best possible
information. Their evolution depends on their ballistic coefficient B. A careful perusal of the
values, where B* = (
0
/2) B, where 0 is the density at a reference altitude 120 Km above
earth, in the two line elements of the USSPACECOM catalogue shows that this is also
varying somewhat with time. Many objects have a large tumbling effect due to the moments
of gravitational, aerodynamic origin acting about their various axes. This changes the
orientation with respect to the free stream velocity leading to variation in the drag coefficient.
95

Usually a standard value of 2.2 is assumed for the C
D
but even this can change due to varying
nature of the gas-surface interaction. Further the atmospheric density varies with solar flux,
magnetic index, time of the day and so on. Even the most sophisticated atmospheric models
represent only the best possible estimated conditions and there are always unmodellable
random variations occurring around the mean. Thus it is an onerous task to model the
environmental density, gravitational anomaly, the aerodynamic coefficients of the body and
many more in order to track each and every one of the debris for a long period of time. The
various orbital elements also continuously evolve randomly with time. Of course this may be
carried out for cataloging purposes with corrections at times based on the measurements.
However this cannot be the task for everyone. Most of the time one desires to have a broad
and reasonably accurate representation of the scenario.

Because of such above uncertain, variable, complex and enormous situation that is prevailing
even for catalogued debris it is felt best to model the scenario in a probabilistic way which
when allowed to vary with time becomes stochastic. Certainly the smaller sized debris less
than 10 cm which are generally not trackable could also be later handled by following the
present approach.

4.5 DATABASE UTILISED FOR THE SIMPLE ENGINEERING
MODEL

The space debris fragments are created at various random times by collisions or explosions
form a large number of objects of varying size, mass, effective area and orbital parameters
namely semi major axis, eccentricity, inclination and so on. As these move in space under the
influence of random environmental forces it is advantageous to characterize the scenario in
terms of suitable probability density functions.

The present approach is based on the statistical distributions of the orbital characteristics of
the environment and is limited to large objects, which are catalogued in USSPACECOM
Two Line Element (TLE) sets. Here we have considered the time from 1999 to 2002 for the
objects

96
present analysis. In line with the general concept of LEO and near circular orbits, here one
deals with the catalogued objects with altitude less than 2000 km and eccentricity < 0.2. It
means an orbit with altitude less than 2000 km and eccentricity being 0 to 0.2, is consistent
with usual definition of near circular orbits. Such objects constitute about 75% of the
catalogued objects. In fact this region is most used by the satellite missions. The number of
objects in this region across the years 1999 to 2002 are respectively 6123, 6014, 6181 and
6102 based on 1
st
January data of each year. Since only about 4 years of data was utilized in
the present analysis it is neither feasible nor proper to bring in the effect of the solar activity
index F
10.7
or the magnetic index A
p
into the present model.

The USSPACECOM TLE data provides the best possible data with the least amount of
uncertainty of the quantities. The parameters are, namely,

n the mean motion,
e the eccentricity,
i the inclination,
B* the ballistic parameter,
the mean anomaly,
the right ascension of ascending node and
the argument of perigee.

The B* is equivalent to (
0
/2).C
D
.A
eff
/m, with
0
being the density of the atmosphere at a
specific altitude say 120 Km above earth, and C
D
is the non dimensional drag coefficient
which can be taken as 2.2 for all practical purposes, A
eff
is the effective area of the object and
m is the mass of the object. The orbital parameters, say, a the semi major axis, h
a
the
apogee height, h
p
the perigee height and its period can be derived from the above
quantities. A detailed description of the TLEs are provided in Appendix 1 as already
mentioned in Chapter 1.

97

4.6 CHOICE OF SUITABLE PROBABILITY DISTRIBUTIONS IN
THE PRESENT SIMPLE MODEL

As mentioned earlier the joint probability distribution of the near circular space debris
scenario in LEO depending on various parameters can be specified in general as

f(h or h
p
, e, i, B, t)

where
height = h < 2000 km;
perigee height = h
p
,
eccentricity = e < 0.2,
inclination = i taken in five bands
B = Ballistic parameter related to B* in TLE set and
t = time.

It may be noted that in particular the distributions in terms of perigee height and altitude have
been provided separately so that the future simulations will have better accuracy in defining
the low earth near circular orbital debris scenario. At this point it may be noted that ORDEM
provides the distribution with respect to altitude only in the model (Kessler et al 1997, Liou
et al 2001).

One may notice that there is debris accumulation peak at some particular altitudes, due to the
increased launch activity in some altitude regions of space in LEO and also due to the
breakup of objects, which creates debris clouds around the breakup point. It has been
reported that most of the on orbit breakups reported that so far (Anz Meador 1997 and
Orbital Quarterly News, January 2002 and January 2004), more than 180 breakups have
occurred in this region of LEO. This results in peaking up of the number density of the
fragments at some altitudes, particularly around 750km, 1000 km and 1500 km. Here we
have considered the marginal probability density functions f(h, t), f(h
p
, t), f(e, t) and f(B, t)
objects

98
where t ranges from the year 1996 to the year 2002 and subsequently have been able to
derive suitably averaged marginal distributions for the proposed model (Ananthasayanam et
al 2002 and 2003).

In the present model the following marginal distributions and their four year averaged values
for 'a' and 'e' and another representative distribution for B in the model denoted by only one
argument are as defined below

f (a, t) where t = 1999, , 2002

f (e, t) where t = 1999, , 2002

f (B, t) where t = 1999, , 2002

and f (a) = (1/4) f (a,t)dt

f (e) = (1/4) f (e,t)dt

f (B) = f (B,2002)

where denotes the sum over the four years. Firstly among the orbital debris the distribution
of the orbital elements namely altitude and eccentricity and the B values for one set of data in
each of the years (the first available set in the month of January) were analyzed to arrive at
their characteristic probability distributions. A quick analysis of the data at other times during
the years do not make much changes.

Here we have considered the marginal probability density functions f(n, h, t), f(n, h
p
, t), f(n,
e, t) and f(n, B, t) where t ranges from the year 1999 to the year 2002 and subsequently have
been able to derive suitably time averaged marginal distributions for the purpose of the
model. Firstly among the orbital debris, the distribution of the orbital elements namely
99

altitude, perigee height, eccentricity and the B values for TLE sets of data in each of the
years were analyzed to arrive at their characteristic probability distributions. It is observed
that the altitude distribution for the number of fragments exhibits peaks and it turned out that
such a feature can be best modeled with a tertiary mixture of Laplace distributions with eight
parameters. The details of the tertiary mixture of Laplace distributions with eight parameters
are provided in Appendix 4. It was noticed that no statistically significant variations could be
observed for the parameters across the years. Hence it is concluded that the probability
density function of the altitude distribution of the debris objects has some kind of equilibrium
and it follows a three component mixture of Laplace distribution (Mclauglin 1999,
Ananthasayanam et al 2002).

For the eccentricity e and the ballistic parameter B values the present analysis showed that
they can be acceptably quite well fitted by Lognormal distributions with two parameters. The
Lognormal distribution characteristics are provided in Appendix 4. In the case of eccentricity
also the describing parameter values do not vary much across the years. But for the
parameters of the B distribution there is some trend across the years which perhaps may be
attributed to causes such as decay effect, miniaturization of space systems and even the
uncertainty in the measurement data of B. However in the absence of definitive cause that
can be attributed for the variation across the years, it turns out to be best to have the most
recent value as the model reference value.

Lastly the same kind of analysis has also been carried out with respect to the various
inclination bands. Here the orbital parameters were analyzed with respect to the inclination
bands as is done in ORDEM (Kessler et al 1997, Liou et al 2001) for near circular orbits in
LEO. The five inclination bands considered here are 0-36deg (in ORDEM they consider 19-
36 deg, and did not consider 0-19 deg), 36-61 deg, 61-73 deg, 73-91 deg and 91-180 deg, and
corresponding to each band, the altitude, eccentricity and B values were modeled. It is found
that the third band shows the models with single Laplace distribution for altitude and
Lognormal for eccentricity and B fit quite well. The number density distributions with
altitude at other bands were modeled using a tertiary mixture of Laplace distributions, with
objects

100
the e and B following once again a Lognormal distribution. As mentioned two kinds of
model namely global and local models in space each providing different insight and
usefulness in orbital debris studies are proposed. These can be summarized as in Table 4.2.

Table 4.2 Summary of the Table of Contents of the Results
for the Various Distributions

Sl. No. Distribution Table Nos.
1 f (n, h , t ) 4.3 and 4.7
2 f (n, h
p
, t ) 4.4 and 4.8
3 f (n , e, t ) 4.5
4 f (n , B, t ) 4.6
5 f (n, e, i, t ) 4.12, 4.16
6 f (n, B, i, t ) 4.13
7 f (n, h , i, t ) 4.10 and 4.14
8 f (n, h
p
, i, t ) 4.10 and 4.14
9 f (n, h , e, t ) 4.17, 4.18, 4.19
10 f (n, h
p
, e, t ) 4.21, 4.22, 4.23
11 f (n, h , B, t ) 4.20
12 f (n, h
p
, B, t ) 4.24
13 f (n, h , e, i, t ) 4.25
14 f (n, h
p
, e, i, t ) 4.26

The first eight rows provide the global description of the orbital debris and the latter ones as
with the denoting the appropriate altitude or perigee height bins leading to the local
101

description of the orbital debris. The distributions in terms of the perigee height and altitude
help later simulation studies to have better accuracy in defining the low earth near circular
orbital debris scenario. It can be noted that the above are the appropriate marginal
distribution of the basic probability distribution defined at the very beginning of this chapter.

4.6.1 DISTRIBUTION OF THE NUMBER DENSITY OF DEBRIS WITH
ALTITUDE: f (n, h, t)

The Figure 4.1a provides the debris scenario with respect to altitude for the tracked debris
based on the first available data in January 2002. It was observed that in all other years from
1999 to 2001, the nature of the distribution is very similar. In every year the distribution
shows more than one peak. It has been reported by Kessler et al (1997) that ORDEM96
under predicts the environment at 1000 km by a factor of 2 and the reason was that the
breakups close to 1000 km, lead to a sharp peak around it but the model averages out sharp
peaks. The present approach using the tertiary mixture of Laplace distributions picks up the
sharp peaks and at the same time not loosing the accuracy at the other heights. The
parameters obtained using one set of data in January of each of the years 1999 to 2002 are
provided in Table 4.3.

Table 4.3 Parameters of the Tertiary Mixture of Laplace Distributions for the Number
Density of Catalogued Debris Objects with Altitude.

Location parameters (km) Scale parameters (km) Data
Year

1

2

3

1

2

3

Weight
p
1

Weight
p
2

2002 775 970 1460 175 80 90 0.7 0.7
2001 790 980 1460 150 85 95 0.7 0.7
2000 780 975 1460 155 80 95 0.7 0.7
1999 795 990 1460 140 75 100 0.7 0.7
objects

102

These parameters were obtained using Genetic algorithm (Goldberg 2000, Deb 1995) by
minimizing a suitable cost function, which is the sum of the squared error between the
observed densities with that of the fitted densities. For easy reference a brief introduction to
the basic genetic algorithm (Deb 1995, Goldberg 2000) utilised herein is provided in
Appendix 7. After analyzing many data sets all of which has a weight between 0.66 and 0.73
based on GA both the weight parameters were rounded off and for simplicity fixed in the
model to be at 0.7 and it was noticed that the location and scale parameters were worked out
which did not deteriorate the fit.

It is worth noting that the location parameters and the weight parameter do not vary
significantly over the years. But the scale parameters show some deviations across the years.
This may be due to the orbit decay effects due to air drag, varying solar activity and other
atmospheric perturbations. It is worth mentioning that even these deviations are not
statistically significant. It means that there is some kind of quasi equilibrium state maintained
across the altitude in this region of LEO as far as the catalogued objects are concerned
though the total number density itself is changing. This phenomenon can be attributed to
various reasons including, the source mechanisms such as launch activities, breakup of
objects in space (Jehn 1996) and rain down effect (Johnson and McKnight 1987) and also to
sink mechanisms such as decay of objects due to drag effects and solar activity variation
(Johnson and McKnight 1987).

4.6.2 DISTRIBUTIONS OF THE NUMBER DENSITY OF DEBRIS WITH
PERIGEE HEIGHT: f(n, h
p
, t)

Here we describe the marginal distributions for the perigee heights for each of the years.
Again a mixture of Laplace distributions was used to fit the model for the number density
with the perigee height. The parameters obtained for the model fits using the genetic
algorithm are provided in Table 4.4. The location, scale and weight parameters are nearly
constant across the years.
103

This table brings out the fact that the location of the peaks and its dispersion on both sides
can be considered as constants, even after accounting for the orbit decay effects due to air
drag, varying solar activity and other atmospheric perturbations. This characteristic adds the
validity of the inference that there is some kind of quasi equilibrium state is maintained in
this region of LEO as far as the catalogued objects are concerned though the numbers
themselves are changing.

Table 4.4 Parameters of the Tertiary Mixture of Laplace Distributions for the Number
Density of the Catalogued Debris Objects for Perigee Height.

Location parameters (km) Scale parameters (km) Data
Year

1

2

3

1

2

3

Weight
p
1

Weight
p
2

2002 710 870 1450 180 140 50 0.5 0.8
2001 710 870 1450 180 140 50 0.5 0.8
2000 710 870 1450 180 130 50 0.5 0.8
1999 710 870 1450 180 130 50 0.5 0.8

It may be noted that the behavior of the number of debris objects in perigee bins is
comparably better than that in altitude bins, in other words the fitted distributions match
better for perigee heights than altitudes. Figure 4.1a shows once again the goodness of fit of
the Laplace distribution for the perigee height. As clearly pointed out by the figures, the
peaks of the altitudes and perigee heights are different for the debris objects and hence a
separate study for the perigee height is justified.

objects

104

4.6.3 DISTRIBUTIONS OF THE ECCENTRICITY OF THE DEBRIS: f (n, e, t)

Eccentricity values for all the years showed a clear left skewing and a high concentration
close to the neighborhood of zero. This nature pointed to a Lognormal distribution fit for the
eccentricities which turned out to be appropriate.

Table 4.5 provides the parameters for the eccentricity in the data for various years. It also
provides the sample mean and standard deviation of the eccentricity across the period 1999 to
2002.

Table 4.5 Parameters of the Lognormal Distribution for the
Eccentricity of Catalogued Debris Objects.

Location Parameter

Scale Parameter

Statistics Data
Year
Estimate

95%
Confidence
limit
Estimate

95%
Confidence
limit
Mean Sigma
2002 -5.15 (-5.19,-5.12) 1.46 (1.44,1.49) 0.0151 0.0248
2001 -5.12 (-5.16,-5.09) 1.46 (1.44,1.49) 0.0155 0.0246
2000 -5.10 (-5.14,-5.06) 1.48 (1.45,1.50) 0.0159 0.0251
1999 -5.03 (-5.07,-4.99) 1.43 (1.41,1.46) 0.0165 0.0257

The Figure 4.1b shows a closer look at the goodness of fit of the model. From Table 4.5 it
can be seen that the fitted lognormal distribution, to the eccentricities of the tracked orbital
105

debris in the present region of interest, have nearly the same location parameter and scale
parameter for the period from 1999 to 2002

4.6.4 DISTRIBUTIONS OF THE BALLISTIC COEFFICIENT OF THE DEBRIS:
f (n, B, t)

As already mentioned, the B* values given in the TLE set is (
0
/2).C
D
A
eff
/m. In the present
analysis the ballistic coefficient B = C
D
A
eff
/m, is related to B
*
as

B = (2/
0
) B
*
= 12.7416 B*

The uncertainty in B as obtained from B* provided by the TLE sets is somewhat
considerable. While analyzing the data, it is found that a large number of data have one
specific value of 0.001. These data were filtered out before further analysis. It may be noted
that in this analysis negative B values and zero B values (due to non availability) are not
considered. Thus about 0.1% values were discarded. As in the case of eccentricities, a high
left skewing is clear for B values and it is found that B also follows a lognormal distribution.
The same nature was found across the years based on the TLE sets. The parameters, obtained
by the genetic algorithm by minimizing the least square error that best fits the distribution are
provided in Table 4.6 which also provides the mean and standard deviations of the B

values
for a natural comparison. Again Figure 4.1b provide the comparison between data and the fit.
This table shows changes in both the location and scale parameters over the years. This
change may be attributed to the following

(i) atmospheric decay effects,
(ii) control on the launch activities and hence less accumulation of large objects ,
(iii) small objects generation due to breakup of large objects and
(iv) miniaturization of the space systems.

objects

106

Table 4.6 Parameters of the Distributions for B Values of Catalogued Debris Objects.

Location Parameter Scale Parameter Statistics Year
of
Data
Estimate

95%
Confidence
limit
Estimate

95%
Confidence
limit
Mean Sigma
2002 -4.89 (-4.94,-4.85) 1.36 (1.34,1.39) 0.02 0.05
2001 -5.29 (-5.33,-5.25) 1.26 (1.24,1.29) 0.01 0.03
2000 -5.36 (-5.40,-5.32) 1.25 (1.23,1.28) 0.01 0.04
1999 -5.55 (-5.59,-5.50) 1.29 (1.26,1.32) 0.02 0.06

4.7 STATISTICAL CORRELATION AMONG THE FRAGMENT
CHARACTERISTICS

It may be noted that for simplicity we have provided the description of the space debris
scenario not in terms of the joint probability distributions of the various characteristics but in
terms of the individual marginal distributions. This misses out the possible correlation among
the various characteristics. After all the debris have been created by some physical process
and the resulting sizes and the velocities that have been imparted to these could be related in
some way or the other. Such a feature helps to simulate the debris scenario in a more realistic
way.

4.7.1 CORRELATION BETWEEN ALTITUDE AND ECCENTRICITY

This is similar to the inter level correlation of temperature with altitude in the atmosphere.
The correlation between these above two parameters are 0.15, 0.13, 0.11 and 0.15
107

respectively for years 1999 to 2002. Thus it turns out that there is only a weak correlation
between altitude and eccentricity. While simulating the environment, since the correlation is
very weak and of the order less than 0.15, one could take the altitudes and eccentricities as
independent.

4.7.2 CORRELATION BETWEEN PERIGEE HEIGHT AND ECCENTRICITY

The correlation between these above two parameters are 0.55, 0.43, 0.51 and 0.45
respectively for years 1999 to 2002. Hence it turns out that there is a reasonable correlation
between the perigee height and the eccentricity since the perigee height is a function of semi
major axis and eccentricity. While simulating the environment, since the correlation is not
small and of the order about 0.50, one need to take the perigee altitudes and eccentricities as
dependent.

4.7.3 CORRELATION BETWEEN ECCENTRICITY AND B

As discussed above, eccentricity and B values follows Lognormal distribution, the joint
distribution of log (eccentricity) and log (B) follows a bivariate distribution. The correlation
between these parameters are 0.32, 0.31, 0.29 and 0.24 respectively for years 1999 to 2002.
So an average value of 0.30 can be taken as the model correlation value. This information is
crucial in simulating the corresponding model B
*
and the eccentricity values.

4.8 THE PROPOSED SIMPLE MODEL FOR THE NUMBER
DENSITY WITH RESPECT TO ALTITUDE, PERIGEE HEIGHT,
THE ECCENTRICITY AND THE BALLISTIC COEFFICIENT

The parameter values in the SIMPLE model are based on the fitted distribution for the
average over the four year period namely 1999 to 2002 providing the relative frequency
objects

108
values for the number of debris in each of the intervals of the corresponding parameters
(i) altitude, (ii) perigee height, (iii) eccentricity, or (iv) ballistic coefficient.

In Table 4.7 the observed frequencies of the number of debris, for the years 1999 to 2002 is
provided for comparison with the proposed model frequencies.

The parameters for the model fit with tertiary mixture of Laplace distributions are

790, 975, 1470, 160, 85, 95, 0.7, and 0.7

respectively for three locations, three scales and two weights. The Figure 4.2 shows the
relative frequencies during the above years together with the model frequency. The figure in
the bottom of the above shows the goodness of fit of the mixture of Laplace distribution
fitting based on these parameters.

As in the case of altitude, in Table 4.8 the observed frequencies of the number of debris with
respect to the perigee height, for the years 1999 to 2002 is provided for comparison with the
proposed model frequencies.

The parameters for the model fit for the perigee height with tertiary mixture of Laplace
distributions are obtained as

(710, 870, 1450, 180, 140, 50, 0.5, and 0.8).

The average parameters for the model fit of eccentricity with the lognormal distribution
based on Table 4.5 are

(-5.12, 1.46)

and corresponding values for mean and standard deviation are 0.0157 and 0.025.
109

Table 4.7 Model Values for the Frequency of the Number Density of the Catalogued
Debris Objects with Altitude.

Observed Relative Frequencies

Altitude
Band (km)

1999

2000

2001

2002

Relative
Frequency
in the
Model
100-200 0.0002 0.0000 0.0003 0.0007 0.0003
200-300 0.0010 0.0012 0.0021 0.0021 0.0015
300-400 0.0073 0.0063 0.0066 0.0066 0.0070
400-500 0.0199 0.0193 0.0161 0.0161 0.0178
500-600 0.0465 0.0474 0.0559 0.0559 0.0496
600-700 0.0710 0.0705 0.0780 0.0780 0.0725
700-800 0.1369 0.1245 0.1403 0.1403 0.1374
800-900 0.1395 0.1325 0.1175 0.1175 0.1318
900-1000 0.1601 0.1691 0.1577 0.1577 0.1593
1000-1100 0.0844 0.0920 0.0887 0.0887 0.0881
1100-1200 0.0480 0.0389 0.0390 0.0390 0.0431
1200-1300 0.0281 0.0259 0.0323 0.0323 0.0289
1300-1400 0.0279 0.0392 0.0336 0.0336 0.0336
1400-1500 0.1230 0.1295 0.1351 0.1351 0.1280
1500-1600 0.0519 0.0540 0.0531 0.0531 0.0526
1600-1700 0.0245 0.0219 0.0229 0.0229 0.0224
1700-1800 0.0124 0.0110 0.0100 0.0100 0.0124
1800-1900 0.0106 0.0075 0.0064 0.0064 0.0077
1900-2000 0.0067 0.0091 0.0053 0.0041 0.0063

objects

110

Table 4.8 Model Values for the Frequency of the Number Density of the Catalogued
Debris Objects with Perigee Height.

Observed Relative frequencies

Perigee
height band
(km)
1999

2000

2001

2002

Relative
frequency
in the
Model
100-200 0.0007 0.0005 0.0016 0.0015 0.0011
200-300 0.0046 0.0035 0.0024 0.0031 0.0034
300-400 0.0165 0.0165 0.0159 0.0159 0.0162
400-500 0.0369 0.0334 0.0324 0.0361 0.0347
500-600 0.1104 0.1001 0.0982 0.0957 0.1011
600-700 0.0862 0.0818 0.0896 0.0929 0.0877
700-800 0.1380 0.1385 0.1545 0.1505 0.1454
800-900 0.1579 0.1553 0.1545 0.1503 0.1545
900-1000 0.1573 0.1601 0.1524 0.1482 0.1545
1000-1100 0.0617 0.0634 0.0612 0.0643 0.0626
1100-1200 0.0129 0.0138 0.0136 0.0134 0.0134
1200-1300 0.0175 0.0185 0.0180 0.0188 0.0182
1300-1400 0.0438 0.0476 0.0451 0.0461 0.0456
1400-1500 0.1315 0.1428 0.1419 0.1428 0.1397
1500-1600 0.0173 0.0173 0.0144 0.0166 0.0164
1600-1700 0.0033 0.0037 0.0034 0.0034 0.0034
1700-1800 0.0003 0.0002 0.0002 0.0000 0.0002
1800-1900 0.0029 0.0028 0.0008 0.0005 0.0018
1900-2000 0.0003 0.0003 0.0000 0.0000 0.0002

111

As already discussed, the B values show a definite trend of increase in mean values as well as
the dispersions, but a definitive trend evaluation with the limited data is not possible. Hence
we propose the parameters of the lognormal distribution fitting for B values as that of the
present year 2002, with location and scale parameters respectively as

(- 4.89 and 1.36).

4.9 ANALYSIS OF THE DATA FOR DIFFERENT INCLINATION
BANDS

Here the previous type of analysis is carried out with respect to different inclination bands.
Similar to ORDEM 96, the SIMPLE model considers 5 inclination bands (joining the first
two bands together) for low earth near circular orbits, that is, we consider the inclination
bands specified by the ranges as (0,36], (36,61],(61,73], (73,91] and (91,180]. Here we
describe the three characteristics of the objects, namely the altitude, perigee height and
eccentricity for each of the bands and arrive at a best possible impressionistic model. Once
again the tertiary mixture of Laplace distributions for altitude and perigee height and
lognormal distributions for eccentricity turn out to be adequate for the description in the
model. Table 4.9 provides the number of objects in each of the bands for the years 1999 to
2002. It may be noted that the number of debris is nearly the same in all the bands through
the above years, except for the fifth inclination band (91, 180] during the years 2001 and
2002.

4.9.1 PARAMETERS OF THE DISTRIBUTIONS FOR THE NUMBER OF DEBRIS
WITH ALTITUDE IN VARIOUS INCLINATION BANDS

The parameter values of the Laplace fit for all the five bands across the years from 1999 to
2002 are provided in the Table 4.10. It is observed that in terms of the altitude above the
earth surface the number of tracked objects in all the inclination bands except for the third
band (61, 76] follows a mixture of Laplace distributions with eight parameters.
objects

112
Table 4.9 Number of Debris with Time in the Different Inclination Bands.

Inclination Bands in deg

Year

Number
of Debris
objects (0, 36] (36, 63] (61, 73] (73,91] (91, 180]
1999 6123 247 184 1313 2713 1666
2000 6014 235 230 1299 2585 1665
2001 6181 221 236 1323 2500 1901
2002 6101 218 220 1247 2446 1970

But for the third band a single Laplace distribution with just two parameters alone is
sufficient. It may be noted that the parameters corresponding to the different bands do not
significantly vary across the years except for band 5. It may be noted that there is a change
in the location parameter from 2000 to 2001 for the last inclination band. It may be attributed
to the breakup of CBERS 1/SACI-1 Rocket body (3
rd
stage of Long March) at an altitude of
741 Km on 11
th
March 2000 generating around 300 catalogued objects (Anz Meador 2001).
The inclination of this rocket body was 98.5373 degrees and comes under the last inclination
band (91, 180]. The debris cloud of this fragmentation event brought the location parameter
from around 820 km to 770 km.

4.9.2 PARAMETERS OF THE DISTRIBUTIONS FOR THE NUMBER OF DEBRIS
WITH PERIGEE HEIGHTS IN VARIOUS INCLINATION BANDS

As in the case of the modeling with altitude, the distributions with the perigee height in each
of the bands are also modeled using the mixture of Laplace distributions. The 8 parameters of
the models for each of the inclination bands for the years 1999 to 2002 are provided in the
above Table 4.11. It may be noted in this table also that the parameters corresponding to the
different bands do not significantly vary across the years.
113

Table 4.10 The Location and Scale Parameters of the Tertiary Mixture of Laplace
Distributions for the Number Density of Debris in Altitude

Parameter values (
1

1

2

2

3

3

p
1

p
2
) Year
Band 1999 2000 2001 2002

(0,36]

(590 161 708
130 1440 297
0.4 0.8)

(570 110 710
110 1440 290
0.4 0.8)

(565 110 745
155 1350 250
0.4 0.8)

(570 120 730
140 1400 210
0.4 0.8)

(36,61]

(551 100 822
140 1475 80
0.2 0.5)

(563 42 830
83 1445 145
0.2 0.5)

(605 57 810
97 1450 72
0.2 0.5)

(587 65 823
100 1440 61
0.2 0.5)

(61,73]

(875 160 )

(917 177)

(875 155)

(920 175)

(73,91]

(725 91 950
80 1450 60
0.4 0.7)

(735 116 978
90 1472 63
0.4 0.7)

(730 110 975
100 1425 70
0.4 0.7)

(745 120 970
100 1450 62
0.4 0.7)

(91,180]

(825 140 1085
120 1514 136
0.8 0.7)

(818 140 1042
100 1496 116
0.8 0.7)

(770 120 1070
70 1470 115
0.8 0.7)

(770 140 1020
141 1475 105
0.8 0.7)

objects

114

Table 4.11 The Location and Scale Parameters of the Tertiary Mixture of Laplace
Distributions for the Number Density of Debris in Perigee Height

Parameter values (
1

1

2

2

3

3

p
1

p
2
) Year
Band 1999 2000 2001 2002

(0,36]

(530 80 690
25 940 70
0.8 0.9)

(530 82 690
25 940 65
0.8 0.9)

(530 85 680
40 930 70
0.8 0.9)

(530 100 670
55 940 50
0.8 0.9)

(36,61]

(520 70 690
90 940 50
0.7 0.9)

(530 75 690
40 940 36
0.7 0.9)

(530 75 662
60 940 60
0.7 0.9)

(520 90 690
45 950 50
0.7 0.9)

(61,73]

(530 70 900
140 1380 250
0.4 0.9)

(535 70 890
140 1410 75
0.4 0.9)

(540 70 880
115 1420 90
0.4 0.9)

(550 50 880
110 1435 100
0.4 0.9)

(73,91]

(520 100 810
150 1440 20
0.2 0.9)

(530 120 820
175 1430 28
0.2 0.9)

(530 120 820
160 1430 28
0.2 0.9)

(540 120 790
160 1430 28
0.2 0.9)

(91,180]

(525 50 830
120 1400 250
0.2 0.9)

(530 50 825
130 1460 100
0.2 0.9)

(540 55 820
130 1440 70
0.2 0.9)

(530 70 810
130 1430 50
0.9)

115

4.9.3 DISTRIBUTION OF THE ECCENTRICITY OF THE DEBRIS IN THE
DIFFERENT INCLINATION BANDS

The parameter values with respect to eccentricity, for all the five bands across the years from
1999 to 2002 are provided in the following Table 4.12.

It was noted that in all these bands and across the years, the eccentricity follows lognormal
distributions. But for the bands one and three there is some deviation from the fit and as an
impressionist approach the deviation may perhaps be not taken into account. However it may
be seen from the above Table 4.12 in each band the location and scale parameters do not vary
much through the years.

Table 4.12 The Location and Scale Parameters of the Lognormal Distributions for the
Eccentricity in the Different Inclination Bands.

Location and Scale Parameters with respect to inclination bands Year
(0,36] (36,61] (61,73] (73,91] (91,180]

1999

(-3.92, 1.62)

(-5.69, 2.39)

(-4.30, 1.31)

(-5.47, 1.30)

(-5.06, 1.32)

2000

(-3.94, 1.64)

(-5.65, 2.36)

(-4.32, 1.35)

(-5.55, 1.27)

(-5.08, 1.32)

2001

(-3.97, 1.70)

(-5.74, 2.34)

(-4.34, 1.35)

(-5.61, 1.25)

(-5.08, 1.30)

2002

(-3.96, 1.68)

(-5.85, 2.36)

(-4.38, 1.36)

(-5.66, 1.26)

(-5.07, 1.28)

objects

116

4.9.4 DISTRIBUTION OF THE BALLISTIC COEFFICIENT OF THE DEBRIS IN
THE DIFFERENT INCLINATION BANDS.

The parameter values for all the five bands across the years from 1999 to 2002 are provided
in the following Table 4.13. It is noted that in all these bands and across the years the ballistic
coefficient can be considered as following lognormal distributions. As noted earlier the trend
for the parameters of the distribution fitting the B varies from year to year across all the
inclination bands

Table 4.13 The Scale and Location Parameters of the Lognormal Distributions for the
Ballistic Coefficient in the Different Inclination Bands.

Location and Scale Parameter values with respect to Inclination bands Year
(0,36] (36,61] (61,73] (73,91] (91,180]

1999

(-5.85, 1.10)

(-6.04, 0.99)

(-5.72, 0.97)

(-5.30, 1.57)

(-5.55, 1.18)

2000

(-5.44, 1.10)

(-5.57, 1.20)

(-5.35, 1.16)

(-5.34, 1.36)

(-5.36, 1.26)

2001

(-5.40, 1.10)

(-5.55, 1.14)

(-5.35, 1.11)

(-5.56, 1.31)

(-5.16, 1.25)

2002

(-4.78, 1.18)

(-4.84, 1.22)

(-4.77, 1.29)

(-5.37, 1.39)

(-4.51, 1.30)

117

4.10 SIMPLE MODEL FOR DIFFERENT INCLINATION BANDS

As already explained in the previous sections, the SIMPLE model approach is utilized to
obtain the model behavior of number densities across the altitude bands and perigee height
bands with respect to the five inclination bands under consideration. For the altitude and
perigee height, a tertiary mixture of Laplace distribution with eight parameters can be
proposed. The eight model parameters, namely 3 for the location, 3 for the scale and the 2
weight parameters are provided in the following Tables 4.14 and 4.15. The Figures 4.3a,
4.3b, 4.3c, 4.3d and 4.3e provide the comparison of the proposed model for altitude and
perigee with those of the year 1999 to 2002 in all the five bands.

For eccentricity, the proposed model is a lognormal distribution with the parameters as
provided in Table 4.16.

Table 4.14 Parameters of the SIMPLE Model for the Number Density with Altitude
Distribution for Different Inclination Bands.

Location parameter (km) Scale parameter (km) Inclination
Band

1

2

3

1

2

3

Weight
p
1

Weight
p
2

(0, 36] 550 710 1450 120 120 250 0.4 0.8
(36, 61] 575 815 1440 60 85 90 0.2 0.5
(61, 73]
*
900 --- --- 160 ---- ---- --- ---
(73, 91] 740 950 1450 100 60 70 0.4 0.7
(91, 180] 815 930 1485 165 130 120 0.8 0.7
* Single Laplace Distribution

objects

118

Table 4.15 Parameters of the SIMPLE Model for Perigee Distribution

Location parameter (km) Scale parameter (km) Inclination
Band

1

2

3

1

2

3

Weight
p
1

Weight
p
2

(0, 36] 530 680 940 90 45 65 0.8 0.9
(36, 61] 525 680 950 80 60 50 0.7 0.9
(61, 73] 540 890 1420 65 125 100 0.4 0.9
(73, 91] 530 800 1430 120 165 28 0.2 0.9
(91, 180] 530 810 1430 80 130 80 0.2 0.9

Table 4.16 Parameters of the Eccentricity Distribution Model

Location and Scale Parameter values with respect to Inclination bands Model
Values (0,36] (36,61] (61,73] (73,91] (91,180]

( , )

(-3.95, 1.33)

(-5.70, 2.37)

(-4.33, 1.33)

(-5.50, 1.28)

(-5.07, 1.31)
Mean and
Sigma

(0.045,0.043)

(0.021,0.034)

(0.028,0.033)

(0.009,0.019)

(0.012, 0.015)

119

4.11 FURTHER ANALYSIS ON THE DISTRIBUTION OF THE
ECCENTRICITY AND BALLISTIC COEFFICIENTS

We next expand the characteristics namely the eccentricity and ballistic coefficient of the
debris with respect to various altitude, perigee height, and inclination bands. Such an above
local characterization in terms of inclination bands is necessary and useful for the simulation
of the debris environment and its further analysis for risk assessment and mission planning.
This is because generally the mission related requirements are confined to a particular
altitude or perigee bin.

4.11.1 DISTRIBUTION OF THE ECCENTRICITY AND BALLISTIC
COEFFICIENT IN VARIOUS ALTITUDE BINS

It may be recalled that the eccentricity and the ballistic coefficient of the debris could be well
represented by lognormal distribution for all the years and hence it provided a model for the
description of eccentricity and ballistic coefficient in LEO. Now in order to characterize the
eccentricity and ballistic coefficient behavior across the altitude bins, a further analysis of the
TLE data of the objects in all the four years were carried out. It was found that both the
eccentricity and ballistic coefficient, in all the four years, could once again be well fitted by a
lognormal distribution. The Figure 4.4 shows the goodness of fit of the lognormal
distribution for two typical altitude bins for the year 2002 and it was noted that the match is
similar for the other three years 2001, 2000 and 1999 and other altitude bins as well and
hence not repeated for the sake of brevity.

The parameters of the lognormal fit, location and scale, for eccentricity in each of the altitude
bins are provided in Tables 4.17 and 4.18. For each of these parameters, it is noted that,
across the years there is not much of a variation. Hence an averaged model values can be
provided for all these parameters in each altitude bins, as shown in these tables.

objects

120
TABLE 4.17 Location Parameters of the Lognormal Fit for Eccentricity

Location parameters in Altitude bins for eccentricity
for various years and in the Model
Mid value
of Altitude
bin (km) 2002 2001 2000 1999 Model
250
350
450
550
650
750
850
950
1050
1150
1250
1350
1450
1550
1650
-3.4373
-4.9136
-4.7157
-4.7421
-4.9241
-5.3675
-5.2117
-5.3264
-4.7701
-4.0565
-3.9378
-4.1258
-5.5432
-4.4933
-3.7515
-4.5074
-4.4191
-4.4590
-4.5463
-4.8754
-5.2102
-5.1311
-5.2843
-4.6875
-4.0681
-3.8554
-4.1747
-5.5782
-4.4733
-3.8193
-5.0494
-4.3859
-4.5295
-4.3518
-4.6640
-5.3631
-5.1083
-5.1355
-4.6721
-4.0357
-3.8709
-4.4602
-5.4725
-4.6303
-3.6596
-4.7783
-4.3250
-4.5830
-4.5845
-4.6709
-5.1683
-4.9707
-5.1770
-4.7538
-4.1471
-3.8852
-4.1992
-5.4028
-4.4179
-3.7196
-4.4431
-4.5109
-4.5718
-4.5562
-4.7836
-5.2773
-5.1055
-5.2308
-4.7209
-4.0769
-3.8873
-4.2400
-5.4992
-4.5037
-3.7375

121


Scale parameters in Altitude bins for eccentricity
for various years and in the Model
Mid value
of Altitude
bin (km) 2002 2001 2000 1999 Model
250
350
450
550
650
750
850
950
1050
1150
1250
1350
1450
1550
1650
1.0237
1.2593
1.1884
1.1326
1.2080
1.2924
1.1592
1.1447
1.0522
0.8862
0.8185
0.9206
1.2536
1.0009
0.8108
1.2557
1.1702
1.1426
1.1348
1.1249
1.2466
1.1091
1.1197
0.9971
0.8297
0.8655
0.9471
1.2905
1.0327
0.8377
0.9317
1.2474
1.1437
1.0294
1.1134
1.2923
1.1411
1.1309
1.0381
0.8971
0.8155
1.0289
1.2867
0.9953
0.8327
1.0429
1.2312
1.1842
1.1113
1.1185
1.2127
1.0960
1.1151
1.0288
0.8924
0.8225
0.9446
1.2770
1.0510
0.8060
1.0635
1.2270
1.1647
1.1020
1.1412
1.2610
1.1263
1.1276
1.0291
0.8764
0.8305
0.9603
1.2770
1.0200
0.8218

objects

122
In order to further demonstrate the stability and consistency of the estimated parameters,
location, scale, mean and standard deviation values, a moving window of bin size 100 km
and a window shift of 50 km was considered with respect to the altitude and the four
parameters were estimated for the eccentricity distribution for the years 1999 to 2002.
Comparison of the individual bin estimates to that of the moving window estimates are
provided in Figures 4.5 and 4.6 and they showed that the estimates are consistent across the
years and independent of moving window shift. Further to model the parameters over all the
bins a polynomial fit for the model parameter values across the altitude bins was generated.
Figure 4.7 provides the comparison of the proposed model values of the location parameters
in altitude bins from the individual bins and the moving windows.

The polynomial considered for the model fit of order n for the various location and scale
parameters are of the following form

where x = mid altitude or perigee height, as the case may be, of the bins in km above the
earth.

A careful analysis of the polynomial fit brought out that the best fit in least square sense for
the parameters namely mean, standard deviation, location parameter and scale parameter is a
9
th
degree polynomial. The coefficients of the 9
th
order model polynomial fit with respect to
eccentricity in altitude bins for each of the above parameters are provided in Table 4.19. A
similar analysis is carried out with respect to the perigee height bins also and it is found that
5
th
degree polynomial fit is sufficient to describe the parameters, which is described in the
next section.

In the case of the ballistic coefficient B, it is observed that, even though the data follows a
lognormal distribution there are appreciable variations across the years for the parameters as
can be seen from the Table 4.20. Hence an average model specification may not be feasible.
But it may be noted from Table 4.20 that a same order namely 9
th
order polynomial fit is
0 1
1
1
... p x p x p x p
n
n
n
n
+ + + +

123

possible for each of the year ranging from 1999 to 2002. Hence we provide here with an
order of best polynomial fit for further analysis of the ballistic coefficient characteristics, or it
may be suggested to use the parameter values from the latest TLEs as the tentative value for
further use.

Table 4.19 The Coefficients of Model Polynomial fit for Eccentricity in Altitude Bins.

Coefficients of 9
th
degree polynomial Para
meter p
9
p
8
p
7
p
6
p
5
p
4
p
3
p
2

p
1
p
0

Location

-0.9335 -0.9034 4.5848 4.9303 -6.3568 -8.4723 1.4400 4.9150
1.0632 -5.2254

Scale

0.2787 0.1617 -1.3316 -0.9423 1.7956 1.6731 -0.4007 -0.9019
-0.3343 1.1398

Mean

-0.0108 -0.0114 0.0557 0.0635 -0.0831 -0.1108 0.0251 0.0665
0.0111 0.0098

Standard
deviation

-0.0037 -0.0114 0.0218 0.0580 -0.0360 -0.0939 0.0096 0.0579
0.0041 0.0165

objects

124
Table 4.20 Coefficients of Polynomial Fits for Location and Scale of B in Altitude Bins
Across the Years.

Coefficients of 9
th
degree polynomial Para
Meter
Year
p
9
p
8
p
7
p
6
p
5
p
4
p
3
p
2

p
1
p
0

Location 2002

2001

2000

1999
-1.4443 -1.1719 7.9625 5.9715 -13.9007 -9.4305 8.0383 4.5337
-1.9943 -5.6651
-0.9576 0.1920 4.3500 -0.0527 -5.4121 -1.5961 0.8445 1.7541
-0.1317 -6.2651
-0.5489 -0.5262 2.9128 3.0463 -4.3567 -5.6469 1.1452 3.5190
-0.2824 -6.3953
-1.3274 -0.3315 6.6136 2.4403 -9.7782 -5.3467 3.8234 3.5828
-0.5686 -6.6254
Scale 2002

2001

2000

1999
0.4434 -0.1902 -2.3961 0.9604 4.3194 -1.4342 -3.1672 0.5359
1.1526 1.5719
-0.5848 -0.3009 3.3252 1.2060 -6.5430 -1.3418 5.2607 0.2880
-1.2483 1.6219
-0.3782 0.2045 2.2186 -0.8372 -4.4875 0.8133 3.3955 -0.1115
-0.3749 1.5827
0.1741 0.1060 -0.8531 -0.9484 1.1080 2.3361 -0.0299 -1.9119
-0.2438 2.0654

125

4.11.2 DISTRIBUTION OF THE ECCENTRICITY AND BALLISTIC
COEFFICIENT IN VARIOUS PERIGEE HEIGHT BINS

As in the case of altitude bins, to characterize the eccentricity and ballistic coefficient
behavior across the perigee height bins, the TLE data of the objects in all the four years were
utilized. It is found that both the eccentricity and the ballistic coefficient, in all the four years,
can be well fitted again by a lognormal distribution. The Figure 4.8 brings out the goodness
of fit of the Lognormal distribution in some of the perigee height bins for the year 2002. The
match of the lognormal fitting is similar for the other years 1999, 2000 and 2001 as well. The
first lognormal probability plot on the top is for each perigee height bin with respect to
eccentricity and that the second at the bottom shows the fit for the ballistic coefficient. While
plotting the ballistic coefficient in lognormal plot, some spurious data as mentioned earlier
were removed.

The parameters of the lognormal fit, location and scale, for eccentricity in each of the perigee
height bins were obtained and tabulated in the Tables 4.21 and 4.22. For each of these
parameters, it is noted that, across the years there is not much of a variation. So an averaged
model can be and was provided for all these four parameters in each perigee height bins.
These model values are also provided in these tables.

Further analysis with respect to perigee height bins was carried out as was done with altitude
bins. In order to demonstrate the stability and consistency of the estimated parameters,
location, scale, mean and standard deviation, a moving window of bin size 100 km and
window shift 50 km is considered for the perigee height and the four parameters are
estimated for eccentricity through the years 1999 to 2002. It is seen that the estimates are
consistent across the years and independent of the moving window shift. The same type of
analysis was carried out for other parameters of eccentricity, namely scale, mean and
standard deviation and the inferences are the same as that of the location. As in the case of
altitude bins, polynomial fits were generated for the model parameters of eccentricity across
the perigee height bins and those are provided in the Table 4.23.
objects

126
Table 4.21 Location Parameters of the Lognormal Fit for Eccentricity in Perigee
Height Bins of Size 100 Km.

Location parameters in perigee heights bins for eccentricity
for the various years and Model
Mid value of
Perigee height bin
(km) 2002 2001 2000 1999 model
250
350
450
550
650
750
850
950
1050
1150
1250
1350
1450
-3.0476
-3.4345
-3.8309
-3.7706
-4.9049
-5.2605
-5.0445
-5.1460
-4.5533
-4.4752
-4.8497
-5.0825
-5.5359
-3.0947
-3.6000
-3.7257
-3.7041
-4.7142
-5.1332
-4.9801
-5.1868
-4.5579
-4.2779
-4.8432
-5.0516
-5.4941
-3.3048
-3.3871
-3.8223
-3.5829
-4.5745
-5.3545
-4.9786
-5.1475
-4.5680
-4.1210
-4.8681
-4.9844
-5.4939
-3.1215
-3.2578
-3.8828
-3.6107
-4.4797
-5.3066
-4.9773
-5.1466
-4.5449
-4.6749
-4.7821
-4.9361
-5.4627
-3.1421
-3.4199
-3.8154
-3.6671
-4.6683
-5.2637
-4.9951
-5.1567
-4.5560
-4.3873
-4.8357
-5.0137
-5.4967

In the case of the ballistic coefficient, it is observed that, even though the data follows
lognormal distribution, there are distinct variations between the parameters, across the years.
So as in the case of altitude bins, an average model prediction may not be possible with
respect to perigee height bins. But it may be noted that a same order polynomial fit is
possible for each of the year ranging from 1999 to 2002. Hence we provide here with an
127

order of best polynomial fit for further analysis of the ballistic coefficient characteristics, or it
is suggested to use the parameter values from the latest TLEs as a model for further
investigation.

in Perigee Height Bins of Size 100 Km.

Scale parameters in perigee heights bins for eccentricity
for the various years and Model
Mid value
of Perigee
height bin
(km)
2002 2001 2000 1999 Model
250
350
450
550
650
750
850
950
1050
1150
1250
1350
1450
0.8916
0.8451
0.9121
0.8918
1.1843
1.2448
1.0603
1.1356
0.9179
0.9445
0.9020
1.0094
1.0459
0.8958
0.8725
0.8864
0.8644
1.1353
1.1957
0.9877
1.1183
0.9229
0.9539
1.1117
0.9778
1.0629
0.8051
0.8489
0.8998
0.8207
1.0953
1.2724
0.9930
1.1059
0.9682
0.8853
1.1309
0.8204
1.0422
0.7291
0.8330
0.9436
0.8164
1.0476
1.2513
1.0293
1.1080
0.9568
1.0009
0.8873
0.8875
1.0426
0.8304
0.8499
0.9105
0.8484
1.1156
1.2411
1.0176
1.1170
0.9415
0.9462
1.0080
0.9238
1.0484

objects

128
Table 4.23 The Coefficients of Polynomial Fit for Eccentricity
in Perigee Height Bins

Coefficients of the polynomial fit of 5
th
degree Parameter
p
5
p
4
p
3
p
2
p
1
p
0

Location 0.1376 -0.4359 -0.6705 1.3005 0.0643 -5.0308
Scale -0.0450 0.0924 0.1809 -0.2890 -0.1132 1.1106

Table 4.24 provides the coefficients of polynomial fits for the parameters of B in perigee
height bins across the years.

Table 4.24 The Coefficients of Model Polynomial Fit for B in Perigee Height Bins

Coefficients of the polynomial fit of 5
th
degree Parameter Year
p
5
p
4
p
3
p
2
p
1
p
0

Location 2002
2001
2000
1999
-0.2202 0.0287 1.1132 -0.4644 -2.6752 -5.6073
-0.1072 -0.0513 0.3418 0.0121 -1.5419 -6.2447
-0.0775 -0.4310 0.3118 0.8555 -1.2922 -6.4960
-0.1767 -0.2442 0.8945 0.4424 -1.8649 -6.5611
Scale 2002
2001
2000
1999
0.3136 0.2593 -1.0137 -0.6131 0.7880 1.6923
-0.3050 0.0658 1.0047 -0.2101 -0.3765 1.7015
0.0755 0.0254 -0.2331 -0.1553 0.3862 1.7407
0.0616 0.1534 -0.0966 -0.6567 -0.0396 1.9738

129

4.11.3 CHARACTERIZATION WITH RESPECT TO INCLINATION BANDS

We describe here the nature of the distribution of eccentricity and ballistic coefficient with
respect to five inclination bands in the SIMPLE model. The inclination bands considered
were (0, 36], (36, 61], (61, 73], (73, 91] and (91, 180] and these inclination bands correspond
to those considered in ORDEM model. In order to characterize fully the nature of the
distribution of eccentricity and B across the altitude and perigee height bins, the eccentricity
and B data were analyzed with respect to the inclination bands for each of the altitude and
perigee height bins. As the first inclination band does not have sufficient number of objects
in the altitude and perigee height bins, the other four inclination bands are considered for the
analysis. It is found that the eccentricity and ballistic coefficient once again follow lognormal
distributions in the inclination bands. In order to demonstrate this nature, we provide Figures
4.9 and 4.10 for two of the five inclination bands with respect to two altitude bins of (700,
800] km and (800, 900] km. These figures clearly bring out the lognormality nature of
eccentricity and B across the inclination bands as well as altitude bands. The lognormality of
the eccentricity and ballistic coefficient is true for all the inclination bands with respect to
other altitude bins and perigee height bins. Tables 4.25 and 4.26 provide the coefficients of
the best polynomial fits for eccentricity with respect to various inclination bands across
altitude bins and perigee height bins.

Considering ballistic coefficients in the inclination bands for each of altitude bins and perigee
height bins, as in the earlier case, providing model fit may not be feasible as the B values
vary randomly with respect to time. However here also it is found that the order of the
polynomial fit for each of the years are same, for example the mean and standard deviations
in all the inclination bands for the three years 2002, 2001 and 2000 can be considered as 3
rd

degree. Similarly location and scale can be best approximated by a 4
th
degree polynomial.

objects

130

th
Degree Model Fit with Respect
To Altitude Bins and Inclination Bands for Eccentricity

Coefficients of the polynomial fit Parameter
p
5
p
4
p
3
p
2
p
1
p
0

Inclination Band (36, 61]
Location
Scale

2.3230 0.3832 -5.6703 -0.9974 2.7117 -4.0512
-0.6852 -0.1937 1.7319 0.3965 -0.8649 0.9032

Location
Scale

0.8701 -0.9338 -2.2973 2.2512 1.8453 -4.3552
-0.2324 0.2181 0.6484 -0.6093 -0.5918 0.9483

Location
Scale

0.9836 0.6920 -2.8260 -0.8660 1.4863 -5.2839
-0.1486 -0.1803 0.4415 0.3001 -0.3224 1.1352

Location
Scale

0.4988 0.2958 -1.4116 -0.6085 1.4829 -4.7948
-0.2204 -0.0216 0.5483 0.0579 -0.3992 0.8837

131

rd
Degree Model Fit with Respect
to Perigee Height Bins and Inclination Bands for Eccentricity

Coefficients of the polynomial fit Parameter
p
3
p
2
p
1
p
0

Location
Scale

1.2039 0.6200 -1.9314 -4.6301
-0.4393 -0.1173 0.5967 1.0536

Location
Scale

1.2881 0.1576 -1.7833 -3.9701
-0.3708 -0.0689 0.5208 0.8771

Location
Scale

0.1121 0.9677 -1.0409 -5.6863
0.2280 -0.2332 -0.3029 1.1582

Location
Scale

-0.3964 0.5344 0.6894 -5.3886
0.0945 -0.0623 -0.2441 1.0793

objects

132
It may be noted that here we characterized the lognormality of the eccentricity and ballistic
coefficient in each of the altitude bins and perigee height bins within the inclination bands
considered. Hence this characterization of eccentricity and ballistic coefficient can be utilized
for any application of further analysis of debris flux and collision risk assessment of the
debris objects in any altitude or perigee height bins and in any inclination band (Anilkumar et
al 2003).

4.12 APPLICATION OF SIMPLE NATURE OF DEBRIS
ENVIRONMENT

The characterization provided here help in a proper assessment of the space debris
environment in LEO as these models of eccentricity and ballistic coefficient in altitude bins
or perigee height bins together with the SIMPLE model of altitudes and perigee heights
describe the statistical nature of the environment. A Monte Carlo simulation of the orbital
parameters, namely perigee height, altitude and eccentricity, based on the SIMPLE
characterization can be used to obtain the closest approach, in a statistical sense, to a target
object. Such an analysis has been carried out and reported in Chapter 6 of this thesis.

Further as the catalogued data consists of only the objects of large size, to obtain the risk
assessment due to other uncatalogued objects, statistical simulation of small objects is
essential. This is a potential area of application of SIMPLE structure as the statistical
simulation of the complete LEO debris environment is possible by expanding the SIMPLE
scenario.

The debris flux and hence the spatial density described by SIMPLE model can be
combined with any of the collision probability assessment tools, such as Poisson density
approach, and hence obtain the probability of collision either with a target object or between
the debris objects.

133

4.13 CONCLUSIONS

The present SIMPLE model provides another engineering model for describing the orbital
space debris scenario in LEO between altitudes 200 km to 2000 km and eccentricity between
0 to 0.2. The present model SIMPLE proposes in terms of both the altitude and perigee
distributions with respect to different inclination bands as well as for the whole region of
inclinations. While other models like ORDEM96 utilizes deterministic (semi empirical)
functions to describe the distributions, SIMPLE uses probability distributions. The number
of parameter values in SIMPLE is, in general, just 8 for each description of altitude or
perigee distributions and in ORDEM96 it is more. SIMPLE model captures closely all the
peak densities without losing the accuracy at other altitudes.

Also SIMPLE structure gives insight into the evolution of the space debris scenario. It
brings out the equilibrium nature of source and sink of the space debris by clearly specifying
the lognormality of the eccentricity and ballistic coefficient and Laplace nature to the number
density with altitudes and perigee height. Space debris environment is a mixture of objects
arising out of many phenomena, including fragmentation due to explosion and collision and
new launch activities. SIMPLE clearly shows that the distributional nature of the orbital
parameters do not vary across the years and the variation of the numerical values of
parameters is small. The processes like explosions or collisions or increased launch activity
in a certain region in LEO such as constellations, change the equilibrium between source and
sink, but this state of non equilibrium appears to be dissipated quickly into background debris
environment to follow once again the SIMPLE structure.

objects

134
0 500 1000 1500 2000
0
100
200
300
400
500
600
700
800
Altitude (km)
N
o
.

o
f

o
b
j
e
c
t
s
0 500 1000 1500 2000 2500
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
x 10
-3
Altitude (km)
L
a
p
l
a
c
e

P
D
F
Model
Observed
0 500 1000 1500 2000
0
100
200
300
400
500
600
700
800
Perigee height(km)
N
o
.

o
f

o
b
j
e
c
t
s
0 500 1000 1500 2000 2500
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
x 10
-3
Perigee height (km)
L
a
p
l
a
c
e

P
D
F
Model
Observed

Figure 4.1a Number Density Distributions in Altitude and Perigee Height
for the Year 2002

CHAPTER 4 A NEW STOCHASTIC IMPRESSIONISTIC LOW EARTH (SIMPLE)
ENGINEERING MODEL OF THE SPACE DEBRIS SCENARIO
135

-8 -6 -4 -2 0
0
200
400
600
800
1000
1200
1400
log(eccentricity)
N
o
.

o
f

o
b
j
e
c
t
s
-8 -6 -4 -2 0
0
20
40
60
80
100
120
140
log(eccentricity)
P
D
F
Observed
Model
-6 -4 -2 0
0
500
1000
1500
2000
2500
log(B)
N
o
.

o
f

o
b
j
e
c
t
s
-6 -4 -2 0
0
10
20
30
40
50
60
70
log(B)
P
D
F
Observed
Model

Figure 4.1b Eccentricity and B Distributions of the Debris Objects for the Year 2002
objects

136
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.05
0.1
0.15
0.2
Altitude in km
R
e
l
.

f
r
e
q
.

M
e
a
s
.

v
s
.

m
o
d
e
l
Model
Observed
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.5
1
1.5
2
x 10
-3
Altitude in km
P
D
F
Model fit
Model

Figure 4.2 Comparison of Model with Year wise and Distribution Fit of
Number Density with Altitude
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.5
1
1.5
2
2.5
x 10
-3
Altitude in km
P
D
F

i
n

i
n
c
.

b
a
n
d

1
Model
Observed
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
1
2
3
4
x 10
-3
Perigee height in km
P
D
F

i
n

i
n
c
.

b
a
n
d

1
Model
Observed

Figure 4.3a Comparison of the Number Density of Debris from Measurements
and Model Distributions for Inclination Band (0, 36]
137

0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
1
2
3
4
x 10
-3
Altitude in km
P
D
F

i
n

i
n
c
.

b
a
n
d

2
Model
Observed
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
1
2
3
4
x 10
-3
P
D
F

i
n

i
n
c
.

b
a
n
d

2
Model
Observed

Figure 4.3b Comparison of Number Density of Debris from Measurements and
Model Distributions in Inclination Band (36,61]
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
1
2
3
x 10
-3
Altitude in km
P
D
F

i
n

i
n
c
.

b
a
n
d

3
Model
Observed
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.5
1
1.5
2
2.5
x 10
-3
P
D
F

i
n

i
n
c
.

b
a
n
d

3
Model
Observed

Figure 4.3c Comparison of Number Density of Debris from Measurements and
Model Distributions in Inclination Band (61, 73]
objects

138
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.5
1
1.5
2
2.5
x 10
-3
Altitude in km
P
D
F

i
n

i
n
c
.

b
a
n
d

4
Model
Observed
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.5
1
1.5
2
2.5
x 10
-3
P
D
F

i
n

i
n
c
.

b
a
n
d

4
Model
Observed

Figure 4.3d Comparison of Number Density of Debris from Measurements
and Model Distributions in Inclination Band (73, 91]
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.5
1
1.5
2
x 10
-3
Altitude in km
P
D
F

i
n

i
n
c
.

b
a
n
d

5
Model
Observed
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
1
2
3
x 10
-3
P
D
F

i
n

i
n
c
.

b
a
n
d

5
Model
Observed

Figure 4.3e Comparison of Number Density of Debris from Measurements
and Model Distributions in Inclination Band (91,180]
139

Figure 4.4 Probability Plots of Eccentricity and Ballistic Coefficient
in Altitude Bins for the Year 2002
objects

140

Figure 4.5 Variation of Location Parameter for Eccentricity
in individual Altitude Bins

Figure 4.6 Variation of Location Parameter for Eccentricity
in Moving Window Altitude Bins
141

Figure 4.7 Comparison of Model Values of Location Parameter for Eccentricity in
Moving Window Altitude Bins and Individual Altitude Bins

Figure 4.8 Probability Plots of Eccentricity and Ballistic Coefficient in
Perigee Height Bins for the Year 2002
objects

142
Inclination Bands in Altitude Bin (700, 800] km for the Year 2002

Inclination Bands in Altitude Bin (800, 900] km for the Year 2002

Chapter 5 Evolution of the orbital debris scenario based on Kalman filter approach

143
CHAPTER 5

EVOLUTION OF THE
ORBITAL DEBRIS SCENARIO
BASED ON KALMAN FILTER APPROACH
(FAST NUMBER DENSITY SIMULATOR AND EQUIVALENT
BALLISTIC COEFFICIENT ESTIMATOR)

5.0 PROLOGUE

In this prologue the basic methodology that is desirable in order to predict the long term
evolution of many complex systems and in particular the scenario of the space debris is
provided. Subsequently the basic concepts of the Kalman filter and its implementation in
particular utilising the constant Kalman gain approach is presented. Such an approach is able
to track the evolution of grouped orbital debris fragments in the semi major axis and
eccentricity bins as well as estimating a suitable equivalent ballistic coefficient, which is an
expansion of the scenario. The proof of the above concept is provided based on
computationally intensive simulated studies.

There are many examples in the literature dealing with long term prediction of very complex
systems such as the atmosphere by Bouttier and Courtier (1999), in economics by
(Makridakis and Wheelwright (1989), population and even the space debris scenario (Rossi
et al 1997). This last one has been briefly considered in Chapters 1 and 2. As mentioned there
such studies are useful up to a certain extent only in providing a qualitative feel for the kind
of scenario that could exist after a long time. In all these studies one has to necessarily
assume a model to describe the time evolution. Obviously the long term predictions depend
on the qualitative structure and the quantitative parameters in the assumed model. It is
possible that at large times the prediction could depart greatly from the actual scenario due to
the sensitivity of the evolution to some of the parameters.
objects

144

The only way that the prediction can be made to follow more closely the real situation is to
update the predicted value of the states, based on the available measurements at later times.
Such updated states can be utilized for further evolution in time and later updated using
subsequent measurements as and when they become available. Another way of interpreting
this is to say that for proper progress theory and experiments should go together! Such
systems are examples of a stochastic corrective process (Narasimha 1975). In these
processes generally the states and even the parameters in the model can be updated based on
measurement (or information) at later times. Such a fusion of the state estimates in general
from the state and the measurement equation as also the parameter updates (if necessary) can
be achieved with appropriate weighting by using the Kalman filter. Tuning the various
parameters in the design of the Kalman filter carries out the above assignment of the
weightage. However people working in the area of Kalman filter know that this is the hardest
task in the filter design and has become a roaring industry (Ananthasayanam 2004). Another
well known problem is the fusion or mixing of the data from GPS and INS (Strang and
Borre 1997).

5.1 INTRODUCTION

The present section describes the methodology proposed for the tracking of the group of
evolving debris fragments of various sizes and also estimate suitable equivalent ballistic
coefficients for each such group of fragments. The reasons for introducing the so called
equivalent ballistic coefficient for a group of fragments are discussed later. Ananthasayanam
(2001) has mentioned some important reasons for compulsively following a probabilistic
estimation theoretic approach to handle problems with uncertain states and noisy
measurements. These are

(i) Uncertainty
(ii) Variability
(iii) Complexity

145
(iv) Enormity and
(v) Even deterministic problems could be modeled as probabilistic and treated as
such.

Some of the most difficult issues in the study of the LEO debris scenario are

(i) The unpredictable on orbit breakups occurring at random location, intensity and
directionality, generating debris clouds made up of many fragments,
(ii) A large number of presently existing debris fragments, running into thousands for
large objects and millions for the smaller ones,
(iii) The uncertainty of the fragments physical characteristics such as the mass, shape,
and size,
(iv) The uncertainty of the more important characteristic such as the ballistic
coefficient, which is responsible for the aerodynamic force on the fragment due to
the air density and thus change its orbital characteristics,
(v) The variability of the perturbations due to air drag, gravitational anomalies, luni
solar perturbations and solar radiation pressure all of which exist all the time,
(vi) These debris clouds take about a few years to merge with the background debris
scenario,
(vii) The uncertainty in the various measurements by radar, optical telescopes and
retrieved satellites and space objects, and
(viii) Any simulated laboratory experiment provides only a sample characteristic of the
breakup.

The scenario of the orbital space debris involving the

(1) Propagation of the state equation describing the orbital characteristics with
time and
(2) Measurements made on them by tracking at various times

objects

146
possess all the above mentioned features. As shown in Figure 2.3 there are large differences
in the number density of the debris among the many models. These models are generally
derived by simulating the various historical breakups, assuming some growth rate due to later
breakups, and also utilising the measured data. The present approach differs from the earlier
studies by providing a systematic way of assimilating the measurement data at various later
times with the present debris scenario. Further it also helps by such an assimilation of the
data to expand the scenario to estimate the equivalent ballistic coefficient. This is just the
right kind of situation for utilising the Kalman filter approach with uncertain or unmodellable
state and measurement equations as well as unknown noise statistics but still be able to
follow the time evolution and derive other debris characteristics.

5.2 PRESENT MODEL

Consider the specific example of any satellite moving in space, which is continuously under
the influence of various perturbing forces. When the orbital parameters are estimated
sufficiently accurately and provided to the user, it is possible track it. However this can help
only for about a week or so after which its orbital characteristics should be once again
estimated. For continuous tracking the update of the orbital parameters at various time
intervals based on the recent measurements are necessary. If this is the prevailing situation
for a single object namely, a satellite, one can imagine the complexity of the task that is
involved in tracking thousands of large and millions of small fragments of the space debris.

5.2.1 CONCEPT OF BINNING

The concept of binning can be understood or interpreted as follows. Consider the molecules
making up a fluid volume. We can think of the fluid volume to be fixed in space though the
molecules move in and out of it. The gross macroscopic property such as air density, mean
velocity or temperature are the ones in which one may be interested instead of the
microscopic individual mass and velocity of each of the molecules making up the fluid
volume. Thus a compression of information is taking place. Similarly when we consider the

147
debris in fixed (a, e) bins, we are handling some representative average (defined
appropriately) property. In the present case we do not have the luxury of a very large (say
millions) molecules, and thus a good situation as in fluid flows. Due to only hundreds and
thousands of debris fragments the error due to coarseness of the (a, e) bin does remain and in
order to improve the estimates and limit the error one needs periodic update through
measurements.

5.2.2 CONCEPT OF EQUIVALENT FRAGMENT AND ITS PROPERTIES IN
TERMS OF THE INDIVIDUAL FRAGMENT CHARACTERISTICS

A simplification that is possible as mentioned in the previous section is to bin the fragments
in discrete intervals of semi major axis and eccentricity. Then instead of handling each and
every fragment, the conglomeration of all the fragments in each of the above bins is handled
as an equivalent fragment (EQF). Thus it becomes necessary to assign suitable,
representative, useful values for the EQF properties in terms of the individual fragment
properties. Such an assignment is to help in following the dynamics of the conglomeration in
terms of the EQF. This presents a deep conceptual problem. Even the well known mean can
be arithmetic, geometric, or harmonic or any other. A priori how well such a mathematically
defined quantity helps to follow the dynamics is not clear. Further the numerical value of
such a defined quantity is later updated in the filter (!) Does this imply the updated values
can be equal to the defined quantity all the subsequent times? As is to be expected it is not so.
Hence at best we can utilize the initially defined and the subsequently updated values to help
us track the conglomeration in some overall, rapid, and efficient way. This is just the reason
for us to call the ballistic coefficient of the EQF as the equivalent ballistic coefficient.

In the present case we initially assign for the EQF the arithmetic mean for the semi major
axis and the geometric mean for both the eccentricity and the ballistic coefficient based on
the corresponding values of all the individual fragments. Further since at all times the limit of
the bin size for the first two do not change it is the value for the ballistic coefficient that has
to be reassessed at various times as the scenario evolves in time. At various times the
objects

148
equivalent ballistic coefficient of each EQF is updated based on the weighted average of the
predicted and measured number density of the fragments in the bins. This weightage is the
Kalman gain, as we will see later on. What then is the advantage of the filter? It will be
shown later that if there is no update of this equivalent ballistic coefficient of these EQF then
there is a drift with time of the predicted number density from the true values and when there
is an update the filter is able to follow closely the true time variation of the number density.

5.2.3 EVOLUTION AND UPDATE OF THE EQF CHARACTERISTICS

The present model carries out the evolution of the debris objects in two steps, namely

(i) The propagation of the EQFs representing all the fragments in the various bins.
The next task is to redistribute the fragments with some weightage around the
adjacent bins based on some simple and reasonable rule. Further if there are any
further breakups these are also accounted for the changed number density in the
various bins.

(ii) Provide an improved estimate of the number density of the fragments in the
various bins and also update the equivalent ballistic coefficient of the EQFs both
by utilizing appropriate constant Kalman gains. The constant gains are obtained
by minimizing a suitable cost function. This minimization is presently achieved
by using the Genetic Algorithm (Goldberg 2000, Deb 1995 and Whitley 1993).

The first step is somewhat similar to the treatment in STAT (STochastic Analog Tool) of
Rossi et al (1994, 1997). The STAT code simulates the time evolution of the debris
environment by considering the number of objects contained in a set of discrete bins of semi
major axis, eccentricity and mass. The evolution algorithm is based upon a set of finite-
difference equations, which take into account launches, retrievals, explosions, collisions and
orbital propagation under the influence of atmospheric drag.


149
In STAT the bins are provided by a three-fold subdivision namely (i) in the semi major axis
(from 6378 to 46378 km), (ii) eccentricity (from 0 to 1) and (iii) mass (from 1 mg to
10,000 kg). The state variables are initialized at the actual population, which is taken from
the CNUCE 1994.0 Orbital Debris Reference Model (1994). Once the variables have been
initialized, the calculation and summation of different contributions at each time step is
performed. It propagates the orbit of an individual representative object for each bin and
therefore the computing time is almost independent of the dimension of the population.

It may be noted that the present approach considers a, log (e) and log (B) for the three fold
division as against a, e, and m of STAT. Presently this third parameter B has been
used since the orbital parameters are sensitive to the air drag and thus change with time.
When each of the fragments change their orbital parameters it is easy to see the (a, e) bin to
which they move. But in the present case for the moves of the representative EQF some kind
of a rule is necessary to account for the dispersion that would have occurred otherwise. This
rule is explained later but it is the same as in Rossi et al (1997). A question that would occur
to anybodys mind is that there could be errors due to discretization and approximations in
specifying the mean values for a and e in the various bins and further there could be errors
in propagation due to forces not taken into account or properly accountable or even
modellable at all. Such errors can be accounted for by process noise in the state equations.
An important feature of the present approach as in STAT is that since the individual
representative object of each bin is propagated the computing time is almost independent of
the debris population size.

It is the second step that is fundamentally new and different in the present approach. The KF
approach helps to not only assimilate the measurement information but also expand the
scenario (Ananthasayanam 2004) that has not been considered in earlier studies. In general
for Kalman filter implementation it is necessary to account for both the process and
measurement noises as shown in Appendix 8. The relative weightage or in other words the
Kalman gains provided by the filter depends on the combination of the statistics of the
process and measurement noises. Thus instead of handling both the kinds of noises it is
objects

150
efficient and expedient to follow the route of constant Kalman gain. This constant Kalman
gain approach used successfully by Philip (2001) for the problem of rendezvous and docking,
and Anilkumar et al (2003) for reentry studies of space objects has been used for the present
evolution of the debris scenario as well. In this second step apart from assimilating the
information content provided at later points of time from the measurements it also expands
the scenario to estimate the equivalent ballistic coefficient for the EQFs.

5.3 ESSENCE OF THE CONCEPT OF USING CONSTANT KALMAN
GAIN

In order to provide such of those readers who are not familiar with Kalman filtering concepts
the necessary fundamentals of the filter (Kalman 1960, 1961 and Gelb 1970), its design
aspects and in particular the basic reason for using the constant Kalman gain approach are all
provided in Appendix 8 through a simple example. For the present consider a single variable
with the state also being the measured quantity. Summarizing the broad ideas one may keep
in mind that when the Kalman gain is unity the prediction follows the measurement and there
is no improvement in predicting the state. Thus the measurement is trusted fully. If the
Kalman gain is zero then there is no update of the state from the initial condition. In this case
the initial state is fully trusted. If the gain is set to very small value there are instabilities
based on using each measurement as can be seen in Figures 3 and 4 of the Appendix 8.
However if an appropriate gain is used it helps to decrease the effect of the noise in the
measurement which means there is filtering. But due to the presence of the process or state
noise the estimation accuracy cannot be decreased below a certain minimum value. Only if
the process noise is zero the estimation accuracy can decrease to zero with more and more
measurements.

The next question is what happens if the signal is not constant but there is some system
dynamics meaning the state is varying with time? Here again the situation is similar if the
gain is unity and zero. However with a small but finite gain the filter takes some transient
time to start following the time varying state and later filter the noise in the measurement,

151
which means the steady state error, is reduced. If the above finite gain is somewhat increased
then it would take a shorter transient time to commence following the time varying signal
and later decrease the state error. As in the case of constant signal the steady state error with
smaller gain has a longer transient with smaller steady state error and vice versa with higher
gain. Obviously any arbitrary time varying signal will be followed through the spells of
transient and attempt to reach a lower steady state error. Thus it is possible to choose a
suitable value for the gain such that the dynamics is followed fairly well and the noise is also
filtered.

It turns out based on the simple experiment in Appendix 8 that the gain over a reasonable
range of values is robust to follow the dynamics of the signal and also reduce the noise in the
data. This is just the essence that would help to follow the subsequent discussion.

What has really happened is that we are using the more robust variable namely the Kalman
gain in place of the different statistics of the process and measurement noise whose
combination eventually leads to the gain. Hence one can consider a simple simulated
scenario, work out the gains and later use it in a slightly different case due to its robustness.
Such a feature has been demonstrated in the earlier work of Philip (2001), and Anilkumar et
al (2003). There is another great advantage in using the constant gain approach namely one
need not propagate the covariance of the state equations, which is numerically highly time
consuming.

5.4 IMPLEMENTATION OF THE CONSTANT KALMAN GAIN
APPROACH

In order to implement the constant Kalman gain approach in the filter, the three-fold
subdivision is of the following:

(i) the semi major axis (from 6378 to 46378 km),
(ii) the eccentricity (from 0 to 1), and
objects

152
(iii) the ballistic coefficient.

The semi major axis is binned in linear scale and the eccentricities and the ballistic
coefficients are binned based on the logarithmic scale. The state variables are initialized
based on the existing scenario at the initial time. Presently a large number similarly of
(10,000) debris objects generated due to an explosion is considered. At later times the
additional fragments due to further breakups are accounted for as source terms.

As mentioned earlier we have binned the debris objects comes earlier into a number of bins
based on the logarithmic scaling of the ballistic coefficient (B = C
D
A/M). The orbital
propagation of the EQF based on its assigned value of suitable a and e is carried out with
the propagator (King Hele 1987) considering only the air drag effects and it could in general
end up in just within another bin. However the various fragments of which it is the above is
made up of could end up in many other bins as well. In order to simulate or mimic such a
feature it is therefore necessary to provide a rule to redistribute the fraction of the EQF
among the bins. Such a rule is to take the ratio between the portion of area covered by the
propagated rectangle and the area of the rectangle itself as shown in Figure 5.1. Finally, the
contribution of the changed number density, and their orbital parameters and the ballistic
coefficient due to the launches and retrievals are also accounted.

5.5 THE REAL WORLD AND THE FILTER WORLD SCENARIO

In order to appreciate the procedure envisaged in the present treatment, the state and the
measurement equations in the real world and filter world scenarios are provided here in the
following Tables 5.1 and 5.2.


153
Table 5.1 Real World Scenario without Binning

The states
variables
Semi major axis, eccentricity and ballistic coefficient of each and every
one of the fragments
The state input Some suitable models for the complex environment
The state noise Random variations in real situation from the above model environment
The initial state Fragment characteristics obtained from the ASSEMBLE or any other
suitable model
The measured
variables
Orbital characteristics of each of the fragments.
The measurement
noise
Measurement errors in the above orbital characteristics due to tracking
and data processing.

Table 5.2 Filter World Scenario with Binning

The state variables Semi major axis, eccentricity and equivalent ballistic coefficient of the
EQF in each of the bins propagated and redistributed and further
accounting for fragments from additional breakups
The state input Some suitable model for the complex environment
The state/process
noise
The unmodellable inaccuracies in assigning the mean or average
values for (a, e, B) for the EQF and its further propagation.
The initial state Fragment characteristics obtained from the ASSEMBLE or any other
suitable model with arithmetic mean values for the semi major axis
and geometric means for the eccentricity for e and B of the
individual fragments making the EQF.
The measured
variables
The semi major axis, eccentricity and ballistic coefficient of each of
the fragments as in the real world but propagated with atmospheric
drag alone and later assigned to the various bins.
The measurement
noise
In the present simulated scenario there is no measurement error since
all the fragments are propagated and based on the changed values of
the orbital parameters (a, e) they are assigned to the appropriate bins.
objects

154

It is very interesting to see that for the present study based on the filter world scenario the
coarser conditions brought about by the binning have formed the state equations and the finer
conditions without binning has led to the measurement equations. The process noise arises
from binning and assigning average values, and there is no measurement noise.

Under these conditions it is useful to invert the role of the above state and measurement
equations (!) Thus we reach a position like in OEM where the state equations have no error
but the measurement equations have an error. Thus in the present inverted roles of the state
and measurement equations the measurement noise arises from binning and assigning
average values. Now as in the example problem treated in Appendix 8 the choice of the
various constant Kalman gains tries to reduce the effect of the measurement or process noise
and estimate an improved number density.

We now explain the intricacies in the estimation of the equivalent ballistic coefficient of the
EQF corresponding to various (a, e) bins. After update the value of B for the EQF at times
can fall outside the bin interval (!) Such a thing is surprising since we mentioned earlier that
range of the B bins are fixed. The equivalent B formed out of individual fragments cannot
have the ballistic coefficient outside this range. However there is subtle approximation that
has occurred. Each time the EQF is propagated it is from the location of the arithmetic and
geometric mean of the (a, e) bin. There is a loss of information with regard to the true (a, e)
coordinates of the individual fragments. The propagation of the EQF is always from the
above initial condition based on the mean. But for a conglomeration of fragments this may
not be the best or appropriate. The EQF could have started its trajectory from the corners or
the edges of the (a, e) bin at which time the redistribution could have been different and so
also the updated ballistic coefficient B.

155

5.6 PROPAGATION OF THE ORBITAL CHARACTERISTICS OF
THE FRAGMENTS IN THE BINS AND THE EQF

This section provides the state equations for the fragments and EQF and their propagation.
The basic dynamics of equations involved in the orbital motion (King-Hele 1987, Wiesel
1989, Sharma et al 2003) are outlined in Appendix 3 providing a detailed description of the
perturbation forces acting on the satellite orbits and the governing equations of the decay due
to atmospheric air drag, the J
2
effects of gravitational anomaly, other effects such as solar
radiation pressure, luni solar perturbations and so on. For a quick reference, the rate of
change over one revolution in the elements a and e due to drag alone for the k
th
revolution
is given by

a
k
= a
k-1
+ a
k-1

e
k
= e
k-1
+ e
k-1

where a
k
, e
k
stand for semi major axis and eccentricity at time instant, i = k-1, k with
c
EXP
e O J J J
e
J J
e
J J
e
J
p
a
rev
e
c
EXP
e O J J
e
J J
e
eJ J
p
a
rev
a
+ +
+ + +
=
+ + +
+ + +
=
)
4
( )
4 2
4
01
5 (
16
3
)
3 1
5 (
8
2
)
2 0
(
2
1
2
)
4
( )
3 1
3 (
4
3
)
2 0
(
4
2
3
1
2
0
2
2

objects

156
with

p
= Density at the perigee height,
c = ae/H, where H is the density scale height,
= (QC
D
A
eff
) / m,
Q = Factor which includes the rotation of the atmosphere (0.9 < Q < 1.1),
m = Mass of the satellite,
Js = Modified Bessel function of first kind of order s, in all cases having
argument c,
C
D
= Drag Coefficient usually taken as 2.2, and
A
eff
= Effective reference area.

5.7 KALMAN FILTER APPROACH (TO ASSIMILATE THE
MEASUREMENT INFORMATION AND EXPAND THE SCENARIO
TO ESTIMATE THE BALLISTIC COEFFICIENT OF THE EQFS IN
THE VARIOUS BINS)

The present section explains how the measurements can be assimilated and the scenario
expanded to obtain the equivalent ballistic coefficient. Assuming that the number of objects
in each three-dimensional (a, e, B) bins are known, the predicted numbers of objects using
the above procedure are updated using Kalman filter approach.

There are a total of (10 x 10 x 10) bins, with 10 divisions in each of the parameters a, e and
B, formed by splitting the range of a, e and B. One important simplification that makes the
problem tractable is that the various slices of the B bins can be considered to be
independent. This is similar to a single object with a constant value of the ballistic coefficient
moving in the atmosphere. In general it is possible for the space objects to change their
orientation in the orbit, or the gas-surface interaction thus altering the drag coefficient and
hence the ballistic coefficient.

157

5.7.1 FILTER WORLD STATE EQUATIONS

Here the attention can be given to various EQFs. Each one of such EQFs are specified by the
parameters

(1) The number of fragments in the particular (a, e) bin,
(2) The equivalent semi major axis
(3) The equivalent eccentricity and
(4) The equivalent ballistic coefficient.

The first parameter is clear and the second is arithmetic, with the third and fourth being the
geometric mean of the fragments in the bins. The (a, e) bins are not changing and the EQF
moves in the (a, e) plane like any other single fragment (SF) and later gets redistributed
based on a certain rule. Thus the states presently considered in each and every one of the (a,
e) bins are

(i) The number of objects N and
(ii) Their equivalent ballistic coefficient B

The measurements are only the number of objects in the above (a, e) bins. The state
equations for the EQF in the various bins between measurements are

dN /dt = (propagation across the (a, e) bins + redistribution + source terms)
+ state noise
dB/dt = 0 + state noise

objects

158

5.7.2 FILTER WORLD MEASUREMENT UPDATE EQUATIONS

The measured quantities are the number density N
M
only. These are obtained by propagating
each and every one of the individual debris fragments. It is this that is used to update the
predicted number density from the EQF propagation and redistribution to an improved value
and also update the ballistic coefficient of the EQF in the various bins based on the equations
such as

N
+
= N
-
+ K
N
( N
M
- N
-
)
and B
+
= B
-
+ K
B
( B
M
- B
-
)

with the superscript (+) and (-) denoting respectively the post and pre updated values. The
constant Kalman gains K
N
and K
B
correspond respectively to the number density and the
equivalent ballistic coefficient. In the constant gain approach the number of gains to be
estimated is thus 200 based on 2 namely K
N
and K
B
for each (a, e) bin. These have to be
obtained by minimizing the cost function

J
D
= (1/N) v
k
T
[]
- 1
v
k,

where the innovation
v
k
= ( N
M
- N
-
) = (measurement - model output),

and is the covariance matrix of the innovation.

These gains have been obtained by using the Genetic Algorithm (Whitley 1993, Deb 1995
and Goldberg 2000, Appendix 7). In the GA each population will thus have 200 sets of gains.
The values of the different parameters utilized in the GA implementation after some trial and
error are provided in Table 5. 3 below.


159
Table 5.3 Parameters Utilized in the Implementation of GA

Parameter Values
Population Size 200
Bit Length 20
Probability of Cross Over 0.90
Probability of Mutation 0.05
Number of generations for Convergence 50
Alternate Criteria for Convergence: Change in J
D

between generations
0.0001

The major part of the simulation time is taken up to simulate the trajectory of each and every
one of the individual fragments, which are about 10,000 in the present case. Later the 100
EQFs have to be propagated in each slice of B to form the cost function. Obviously this has
to be repeated over the population size and generations. It takes about 100 hours of computer
time in a MATLAB environment (though not CPU) using a 1.7 MHz PC to obtain the results
of each of the test cases discussed here. Though such a computing time is not presently
uncommon the present interest is to derive the constant gains by following one single
breakup alone and subsequently using the same gains in order to demonstrate the adequacy to
follow the evolution with more breakups, and source terms. Such an approach shows the way
in which a certain model of the space debris scenario can be updated.

5.8 CASE STUDIES AND THE ANALYSIS OF THE RESULTS

Firstly we consider the evolution of the debris clouds generated due to one single explosion.
Secondly we add more breakups and further source terms at arbitrary times during the
evolution time.

objects

160

5.9 EVOLUTION OF DEBRIS OBJECTS GENERATED DUE TO
EXPLOSIONS

In order to study the effect and the soundness of the above approach, a single explosion is
simulated using the ASSEMBLE model (Anilkumar et. al 2002) given in Chapter 3.
Considering an explosion, at a typical altitude of 800 km and eccentricity 0.00045, simulates
the 10,000 objects of different ballistic coefficients. These objects were propagated by taking
into account only the atmospheric drag effects (King Hele 1987) for a period of 600 days and
the basic measurement data of the number of objects in the three-dimensional (a, e, B) bins
are generated.

5.9.1 EVOLUTION OF A SINGLE BREAKUP

Based on this model three different possible approaches to follow the number density of the
fragments are

(i) Updating number of objects N alone,
(ii) Updating only the equivalent B, and
(iii) Updating both the states N and the equivalent B.

In all the subsequent figures the symbol (o) denotes the measurements which in this case
becomes the state and with no error at various times. Hence the measurements represent the
true values of the number density. The state with noise due to error as mentioned earlier
becomes the measurement with noise. Due to the filtering process the error can be reduced as
best as is possible but it cannot be reduced beyond a certain value due to continuous
occurrence of error due to binning, propagation and redistribution. The dashed line (----)
represents the results without any update, meaning the Kalman gain is zero and thus at all
times the update equals the prediction. The continuous line shows the present results from the
filter.

161

(i) Updating Only the Number Density

The first case led to most of the constant gains being close to unity of around 0.95 and the
results slowly diverge away from the true values. Hence without the estimation of the
important parameter namely the equivalent ballistic coefficient the results do not have much
value.

(ii) Estimating the Number of Objects by Updating the Ballistic Coefficient Alone

The bins with large B have smaller number of fragments and the bins with very small
values of B are not sensitive to drag and thus hardly change their orbital parameters and
hence it is not easy to update their values. Only some of the intermediate bins with a fair
number of fragments and responding to the drag are amenable for appropriate updates of the
B are the ones discussed in the following sections. Thus the present results are mostly
discussed across the (i) semi major axis bins between the intervals (400,550), (550,700),
(700,850), and (850,1000) which have generally have a reasonably large number of
fragments.

The Figure 5.2 provides the results for the above case by updating B alone for the lower B
bin (0.0301, 0.0818). Here the number density is not considered as a state and the only state
is B. Even though the number densities of the fragments are not updated, in such of those
cases where there are many fragments (as between 700 km and 850 km) the filter estimates
are close to the measurements. Also note that there is a definite improvement in the
estimation for the number of objects over the one without filter update. The next Figure 5.3
corresponding to the medium B bin (0.0818, 0.2226) shows the estimates are better with
filtering than without filtering and in particular with more fragments. The Figure 5.4 for the
higher B bin (0.6056, 1.6476) shows a similar behavior as in the previous case.

objects

162
(iii) Estimating of the Number of Objects by Updating both the Number Density and
the Ballistic Coefficient

The next three Figures 5.5 to 5.7 show the results for the third case where the filter updates
have been provided for both the number density and the ballistic coefficient and filter results
are closer to the true values. The Figure 5.5 provides the results for the above case updating
for the lower B bin (0.0301, 0.0818). It may be noted that at all perigee altitudes the filtered
estimates are closer to the true number of objects shown by the (o) symbol. The Figure 5.6
for the medium B bin (0.0818, 0.2226) shows consistently the results with filtering are
much better and closely follow the true number density than without filtering. The Figure 5.7
for the higher B bin (0.6056, 1.6476) shows a very good tracking behavior of the filter as in
the previous case.

Thus one can summarise the best results are obtained when the number density and the
equivalent ballistic coefficients are both updated.

5.9.2 VARIATION OF THE CONSTANT KALMAN GAIN FOR THE NUMBER OF
FRAGMENTS

The variation for the Kalman gain across the semi major axes bins has been shown for four
ballistic bins in Figure 5.8. It may be noted that the Kalman gains for the number density are
always between zero and unity. This happens obviously since the state N is also the
measured quantity. The gains show some variations across the semi major axes.

However one may hope and see that some constancy appears to exist for the gains across the
a bins. Next assuming a single gain for the number density across all the a bins the GA
was run. It turned out that the filter updates were able to track the number of objects quite as
well as in the previous three cases (Figures 5.5 to 5.7) which had varying gains across the a
for the number density. The gains in this simpler case turned out to be between 0.4 and 0.6
for various ballistic coefficient ranges.

163

This happy behavior can be explained as follows. In Kalman filter the constant gain are more
robust variables than the values of the statistics of the initial state, process noise, and
measurement noise covariance values. The different combinations of the above noise
statistics can lead to the same gain. Further even the constant Kalman gain appears to be
robust in its own right in the following way. When the filter was run with modest changes in
the constant gains generally one may notice that there is not much of a change in the results.
Such a behavior has been reported by Phillip (2001) while dealing with the rendezvous and
docking problem. The advantage of such a result is that much fewer numbers need to be
stored to run the Kalman filter and in particular on line real time applications.

In contrast the behavior of the constant gains for the ballistic coefficient shown in Figure 5.9
are somewhat different. Firstly the gains can be positive or negative at various values of the
semi major axes. This happens because the ballistic coefficient is not a state that has also
been measured. Secondly it is not possible to obtain a single value for the constant gain for
the update of the ballistic coefficient all across the semi major axes. Though formally such a
value was estimated the match of the results for the number density was noticed to be far
from satisfactory. The Figure 5.10 shows variation of the estimated ballistic coefficient with
time in four typical bins. It may be noted that the estimates are generally within the limits of
the ballistic coefficient bins. However at times they move somewhat outside the limits of the
bin values. This can be explained as perhaps due to the initial condition for the propagation
of the EQF is always from the specified mean values of the (a, e) bins. It could have started
from anywhere including the corners or the sides. The subsequent propagation and
redistribution error that this implies when compared to the true initial conditions of the
individual fragments could be responsible for such a behavior.

Further we wish to reiterate as mentioned earlier even the appropriateness of the mean as is
presently defined cannot be expected to describe correctly the propagation and redistribution
of the EQF at all times. In this sense the best interpretation is to think of the various means
defined presently as a formal mathematical specification having some ability to mimic the
objects

164
dynamical behavior. At best it can be an acceptable choice that helps in mathematical
tractability and provide reasonable results.

5.9.3 EVOLUTION OF A SINGLE BREAKUP FOLLOWED BY ANOTHER
BREAKUP

This forms the test case for the results and in particular the various constant Kalman gains
derived from the previous case of an evolving breakup followed for 600 days. The values of
the constants have been utilized even in the present scenario namely after the first breakup at
the beginning, the second breakup was introduced after 120 days. One might be tempted to
think that should all the results including the constant Kalman gains be reworked? The
answer to that is provided from the simple dynamical scenario considered in Appendix 8. It
turns out to be not necessary to rework the results for the constant gains. The derived values
from the earlier case are able to follow the present evolution involving two breakups and are
demonstrated in the latter sections. Slower changes in the dynamical behavior can be tracked
using the earlier values of the gains. After 10,000 fragments were introduced at the beginning
and additionally 300 fragments were introduced after 120 days. Such a situation is similar to
what is happening in real world scenario of debris creation. The debris is growing but not at a
too rapid a rate when compared to the existing population.

The constant Kalman gains obtained based on the debris cloud evolution have been utilized
even after adding the source terms. The robustness of the constant gains, in other words the
performance of the filter around a range of the estimated gains, is the reason behind the
above behavior, even if the results are non optimal, they will be sufficient for the proper
estimation. The behavior of the filter is shown in Figures 5.11, 5.12 and 5.13 for the three
ballistic coefficient bins, namely lower, medium and higher, respectively. It may be noted
that the filter tracks the dynamic variation very well in particular soon after the new breakup
in all the ballistic bins.


165

5.9.4 EVOLUTION OF A SINGLE BREAKUP FOLLOWED BY MORE THAN
ONE BREAKUPS

Further the concept of utilizing the constant gains obtained for a single breakup for tracking
slowly varying further breakups is demonstrated with the introduction of two breakups at
different time steps, namely at 120 days and another at 330 days. The Figures 5.14 and 5.15
for two typical lower and higher bins show the behavior of the filter in both these cases can
be described as quite good. It can be noticed that the filter without any update is just not able
to follow the true number density values.

5.9.5 EVOLUTION OF A SINGLE BREAKUP FOLLOWED BY MORE THAN
ONE BREAKUPS AND SOME LAUNCH ACTIVITIES

Further in order to simulate the real world scenario of space debris sources, some launch
activities; say satellites and rocket bodies are introduced during the evolution of the breakup
clouds. Here we added at random times a maximum of 5 objects (0 to 5) together with the
latter two breakup sources. It may be noted that once again the constant gains obtained using
the primary debris clouds alone are the ones that are continued to be utilized in all the later
cases as well. The results of this study are provided in the Figures 5.16 and 5.17. All these
figures show that the present model is able to track the number of objects in the evolution
process and also update the ballistic coefficient B. The behavior of the filter even in such
cases is quite good.

5.10 INFERENCE OF FRAGMENT PROPERTIES FROM THE EQF
CHARACTERISTICS

At this stage one might think whether it would not be possible to infer the fragment
characteristics from that of the EQF. One may see straightaway that in forming the EQF there
is some loss of information. Hence if the inverse problem has to be solved then it would
objects

166
involve putting some information, innovation, or assumptions into the EQF characteristics.
Perhaps if there are some well known asymptotic properties like in a gas in thermodynamic
equilibrium at a given temperature one can then say that the distribution of the individual
velocities follows a Maxwellian distribution. But no such results are presently available in
the present situation for orbital debris. Also we have seen earlier that some of the
characteristics of the debris obey certain distributions such as Normal, Lognormal, or
Laplace and so on. Generally with just one single parameter at our disposal it is not possible
to hazard a guess regarding the probability distribution.

If instead of using the constant Kalman gain approach the more involved approach of
working with the statistics of the initial state, process, and measurement noise covariances
we could have arrived at another parameter characterizing the dispersion of the characteristic
parameters. Here we have the subtle feature of the Kalman filter describing the result in
terms of the normal distribution with the estimate and the covariance for linear systems. For
nonlinear systems the normal distribution approximates the behavior of the estimates. Even
then the computational burden of handling the present problem in terms of the above
statistics is far more. Such a route would appear defeat the very purpose of simplicity of the
present constant gain Kalman filter approach. Perhaps the present approach is just about the
simplest providing the best possible results.

5.11 CONCLUSIONS

A novel innovative approach based on the propagation of an equivalent object in a three
dimensional bin of semi major axis, eccentricity, and the ballistic coefficient (a, e, B)
together with a constant gain Kalman filter technique is described in this report. This new
approach propagates the number density in a bin of a and e rapidly without propagating
each of the space fragments in the above bins. Further this approach expands the scenario to
provide suitable ballistic coefficient values for the EQFs in the various bins. Also it is able to
assimilate the information with the passage of time from other breakups as well. The heart of
the technique is to use a constant Kalman gain, which is nearly optimal and able to track the

167
dynamically evolving fragment scenario and further expand the scenario to provide
appropriate average time varying equivalent ballistic coefficients for the various bins.
objects

168

Figure 5. 1 Graphical Representation of the Orbital Propagation Algorithm in the
Eccentricity e vs. Semi Major Axis a Space.
Log( e)
a

169

Figure 5.2 Updating the Ballistic Coefficients only Using Constant Gains
for B bin (0.0301, 0.0818)
(o) True; (-) Filter; (---) No update
objects

170

for B bin (0.0818, 0.2226)


171

for B bin (0.6056. 1.6476)

objects

172

Figure 5.5 Updating the Ballistic Coefficients and Number of Objects Using
Constant Gains for B bin (0.0301, 0.0818)

173

Figure 5. 6 Updating the Ballistic Coefficients and Number of Objects Using
Constant Gains for B bin (0.0818, 0.2226)
objects

174

Figure 5.7 Updating the Ballistic Coefficients and Number of Objects Using
Constant Gains for B bin ( 0.6056, 1.6476)


175

Figure 5.8 Kalman Gains for Number of Objects (K
N
)

objects

176

Figure 5. 9 Kalman Gains for Ballistic Coefficients (K
B
)

177

Figure 5.10 The Estimated Ballistic Coefficients
objects

178

Figure 5.11 One Source of Explosion Included. Updating the Ballistic Coefficients and
Number of Objects Using Constant Gains for B bin (0.0301, 0.0818)

179


objects

180


181

Figure 5.14 Comparison between Estimated and Observed Number of Objects:
Two Sources of Explosions Included. Updating the Ballistic Coefficients and Number of
Objects Using Constant Gains for B bin (0.0301, 0.0818)
objects

182

Two Sources of Explosions Included. Updating the Ballistic Coefficients and Number of
Objects Using Constant Gains for B bin (0.6056, 1.6476)

183

Two Sources of Explosions and Some New Launches Included. Updating the Ballistic
Coefficients and Number of Objects Using Constant Gains for B bin (0.0301, 0.818)

objects

184

Figure 5.17 Comparison between Estimated and Observed Number of Obects:
Two Sources of Explosions and Some New Launches Included. Updating the Ballistic
Coefficients and Number of Objects Using Constant Gains for B bin (0.6056, 1.6476)
Chapter 6 A reference collision probability estimation model for target risk assessment due to orbital debris

185
CHAPTER 6

A REFERENCE COLLISION
PROBABILITY ESTIMATION
MODEL FOR TARGET RISK ASSESSMENT DUE TO
ORBITAL DEBRIS

6.1 INTRODUCTION

Any object placed in an orbit around the earth has some small risk of colliding with other
orbiting objects, including the debris fragments. For mission planning, safety or mitigation
reasons, it is necessary to estimate the level of risk by determining the probability that a
given target spacecraft object will collide with currently orbiting debris. The presently
available database for such an assessment is the catalogued data of the orbital parameters of
trackable objects through USSPACECOM.

Many different approaches with varying levels of sophistication have been developed in the
literature to estimate the probability of collision of target objects with orbital debris. As is
well known in probabilistic calculations (Gnedenko 1973) it is essential to define
unambiguously the sample space scenario in order to draw proper and not conflicting
conclusions from such analysis. In order to achieve the above needed clarity we formally
define in the next section certain quantities used in the later discussions for the benefit of the
reader. This is followed in the subsequent section by a discussion regarding the various
earlier approaches that could be broadly classified in terms of the scenario and the
sophistication into five levels including the present reference model for estimating the
collision probability.

objects
186
6.2 DEFINITION OF USEFUL TERMS

The target bodies such as an orbiting satellite, Space Shuttle, or the International Space
Station (ISS) are generally much larger in size than the various debris fragments. Their
projected area is denoted by A. It is useful at times to calculate the collision probability for a
unit projected area of the target body.

The orbital debris fragments being much smaller in size than the target body can be
considered to be like a stream of particles and if they move towards the target then a collision
can be said to have occurred.

The spatial number density S of the orbital debris denotes their number for unit volume. This
is expressed in the unit (km
-3
).

The relative velocity between the target body and the debris fragment is written as V and
expressed in the unit (kms
-1
).

The flux of the debris is the number of fragments sweeping a given area per unit time. It is
given by the product S.V and its usual unit being (m
-2
yr
-1
). Hence this represents the number
of debris inside a volume swept by them during a period of one year. This can also be
interpreted as the number of fragments moving towards a unit area per unit time.

If the above flux is multiplied by the projected area of the target body A it gives the
quantity A.S.V. Obviously since the fragments are treated like small particles it is equal to
the number of fragments crossing the target implying a collision. Its unit is yr
-1
. Then if we
wish to know the number of collisions (in other words the collision frequency) during the
period of as many years it is given by A.S.V.T where T is in years.

Now to summarizing the above we have,


187
Spatial number density = S, with units (km
-3
)
Flux of the debris = S.V, with units (km
-2
yr
-1
)
Frequency of collision = A.S.V, with units (yr
-1
)
Number of Collisions = A. S.V.T. ( nondimensional)

The third one represents the rate of occurrence of a phenomena which in this case is the
frequency of collision. If this is considered as very small representing the mean value then it
can be assumed to follow the Poisson probability distribution for no collision and for single
and multiple collisions. Hence it can be interpreted as the probability of collision. Thus the
quantity
A.S.V.T

is fundamental to all the further discussions regarding the various approaches for collision
probability calculations and they can be understood easily with the above in mind. The next
section considers the various approaches for estimating the probability of collision classified
into various levels.

6.3 DIFFERENT APPROACHES FOR ESTIMATING COLLISION
PROBABILITY

There are many approaches in the literature for estimating the probability of collision for a
target due to the debris fragments. These range from very simple as spelt out in deriving the
above to many sophisticated approaches requiring involved numerical calculations. A
detailed literature survey (Johnson 1987, Sehnal 1985, Vedder et al 1991 and Chobotov et al
1997) provides different approaches to the calculation of the risk. Some approaches are
discussed below. It is well known that any two models would not provide exactly matching
results (Jehn 1996, Johnson 1987) due to the differences in the underlying assumptions on
probability distributions, source models and terms and evolution models. The Table 6.1
provides a summary of representative approaches including the present reference proposal.

objects
188
Table 6.1 Summary of the Broad Nature of Various Approaches
for Collision Risk Assessment. P = Probabilistic, D = Deterministic.

Level Evolution of
the Debris
Scenario
Use of
B
Environment Approach
P /D
Usefulness

Kinetic Theory of
Gases

No

No

Considered
Uniform over
all Space and
Time

P

Feel for the
order of
magnitude of
collision risk

MC Analysis for the
Closest Distance.

No

No

Keplerian

P

Risk
Estimation for
an Operational
Spacecraft

The Proposed
Reference Model

Yes

Yes

Dynamic but
Simple

P

Preliminary
Design

Long Term
Propagation of TLE
Data Set

Yes

Yes

Dynamic but
Involved

D

Launch
window

Collision Avoidance
(COLA). Short Term
Propagation of TLE
Data Set

Yes

Yes

Dynamic but
Involved

D

Actual Day of
Launch
Operations


189
The main differences among these involve

(i) Considering the scenario in simple uniform distribution or that of the orbiting
debris in Keplerian orbit,
(ii) Handling the problem either as a probabilistic one or a deterministic one,
(iii) If a time evolution of the scenario is considered or not,
(iv) The eventual usefulness of the approach for getting a feel for the order of
magnitude, preliminary design, obtaining the launch window, or checking for the
collision just before launch.

We now discuss each one of the above in detail.

6.3.1 COLLISION PROBABILITY AS A POISSON PROCESS

This is perhaps the simplest approach without requiring much information. There is no
evolution of the debris scenario that is assumed to be static and uniform all over the space.
The fragments can move in any direction with equal probability but all of them are assumed
to have the same velocity V.

6.3.2 MONTE CARLO ANALYSIS TO OBTAIN THE DISTRIBUTION OF
CLOSEST APPROACH

In this approach the closest distances are calculated based on either the propagation of the
debris objects with respect to a target spacecraft at different time intervals during an orbital
period of the target body. Identifying an accurate propagator and propagating all debris
objects to an epoch considered for the spacecraft are tedious and time consuming. The basic
concept of this approach is to observe the frequency of debris approach distances to a target
on a larger, observable scale, and then to use statistical methods to extrapolate to collision
proximity distances.

objects
190
This approach has been suggested by (Vedder et al 1991). Here unlike in the previous case
all the debris move in a Keplerian elliptical path. By simulating the position of both the
target and the debris which are considered to be equally likely in their mean anomaly
position samples the distribution of the minimum distance between the target and all the
debris fragments are worked out. Next a suitable probability distribution is fitted which turns
out to follow a Weibull distribution. By calculating the parameters of the Weibull distribution
the extrapolation to very small distance of closest approach is worked out. Then this
probability for distance is transformed to derive the time the debris spends inside the sphere
of influence of the target. Later the above probability is converted to the frequency of
collision.

The generalized extreme value distribution using the Monte Carlo method approach thus
comprises of the following steps.

(1) Generate via Monte Carlo the closest approach distance between the target
spacecraft and the nearest debris object.
(2) Fit a Weibull probability density function to the closest approach data set obtained
from Step 1.
(3) Estimate the collision probability and time to collision between the spacecraft and
the debris objects under consideration.

Hence the closest approach obtained at different time intervals are modeled using the
Weibull distribution as it is a good model approach for these minimum distances and can be
used to calculate the time interval between collisions. The Weibull cumulative distribution
function

is

) ) / ( exp( 1 ) (

x x F =

where
F(x) = Cumulative probability of the minimum distance being < x

191
x = Distance between satellites
= Scale parameter
= Shape parameter

The probability of collision equal to the cumulative probability F(x) for a distance x
multiplied with the number of possible encounters. The time between collisions is simply the
inverse of the probability of collision per year.

It is observed that an MC approach based on randomly varying mean anomalies of debris
objects provide nearly the same results as in the former case (Johnson 1987, Vedder et al
1991 and Chobotov et al 1997).

It can be noted that the above analysis is for a static picture of the debris fragments existing
at a given time. But the slightly more realistic picture is to consider the debris objects in
Keplerian orbit. The evolution of the debris orbital characteristics is not considered. Hence
the information regarding their ballistic coefficient B is not needed. This may be acceptable
if the estimates are made at such altitudes of the ISS wherein the change in the orbital
scenario threatening the target object does not change over a period of time.

There are however some shortcomings of this approach. Consider all the debris to be in
circular orbit. Obviously the debris in circular orbit too decay and could collide with the
target object. However from this approach the distribution of the minimum distance to the
target object is linear (!). This approach does not provide any weightage for the relative
shorter or longer times spent by the debris with varying semi major axes or even the number
of debris. If a weightage is given for the latter feature then the distribution becomes that of
the number density of the fragments at varying distances from the target.

Hence this approach is involved extensive Monte Carlo simulation, fitting extreme value
distribution for the closest distance, converting it into time, and further not quite realistic in
predicting the time varying collision risk due to the evolving debris scenario.
objects
192
6.3.3 A NEW APPROACH FOR THE COLLISION PROBABILITY ASSESSMENT

The present new approach for the collision probability assessment utilises the simple closed
form solution of Wiesel (1989) for the lifetime estimation of decaying space objects. This
method assumes an exponential model for the density variation in the atmosphere, and takes
into account only aerodynamic drag in changing the semi major axis of a body in circular
orbit.

Thus this method accounting for the orbital decay of circular orbits due to air drag by
utilising the ballistic coefficient B which is important for the evolution of the debris
scenario appears to be a good choice between simplicity and accuracy. As mentioned earlier
in Chapter 3 a reference model should permit ease of numerical calculation and represent a
reasonably realistic scenario. Thus the above choice for the environmental scenario, and the
B seems just about simple and appropriate. The decay expressions for the eccentric orbits
are more involved. Further accounting for other perturbations due to the solar and magnetic
indices, gravitational anomalies, luni solar perturbations, solar radiation pressure, and other
effects would make the calculation complex.

Satellites in low earth orbits have finite lifetimes due to the effect of atmospheric drag.
Although the atmosphere at several hundred kilometers of altitude is an almost perfect
vacuum by earthly standards, the satellite must travel at very high speed through this low-
density medium for years. The effects of drag are cumulative and eventually become
significant. A satellite in low orbit suffers an acceleration due to drag given by

a
d
= Bv
2
= C
D
(A/m) + V
2

Where B is the ballistic coefficient, defined as B = C
D
A/m, with C
D
as the non dimensional
drag coefficient, A is the reference cross sectional area of the satellite and m is its mass,
is the atmospheric density and V is the satellite velocity.


193
The first effect of air drag on a satellite orbit is to circularize the orbit. This occurs since the
effect of drag is much more pronounced near perigee, when the satellite is deepest in the
atmosphere. The satellite is slowed at this point, lowering the apogee height until it nearly
equals the perigee height. For simplicity it is assumed that the orbits are circularized and
need to study only on the decay of semi major axis a due to the drag effect. With a circular
orbit, all the air drag force is in the tangential direction. Assuming the exponential density
model given as
=
0
e
-(r-Re)/h

where
0
is a fictitious base density of the atmosphere, R
e
is the earths radius, and h is
called the atmospheric scale height.

Assuming the circular orbit, the satellite velocity is given by

Air drag will change the specific energy of the orbit, since the drag acceleration a
d
does work
on a satellite at a rate = -V a
d
. The negative sign is necessary because the drag is in the
opposite direction to the velocity. This specific energy is related to the semi major axis by

E = - / 2a.

So by calculating the rate of change of the energy and equating this to the work rate due to
the air drag, we obtain

From the above equations, the rate of change of the semi major axis becomes
r
V

=
d
Va
dt
da
a dt
dE
= =
2
2
objects
194

The variables in this equation can be separated to yield the two definite integrals

In order to obtain an approximate solution to the above integral equation, the variables on the
left side of the above equation are changed to altitude above the surface, H= a-R
e
. Hence it
provides

The radical in the denominator is proportional to the satellite velocity, which does not greatly
increase during the last few hundred kilometers of decay. Thus R
e
+ H can be approximated
as R
e
alone. With this simplification, the integral by a simple process yields

From this simplified equation the ballistic coefficient B required could be estimated for a
satellite at an altitude H
0
to reach a target altitude H in specified time duration of say one
year. Hence it is now simple to formulate a three dimensional table of (H
0
, H, B) for a
specified time duration t. As an example, the Table 6.2 given below shows this three way
table formulated for one year duration. This table provides a three dimensional view of the
ballistic coefficients required by the debris objects to pass through or impact the target
altitude in an year duration. Here target objects are considered at altitudes 200 km to 290 km
in an interval of 10 km, and debris altitudes are considered from 500 km to 590 km in an
interval of 10 km. It may be pointed out that such a concept of generating a three way table
h R a
e
e B a
dt
da
/ ) (
0

=

=
t
t
a
a
h R a
dt B da
a
e
e
0 0
0
/ ) (

) (
0 0
/
0
t t B dH
H R
e
H
H e
h H
=
+

( ) ) (
0 0
/ /
0
t t B e e
R
h
h H h H
e
=

195
can be carried out using any sophisticated propagation model and used later for collision risk
calculations.

Table 6.2 Ballistic Coefficient Values from the Three Way Table for (H
o
, H, B)
for One Year Duration

Altitude of Debris Objects (km) Target
Body
Altitude
(km)

500 510 520 530 540 550 560 570 580 590
200
210
220
230
240
250
260
270
280
290
0.2186 0.2881 0.3799 0.5008 0.6603 0.8704 1.1476 1.5129 1.9945 2.6294
0.1496 0.1944 0.2526 0.3282 0.4265 0.5542 0.7201 0.9358 1.2159 1.5800
0.1125 0.1444 0.1854 0.2380 0.3055 0.3921 0.5034 0.6462 0.8296 1.0649
0.0908 0.1154 0.1466 0.1862 0.2366 0.3006 0.3819 0.4853 0.6165 0.7832
0.0775 0.0976 0.1229 0.1548 0.1949 0.2454 0.3090 0.3892 0.4900 0.6170
0.0691 0.0864 0.1080 0.1350 0.1687 0.2108 0.2634 0.3291 0.4111 0.5137
0.0640 0.0794 0.0986 0.1224 0.1519 0.1886 0.2341 0.2905 0.3605 0.4474
0.0609 0.0752 0.0928 0.1146 0.1414 0.1744 0.2152 0.2655 0.3276 0.4041
0.0594 0.0730 0.0897 0.1101 0.1352 0.1659 0.2037 0.2499 0.3067 0.3764
0.0592 0.0724 0.0885 0.1082 0.1322 0.1615 0.1973 0.2411 0.2945 0.3597

The table can be read in the following way. A debris objects at an altitude of 500 km can pass
through or having a collision within a duration of one year with a target altitude of 200 km if
its ballistic coefficient is more than 0.2186 and will pass through a target altitude of 290 km
only if its B value is more than 0.0592. Similarly A debris objects at an altitude of 590 km
can pass through or having a collision within a duration of one year with a target altitude of
200 km if its ballistic coefficient is more than 2.6294 and will pass through a target altitude
of 290 km only if its B value is more than 0.3597.
objects
196
6.3.4 LONG TERM EVOLUTION OF THE DEBRIS USING SOPHISTICATED
ENVIRONMENT MODELS

It is possible to utilize the TLE data sets and together with the sophisticated environment
models in order to predict the debris scenario much ahead of the launch date. This would
provide information regarding any possible increase of collision risk between the present
time and the launch date. This corresponds to the planning of the launch using the wind data
for many months in advance. However such a procedure would call for large computing time.

6.3.5 COLLISION AVOIDANCE (COLA)

The above method known as COLA is used for operational purposes and in particular just a
few hours before launch. This method carries out a detailed evolution of the trackable
objects whose latest data (perhaps before one week) given by the TLE sets together with
highly sophisticated propagation model involving all the perturbations for a few hours before
the launch. There could be some uncertainties of the data from the TLE but by providing
larger margins for critical closest distances the operational safety is ensured. Such a
procedure is similar to using the wind information a few hours before the launch and clearing
the launch.

6.4 RISK ANALYSIS

The hazard due to space debris is generally a function of the size of the spacecraft, the size
and number of the orbital fragments in the orbital regime traversed by the spacecraft and the
duration of the mission. Another factor affecting the collision hazard is the relative velocity
of the spacecraft with respect to the debris objects. The relative velocity multiplied with the
relevant area of cross section denotes the volume swept per unit time which if larger means
higher probability of collision. In the present study an average relative velocity of 10 km/s is
assumed.


197
6.5 APPROACH BASED ON MONTE CARLO TECHNIQUE AND
EXTREME VALUE DISTRIBUTION

In order to demonstrate the applicability of the above approach, we provide here the result of
a test case. The basic debris objects data presently utilized is from the USPACECOM
catalogue. The catalogued data set used is the one that contains 8367 objects as observed in
January 2003. In this case 2000 MC simulation were carried out by randomly simulating the
mean anomaly of the target body and the debris objects. For a specified target orbit of 500
km altitude and inclination 0 deg., it is noted that the scale and shape parameters, have
converged and settled to the values 408 and 3.2402. The Figures 6.1 and 6.2 provides the
nature of convergence of scale and shape parameters for two cases of target altitude 500 km
with inclinations 0 deg and 28.5 deg. The Figure 6.3 provides the probability density function
and the cumulative distribution function of the Weibull distribution fit of the closest
approach distances. The Figure 6.4 shows the goodness of fit of Weibull distribution through
a plot of minimum distances in a Weibull probability paper. It may be noted from these
figures that the Weibull distribution fits the closest approach distances.

6.5.1 COLLISION PROBABILITIES AND TIME BETWEEN COLLISIONS

Given a characteristic collision distance, say r
c
, and a characteristic collision time, say t
c
, it
now follows that the estimated number of collisions per unit time (Vedder et al 1991 and
Chobotov et al 1997) is given by

( ) { } [ ] ( ) [ ]

+ + =
=
1
2 / / 2 / 1 ) )(
1
1 (
) ( / ) (
c
r Vr
c
t E
c
r F N

where Vr is the averaged relative velocity and (.) is the gamma function.

objects
198
This formulation results in expected time between collisions as

( ) [ ] ( ) { } [ ]
+ =
=
1
2 / 1 / 2 / ) (
1
)
1
1 (
/ 1
c
r Vr
N
c
T

where r
c
is the minimum distance of closeness assumed as collision.

It can be easily seen that the estimated times between collisions strongly depends on value.
It can be inferred from the above expression that the large values implies a large interval
between collisions, and values less than 3 for points to the real threat to the spacecraft
missions in the near future due to the catalogued objects itself.

6.6 APPLICATION OF THE PROPOSED REFERENCE COLLISION
PROBABILITY MODEL

Now we propose a strategy to obtain the collision risk of debris clouds using the above table.
In order to assess whether a debris object has a collision probability with a target altitude, the
required ballistic coefficient for a collision is calculated from the look up table and if the
ballistic coefficient of the debris object is more than the calculated value, the object will
sweep through the target altitude within a year, otherwise it does not pose a threat to the
target. This process will be continued for all the debris objects in the cloud and estimate the
number of objects reaching or sweeping through the target altitude in a year. Hence the
probability of collision in a year due to the debris clouds can be obtained.

In order to demonstrate the effectiveness of the above strategy and to assess the collision risk,
a three way table similar to the one shown above is generated with target altitudes and debris
altitudes ranging from 200 km to 1200 km in one km interval. The Figure 6.5 provides a

199
three dimensional view of the ballistic coefficients required by the debris objects to pass
through or impact the target altitude for one year duration.

Sets of clouds of 6000 debris objects due to an explosion at different altitudes, say 440 km,
640 km, 825 km, and 1100 km, were simulated using ASSEMBLE model, which provides
information such as semi major axis, eccentricity, inclination and ballistic coefficient. The
approach outlined above is utilized to obtain the collision probability with different target
orbits. The Table 6.3 provides the relative frequency of objects sweeping through different
target altitudes obtained through this approach. Further the lookup table is used to obtain the
number of objects sweeping through different altitudes in consecutive years, which will give
insight into the evolution of the debris scenario. It may be worth noting that the one year
three way look up table is sufficient enough to obtain the evolution details in any consecutive
years, by applying the table reading process iteratively. This is an important and useful
aspect!

Further for analyzing the debris cloud behavior, the look up table is used to obtain the
number of objects remained (not decayed) in the space. The results with the second set of
6000 objects are provided here. It is noted that within 5 years nearly 80 % of the objects had
decayed. The Figure 6.6 shows the number of objects remained at various time up to 5 years.
The Table 6.4 provides the number of debris objects swept through different target altitudes
during the first 6 years. The Figure 6.7 provides the graphical representation of this
information in a relative frequency scale, which is scaled by 6000.

Further to validate the above approach, the spatial densities of the catalogued objects
sweeping through 10 km bins of target altitudes are obtained. A TLE set of 8367
USSPACECOM catalogued objects is utilized in this study. Applying the kinetic theory of
gases with an averaged relative velocity of 10 km/sec, the collision probabilities at different
altitude bins are also estimated. The collision probability estimated by this approach is
provided in the Figure 6.8.

objects
200
Table 6.3 Relative Number of Objects Sweeping Through the Target in One Year.

Relative Number of Objects Sweeping through the Target in a Year

Altitude of
the
Target Body
(km) 440 640 825 1100
200
250
300
350
400
450
500
550
600
650
700
750
800
850
900
950
1000
9.9 E-1
9.9 E-1
9.9 E-1
9.7 E-1
8.4 E-1
3.0 E-1
8.3 E-3
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
2.6 E-2
1.8 E-1
2.7 E-1
2.5 E-1
1.9 E-1
1.4 E-1
1.3 E-1
1.9 E-1
3.3 E-1
2.4 E-1
1.3 E-1
5.3 E-2
1.4 E-2
4.5 E-3
8.0 E-4
8.0 E-4
5.0 E-4
1.1 E-3
4.6 E-3
6.8 E-3
6.6 E-3
5.7 E-3
4.0 E-3
3.9 E-3
4.9 E-3
1.0 E-2
4.1 E-2
1.5 E-1
4.2 E-1
6.2 E-1
1.2 E-1
2.8 E-2
1.1 E-2
4.7 E-3
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
1.5 E-3
6.5 E-3
1.8 E-2
3.5 E-2
6.6 E-2

201
Table 6.4 Number of Debris Objects Swept through Different Target Altitudes During
the First 6 Years (Debris Clouds of 6000 Objects Simulated at 640 km Altitude)

At the end of the years Target
Altitude
(Km)

1

2

3

4

5

6
200
250
300
350
400
450
500
550
600
650
700
750
800
850
900
950
1000
124
860
1349
1231
905
664
642
938
1751
1257
743
300
79
27
5
3
3
453
1570
1982
1840
1494
1243
1231
1380
1253
664
275
60
23
7
5
2
4
527
1645
1971
1837
1516
1270
1274
1133
768
350
94
24
7
4
6
2
0
463
1335
1556
1429
1198
998
963
744
440
116
29
8
2
5
1
0
0
295
817
944
871
717
625
585
449
191
41
8
5
2
2
0
0
0
146
432
487
450
380
319
301
206
64
23
5
2
2
1
0
0
0

objects
202
Table 6.5 Spatial Densities and Collision Probabilities at Different Target Altitudes
from the Catalogued Objects

Target Altitude (km) Spatial Density (1/Km
3
) Collision probability (1/ m
2
/ year)
400
425
450
475
500
525
550
575
600
625
650
675
700
725
750
775
800
825
850
875
900
925
950
975
1000
8.7 E-10
8.6 E-10
5.1 E-10
1.0 E-9
1.9 E-9
2.5 E-9
4.0 E-9
6.1 E-9
8.8 E-9
1.1 E-8
5.3 E-9
6.7 E-9
7.9 E-9
9.5 E-9
1.1 E-8
2.9 E-8
1.4 E-8
8.4 E-9
1.1 E-8
1.0 E-8
1.2 E-8
1.4 E-8
1.2 E-8
2.9 E-8
1.3 E-8
2.7 E-7
2.7 E-7
1.6 E-7
3.2 E-7
5.8 E-7
7.9 E-7
1.3 E-6
1.9 E-6
2.8 E-6
3.5 E-6
1.7 E-6
2.1 E-6
2.5 E-6
3.0 E-6
3.5 E-6
9.2 E-6
4.3 E-6
2.7 E-6
3.3 E-6
3.2 E-6
3.6 E-6
4.5 E-6
3.7 E-6
9.1 E-6
4.1 E-6
The high collision risk is clear from the peaks at the altitudes around 600 km, 800 km and
1000 km. The Table 6.5 provides the spatial densities and the collision probabilities obtained

203
by the present approach for the catalogued objects. It may be noted that the spatial density
values are comparable to those given in Chapter 4.

6.7 COMPARISON OF THE PRESENT RESULTS WITH OTHER
MODELS

Here we compare the collision probability obtained through the present approach with that of
obtained from ORDEM 96 and MASTER 2000 analyst application. It may be noted that only
TLE background is considered for MASTER 2000 application and the epoch considered is
the same as that of the TLE sets used for the present approach validation. For ORDEM 96
application the objects of size above 8cm were considered. The collision probabilities are
estimated based on a 10 km bin size for altitudes. Table 6.6 provides a comparison between
different models. It may be noted from Table 6.6 that the results are comparable, and match
is good in lower altitudes.

Table 6. 6 Comparisons of Collision Probabilities
at Different Target Altitudes by Various Approaches.

Collision Probability Sl.
No
Target
Altitude
(km)
Target
Inclination
(deg)
Present
Model*
Poisson
Model*
Extreme
Value
MC
MASTER
2000
ORDEM
96
1. 500 28.5 5.8e-7 6.4e-7 2.8e-7 3.9e-7 3.9e-7
2. 600 28.5 2.8e-6 2.7e-6 1.8e-6 8.8e-7 9.6e-7
3. 700 28.5 2.5e-6 2.6e-6 3.1e-6 8.8e-7 2.1e-6
* inclination is not taken into account

The differences in the collision probabilities can be attributed to the fact that any two models
would not provide exactly matching results (Jehn 1996, Johnson 1987) due to the differences
in the underlying assumptions on probability distributions, source models and terms and
objects
204
evolution models. Essentially this is due to the differences in the specification of the sample
space. However most estimates are well within one order of magnitude. The different
characteristics between the models are already provided in Table 6.1 and may be noted that
MASTER and ORDEM models come into fourth level in the table. It may be noted that the
present model and the Poisson model provide conservative values of the collision
probabilities which are quantitatively higher than MASTER and ORDEM predictions as the
inclination was not taken into account.

Finally it may be pointed out here that the real tracked data on the debris cloud generated out
of an explosion is not freely available and hence the utility and quantitative assessment of the
above proposed reference collision probability model estimation could be carried out only by
making use of the simulated debris cloud data.

6.8 CONCLUSIONS

A new approach for the collision probability assessment utilizing the closed form solution of
Wiesel (1989) by the way of three dimensional look up table, which takes only air drag effect
and an exponential model of the atmosphere, is presented. This approach can serve as a
reference collision probability assessment tool for LEO debris cloud environment. This
approach takes into account the dynamical behavior of the debris objects propagation and the
model utilized is simple for quick assessment of collision probability. This chapter also
brought out a broad comparison of presently available collision probability assessment
algorithms based on their complexities, application areas and sample space on which they
operate. Further the quantitative assessment of the collision probability estimates between
different presently available methods is carried out and the obtained collision probabilities
match qualitatively. The present reference model estimates can be refined by utilising more
accurate environment modeling as well.


205

Figure 6.1 Convergence of Scale and Shape Parameters in Weibull Distribution Fit for
Target Altitude 500 Km and Inclination 0 deg.

Figure 6.2 Convergence of Scale and Shape Parameters in Weibull Distribution Fit for
Target Altitude 500 Km and Inclination 28.5 deg.

objects
206

Figure 6.3 PDF and CDF of the Weibull Distribution Fit

Figure 6.4 Goodness of Fit of the Weibull Distribution
for Closest Approach Distances

207
500
520
540
560
580
600
200
250
300
0
0.5
1
1.5
2
2.5
3
Target A
ltitu
de (km
)
D
ebris O
bjects A
ltitude (km
)
B
a
l
l
i
s
t
i
c

C
o
e
f
f
i
c
i
e
n
t

(
m
2
/
k
g
)

Figure 6.5 Ballistic Coefficients of the 3D Table (H, H
0
, B)

Figure 6.6 Number of Objects Remaining Out of
6000 Objects Simulated Breakup at 650 km Altitude
objects
208

Figure 6.7 Relative Frequency of Objects Sweeping through Different Altitudes Due to
the Simulated Debris Cloud of 6000 objects (Simulated Breakup at 640 km Altitude)

Figure 6.8 Collision Probabilities at Different Altitudes from Catalogued Objects.
Chapter 7 Prediction of reentry of debris objects: constant gain Kalman filter approach

209
CHAPTER 7

PREDICTION OF REENTRY
OF DEBRIS OBJECTS:
CONSTANT GAIN KALMAN FILTER APPROACH

7.1 INTRODUCTION

Space debris studies demand assessment of orbital life times of a large number of objects
resulting from on orbit fragmentation or collision. In particular an accurate estimation of the
orbital decay of objects during the final stages of reentry is of considerable importance. This
helps to predict the reentry time and thus plan proper hazard assessment and mitigation
strategies.

The database available for the prediction of orbital lifetime and reentry of large debris objects
is the set of Two Line Elements (TLEs) provided by agencies like USSPACECOM. As
already mentioned in the previous chapters the TLE sets provide information regarding
orbital parameters together with rate of mean motion decay and a parameter B* related to the
ballistic coefficient B as

B* = B (
o
/2) = C
D
A
eff
/m

Where B* is an adjusted value of B using the reference value of atmospheric density
o
.
The parameter
0
is the density at a reference altitude 120 km above earth, C
D
is the non
dimensional drag coefficient, m is the mass and A
eff
is the effective area of cross section of
the object. Since A
eff
~ L
2
and m ~ L
3
, where L is a characteristic length, larger B means
small area to mass ratio and thus its orbit decay faster. The ballistic coefficient B represents
how sensitive is an object to air drag.
objects

210
In general the physical parameters of the objects like mass, area of cross section, shape and
dimensions are not available accurately. Further the atmosphere in which the objects decay
also varies randomly. Besides, the tumbling effect of the body and the molecular gas surface
interaction lead to an uncertain and varying aerodynamic drag coefficient that further makes
the prediction of reentry time a complex and involved problem.

The governing equations of motion of reentry objects are the same set of equations already
provided in Chapter 5. The reentry objects are mainly affected by the atmospheric drag,
earths oblateness, solar activity index F
10.7
and magnetic index A
p
. However the orbital
propagator utilized in this study is a very simple model (King-Hele 1987) as given in
Appendix 3, which accounts for only the atmospheric drag effect. The present propagator
assumes a mean atmospheric condition

as provided by the US Standard Atmosphere (1976).
This model estimates only the semi major axis and eccentricity decay with respect to one
revolution assuming a constant scale height during one revolution. This model is sufficient
for the reentry prediction as the decay of the object is mainly governed by the air drag only.
The effects of other orbital perturbations together with the variations in the atmospheric
density can be considered as random uncertainties in modeling which are accountable
through the process noise and the Kalman filter is thus able to handle it through the proper
gains as will be demonstrated in the later sections.

In all the prediction exercises, when the semi major axis of the object reaches a height of 120
km above the earth, it is considered to have reentered the atmosphere. This assumption is
appropriate as a reference condition since there are significant variations in the atmospheric
properties above 120 km with solar, magnetic activity and local time than below this height.
Also effectively a diffusive equilibrium predominates beyond 120 km (Whitten et al 1990).

As mentioned in the earlier Chapter 5, the problem of fusing the information from the state
propagation with measurement information can be formulated as a Kalman filter problem.
Ananthasayanam (2004) has stated the five various levels of solving the Kalman filter

211
problem as provided in Appendix 8. Once again as in Chapter 5, we use the formalism of
constant gain Kalman Filter approach to handle the reentry problem.

A new technique for the reentry prediction of debris objects utilising the Kalman filter with
constant gains, which are obtainable by the minimization of a suitable cost function (Philip
2001 and Anilkumar et al 2003) is proposed in this chapter. The states considered are the
semi major axis, eccentricity and ballistic coefficient. The measurements are the apogee
height and perigee heights, which are equivalent to semi major axis and eccentricity. The
chosen constant Kalman gains are able to account for the modeling and measurement errors.
The details of this approach to study four recent reentries of debris objects, namely

(i) US Sat No. 25947,
(ii) SROSS-C2 Satellite,
(iii) COSMOS 1043 rocket body and
(iv) COSMOS 389 Satellite.

are provided. These objects reentered the earths atmosphere during the last three years, on
4
th
March 2000, 12
th
July 2001, 19
th
January 2002 and 24
th
November 2003 respectively. A
comparison of the eventually observed reentry with the predicted reentry using the present
approach is also discussed (Sharma et al 2003, Anilkumar et al 2002).

Measurement Equations

For the present analysis the set of discrete measurement equations are

(h
a
)
k
= a
k
(1 + e
k
) +
1

(h
p
)
k
= a
k
(1- e
k
) +
2
with
[
1
,
2
]
T
~ N (0, R)

objects

212
where (h
a
)
k
, (h
p
)
k
denote the apogee and perigee heights respectively at time instant t
k

and w
1
, w
2
corresponds to noise in the apogee height and perigee height measurements with
R, the measurement noise covariance matrix assumed as constant.

7.2 FILTER WORLD SCENARIO: STATE EQUATIONS

The measurements are available in terms of the orbital parameters a and e, in both the
simulated cases and the tracked TLE elements. But in the filter implementation process, the
transformed variables namely the apogee and perigee heights (equivalent states of semi major
axis and eccentricity) are utilized.

The state equations governing the state the variables (a, e, B), are
a
k
= a
k-1
+ a
k -1
= a
k-1
+
1
(a
k-1
, e
k-1
) B
k-1

e
k
= e
k-1
+ e
k -1
= e
k-1
+
2
(a
k-1
, e
k-1
) B
k-1

B
k
= B
k-1

where
1
,
2
are the functional forms obtained from expressions of King-Hele (1987) which
depend on ballistic coefficient B , the semi major axis a and eccentricity e. The suffix k
denote for the time instant. The details are provided in Appendix 3.

7.3 FILTER WORLD SCENARIO: MESUREMENT EQUATIONS

The predicted apogee and perigee heights are

(h
a
-
)
k
= a
k
(1 + e
k
) and
(h
p
-
)
k
= a
k
(1 - e
k
)

213

The measurements at time t
k
are of the form

(h
a
)
k
= (h
a
-
)
k
+
1
and
(h
p
)
k
= (h
p
-
)
k
+
2

where
1
,
2
are respectively the measurement noises in apogee height and perigee height. As
there are no definite information on the behaviour or the distribution of the measurement
noises in apogee height and perigee height values derived from the mean motion and
eccentricity in TLE sets, we assume here for them to be white Gaussian noises.

The predicted values of these heights in the state equations are updated by utilizing the
measured values of the apogee height and perigee height. These are

k
p
h
p
h
a
h
a
h
k k
k k
k k
k
B
p
h
a
h
k
B
p
h
a
h
(
(
(
(
(
(
(
+
|
|
|
|
|
|
.
|
\
|
=
|
|
|
|
|
|
.
|
\
|
+
+
+
32

31
22

21
12

11

where B is the ballistic coefficient, and h
a
, h
p
respectively stand for measured apogee height
and perigee height. The superscript (+) correspond to the updated values, superscript (-) for
the predicted values and suffix k denotes the time instant.

7.4 ADAPTIVE EXTENDED KALMAN FILTERING APPROACH
WITH TIME VARYING GAINS

Assuming that the measurements are available at N discrete time instants, the cost function
J
4
, fundamental to the Kalman Filter as suggested by Sorenson (1970), based on the
innovation is defined as (Gemson et al 1998 and Anilkumar et al 1999)
objects

214

J
3
= (1/N)
k
T
[ ]
- 1

k,

where the innovation
k
= (measurement - model output), is the covariance matrix of the
innovation defined in Appendix 8. The above cost function can be defined as

J
4
= (1/N)
k
T
[ ]
- 1

k

by treating as constant with time.

The measurement Z is related to the state x, as

Z = Hx

where H is the measurement matrix.

Since the unknown parameter B in the state equations is treated as another augmented state
this procedure is known as the extended Kalman Filter (EKF). There are many approaches to
deal with the minimization problem (Gemson 1991 and Anilkumat et al 1999). In general
varies as a function of time depending on the initial state, process, and measurement noise
covariance, P
o
, Q, and R respectively. The above cost function J
3
has to be minimized based
on an appropriate choice of P
0
, Q and R. This is of course is quite difficult.

We have utilized Myers and Tapleys (1976) heuristic estimators for the estimation of the
state and measurement noise processes occurring in the Kalman filter and P
0
has been
calculated from the deviations in the assumed initial states to those obtained from the first
TLE under consideration.


215
7.4.1 ADAPTIVE PROCEDURE OF MYERS AND TAPLEY
FOR ESTIMATING Q AND R

Myers and Tapley (1976) derived heuristic estimators for the first and second order moments
of the state and measurement noise covariance arising in the Kalman filter. These estimators
are obtained from the state and observation samples generated based on the operation of the
Kalman filter through the data.

For a discrete linear system Myers and Tapley algorithm has the following structure. For the
sample values of measurement and state noises they proposed that

+

+
=
+
=
1

1
k
X
k
X
k
q
k
X
k
H
k
Z
k
r

where is the state transition matrix as shown in Appendix 3 and H is the measurement
matrix. The superscripts (-) and (+) are for the predicted and updated estimates, N is the
number of time steps at which the measurement sample is available with suffix k denoting
the time instant.

Then the estimate for the mean and covariance of the measurement noise are respectively

=
=
N
k
T
r
k
r r
k
r
N
R
N
k
k
r
N
r
1
] ) )( [(
1
1
1

1

Similarly for the state Noise
] ) ( ) [(
1
1
1

1
T
q
k
q q
k
q
N
Q
N
k
k
q
N
q

=
=

objects

216
The present formulation differs slightly from those of Myers and Tapley in neglecting the
additional terms involving the state covariances before and after update.

Extensive experiments have been conducted with the above adaptive technique. It turned out
that generally the Kalman gain matrix elements tend to a constant value after a transient. This
observation of the Kalman gain reaching a nearly constant value provides a different possible
approach. Thus instead of tuning P
0
, Q and R a smaller number of Kalman gain elements can
perhaps be used to solve the problem. The good statistical consistency of the results between
the adaptive technique and the constant Kalman gain approach provided the confidence in the
latter method, which is described below.

7.5 CONSTANT KALMAN GAIN APPROACH

Generally after the initial transients the Kalman Gain matrix tends to a constant value in
many such problems (Philip 2001 and Ananthasayanam 2004) and is also noticed in the
present case as well. This implementation requires the estimation of such constant gains.
Such a set of constant Kalman gains can also be worked out from the previous cost function
J
4
by utilizing a proper optimization algorithm if it is assumed that the above is a constant.
For this purpose the Genetic Algorithm (Goldberg 2000, Deb 1995 and Whitley 1993) has
been utilized in the present work. There could be slight differences in the gain values due to
the relative periods of transients and the steady state conditions. The can be estimated
similar to the estimation of R in the Method of Maximum Likelihood Estimation with
measurement noise alone (Philip et al 1995) or better known as output error method (OEM).

The near optimal constant gains can be obtained by minimizing the above cost function and
this is the presently proposed algorithm to handle the reentry problem. As mentioned at
several places earlier the advantage of the constant Kalman gain approach based on the
minimization of the cost J
4
is that one need not propagate the state covariance equations thus
reducing the computational load. The results from the adaptive filtering technique and the
constant Kalman gain approach do not differ much.

217

7.6 GENETIC ALGORITHM

The GA fundamentals, its features and other implementation aspects have been discussed in
Appendix 7 and Chapter 5. The values of the parameters considered for the present study are
given in the following Table 7.1. These values were arrived at after some trials.

Table 7.1 Parameters Utilized in the Implementation of GA for the Reentry Problem

Parameter Values
Population Size 100
Bit Length 20
Probability of Cross Over 0.90
Probability of Mutation 0.05
Number of generations for Convergence 50
Alternate Tolerance for Convergence:
Change in J
3
between generations
0.0001

7.7 RESULTS FROM THE ADAPTIVE KALMAN FILTERING
APPROACH

Here the Case 1 of the Table 7.2 is considered. The propagation of an object with an initial
semi major axis 250 km above the earth and eccentricity 0.001 with a ballistic coefficient of
0.012 kg m/s
2
was considered first. The propagator utilized herein predicted a life of 10.1893
days for its reentry.

A life time of 10.2278 days was predicted by the implementation of the Kalman filter on a
prediction epoch of 5.5182 days. The Figure 7.1 shows that the estimated ballistic coefficient
objects

218
reaches the true value of 0.012 in about 2 days. The cost function also decreases from 7 to
around 2.5, which is close to the number of measurement channels namely 2. The errors in
the measurements apogee height and perigee height tend to zero. Such simulation studies
provide one with confidence in the algorithm to process the real data.

The typical reentry data of COSMOS 389 satellite is shown in Appendix 1. An adaptive
extended Kalman filter (AEKF) was run on the data for different lengths of time. Based on
such runs the values of Q and R were estimated and these are shown in the Figure 7.2. It was
observed here that the Kalman gains tend towards a constant value after a short transient in
all the above cases. The constant values so obtained were

[0.21 0.85 0.67 0.55 0.0019 0.0021]

and the initial ballistic coefficient utilized was 0.015. The estimated process noises in the
states are shown in the Figure 7.3 for varying length of the processed data. The next Figure
7.4 compares the estimated and measured apogee and perigee heights. The present study
concentrates further only the utilization of the constant Kalman gain approach for the reentry
time prediction.

7.7 RESULTS AND DISCUSSIONS: CONSTANT KALMAN GAIN
APPROACH

A set of constant Kalman gains can also be worked out from the above cost function J
4
by
utilising the Genetic algorithm if is assumed to be a constant. There could be slight
differences in the present gain values from the earlier ones based on AEKF due to the relative
periods of transients and the steady state conditions. The can be estimated similar to the
estimation of the pure measurement noise only case in the Method of Maximum Likelihood
Estimation (Philip and Ananthasayanam 1995) which is the OEM.


219
The near optimal constant gains can be obtained from the state updates so as to minimize the
cost function J
4
and this is the presently proposed algorithm to handle the reentry problem.
The implementation of the above constant Kalman gain filter was carried out in a MATLAB
environment. With this implementation, the reentry prediction accuracy and the state
estimation accuracy were verified for some simulated cases. Using these predictions of the
measurements and states, the noise characteristics are also obtained based on the expressions
given section 7.4.

7.8.1 SIMULATED CASE STUDIES

A number of further simulation experiments were carried out to assess the efficiency of the
present approach. The salient features of the cases are summarized in Table 7.2. The present
approach outlined in this chapter was applied to all the above cases for the prediction of the
states as well as the reentry time. The details of the results obtained for the above cases are
given in the next Table 7.3.

From these tables it is observed that the gains for the update of apogee and perigee heights
are less sensitive while the gains, which update the ballistic coefficient, is sensitive. Also it is
true that this approach estimates the process and measurement noise within acceptable limits
of accuracy. The various simulation cases considered bring out the efficiency of this
algorithm in possible combinations of no error, process noise alone, measurement noise alone
and both presents. The prediction of the reentry with earlier data, the prediction at earlier
epoch, also provided good results with this approach. Now we bring out the details of the
case studies with the real TLE data set for some more recent reentries.

7.8.2 CASE STUDY WITH US SAT NO. 25947, SOYUZ

The Satellite No. 25947, with international designation 1999-058E, is a rocket body (Starsem
Soyuz Ikar 3
rd
Stage, Block 1 of Soyuz 11A511 U launcher), launched on October 18,
1999. Its mass is about 2300 Kg and is nearly cylindrical in shape, with end domes, of
objects

220
dimensions 2.7 meter diameter and 8.1 meter long. This object has been the test case for the
third IADC Reentry campaign. The sets of 72 orbital elements were made available for
reentry prediction during Feb. 2, 2000 to March 3, 2000. The actual reentry of this body
occurred on March 4, 2000 at 5 hrs 50 minutes, 64.243 days calculated from January 1, 2000.

Starting from the 22
nd
TLE set, the present Algorithm is utilised for the prediction of the
reentry time. As is known, the present algorithm utilises constant Kalman gain. A total of 6
gains exists with two each for the three states considered namely, apogee height, perigee
height and ballistic coefficient with respect to the measurements of the apogee and perigee
heights. An important parameter in this implementation is the initial assumed value of
ballistic coefficient B
0
. Starting with different B
0
values, the performance of the algorithm is
provided in the Figure 7.5, by considering a set of constant values for Kalman gains. From
this figure, it is clear that the proper initial value of B
0
is nearly 0.040. It can be noted that,
the strategy that can be utilised for this selection may be the variance reduction in the reentry
prediction with respect to individual TLE elements. From this figure it is clear that the value
0.040 as B
0
, provides the minimum variance in the reentry prediction based on different
sample sets of TLE elements.

The Figure 7.6 provides the ballistic coefficient update in each of the previous cases as the
available TLE set numbers increase. As already discussed, the ballistic coefficient (as a state)
gets updated online. This is expected as the body may be in tumbling position, with irregular
shape and with varying gas molecule and surface interaction. Also it may be noted that the
model utilised here for propagation is a very simple one which considers only atmospheric
drag effect and with mean atmospheric conditions. The uncertainty in this model is also
contained in the predicted ballistic coefficient. The Figure 7.6 clearly brings out the
robustness of the approach as all the ballistic coefficients converge to the same value at
individual TLE Element after filtering beginning with different values of B
0
.

The procedure for fixing the approximate B
0
, is extended to obtain the constant Kalman gains
also. Here we minimised the cost function, J
4
in order to obtain the Kalman Gains. The

221
Figure 7.7 provides the performance of the algorithm with different sets of Kalman gains. A
careful study of the costs for each set, provided a set of constant gains as 0.6, 0.2, 0.2, 0.6,
0.00014 and 0.0001, which is near optimal. Hence a final prediction is made on the reentry of
Sat No. 25947 with the above Kalman gains and with B
0
equal to 0.40. A graphical
representation of this final prediction is provided in the Figure 7.8. The details of the results
are as given below.

Actual Reentry = 64.243 days from Jan. 1, 2000
= March 4, 2000 at 5 hrs 50 minutes
Near Optimal B
0
= 0.040
Near Optimal Kalman Gains = [0.6 0.2 0.2 0.6 0.00014 0.0001]
Prediction with Latest Elset = 64.3257 days from Jan. 1, 2000

The prediction depends on the epoch at which the last TLE is available at which the reentry
is predicted. In principle the optimal gains can be found out as and when the data is
available. The present approach is exercised from TLE set number 22 (about 12 days prior to
actual reentry) to the last available TLE by predicting the reentry. The results are

Mean Prediction = 64.2323 days from Jan. 1, 2000
RSS Error on Prediction from Actual Reentry
= 1 hr 51 minutes
Standard deviation based on the prediction from present technique
= 2 hrs 09 minutes

Further the noise in the measurements and the states are estimated using the adaptive
procedure outlined above based on the cost function J
3
. The ballistic coefficient estimated
sequentially and the corresponding cost J
3
are provided in the Figure 7.9. The Figure 7.10
brings out the estimated measurement noise levels by adaptive estimation of the noises. The
objects

222
estimated process noise values for the three states namely the semi major axis, eccentricity
and the ballistic coefficient are given in the Figure 7.11 and the Figure 7.12 provides a close
look at the measurement updates together with the observations in the present approach. It
may be noted that the cost in the Figure 7.9 is converging close to 2, and that is what
expected from a good filtering technique equal to the number of measurement channels
(Ananthasayanam 2001).

7.8.3 CASE STUDY WITH SROSS-C2 SATELLITE

Satellite No. 23099 is Indias SROSS-C2 satellite (International designation 1994 027 A)
launched on the 4
th
May 1994 initially into an orbit of 430 x 620 km size at nearly 46 deg.
inclination. Its weight is about 113 Kg and cylindrical in shape of diameter 0.32 m and length
of 0.86 cm. It is understood that its reentry had occurred on July 12, 2001 at 3 hrs, 6 minutes
193.129 days reckoned from Jan. 1, 2001. A total of 28 Elset were available. Starting from
the 2
nd
Elset, the present Algorithm is utilised for the prediction of the reentry time. As in the
case of Sat No. 25947, here also a thorough study was made on the sensitivities of B
0

parameter and the Kalman gains. The details of the results are as given below.

Actual Reentry = 193.129 days from Jan. 1, 2001
=July 12, 2001 at 3 hrs 06 minutes
Near Optimal B
0
= 0.014
Near Optimal Kalman Gains = [0.2 0.5 0.2 0.5 0.0006 0.0006]
= July 12, 2001 at 3 hrs 36 minutes

7.8.4 CASE STUDY WITH COSMOS 1043 OBJECT

The Fourth IADC reentry campaign for the identified space object (USSPACECOM
Catalogue Number 11056), a Russian Vostok (SL-3/A-1) Upper stage 1043 R, started on 10
th

January 2002. The object was of cylindrical shape (diameter 2.6 m, length 3.0 m, having a

223
dry mass of 1400 Kg. The COSPAR ID No. of this rocket body is 1978-094B. It deployed its
payload (COSMOS 1043) into an initial orbit of 622 km x 635 Km at 81.2 degrees
inclination on 10
th
October 1978. The number of TLE sets available for this object was 50 for
off line prediction, provided by IADC during the reentry campaign.

As in the earlier cases, the present algorithm is utilised for prediction of the reentry time. A
complete analysis was made on the sensitivities of B
0
parameter and the Kalman gains. The
details of the results are as given below.

Actual Reentry =19.923 days from Jan. 1, 2002
= January 19, 2002 at 22 hrs 09 minutes
Near Optimal B
0
= 0.010
Near Optimal Kalman Gains =[0.6 0.2 0.2 0.6 0.0004 0.0004]
=January 19, 2002 at 22 hrs 35 minutes

7.8.5 CASE STUDY WITH COSMOS 389 SATELLITE

The IADC reentry prediction campaign in 2003 was conducted recently (first campaign in
2003) with the identified object as COSMOS 389 satellite (USSPACECOM Catalog Number
04813). The Cosmos 389 was a Soviet ELINT (Electronic and Signals Intelligence) satellite
launched from the Plesetsk cosmodrome with an on orbit dry mass of 3800 kg., with
unknown dimensions, launched on 18
th
December 1970 in an orbit with Perigee of 543 km.
Apogee of 568 km. with Inclination of 81.2 deg. A picture of the satellite reconstructed by
US is given in the Figure 7.13. The Campaign started on Nov. 07, 2003 and expected a
reentry around Nov. 22, 2003. From Nov. 07 2003 onwards about 54 TLEs were obtained.
These TLEs together with the details on the formats and description of the parameters
available in TLE sets are provided in Appendix 1.

objects

224
For this reentry campaign, seven different agencies including two from INDIA (one from
ISAC and other from VSSC) participated. They are USA, Germany, Russia, Italy, Japan and
India.

The present approach of constant gain Kalman filter was utilized to predict the reentry of the
object. The comparisons of the results with the predictions from the other agencies (Sharma
et al 2003) are shown in the Figure 7.14. The Table 7.4 provides a comparison of the reentry
prediction of the constant gain Kalman filter approach with other agencies, from the last
predictions made. It is learnt that the actual reentry occurred around 24
th
Nov. 2003 at 22: 30
hrs. It may be seen that the present predictions match well with the actual reentry and is
comparable or better than many predictions by different agencies. This figure clearly brings
out the efficiency and the robustness of the prediction procedure using constant gain Kalman
filter algorithm.

7.9 CONCLUSIONS

Here a new approach has been presented utilizing the Kalman filter formalism with constant
gains for an efficient online prediction of the reentry time of space debris objects. Illustrative
examples, which show the efficiency of the proposed methodology with actual reentries of
four recent space debris objects, are also provided. This methodology utilized only a mean
atmosphere and a simple propagator considering only atmospheric drag effects. In particular
even the atmosphere characteristics as in the latest MSIS (1986) can also be used. However
the robustness of the constant gains have the ability to handle this mismatch as well, just as
the inaccuracies in modeling B, atmospheric density variations, measurement noise and other
errors.

225

Table 7.2 Details of the Various Case Studies

Case
No.
Initial
a (km)
above
the
earth
Initial
e
Actual
Ballistic
coefficient
in kg m/s
2

Actual
Reentry
in days
Predic-
tion
epoch
Measu-
rement
Noise
3
Process
noise
3
Initial
Value
of B =
B
0

1. 250 0.001 0.012 10.1893 5.5182 0., 0. 0., 0., 0. 0.015
2. 250 0.001 0.012 10.1893 7.5545 0., 0. 0., 0., 0. 0.015
3. 250 0.001 0.012 10.1893 9.5193 0., 0. 0., 0., 0. 0.015
4. 250 0.001 0.012 10.1893 9.5193 0., 0. 0., 0., 0. 0.010
5. 250 0.001 0.012 10.1893 9.5193 0., 0. 0.5,
0.00005,
0.0
0.015
6. 250 0.001 0.012 10.1893 9.5193 0., 0. 1.0,
0.00005,
0.0
0.015
7. 250 0.001 0.012 10.1893 9.5193 0., 0. 0., 0., 0. floated
8. 300 0.005 0.012 35.3523 19.050 0., 0. 0., 0., 0. 0.015
9. 300 0.005 0.012 35.3523 29.006 0., 0. 0., 0., 0. 0.015
10. 300 0.005 0.012 35.3523 29.006 0., 0. 0., 0., 0. 0.010
11. 300 0.005 0.012 35.3523 29.006 0.5, 0.5 0., 0., 0. 0.015
12. 300 0.005 0.012 35.3523 29.006 0.5, 0.5 0.5,
0.0005,
0.
0.015

objects

226
Table 7.3 Results of the Various Case Studies

Case
No.
Predicted
Reentry in
days
Error in
Prediction
in hours
Optimal Gains Estimated Q

Estimated R

1. 10.2241 0.84 0.19, 0.89, 0.58, 0.43,
0.0019, 0.0017
0.14,
0.000003,
0.0005
0.15, 0.12
2. 10.2217 0.76 0.92 , 0.62, 0.93, 0.08,
0.0016, 0.0017
0.13,
0.0000062,
0.00038
0.17, 0.10
3. 10.2242 0.84 0.42, 0.68, 0.92, 0.09,
0.0014, 0.0020
0.11,
0.0000021,
0.00038
0.12, 0.10
4. 10.2210 0.78 0.92, 0.30, 0.97, 0.13,
0.0019, 0.0015
0.07,
0.0000015,
0.00024
0.08, 0.06
5. 10.2283 0.94 0.02, 0.82, 0.41, 0.46,
0.0017, 0.0006
0.22,
0.0000014,
0.0002
0.30, 0.17
6. 10.2377 1.16 0.10, 0.33, 0.29, 0.16,
0.0010, 0.0004
0.26,
0.000005,
0.0005
0.29, 0.24
7. 10.2238 0.83 0.05, 0.60, 0.82, 0.41,
0.0008, 0.0015
0.0011,
0.00000002,
0.0000009
0.00097,
0.0012
8. 35.3523 0.09 0.37, 0.71, 0.81, 0.51,
0.0091, 0.016
0.0414,
0.000002,
0.0001
0.055, 0.028
9. 35.3611 0.20 0.40, 0.70, 0.63, 0.68,
0.0082, 0.0085
0.035,
0.000002,
0.0001
0.048, 0.022
10. 35.3610 0.20 0.40, 0.70, 0.63, 0.68,
0.0082, 0.0085
0.023,
0.0000013,
0.00007
0.032, 0.015
11. 35.45 2.4 0.04, 0.14, 0.05, 0.19,
0.0013, 0.0023
0.04,
0.0000004,
0.0006
0.037, 0.043
12. 35.62 6.3 0.14, 0.19, 0.35, 0.59,
0.0013, 0.0023
0.04,
0.0000004,
0.0006
0.13, 0.23


227
TABLE 7.4 Comparisons of Last Predictions made by Various Agencies: COSMOS 389
Satellite Reentry Predictions (UTC)
(Actual Reentry Time 24/11/2003, 22:30 hrs)

Sl.
No.
Agency Prediction
Epoch
(UTC)
Remaining
Life
(Hrs)
Predicted
Reentry
(UTC)
Error in
Prediction
(Hr:Min)
Uncertainity
(hrs)
1 VSSC
India
24/11/2003
12:01
10.500 24/11/2003
23:14
-0:44 +1:50,
-1:30
2 ISAC
India
24/11/2003
12:01
10.500 25/11/2003
00:01
+1:31 4:05,
-1:38
3 Klinkrad
Germany
24/11/2003
12:01
10.500 24/11/2003
23:58
+1:28 2:31,
-2:32
4. Johnson
USA
24/11/2003
01:28
21.033 25/11/2003
01:57
+3:27 +5:00,
-5:00
5. Nonaka
Japan
20/11/2003
22:08
96.367 25/11/2003
10:35
+12:05 28:12,
-19:07
6. Ivanov
Russia
24/11/2003
10:38
11.867 25/11/2003
00:40
+2:10 3:42,
-2:02
7.. Pardini
Italy
21/11/2003
08:53
13.617 24/11/2003
23:54
+1:24 4:49,
-2:51
8. Present
Constant
Gain
Kalman
Filter
24/11/2003
12:01
10.500 24/11/2003
23:18
-0:48
----------

objects

228
Figure 7.1 Simulation Case 1 : Ballistic Coefficient, Cost and Errors in Apogee and
Perigee Heights Using Adaptive Filtering Technique

Figure7.2 Simulation Case 1 : Measurement Noise Estimates in Apogee and
Perigee Heights Using Adaptive Filtering Technique

229
Figure 7.3 Simulation Case 1: Process Noise Estimates in Semi Major Axis, Eccentricity
and Ballistic Coefficient Using Adaptive Filtering Technique
Figure 7.4 Simulation Case 1 : Measurement and Estimates in Apogee and Perigee
Heights Using Adaptive Filtering Technique

objects

230

Figure 7. 5 Sensitivities of the Reentry Prediction with Different B
0
Values
(Sat. No. 25947)

Figure 7.6 Estimate of Time Varying B with Different B
0
Values (Sat. No. 25947)

231

Figure 7.7 Reentry Prediction with Different Sets of Kalman Gains (Sat. No. 25947)

Figure 7.8 Final Prediction of Reentry Time with Near Optimal Kalman Gains
(Sat. No. 25947)
objects

232

Figure 7.9 Estimated Ballistic Coefficients and the Cost J

Figure 7.10 Sat25947 Reentry prediction: Measurement Noise Estimates in Apogee and
Perigee Heights from the Adaptive Filtering Technique

233

Figure 7.11 Sat25947 Reentry Prediction: Process Noise Estimates in a, e and B from
the Adaptive Filtering Technique

Figure 7. 12 Sat25947 Reentry Prediction: Observed Measurements and Estimated
Values of Apogee and Perigee Heights
objects

234

Figure 7.13 The US Reconstruction of the Soviet ELINT Spacecraft.

Figure 7.14 Comparisons of COSMOS 389 Satellite Reentry Predictions
by Different Agencies
Chapter 8 Conclusions and suggestions for further work
235
CHAPTER 8
CONCLUSIONS AND
SUGGESTIONS FOR FURTHER WORK

The present thesis has been able to provide newer perspectives for five major issues in space
debris studies namely (i) breakup modeling, (ii) environment modeling, (iii) evolution of the
debris environment, (iv) collision probability analysis, and (v) reentry prediction.

The breakup modeling ASSEMBLE approach is presented here to describe the semi
stochastic characteristics of the orbital fragments generated by a breakup. Subsequently it has
been applied as the inverse problem of simulating some typical well known historical on
orbit fragmentation events. All the simulated results compare quite well with the
observations both at the time of breakup and at a later epoch than earlier models. Such an
improved feature has been possible since ASSEMBLE utilizes the observational TLE data
rather than assuming or obtaining incremental velocities of the fragments based on intuitive
empirical relations or mathematical probability distributions. Any quantitative comparison
with other models is tricky since they could be based on different concepts. Even a
comparison with a simulated or real data can only be for a smaller number among a large
number of characteristics that are important for various features of the fragments.

The SIMPLE is an engineering model for describing the orbital space debris scenario in
LEO. This model provides the distribution of the orbital characteristics with respect to the
altitude and perigee heights in different inclination bands as well as for the whole region of
inclinations. While other models utilize deterministic semi empirical functions to describe the
distributions, SIMPLE uses probability distributions. The number of parameter values here
is, in general, just 8 for each description of altitude or perigee distributions and this model
follow closely all the peak number densities without losing the accuracy at other heights.

The evolution of the debris clouds has been treated through a novel innovative Kalman filter
formalism. This approach is based on the propagation of an equivalent object representing
objects
236
the debris fragments in a three dimensional bin of semi major axis, eccentricity, and the
equivalent ballistic coefficient (a, e, B). This new approach propagates the number density in
a bin of a and e rapidly without propagating each of the fragments in the above bins.
Further this approach expands the scenario to provide suitable ballistic coefficient values for
the EQFs in the various bins. Also it is able to assimilate the information with the passage of
time at an update including further breakups as well. Based on the number density
measurement at a later epoch for various bins the update of the number density and the
equivalent ballistic coefficient utilizes a constant Kalman gain.

A new approach that can serve as a reference collision probability assessment tool for LEO
debris cloud environment has been proposed. Finally a technique has been presented utilizing
the Kalman filter formalism with constant gains for an efficient online prediction of the
reentry time of space debris objects. Illustrative examples with actual reentries of four recent
risk objects demonstrate the efficiency of the proposed methodology.

The studies on the sensitivity effect of various parameters involved in breakup modeling and
the steady state scenarios would be highly worthwhile for a future study. The SIMPLE
engineering model for LEO environment can be extended along similar lines to the GEO
environment. Present reentry studies concentrate only on the prediction of reentry time. The
subsequent prediction of the reentry location is an involved and challenging task. The
generation of an indigenous database and the full scale development of an Indian engineering
model for space debris environment can also be taken up.

The application of constant gain Kalman filtering approach for the fusion or assimilation of
the data is innumerable such as for onboard applications or GPS/INS data. The constant
Kalman gains can be estimated by utilising all the available collected data in a problem. The
most exciting application could be in the massive and challenging task of atmospheric data
assimilation involving tens of thousands of both states and measurements, thus dispensing
with the delicate task of choosing or estimating the noise statistics and following the more
robust route of constant Kalman gains.
Appendix 1 Two line elements (tle) format
237
APPENDIX 1

TWO LINE ELEMENTS (TLE) FORMAT

For the benefit of the reader the material is reproduced from the web site
http://celestrak.com.

A NORAD/ USSPACECOM two line element set consists of two 69 character lines of data
which can be used together with NORAD's SGP4/SDP4 orbital model to determine the
position and velocity of the associated space object. The only valid characters in a two line
element set are the numbers 0-9, the capital letters A-Z, the period, the space, and the plus
and minus signsno other characters are valid.

Figure 1 shows what type of character is valid for each column. Columns with a space or
period can have no other character. Columns with an 'N' can have any number 0-9 or, in
some cases, a space. Columns with an 'A' can have any character A-Z or a space. The column
with a 'C' can only have a character representing the classification of the element set
normally either a 'U' for unclassified data or an 'S' for secret data (of course, only unclassified
data are publicly available). Columns with a '+' can have either a plus sign, a minus sign, or a
space and columns with a '-' can have either a plus or minus sign (if the rest of the field is not
blank).

1 NNNNNC NNNNNAAA NNNNN.NNNNNNNN +.NNNNNNNN +NNNNN-N +NNNNN-N N NNNNN
2 NNNNN NNN.NNNN NNN.NNNN NNNNNNN NNN.NNNN NNN.NNNN NN.NNNNNNNNNNNNNN

Figure 1. Two Line Element Set Format

Further restrictions are placed upon the values in each column as the individual fields of data
are defined. Tables 1 and 2 define each of the individual fields for lines 1 and 2, respectively.
Many of these bear additional explanation.

objects

238

Table 1. Two Line Element Set Format Definition, Line 1

Field Column Description
1.1 01 Line Number of Element Data
1.2 03-07 Satellite Number
1.3 08 Classification
1.4 10-11 International Designator (Last two digits of launch year)
1.5 12-14 International Designator (Launch number of the year)
1.6 15-17 International Designator (Piece of the launch)
1.7 19-20 Epoch Year (Last two digits of year)
1.8 21-32 Epoch (Day of the year and fractional portion of the day)
1.9 34-43 First Time Derivative of the Mean Motion
1.10 45-52 Second Time Derivative of Mean Motion (decimal point assumed)
1.11 54-61 BSTAR drag term (decimal point assumed)
1.12 63 Ephemeris type
1.13 65-68 Element number
1.14 69
Checksum (Modulo 10)
(Letters, blanks, periods, plus signs = 0; minus signs = 1)

Column 1 of each line of the two line element set indicates the line number (and hence the
format) for that line. The next field on each line (Fields 1.2 and 2.2) indicates the satellite
numberactually, the NORAD Catalog Numberof the object the data is for. The NORAD
Catalog Number is a unique identifier assigned by NORAD for each earth orbiting artificial
satellite in their SATCAT (Satellite Catalog). For a valid two line element set, fields 1.2 and
239
2.2 must be identical. As mentioned above, field 1.3 indicates the security classification of
the dataall publicly available data will have a 'U' in this field to indicate unclassified data.

The next three fieldsFields 1.4 through 1.6define the International Designator of the
object. This identifier is an additional unique designation assigned by the World Data
Center-A for Rockets and Satellites (WDC-A-R&S) in accordance with international treaty
(1975 Convention on Registration of Objects Launched into Outer Space). The WDC-A-
R&S works together with NORAD and NASA's National Space Science Data Center
(NSSDC) in maintaining this registry. Although there have been some changes in format
since it was first used back in the late 1950s (see "Space Surveillance" in Satellite Times
Volume 4 Number 1), the International Designator indicates the year of the launch (Field 1.4
only gives the last two digits), the launch of that year (Field 1.5), and the piece of that launch
(Field 1.6) for each object. These three fields can be left blank, but all must be present if any
exists. Finally, Field 1.6 can be either right or left justifiedthe latter is preferred.

As an aside, there are some significant differences between NORAD's Catalog Number and
the International Designator. For example, NORAD assigns a catalog number based upon
when the object was first observed, whereas the International Designator is always tied to the
original launch. For example, the 81
st
launch of 1968 carried four payloads into orbit: OV2-5,
ERS 21 and 28, and LES 6. Together with the Titan 3C transtage rocket body, these objects
were assigned International Designators 1968-081A through E and Catalog Numbers 03428
through 03431. Just this past October, however, NORAD cataloged two additional pieces
associated with this launch as Catalog Numbers 25000 and 25001they have the
International Designators 1968-081F and G.

The next two fields (Fields 1.7 and 1.8) together define the reference time for the element set
and are jointly referred to as the epoch. Field 1.7 is the two digit year (more on this later) and
Field 1.8 is the day of that year. The epoch defines the time to which all of the time varying
fields in the element set are referenced.
objects

240
While talking about the epoch, this is perhaps a good place to answer the other time-related
questions. First, how is the epoch time format interpreted? This question is best answered by
using an example. An epoch of 98001.00000000 corresponds to 0000 UT on 1998 January
01in other words, midnight between 1997 December 31 and 1998 January 01. An epoch of
98000.00000000 would actually correspond to the beginning of 1997 December 31strange
as that might seem. Note that the epoch day starts at UT midnight (not noon) and that all
times are measured mean solar rather than sidereal time units (the answer to our third
question).

Field 1.9 represents the first derivative of the mean motion divided by two, in units of
revolutions per day
2
, and field 1.10 represents the second derivative of the mean motion
divided by six, in units of revolutions per day
3
. Together, these two fields give a second-
order picture of how the mean motion is changing with time. However, these two fields are
not used by the SGP4/SDP4 orbital models (only by the simpler SGP model) and, therefore,
serve no real purpose.

Field 1.11 represents something called B* (BSTAR), which is an SGP4-type drag coefficient.
In aerodynamic theory, every object has a ballistic coefficient, B, that is the product of its
coefficient of drag, C
D
, and its cross-sectional area, A, divided by its mass, m.
B = C
D
A/m
The ballistic coefficient represents how susceptible an object is to dragthe higher the
number, the more susceptible. B* is an adjusted value of B using the reference value of
atmospheric density,
o
.
B* = B
o
/2
B* has units of (earth radii)
-1
.

Fields 1.10 and 1.11 have a somewhat different format that the other fields. In particular,
they use a modified exponential notation with an implied leading decimal point. This
convention is inherited from FORTRAN where all such numbers range from 0 to less than 1.
The first six columns of each field represent the mantissa and the last two represent the
241
exponent. For example, the value -12345-6 corresponds to -0.12345 10
-6
. Each of these two
fields can be blank, corresponding to a value of zero.

Field 1.12 represents the ephemeris type (i.e., orbital model) used to generate the data.
Spacetrack Report Number 3 suggests the following assignments: 1=SGP, 2=SGP4, 3=SDP4,
4=SGP8, 5=SDP8. However, this value is used for internal analysis onlyall distributed
element sets have a value of zero and are generated using the SGP4/SDP4 orbital model (as
appropriate). Field 1.13 represents the element set number. Normally, this number is
incremented each time a new element set is generated. In practice, however, this doesn't
always happen. When operations switch between the primary and backup Space Control
Centers, sometimes the element set numbers get out of sync, with some numbers being
reused and others skipped. Unfortunately, this makes it difficult to tell if you have all the
element sets for a particular object. The last column on each line (fields 1.14 and 2.10)
represents a modulo-10 checksum of the data on that line. To calculate the checksum, simply
add the values of all the numbers on each lineignoring all letters, spaces, periods, and plus
signsand assigning a value of 1 to all minus signs. The checksum is the last digit of that
sum. Although this is a very simple error-checking procedure, it should catch 90 percent of
all errors. However, many errors can still sneak through. To eliminate these, all data posted
on the Celestial WWW site not only pass the checksum test, but must also pass both format
and range-checking tests (as described in this article).

Line 2 consists primarily of mean elements calculated using the SGP4/SDP4 orbital model.
The definitions for fields 2.3 through 2.8 can be seen in table 2 below. Fields 2.3, 2.4, 2.6,
and 2.7 all have units of degrees and can range from 0 up to 360 degreesfield 2.3
(inclination) only goes up to 180 degrees. The eccentricity (field 2.5) is a unitless value with
an assumed leading decimal point. For example, a value of 1234567 corresponds to an
eccentricity of 0.1234567. The mean motion (field 2.8) is measured in revolutions per day.
The final field on line 2, prior to the checksum, is the rev number. Since there are several
conventions for determining rev numbers, this field also bears some clarification. In
NORAD's convention, a revolution begins when the satellite is at the ascending node of its
objects

242
orbit and a revolution is the period between successive ascending nodes. The period from
launch to the first ascending node is considered to be Rev 0 and Rev 1 begins when the first
ascending node is reached. Since many element sets are generated with epochs that place the
satellite near its ascending node, it is important to note whether the satellite has reached the
ascending node when calculating subsequent rev numbers.

Table 2. Two-Line Element Set Format Definition, Line 2

Field Column Description
2.1 01 Line Number of Element Data
2.2 03-07 Satellite Number
2.3 09-16 Inclination [Degrees]
2.4 18-25 Right Ascension of the Ascending Node [Degrees]
2.5 27-33 Eccentricity (decimal point assumed)
2.6 35-42 Argument of Perigee [Degrees]
2.7 44-51 Mean Anomaly [Degrees]
2.8 53-63 Mean Motion [Revs per day]
2.9 64-68 Revolution number at epoch [Revs]
2.10 69 Checksum (Modulo 10)

In general, any number smaller than the maximum field size can be padded with either
leading spaces or leading zeros. In other words, an epoch can be represented as either
98001.12345678 or 98 1.12345678 or an inclination can be represented as 28.1234 or
028.1234. Convention uses leading zeros for fields 1.5 and 1.8 and leading spaces elsewhere,
but either is valid.

243
Obviously, there are a few limitations with the current two-line format. First and foremost is
the need for a four-digit year in fields 1.4 and 1.7. Next, there is a need for a more robust
form of error checkingperhaps a 16-bit CRC. Such a checksum could be applied to both
lines together, not only detecting errors within the data but also mismatched lines 1 and 2. If
such changes were made, it might also be wise to increase the field size for the Catalog
Number to six or seven digits to support the eventual cataloging of smaller debris. The
International Designator format seems to suffice for the foreseeable future, with a four-digit
year, up to 999 launches (the most we've had to date in any one year was 129 in 1984), and
up to 13,824 pieces (the record holder today is 1994-029 with 672 pieces). Of course, the
cataloging of smaller debriswhich we may be unable to correlate with the original
launchstill presents potential problems.

New element sets are generated by NORAD on an as-needed basis rather than according to
an established timetable. How often these updates occur depends upon a number of factors
such as the orbit type or maneuvering capability of the satellite. For example, a satellite in
low-earth orbitsuch as the US space shuttlewould have its element sets updated several
times a day because of the somewhat unpredictable results of atmospheric drag as it varies its
attitude and the maneuvering being performed. A satellite in a low-drag orbit which doesn't
maneuversuch as LAGEOS IImight only need updates once or twice a week. Objects
such as rocket bodies, defunct payloads, or other space debris, won't be updated as
frequently, eitherunless there is a prediction of a close approach with an operational
payload. Special-interest objectssuch as a large object reentering the earth's atmosphere
normally get special treatment.

Two line elements utilized for the re-entry prediction of the COSMOS satellite is provided
here below as an example for the two line elements. It may be noted that some days prior to
the re-entry the observations are merely one in a day, while there are four or more TLEs from
different agencies just one or two days prior to the re-entry.
objects

244
COSMOS 389 (Cat.No.04813) TLE Sets

1 04813U 70113A 03250.72799083 .00082693 00000-0 50004-3 0 07845
2 04813 081.1276 331.4671 0003300 343.8364 016.2571 15.77316097780198
1 04813U 70113A 03252.75794183 .00072310 00000-0 43291-3 0 07885
2 04813 081.1289 328.8662 0003830 320.6709 039.3882 15.77564342780511
1 04813U 70113A 03254.72411211 .00074770 00000-0 44180-3 0 07916
2 04813 081.1302 326.3501 0003118 325.8819 034.3660 15.77858245780821
1 04813U 70113A 03257.64085526 .00077828 00000-0 45026-3 0 07957
2 04813 081.1259 322.6083 0004407 299.0344 061.0123 15.78318560781287
1 04813U 70113A 03260.24005983 .00065412 00000-0 37405-3 0 07998
2 04813 81.1280 319.2775 0004464 292.7746 67.3444 15.78608403781698
1 04813U 70113A 03262.64832624 .00094733 00000-0 52752-3 0 08041
2 04813 81.1217 316.1834 0003359 301.3510 58.8429 15.79137254782071
1 04813U 70113A 03264.67573400 .00102445 00000-0 55892-3 0 08071
2 04813 081.1255 313.5817 0004468 270.6554 089.3780 15.79566706782395
1 04813U 70113A 03266.63936392 .00099993 00000-0 53673-3 0 08104
2 04813 081.1260 311.0589 0003906 262.7172 097.3746 15.79936192782700
1 04813U 70113A 03268.66576158 .00110897 00000-0 58211-3 0 08132
2 04813 081.1232 308.4516 0003420 255.8297 104.3425 15.80404708783025
1 04813U 70113A 03272.40018923 .00124382 00000-0 62450-3 0 08190
2 04813 81.1225 303.6444 0003463 224.7978 135.1687 15.81344921783610
1 04813U 70113A 03273.60245110 .00114139 00000-0 56776-3 0 08220
2 04813 081.1201 302.0944 0004262 234.5791 125.5058 15.81567992783803
1 04813U 70113A 03275.56351887 .00120544 00000-0 58615-3 0 08238
2 04813 081.1219 299.5654 0004752 214.4226 145.7989 15.82042304784112
1 04813U 70113A 03278.59857437 .00148180 00000-0 68826-3 0 08293
2 04813 081.1221 295.6486 0004401 210.9595 149.1437 15.82965581784595
1 04813U 70113A 03280.55792911 .00134048 00000-0 61003-3 0 08329
2 04813 081.1223 293.1174 0005313 191.9744 168.0177 15.83429471784901
1 04813U 70113A 03283.46424400 .00135765 00000-0 59611-3 0 08359
2 04813 81.1206 289.3566 0005744 186.6670 173.4303 15.84182942785363
1 04813U 70113A 03285.54838630 .00130766 00000-0 56075-3 0 08384
2 04813 081.1214 286.6599 0006317 180.1006 180.0177 15.84687674785698
1 04813U 70113A 03287.50559406 .00131036 00000-0 54880-3 0 08421
2 04813 081.1197 284.1220 0006855 168.2530 191.9043 15.85177317786006
1 04813U 70113A 03290.53454662 .00169181 00000-0 67084-3 0 08440
2 04813 81.1242 280.1920 0006917 155.4282 204.7739 15.86230107786486
1 04813U 70113A 03292.48975154 .00170429 00000-0 65472-3 0 08468
2 04813 081.1240 277.6546 0008560 161.7941 198.3550 15.86861753786797
1 04813U 70113A 03295.38954093 .00185414 11011-4 68772-3 0 08519
2 04813 81.1241 273.8848 0009687 149.5616 210.9386 15.87920975787258
1 04813U 70113A 03297.46829666 .00222014 16493-4 78517-3 0 08578
2 04813 81.1168 271.1727 0007910 139.6535 220.5355 15.88916457787580
1 04813U 70113A 03300.36390218 .00251431 21914-4 82695-3 0 08614
2 04813 81.1150 267.3933 0009304 133.8349 226.1479 15.90345646788045
1 04813U 70113A 03302.43925989 .00349583 44916-4 10648-2 0 08658
2 04813 81.1203 264.6816 0009994 134.5091 225.7002 15.91850879788374
1 04813U 70113A 03303.94703887 .00546453 11850-3 15209-2 0 08749
2 04813 081.1160 262.7070 0011532 142.5110 217.6951 15.93579349788611
1 04813U 70113A 03306.95873938 .00352460 48192-4 86745-3 0 08835
2 04813 081.1199 258.7460 0009832 124.3460 235.8639 15.95898502789092
1 04813U 70113A 03308.96418065 .00422766 72443-4 95295-3 0 08893
245
2 04813 081.1184 256.1009 0009671 120.1266 240.1001 15.97522996789417
1 04813U 70113A 03310.71723168 .00415499 00000-0 86221-3 0 8937
2 04813 81.1185 253.7858 0009558 116.9521 243.3257 15.98996179789691
1 04813U 70113A 03310.90497590 .00400739 66115-4 82772-3 0 8947
2 04813 81.1179 253.5361 0009926 121.1211 239.0993 15.99087488789722
1 04813U 70113A 03310.96755248 .00401319 66364-4 82664-3 0 08952
2 04813 081.1178 253.4534 0009936 120.9248 239.2976 15.99136409789732
1 04813U 70113A 03311.21784190 .00400619 66297-4 81665-3 0 8956
2 04813 81.1182 253.1220 0009778 121.5734 238.6499 15.99324569789770
1 04813U 70113A 03311.28040834 .00405115 67933-4 82361-3 0 08962
2 04813 81.1181 253.0390 0009787 121.3302 238.8884 15.99372479789789
1 04813U 70113A 03311.96851973 .00389440 62951-4 76860-3 0 08998
2 04813 081.1184 252.1282 0009676 118.2178 242.0058 15.99897595789898
1 04813U 70113A 03312.90650518 .00369769 56952-4 70196-3 0 09000
2 04813 081.1182 250.8847 0009955 117.0924 243.1365 16.00572071790048
1 04813U 70113A 03313.96907632 .00378960 60679-4 68826-3 0 09190
2 04813 081.1194 249.4756 0009865 115.5516 244.6793 16.01345369790213
1 04813U 70113A 03314.90616515 .00416542 75046-4 72314-3 0 09057
2 04813 081.1191 248.2302 0009737 113.4854 246.7446 16.02129943790367
1 04813U 70113A 03315.96751417 .00525062 12530-3 85312-3 0 09097
2 04813 081.1179 246.8106 0008399 101.0334 259.1956 16.03298603790532
1 04813U 70113A 03316.96574813 .00447128 90255-4 68230-3 0 09133
2 04813 081.1222 245.4916 0009547 110.0432 250.1970 16.04317379790697
1 04813U 70113A 03317.96338564 .00473646 10372-3 68011-3 0 09261
2 04813 081.1172 244.1579 0009927 108.0841 252.1528 16.05319136790858
1 04813U 70113A 03318.28292739 .00595494 00000 0 11361-2 0 00007
2 04813 81.1130 243.7330 0007641 124.5054 281.5359 16.05801306000006
1 04813U 70113A 03318.58653529 .00659123 21514-3 89919-3 0 9199
2 04813 81.1187 243.3250 0009276 109.2014 251.0234 16.06202743790952
1 04813U 70113A 03318.92267805 .00622954 00000 0 10753-2 0 00000
2 04813 81.1152 242.8781 0007395 91.1929 51.6342 16.06615149000003
1 04813U 70113A 03319.95644028 .00666900 22733-3 81888-3 0 09377
2 04813 081.1193 241.4923 0009350 110.3599 249.8684 16.07893370791171
1 04813U 70113A 03320.21457952 .00734390 00000 0 11473-2 0 00007
2 04813 81.1134 241.1454 0007480 132.5148 281.1607 16.08307530000008
1 04813U 70113A 03320.76518815 .00726583 28057-3 82950-3 0 9259
2 04813 81.1159 240.4046 0009217 108.1032 252.1335 16.09056205791302
1 04813U 70113A 03320.95173895 .00739542 29331-3 82964-3 0 09299
2 04813 081.1157 240.1542 0009261 107.7351 252.4946 16.09331281791330
1 04813U 70113A 03321.20959122 .00763669 00000 0 10737-2 0 00006
2 04813 81.1134 239.8137 0006735 126.9817 286.3209 16.09732187000008
1 04813U 70113A 03322.20381182 .00765924 00000 0 10406-2 0 00001
2 04813 81.1126 238.4747 0006149 134.0085 279.7568 16.11314589000006
1 04813U 70113A 03322.50504095 .00890780 46942-3 85026-3 0 9271
2 04813 81.1134 238.0544 0007736 113.3568 246.9222 16.11875236791589
1 04813U 70113A 03322.51793472 .00899684 00000 0 11166-2 0 00009
2 04813 81.1153 238.0503 0002383 146.6465 288.3970 16.11895199000003
1 04813U 70113A 03323.12561718 .01031289 67765-3 91233-3 0 9281
2 04813 81.1115 237.2286 0006665 111.5337 248.6883 16.13080900791687
1 04813U 70113A 03323.19692121 .00989416 00000 0 11446-2 0 00007
2 04813 81.1106 237.1330 0006185 133.4271 280.5959 16.13202548000005
1 04813U 70113A 03323.50435185 .00000000 00000 0 00000 0 0 01
2 04813 81.1159 236.7152 0006966 106.3650 292.1817 16.13686464 02
1 04813U 70113A 03323.51066954 .00996600 00000 0 11085-2 0 00004
2 04813 81.1119 236.7073 0001860 156.3576 278.8780 16.13797065000000
objects

246
1 04813U 70113A 03323.55974904 .00916460 52249-3 76711-3 0 9303
2 04813 81.1106 236.6412 0006854 116.7922 243.3804 16.13843733791752
1 04813U 70113A 03323.93168725 .01010183 66794-3 80344-3 0 09351
2 04813 081.1112 236.1367 0007011 115.9761 244.2219 16.14620114791815
1 04813U 70113A 03324.24905095 .01082780 00000 0 10957-2 0 00006
2 04813 81.1066 235.7135 0021155 123.2501 280.9023 16.15239687000006
1 04813U 70113A 03324.50249334 .01097190 00000 0 88149-3 0 00002
2 04813 81.1086 235.3670 0003232 170.3220 266.7985 16.15827016000003
1 04813U 70113A 03324.73693552 .01137051 00000-0 78499-3 0 9347
2 04813 81.1096 235.0472 0007309 104.6741 255.5825 16.16425166791948
1 04813U 70113A 03324.88591270 .01780856 00000 0 12622-2 0 00002
2 04813 81.0987 234.8511 0017380 99.3500 47.6485 16.15876075000000
1 04813U 70113A 03324.92257995 .01292170 -92825-6 85457-3 0 9355
2 04813 81.1105 234.7916 0007885 131.2622 229.0098 16.17012618791977
1 04813U 70113A 03325.35560345 .02280422 -93688-6 12478-2 0 9368
2 04813 81.1084 234.2020 0006339 96.9747 264.9318 16.19598243792049
1 04813U 70113A 03325.54062494 .02104766 -93516-6 10914-2 0 9371
2 04813 81.1053 233.9524 0005670 102.2878 257.8933 16.20313330792074
1 04813U 70113A 03325.55418544 .02107370 00000 0 12249-2 0 00009
2 04813 81.1153 233.9410 0007122 143.1373 296.1471 16.20299417000004
1 04813U 70113A 03325.84932404 .01877319 -92902-6 89790-3 0 9390
2 04813 81.1056 233.5269 0005033 125.5276 234.6439 16.21366514792125
1 04813U 70113A 03325.91103990 .01890889 -92680-6 88999-3 0 9400
2 04813 81.1030 233.4409 0004778 129.7803 230.3884 16.21575120792132
1 04813U 70113A 03326.21949397 .01981157 -93206-6 85457-3 0 9414
2 04813 81.1051 233.0211 0005054 121.8200 238.3377 16.22672338792184
1 04813U 70113A 03327.45763889 .00000000 00000 0 00000 0 0 04
2 04813 81.1063 231.3130 0003472 127.0351 272.9554 16.28879919 01
1 04813U 70113A 03327.52470383 .03336303 00000 0 94049-3 0 00007
2 04813 81.1070 231.2241 0007673 159.1829 273.9243 16.29246495000006
1 04813U 70113A 03327.90033565 .00000000 00000 0 00000 0 0 05
2 04813 81.1031 230.7020 0002953 126.6548 350.1871 16.32283301 03
1 04813U 70113A 03328.18674196 .05388918 -91023-6 75431-3 0 9522
2 04813 81.1044 230.3031 0002835 124.0428 236.1191 16.35090307792507
1 04813U 70113A 03328.25554295 .05582662 00000 0 93407-3 0 00003
2 04813 81.0969 230.2110 0001741 120.2455 284.7648 16.36064712000007
1 04813U 70113A 03328.30909008 .05955443 -85379-6 71097-3 0 9534
2 04813 81.1005 230.1326 0003298 191.6005 168.6205 16.36590692792524
1 04813U 70113A 03328.37020313 .06523625 -10309-5 72037-3 0 9555
2 04813 81.1125 230.0266 0001880 68.3933 291.6705 16.37323901792535
1 04813U 70113A 03328.37696710 .05795658 00000 0 92883-3 0 00003
2 04813 81.1021 230.0347 0001768 127.7421 272.2141 16.37442032000007
1 04813U 70113A 03328.44316944 .06140315 00000 0 92861-3 0 00000
2 04813 81.1052 229.9501 0002380 188.3448 241.6980 16.38257475000000
1 4813U 70113A 03328.50103009 .00000000 00000 0 00000 0 0 03
2 4813 81.1053 229.8631 0002893 146.4111 264.8096 16.39339929 06
1 04813U 70113A 03328.50355179 .28200639 -50247-6 23294-2 0 9547
2 04813 81.1030 229.8594 0002191 228.5642 197.5056 16.39884351792551
1 04813U 70113A 03328.68993286 -.70063799 -10963-5 -23216-2 0 9551
2 04813 81.1014 229.6006 0007681 183.5405 263.4026 16.44170775792582
Appendix 2 Some of the empirical number mass relationships in the literature on orbit breakups
247
APPENDIX 2

SOME OF THE EMPIRICAL
NUMBER MASS RELATIONSHIPS IN
THE LITERATURE ON ORBIT BREAKUPS

Various relationships are available in the literature for Number Mass relationships of the
fragments generated out of an on orbit breakup or collision. Some of them are

1. For low intensity explosion Su Kessler relationship (1985) is

where M

is the mass of the satellite in grams and m= mass of the fragment in grams
These equations are derived with a reference satellite mass of 1000 kg.

2. The rescaled relationship of the above relation if the satellite mass differs
considerably from 1000 kg.(Reynolds, 1991)

with f
m
is the ratio of the reference mass 1000 kg to the satellite mass M
.

3. For explosions, the relationship by Bess (1975)

) exp(
0
m C N N
m
=
<
>
=
g 1936 m for ), 05756 . 0 exp( 10 69 . 8

g 1936 m for ), 02056 . 0 exp( 10 71 . 1
4
4
m M
m M
N
m
<
>
=
m m
m m
m
f f m
f f m
N
/ g 1936 m for ), 05756 . 0 exp( 869
/ g 1936 m for ), 02056 . 0 exp( 171
objects

248
where N
0
is the total number of fragments and C controls the slope of the curve.

High intensity explosions will have larger values for C and N
0
than low intensity
explosions. C ranges from 0.23 to 0.44 and N
0
ranges from 6921 to 33880 (Johnson
and McKnight 1991) 0.4 for C and 25000 for N
0
(Culp et. al 1989)

4. Master model (Final report of MASTER 2000)

where

N
m
= number of objects of mass larger than m
N
0
= total number of fragments generated
m = mass of the fragments
B = explosion intensity parameter
K = calibration parameter, fixed to 4
c = slope parameter, 0.55

5. Modified power law Applicable to explosions and collisions (Badhwar and Anz
Meador 1989)

where
N
m
= cumulative number of fragments with mass m or greater and
m = mass of the fragment.

c
b m A
m
N

+ = ) (
|
|
.
|
\
|
+
=
c
c
m
m K
m BK
N
exp
kg m m N N N
m m
/ , /
0
= =
249
A is a normalization constant; cut off b is a parameter which governs the mass
values around which the translation takes place from a power law (high mass values)
to a rather flat function. The parameter C, characterizes the steepness of the decrease
towards high mass values.

6. The relationship given by Culp and Mc Knight (1986)
where
N
m
is the cumulative number of fragments with mass m or greater,
m is the mass of the fragment b is the differentiating term,
SF is the shift factor = 10/m
i
with
the spacecraft density and
m
i
is the mass of the smallest fragment.

For low intensity explosion, N
0
=1100, b=10.5 and SF=0.005

7. Relationship with intensity of explosion (Rex 1990)
where
N
m
= cumulative number of fragments with mass m or greater,
m = mass of the fragment,
N
0
= total number of fragments generated,
B = explosion intensity parameter and
K = calibration parameter, fixed to 2

8. Evolve model : High intensity explosion (Reynolds 1991)

) exp(
0
m SFb N
m
N =
|
|
.
|
\
|
+
=
m K
m BK
N N
m
exp
0
objects

250
>
< <
<
=
l
l s m
s s m
m
m m
m m m m g
m m m g
N
for , 0
for , /
for , /

where

9. Sternberg model (1973) for high intensity explosion

10. For collision induced fragments (Culp and McKnight 1985)

where SF=0.1088 and b=0.9067, N
0
= 100

11. Power law distributions by Kessler and Cour Palais (1978) where

N
m
= cumulative number of fragments with mass m or greater
M = mass of the fragment
m
e
= total ejected mass
material explosion intensity high the of mass total
m) 1 ( cutoff mass large
mm) 1 ( cutoff mass small
) / log( /
=
=
=
=
tot
l
s
s l tot m
m
m
m
m m m g
>
< <
< <
=
2
for ,
3
/ exp(
3
2 1
for ,
2
/ exp(
2
1
for ,
1
m m C m N
m m m C m N
m m
cut
m
p
m N
m
N
) exp(
0
m SFb N
m
N =
8 . 0
8 . 0
|
|
.
|
\
|
=
e
m
m
m
N
251

12. Two slope power law of (Frye 1991)

where m

is the largest single fragment mass remaining after impact and M
t
is the total
mass

13. Mclay model (1992) with four degrees of freedom

The variables A, B and C that influence the shape of the curve can be fitted to
approximate observational data. N
0
simply shifts the curve vertically controlling the
total number of fragments.

14. Parabolic distribution model by McKnight (1991)

The constant e identifies the lowest mass the curve is appropriate for. The term c,
shifts the curve up and down along the cumulative number axis.
Ranges of c = (2.1,4.4); d = (0.1,0.8) and e = (3.0,7.4)

< <
|
|
.
|
\
|
<
|
|
.
|
\
|
=
l t
n
t
t
t
m m / M
M
m
A
/ M m
M
m
A
N
1000 for
1000 for
2
2
3 / 2
1
B
c m
Am
N N |
.
|
\
|
+
= 1
0
2
log log
|
|
.
|
\
|
+ = e
M
m
d c N
t
objects

252
15. Hyperbolic model by McKnight and Brechin (1990)

where a, b, c, d, e are constants, n is the log of the number of pieces greater than m,
and m is the log of the mass. One specific set of values are a = 0.7, b = -8.2, c = 4.2,
d = -9.6, e = 12.8

16. The distribution law suggested by Ojakangas (1990)

where A=(1-b)/b, and b should be between 0.5 - 1.0

McKnights relation (1994) between A and b is A = 1.6290 - 1.6636b
2 2
n em dmn cn bm a = + + + +
b
t
M
m
A N
|
|
.
|
\
|
=
Appendix 3 Basics of orbital motion
253
APPENDIX 3

ORBITAL PARAMETERS AND
THEIR PERTURBATIONS

ORBITAL PARAMETERS

The orbital parameters characterizing the satellites are the well known six quantities as
shown in Figure 1. These are (i) a, the semi-major axis, (ii) e, the eccentricity, (iii) i, the
inclination, (iv) , the right ascension of the ascending node, (v) , the argument of the
perigee, and (vi) , the true anomaly. These describe respectively the size, shape, and
inclination of the orbital plane in inertial space, the angular distance from the vernal equinox
to the ascending node, and the angular distance between the ascending node and perigee, and
the angular distance from perigee to the location of the satellite in the orbit. These can be
split into the first, second, and the sixth denoting the position of the satellite in the orbit and
the others denote the orientation of the orbital plane in the inertial space. The definitions of
these orbital parameters are graphically shown in the Figures 1 and 2.

Figure 1 A Graphical Representation of all the Orbital Elements of a Satellite.
New perspectives for analyzing the breakup, environment, evolution, collision risk and
reentry of space debris objects

254
The size and shape of the orbit are described by the semi major axis a and the eccentricity
e. Figure 2 provides a graphical representation of these elements.

Figure 2 A Graphical Representation of a, e and < << <.

PERTURBATIONS OF SATELLITES

The types of perturbations to the satellites orbital parameters due to air drag, gravitational
anomalies, luni-solar perturbations, solar radiation pressure, unexpected thrusting caused by
out-gassing or malfunctioning thrusters and other effects are usually of three types. These are

(i) Short term periodic perturbations,
(ii) Long term periodic perturbations, and
(iii) Secular perturbations.

The first one is with a period less than the orbital period, and the second are those with a
period greater than one orbit, both of which impart oscillatory perturbations. The third type
of perturbations causes divergence of the orbital parameters over time.

Earth

a
ae
C <
S (Satellite)
P (perigee) A
(apogee)
255

SUMMARY OF PERTURBATIONS

There are three main sources of perturbations namely gravitational, atmospheric and luni
solar affecting the different orbital elements in varying ways. The table provided below gives
a brief account of these.

Table 1 Orbital perturbations of close earth satellites

Secular Periodic Sources
Large Small Moderate Small
Gravitational , --- e i, ,
Atmosphere a, e i --- ,
Luni solar --- --- --- a, e, i, ,

Secular means aperiodic variation with time, periodic variations have periods between one
month and one year. Secular perturbations can make large variations in orbital parameters
while periodic perturbations can make only moderate variations.

THE THREE PHASES OF ORBIT DECAY

The effect of the air drag on the perturbation of a near earth satellite is as follows. Firstly the
eccentric orbit due to the effect of the higher air drag experienced at the perigee keeps
loosing its apogee altitude more rapidly than the perigee. Thus the eccentric orbit tends to a
circular orbit. In the next phase the size of the circular orbit shrinks continuously in size. In
the third phase the rapidity with which the altitude decrease is that the satellite reenters the
atmosphere very rapidly and in this process becomes fragmented and these get burnt out
almost completely before reaching the ground. Some well known spacecraft are the Skylab,
and recently the Mir. Also it should be noted that the orbital debris created by the exploded
spent stages or other satellites are continuously reentering the atmosphere.

256

BASIC EQUATIONS OF ORBITAL DECAY DUE TO AIR DRAG

Consider a particle S moving under the attraction of a force :/r
2
per unit mass, which varies
inversely with the square of the distance r from the centre of attraction. If the position vector
of the particle relative to the centre of attraction is r, we have

0
3
= +
r
r
r
t

and the angular momentum h is given by

n p h
2 / 1
) ( =

where n is the unit vector perpendicular to the orbit plane, p is the semi latus rectum defined
as
2 2
/ ) 1 ( h e a p = =
The energy equation is
|
|
.
|
\
|
=
a r
r r
2
1 1
2
1
& &
where the dot () denotes the scalar product.

Suppose now that the orbit perturbed by a force f per unit mass, so that
f
r
r
r = +
3
t

The plane of the orbit is now liable to variation. By differentiating the energy equations and
introducing the drag forces as the perturbation force f, we obtain (King-Hele 1987)
dE e E
E e
E e
a x
and
dE
E e
E e
a a

) (cos
cos 1
cos 1
) cos 1 (
) cos 1 (
2 / 1
2
0
2
2
0
2 / 1
2 / 3
2
+
|
.
|
\
|
+
=
+
=

257
where x = ae.

Expanding the above in a power series of ecos(E) (for e<0.2), we obtain modified Bessel
functions of the first kind of order j, J
j
defined as

Hence utilizing the modified Bessel functions and a simple model for air density, assuming
the density depends solely on the distance r from the earths centre and varies exponentially
with r, we get the rate of change over one revolution in the elements due only to drag as

c
EXP
e O J J J
e
J J
e
J J
e
J
p
a
rev
e
c
EXP
e O J J
e
J J
e
eJ J
p
a
rev
a
+ +
+ + +
=
+ + +
+ + +
=
)
4
( )
4 2
4
01
5 (
16
3
)
3 1
5 (
8
2
)
2 0
(
2
1
2
)
4
( )
3 1
3 (
4
3
)
2 0
(
4
2
3
1
2
0
2
2

where
p
= density at perigee,
c = ae/H, H is the density scale height,
= (QC
D
A) / m,
Q = factor which includes the rotation of the atmosphere (0.9 < Q < 1.1),
2
0
)) cos( exp( ) cos(
2
1
) ( d c j c
j
J

258
m = mass of the satellite,
Js = modified Bessel function of first kind of order s, in all cases having argument c,
C
D
= Drag Coefficient usually taken as 2.2, and
A = Effective reference area.

Hence at k
th
revolution

1 1
1 1
=
k
e
k
e
k
e
k
a
k
a
k
a

Where a
j
, e
j
stands for semi major axis and eccentricity at time instant t
j
, j = k-1, k.

The Model for the Density Variation in the Atmosphere

The model of the atmosphere used here would assume a spherically symmetrical distribution
of particles in which the density varies exponentially according to

where h = Height above the surface of the earth,

0
= Density at height h
0

H = Scale height

(
(

=
H
h h )
0
(
exp
0

Appendix 4 Statistical distributions
259
APPENDIX 4

STATISTICAL DISTRIBUTIONS

Laplace Distribution (McLaughlin 1999)

Laplace distribution is a bilateral exponential distribution, that is, the signed analogue of
exponential distribution. Errors of real valued observations are often follows Laplace
distribution. This is a distribution useful for random variables from a distribution that is more
peaked than a Normal distribution; characterized by two parameters: Location parameter - ,
and Scale parameter -

PDF:
CDF:

Mean = ; variance = 2
2

Binary Mixture of Laplace Distributions

It is a binary mixture of two Laplace distributions has a weighting parameter p. It is very
often referred as double exponential distribution. The interpretation of a binary mixture is
that it represents an undifferentiated composite of two populations having the respective
|
|
.
|
\
|
=
x
x f exp
2
1
) (
>
|
.
|
\
|
|
.
|
\
|
=

y
x
y
x
x F
, exp
2
1
, exp
2
1
) , , (
objects

260
parameters. It has five parameters, namely two location parameters
1
and
2
, two scale
parameters
1
and
2
and a weight parameter p.

PDF:

CDF:

with mean = p
1
+(1-p)
2
; variance = p{2
1
2
- (p-1)(
1
-
2
)
2
}-2(p-1)
2
2

In the same way more than one distribution can be combined to obtain a mixture of
distributions. For example the pdf of tertiary mixture is of the form

Obviously such mixtures have several local maxima and minima.

Log Normal Distribution (McLaughlin 1999)

The Lognormal distribution is the distribution of random variables, which in log space is
following a Normal distribution. The log normal distribution is always right skewed. This is a
two parameter distribution, with mean and standard deviation of the variable in the natural
logarithmic space. It has two parameters namely a location parameter and a scale
parameter

.

|
|
.
|
\
|
+
|
|
.
|
\
|
=
2
2
2 1
1
1
exp
2
) 1 (
exp
2
) (
x
p
x
p
x f
) , , ( ) 1 ( ) , , ( ) , , , , , (
2 2 1 1 2 2 1 1
x F p x pF p x F + =
|
|
|
.
|
\
|

+
|
|
|
.
|
\
|
|
|
|
.
|
\
|

+
|
|
|
.
|
\
|

=
3
3
exp
3
2
)
2
1 (
2
2
exp
2
2
)
1
1 (
1
1
exp
1
2
1
2
) (
x
p x p x p
p x f
Appendix 4 Statistical distributions
261
PDF:

with Mean = exp (+
2
/2) and variance = exp (2+
2
)(exp(
2
)-1)

Weibull Distribution

Weibull distribution is one of the most widely used extreme value distribution in particular
in reliability analysis.

Weibull Probability Density Function

The Weibull distribution has three parameters generally, the location parameter, the 'shape'
parameter and the 'scale' parameter.

The Three-Parameter Weibull Distribution (www.weibull.com, McLaughlin 1999)

The three-parameter Weibull pdf is given by,

where,
|
|
.
|
\
|
|
.
|
\
|
=
2
) log(
exp
2
1
) (

x
x
x f
|
.
|
\
|
|
.
|
\
|

=
x
e
x
x f
1
) (
< < > > , 0 , 0 , , 0 ) ( x x f
objects

262

with

= Location Parameter
= Scale Parameter
= Shape Parameter

The Two-Parameter Weibull Distribution

The two-parameter Weibull PDF is obtained by setting = 0 and is given by,

|
.
|
\
|
|
.
|
\
|
=
x
e
x
x f
1
) (

The formula for the cumulative distribution function of the Weibull distribution is

The estimates of the parameters of the Weibull distribution can be found graphically via
probability plotting paper, or analytically, either using least squares or maximum likelihood.

0 , 0 ; 0 ,
) (
1 ) ( > >
x
x
e x F
Appendix 5 Delta velocity relationships in a breakup
263
APPENDIX 5

DELTA VELOCITY RELATIONSHIPS
IN A BREAKUP

The relation used for the calculation of the delta velocity (V) imparted to the individual
fragments is given by (Culp and McKnight 1985), with the suffix 0 stands for the parent
body and the suffix i for the i
th
fragment,

V
i
=[V
0
2
+ V
i
2
2V
i
V
0
cos (
i
) cos (di)]
1/2

Where
V = the velocity ,
i = inclination and
f = true anomaly
d
i
= | i
0
-i
i
|

cos (
i
) = sin (
0
) sin (
i
) + cos (
0
) cos (
i
)

with
| |
) cos( 1
) sin(
tan
0
1
i
i i
i i
i
i i di
f e
f e
=
+
=

As the altitude of the parent body and the fragments at the instant of breakup is same, the true
anomalies of the fragments and the parent body can be calculated utilizing the equation for
radial distance
objects

264

r = a (1 - e
2
) / [1 + e cos (f)]

where
r = radius vector at the break up, valid for the fragments and parent body
a = semi major axis and e = eccentricity of the parent body or the fragments as
the case may be.

Appendix 6 Components of delta velocities from Lagranges planetary equations

265
APPENDIX 6

COMPONENTS OF DELTA VELOCITIES
FROM LAGRANGES PLANETARY
EQUATIONS

The delta velocity components are derived assuming impulsive fragmentation, uses equations
developed from Lagranges planetary equations (Jehn 1996 and Johnson and McKnight
1987). The equations for the changes in the three orthogonal velocity components can be
obtained from

directions normal and e transvers radial, along velocity delta , ,
parent the of perigee of argument
breakup of at time parent the of anomaly true
up break at distance radial
satellite parent the of motion mean
n inclinatio and ty eccentrici axis, major semi in changes , ,
) 1 (
) cos(
) e - (1

2
) sin(

) 1 (
) sin(
) 1 ( 2
2 / 1 2 2
2
2
2
2
2 / 1 2
=
=
=
=
=
=
(
+
=
(
(
(
(
(
(
(
=
|
.
|
\
|
=
|
.
|
\
|
n t r
n
t
r
V V V
f
r
n
i e a
where
e na
f r
V i
er
a
r
a
f
r
a
e
f e
n
e
A
where
V
V
A
e
a

These linear relations between the changes in the satellite velocity components and the
Keplerian elements are not valid for near circular orbits (Jehn 1996), as the expression for
objects

266
V
r
contains the eccentricity term in the denominator. It is also true that if the breakup
occurs at the perigee point or apogee point, the equations for V
r
and V
t
have singularities.
It can be noted that the true anomaly or eccentricity does not affect the normal velocity
component and for Vn calculation, the results generally agree quite well regardless of
which particular method of calculation is used (Johnson and McKnight 1987).

For breakup in a near circular orbit, Vr is calculated utilizing the following equations (Jehn
1996).

For a given semimajor axis a and eccentricity e,

where V
r
0
is the initial radial velocity component.

|
|
.
|
\
|
=
a r
V
b
1 2
2
) 1 (
2
2
2
e
r
a
V
b
t
=

2 2 2
t r
V V V =
0 2
r r r
V V V =
Appendix 7 Genetic algorithms
267
APPENDIX 7

GENETIC ALGORITHMS

Genetic Algorithms (Kalianmoy Deb 1995, GoldBerg 2000)) have received a great deal of
attention regarding their potential as an optimization technique for complex functions. These
are search algorithms based on the mechanics of natural genetics selection. They combine the
survival of the fittest with a structured yet randomized information exchange to form a search
algorithm. The genetic algorithms (GA) can be considered as a stochastic optimization
techniques where search methods model natural phenomena of genetic inheritance and
Darwinian strife for survival. The metaphor underlying the genetic algorithms is that of
natural evolution. The GAs are different from the traditional algorithms, because

1. GAs work with a coding of the parameter set, not the parameters themselves.
2. GAs search from a population of points, not a single point
3. GAs use payoff (objective function) information, not derivatives or other
auxiliary knowledge
4. GAs use probabilistic and not deterministic transition rules.

A simple genetic algorithm that yields good results in many practical problems is composed
of three operators namely
(1) Reproduction
(2) Crossover and
(3) Mutation

Reproduction is the process by which the proper parents are selected, in accordance with
their fitness, for possible mating to generate off springs. Generally this process is carried out
probabilistically by taking into consideration of the fitness values of the individuals in the
population.

objects

268
Crossover combines the features of two parent solutions to form two offspring by swapping
corresponding segments of parents. Mutation arbitrarily alters one or more bits of the
selected member of the population. The steps for the simplest form of GA are as given below

1. Choose coding to represent problem parameters, Select the criteria for
reproduction , Crossover and Mutation
2. Input the initial population size, probabilities of cross over and mutation,
search domain of the variables, termination criteria or maximum number of
iteration as Tmax
3. Set T=0, Generate initial population from the search domains randomly
4. Evaluate each string of the population for fitness
5. If Termination Criteria is satisfied, or T >T
max
, then STOP
6. Perform reproduction on the population
7. Perform crossover on the population
8. Perform mutation on the population
9. Evaluate the strings of the new population
10. T = T + 1, go to step 4

Here we have constantly assumed probabilities of crossover and mutation as 0.90 and 0.05.

Remarks

It is possible to characterize GA as a search and update based on sets of parameters
rather than a single parameter set, as is generally the case in a deterministic optimization
algorithm. This corresponds to a group of people searching and climbing to the to the top of
the hill rather than a single person searching and trying to make it to the hilltop. Further as in
any optimization algorithm there are tunable parameters for the successful application of the
technique, these are the size of the initial population, convergence criteria, probabilities of
crossover and mutation. These will have to be adjusted depending on the problem.
Appendix 8 A brief introduction to Kalman filter and its implementation using constant gains
269
APPENDIX 8

A BRIEF INTRODUCTION TO
KALMAN FILTER AND ITS
IMPLEMENTATION USING CONSTANT GAINS*

This Appendix provides a brief introduction to Kalman Filter (Ananthasayanam 2004).
Subsequently the reasons and advantages for its implementation using the constant Kalman
gains (Philip 2001, Anilkumar et al 2003) is demonstrated by applying it to a simple
problem. Such a Kalman filter implementation has been utilized in the present work to track
groups of debris fragments and also estimate a meaningful value for their ballistic coefficient.

The Kalman Filter is one of the most interesting and useful innovations of the twentieth
century. This filter owes its origin to the least squares solution proposed by Gauss (1963).
However the scale and magnitude of the many difficult and interesting problems that it has
been able to handle could not have been contemplated by Gauss and anticipated by Kalman
(1960, 1961) himself. Due to the seemingly unpretentious fact of splitting the state and
measurement equations and switching between the state propagation and its update using the
measurements very interesting outcomes have been shown to be possible. Any amount of
deep study and understanding of the state or the measurement equations separately may not
be able to comprehend the exciting possibilities and abilities when both are combined
together. This is similar to the components of a watch, or the cells in an organism leading
respectively to the time keeping ability or life which are not existing in the individual
components.

The Kalman filter (Kalman 1960, 1961) has found applications in many branches of
engineering (Gelb 1970, Maybeck 1982, Candy 1986, Gemson 1991, Brown et al 1997). In
particular the added impetus it has received from aerospace related applications has been
* Extracted from the unpublished lecture notes of Ananthasayanam on State and Parameter
Estimation Techniques at the Indian Institute of Science, Bangalore, INDIA
objects

270
phenomenal (Ananthasayanam 2004, Klein 1989, Hamel et al 1996, Gemson et al 1998). A
complete appreciation of the aesthetics, beauty, and truth of the Kalman filter is available in
Ananthasayanam (2004).

PRINCIPLE OF LEAST SQUARES

SIMPLE LINEAR REGRESSION

Consider the simplest problem of obtaining the linear regression relation

y
i
= x
i
+ , i = 1, 2, 3,.

between the two variables x and y. It is generally assumed that the value of x
i
is exact
but the measured values of y
i
namely

z
i
= y
i
+ w

have an additive noise w. The prediction y
i
from the assumed linear regression model is
compared with the measurements z
i
at the various x
i
s and a cost function

J
1
= (1/N) (y
i
- z
i
) . (y
i
- z
i
)
T

is formed (with the superscript T referring to the transpose of the appropriate quantities, and
denoting the sum over all the N values of i). The above difference (y
i
- z
i
) is known as
the innovation and some call it as residue. The resulting set of equations formed by
differentiating J
1
with respect to the parameters and will have to be solved to estimate
them. The above approach is known as the ordinary least squares (OLS).

271
SYSTEMS WITH MEASUREMENT NOISE ONLY

In many applications it may be necessary to solve the governing differential equations of
motion for many states and these are available only numerically. Also in such cases more
than one measurement may be available. In such cases the cost function is modified as

J
2
= (1/N) (y
i
- z
i
) R
-1
(y
i
- z
i
)
T

where y
i
is the predicted measurement vector from the model state equation and z
i
is the
measurement data vector. The second summation is over as many measurement channels.
This is known as the weighted least squares approach (WLS). In fact MMLE shows that if
the weights R are unknown they can be estimated as

(Ananthasayanam 2000, Philip and
Ananthasayanam 1995)

R = (1/N) ( y
i
- z
i
) ( y
i
- z
i
)
T

where the summation is over the appropriate measurement channel. Such a technique in ET
with only the measurement noise being present (without process noise) is known as the
output error method (OEM). In fact most of the early work on estimating the aerodynamic
stability and control derivatives of airplanes using flight test data centered mainly around this
technique (Klein 1989, Hamel et al 1996).

SYSTEMS WITH MEASUREMENT AND PROCESS NOISE

The most general problem in the state and parameter estimation technique consists of
qualitatively modeling the system, measurement, process and measurement noise
characteristics and to quantitatively determine all the unknown parameters in the above by
combining the information in the model output with the measurement in some suitable
optimal sense. A general nonlinear continuous system and discrete measurement model
(Ananthasayanam 2000, Anilkumar 1999) can be described by
objects

272
dX/dt = F(X, U, , t) + w(t); X(0) = X
0

Z = G(X, U, , t
k
) + v(t
k
); k = 1,2,N
where

X is the state variable of order n,
Z is the measurement variable of size m,
is the set of unknown parameters of order p,
U is the control input vector of size q,
w is the process noise vector of order n and
v is the measurement noise vector of order m.

The further assumptions made regarding the noise characteristics are that w and v are
uncorrelated assumed to be zero mean, white, and Gaussian distributed whose covariance
matrices Q and R respectively are positive semi definite (Sorenson 1970).

The measurement data available at various discrete times are denoted by a subscript k. The
time evolution of the estimate and covariance of state variable X in terms of the transition
matrices and obtained by suitable local linearisation are

X
k
-
=
k-1
X
k-1
+
k-1
U
k-1
; X(0) = X
0

P
k
-
=
k-1
P
k-1

k-1
T
+ Q
k
; P(0) = P
0

Together with the measurement at a time step

Z
k
= H
k
x
k
+ v
k

The superscript (-) and (+) indicate respectively the estimates before and after using the
measurement information. Thus we have at time t
k
an estimate

X
k
-
with covariance P
k
-

273
and another from the measurement namely

Z
k
with covariance R
k

both of which will have to be combined suitably. The above two estimates when combined
optimally (to provide an unbiased and minimum variance estimate) give

X
k
+
= X
k
-
+ K
k
[ Z
k
H
k
X
k
-
] = X
k
-
+ K
k

k

P
k
+
= [ I K
k
H
k
] P
k
-
[ I K
k
H
k
]
T
+ K
k
R K
k
T

where
k
is the innovation and the Kalman Gain

K
k
= P
k
-
H
k
T
[ H
k
P
k
-
H
k
T
+ R
k
]
-1

The above represents the Kalman filter update equations using the measurement. Due to the
linear state and the measurement models and the assumed noise characteristics this
innovation has been shown by Kailath (1969) to be Gaussian and white. After the
problem is solved by a using a suitable numerical algorithm, the innovation is checked for the
above property which when satisfied indicates that (1) the model is acceptable in the
statistical sense and (2) the numerical algorithm has converged. However it is useful to note
that in general practice the above test behaves more like a necessary condition than a
sufficient condition! Otherwise it points to the deficiency in any one or both of the above
aspects. Further whiteness indicates that all the information content from the data has been
fully extracted. If one handles simulated data then the difference between the true and
estimated quantities is an additional check for the consistency of the solution.

The Kalman filter can perhaps be described as an algorithmic approach that assimilates the
external measurement information with possible uncertain system and measurement models
based on probabilistic weighted linear addition of the predicted state and the measurement
data to adapt both the state and measurement models in a statistically consistent way.
objects

274

When a linear system is driven by white noise the Gaussian distributed state X
-
evolves with
modified mean and covariance. The linear measurement equation also provides a Gaussian
distributed Z. Then the conditional probability distribution of X given Z namely p (X| || |Z) is
also a Gaussian. Thus linear systems, measurements, and white noise match to keep X and Z
as a Gaussian. But if either the system or the measurement is nonlinear the Gaussian structure
is altered and then one resorts to quasilinearization at every point, or over the full data as is
appropriate to deal with the problem. The innovation follows a Gaussian distribution whose
probability when maximized leads to the cost function (Maine and Iliff 1981)

J
3
= (1/N)
k
[ H
k
P
k
-
H
k
T
+ R
k
]
-1

k
T

THE DESIGN OF THE KALMAN FILTER

The design of a Kalman filter is to choose suitable values of X
0
, P
0
, Q and R in order to
minimize J
3
with known, inaccurately known, or unknown state and measurement model
together with the input as well as computational errors for

(i) Time varying full matrices Q and R,
(ii) Time varying diagonal matrices Q and R,
(iii) Time invariant diagonal matrices Q and R,
(iv) Work with constant Kalman gain matrix,
(v) Work with important Kalman gain elements

The first one is greatly unwieldy, the second also equally difficult, and the third is generally
the most prolific and preferable. The fourth and fifth ones provide acceptable and more easily
implementable solutions though not optimal.

275
CONSTANT GAIN KALMAN FILTER APPROACH

In many present day problems there are

(1) Unmodellable or modeling errors in the state and measurement equations or
(2) Persistent random unknown control inputs.

When there are no features of the above type the estimation error keeps decreasing
continuously with more measurement data. An example of this is a constant signal! However
such a scenario is a rarity! Even in such cases if the Kalman gain is held fixed after some
amount of data (or the gain is fixed from the very beginning of data) then the error reaches a
steady state corresponding to the above gain. Also when the situation is of the first two kind
then also the Kalman gain reaches a constant value. Assuming that the transients are not too
long the above cost function can be simplified as

J
4
= (1/N) v
k
T
[ ]
- 1
v
k

Now one can see the similarity of this cost function to the OEM with only the measurement
noise existing in the system. Here one needs to propagate only the state equations (and the
covariance equations need not be considered at all) alone and update the states based on the
measurement with the Kalman gains. The Kalman gain which is zero in the classical OEM is
non zero in the present approach and accounts for the process noise! These Kalman gains can
be determined by an appropriate optimization routine so as to minimize the above cost
function J
4
. In the present work the Genetic Algorithm (GA) has been used for such a
minimization. Now similar to the case with measurement noise alone existing in the system
as in the OEM it is possible to estimate the suitable covariance of across the various
measurements. It should be remarked that in general one should handle in the first instance
the estimate and the covariance equations, try to seek the best possible filter parameters by
(say) an adaptive technique to minimize J
3
and obtain the steady state gains. Subsequently
these gains should be compared with the ones by minimizing J
4
and compare the results to
have confidence in the present constant gain approach. Such a confidence has been provided
objects

276
by the work of Philip (2001) on rendezvous and docking, and Anilkumar et al (2003) on
reentry of space objects.

KALMAN FILTER TO COMBINE TWO ESTIMATES OF A QUANTITY

Consider the simplest case of combining two estimates denoted by x
-
and z possessing
respectively S
1
and S
2
as their uncertainties. The update x
+
based on the above estimates is
written as a linear combination like

x
+
= k
1
x
-
+ k
2
z

The above is a probabilistic addition of the estimates with the weights determined from their
uncertainties rather than on their sample values. This means that the weights for the above
mixing are in an overall sense based on all possible combinations of x
-
and z over an
ensemble. The expected taken value over the ensemble gives

E (x
+
) = k
1
E (x
-
) + k
2
E (H x + v)

E (x
+
) = k
1
E (x
-
) + k
2
H E (x) + k
2
E (v).
Thus
x
t
= k
1
x
t
+ k
2
H x
t

leading to k
1
= ( 1 - k
2
H ) since the measurement noise has zero mean and x
t
denotes the true
value. This gives

x
+
= ( 1 - k
2
H ) x
-
+ k
2
z = x
-
+ k
2
( z H x
-
)

This k
2
= K can be called as the Kalman gain. One can straightaway see that the filter
operates between the following two limits. When K = 0 it ignores the measurement
information and does not update the state at all. If K = 1 it always follows the measurement
277
data thus ignoring the information from the state model. The other way of interpreting K is
when measurement is highly uncertain the filter follows the state model and if the state is
highly uncertain it follows the measurement. Since x
-
and z are independent random variables
with variances

S
1
2
and S
2
2
the resulting x
+
will have a variance

S
+2
= (1 - KH)
2
S
1
2
+ K
2
S
2
2

The minimum value of S
+2
with respect to K provides

K = S
1
2
/ ( S
1
2
+ S
2
2
) and S
+2
= (1 KH) S
1
2

If the same quantity x is also the one that is measured then H = 1.

It may be noted that the estimates to be combined are sample values but the weights with
which they are to be combined depend on the ensemble properties like S
1
and S
2
. The
estimates are provided by the model and measurement data but the S
1
and S
2
will have to be
provided by the analyst. In very complex Kalman filter problems these are chosen manually
instead of using optimization techniques and in turn these provide the Kalman gains.

KALMAN FILTER TO ESTIMATE A CONSTANT FROM NOISY DATA

In order to demonstrate the effectiveness of the constant gain Kalman filter approach, we
provide some simulation results for a constant signal corrupted with Gaussian white noise.
Subsequently the constant signal is varied to a sinusoidal one. Based on the variation of
constant Kalman gain we study how the estimation accuracy during the constant phase and in
the later sinusoidal phase changes. It is easy to cast the estimate based on every measurement
x
i
in a sequential form. For a constant signal after processing N measurements with standard
deviation of the noise as the estimate is

E(x) = (x
1
+ x
2
+ x
3
+ . . . x
N
) / N
objects

278
The uncertainty
N
in the estimate after processing N data is given by
2 2 2 2
N

1
. . . . .
1
+ + + =
The value of the Kalman gain to assimilate the N
th
data is K = 1/N. We can now discuss the
results of some experiments as follows.

Figure 1 provides the simulation results averaged over 500 sampled data sets with different
constant Kalman gains used from the very beginning to process the data. Kalman gains. The
variation of the estimates optimal uncertainty / N
1/2
is also shown in the figures. It may be
noted that very small gains takes longer transients and large gains nearly follows the
measurements. It can be noted that a constant Kalman gain of 0.1 behaves well in this case.
Further to demonstrate the efficiency of the present approach in the dynamic situations also,
Figure 2 provides the observations simulated by assuming no dynamics up to 500 seconds
and the signal after 500 seconds has a sinusoidal behaviour and measurement noise values
are added to the signal. The constant gain Kalman filter approach is utilised once again to
track the signal with different constant gains, namely 0.01, 0.1, 0.2, 0.5, 1.0. The optimal
gain is 1/N where N is the number of measurement data to be processed. The Figure 3
provides the results of the constant gain Kalman filter implementation. A zoomed version of
this figure is provided in Figure 4. It may be noted that for smaller gains, the transient takes
more time to settle and there are oscillations in the residual errors. Large Kalman gains
transients are faster converging but does not provide proper filtered output as Kalman gain of
unity follows the measurements and there is no filtering. There is a range of constant gain,
which after a transient follows the signal and later filters the noise in the measured data. Here
it can be observed that 0.5 is the near optimal gain in this case. Another important feature is
that even if the constant gain is varied somewhat away from 0.5 the filter is able to track the
signal. It is this robust gain obtained from simulated results that are utilised even when there
is some difference in the dynamical property of a signal in a real scenario.

In the above experiments the dynamics of the state during the sinusoidal variation was not
taken into account. In other words, state prediction at a later time step k was just the updated
279
value of the state at the previous time step k-1. Further to show that the present concept of
constant gain Kalman filter approach will perform even better if the dynamics of the system
is modelled properly in the state equation, the following experiment was carried out. In this
experiment the state prediction is based on propagating the updated state using the
appropriate sinusoidal variation. The constant gain Kalman filter approach was once again
applied. The Figure 5 gives the actual state together with the predicted states before update. It
may be seen that the predicted state before update follows the dynamics of the actual state.
Figure 6 provides the errors in the estimation. It clearly shows that the behaviour is better
over even a much larger range of gains and even for small gains, in the sense that the
fluctuations in the errors than in Figure 5 have come down. However the very small gain
such as 0.01 leads once again to fluctuations in the estimates.

Thus the lesson from the present set of experiments is that there is some range of constant
Kalman gain, which is able to track the signal and also reduce the effect of noise. The range
is such that even when there is some variation in the dynamics of the signal the filter
performance is quite acceptable.

When further a suitable model is used for the dynamics instead of assuming the signal to be
constant between measurements the performance is even better. In such cases the range of
constant Kalman gain is further enlarged. The constant gain approach involves propagating
only the state equation and not the covariance equations, thereby saving enormous computing
time. In particular it helps in on line real time applications. Such a robust property of the
constant Kalman gain can be utilised to handle more involved problems as well.

objects

280
10
0
10
1
10
2
10
3
10
-2
10
-1
10
0
r
o
o
t

m
e
a
n

s
q
u
a
r
e

e
r
r
o
r
No of time steps
averaged over500 ensembles
K = 0.01
K = 0.1
measurement error variance
sigma/sqrt(n)

Figure 1 Variation of the Error for Different Constant Kalman Gains.
0 100 200 300 400 500 600 700 800 900 1000
0.96
0.98
1
1.02
1.04
1.06
1.08
1.1
1.12
1.14
O
b
s
e
r
v
a
t
i
o
n
s
No. of time steps

Figure 2 Measured Data for Constant Signal Followed by Sinusoidal Variation.
281

Figure 3 Residual Error with different Kalman Gains

Figure 4 Zoomed Residual Error with different Kalman Gains.
objects

282

Figure 5 Actual Signal and the State prediction before Update

References
283
REFERENCES

1. Allahdadi, F., Space Debris Detection and Mitigation, SPIE Proceedings Volume
1951, 1993.
2. Ananthasayanam, M., R., A Relook at the concepts and competence of the Kalman
filter, AIAA-2004-571, AIAA Atmospheric Flight Mechanics Conference, Reno, 2004.
3. Ananthasayanam, M. R., Anilkumar, A. K., Subba Rao, P. V., and Adimurthy, V.,
Characterization of Eccentricity and Ballistic Coefficients of Space Debris in Altitude
and Perigee Bins, IAC-03-IAA5.p.04, IAF Conference, Bremen, October 2003 and also
to be published in the Proceedings of IAF conference, Science and Technology Series,
2003.
4. Ananthasayanam, M. R., Anilkumar, A. K., and Subba Rao, P. V., "A Proposed
Stochastic IMPressionistic Low Earth (SIMPLE) Engineering Model of the Space Debris
Scenario", Fluid Mechanics Report, 2002 FM 5, Dept. of Aerospace Engineering, Indian
Institute of Science, Bangalore, April 2002.
5. Ananthasayanam, M. R., Anil Kumar, A. K., and Subba Rao, P. V., A New Stochastic
Impressionistic Low Earth (SIMPLE) Model of the Space Debris Scenario, COSPAR
02-A-01772, 2002.
6. Ananthasayanam, M. R., A New Approach to the Modeling of Orbital Space Debris,
Key Note Address in Workshop on Space Debris Software Tools, VSSC, Trivandrum,
2001.
7. Ananthasayanam, M.R., Fascinating Perspectives of State and Parameter Estimation
Techniques, Proc. AIAA Conference on Atmospheric Flight Mechanics AIAA-2000-
4319, 2000.
8. Anderson, B. J., "Natural Orbital Environment Guidelines for Use in Aerospace Vehicle
Development", NASA TM 4527, 1994.
objects

284
9. Anderson, B. J., Review of Meteoroids/Orbital Debris Environment, NASA
SSP 30425, Revision A, 1991.
10. Anilkumar, A. K., Ananthasayanam, M. R., Subba Rao, P. V., and Adimurthy, V. , On-
Orbit Collision Probability in LEO Using SIMPLE Model, IAC-03-IAA5.2.09, IAF
Conference, Bremen, October 2003 and also to be published in the Proceedings of IAF
conference, Science and Technology Series, 2003.
11. Anilkumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., Prediction of Re-
entry of Space Debris Objects: Constant Gain Kalman Filter Approach, AIAA 2003-
5393, Proceedings of AIAA AFM Conference, TEXAS, August 2003.
12. Anilkumar, A. K., Initiatives and future challenges in Space Debris studies,
Presentation in one day colloquium Aero Vision 2003, VSSC, Trivandrum, 2003.
13. Anilkumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., Simulation of Some
Historical On-Orbit Breakups Using ASSEMBLE Model, AIAA-2003-5760,
Proceedings of 41
st
Aerospace Science Meeting & Exhibit (Atmospheric Environment),
Reno, January 2003 and also Fluid Mechanics Report, 2002 FM 7, Dept. of Aerospace
Engineering, Indian Institute of Science, Bangalore, April 2002.
14. Anilkumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., " Simple Model:
Characterization of Eccentricity and Ballistic Coefficient of Space Debris in Altitude and
Perigee Bins", Fluid Mechanics Report, 2002 FM 6, Dept. of Aerospace Engineering,
15. Anilkumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., Analysis and
Modeling of PSLV-TES Mission Spent Upper Stage Breakup in LEO, Fluid Mechanics
Report,. 2002 FM 4R, Department of Aerospace Engineering, Indian Institute of Science,
Bangalore, February 2002 and also presented in In House Symposium, Department of
Aerospace Engineering, Indian Institute of Science, Bangalore, March 2002.
16. Anil Kumar, A. K., Ananthasayanam, M. R., and Subba Rao, P. V., A New Modeling
Approach for Orbital Breakup in Space, COSPAR 02-A-01843, 2002.
17. Anil Kumar, A. K., and Subba Rao, P. V., Re-entry Prediction Accuracy Improvement
Using Genetic Algorithm, COSPAR 02-A-01372, 2002.
References
285
18. Anil Kumar, A.K., M.R.Ananthasayanam, Philip, N.K., and Subba Rao, P.V., Adaptive
Kalman Filter Tuning and Controlled Random Search for MMLE with Process Noise.
Proc. AIAA Conference on Flight Mechanics. AIAA-99-4175, 1999.
19. Anz-Meador, P., A Decade of Growth, The Orbital Debris Quarterly News, Volume 5,
Issue 4, October 2000.
20. Anz-Meador, P. D., History of On-Orbit Satellite Fragmentations, JSC 29517, NASA,
July 2001.
21. Anz-Meador, P., HAPS Debris Separation Velocity Distribution, The Orbital Debris
Quarterly News, Volume 5, Issue-2, April 2000.
22. Badhwar, G. D., and Anz-Meador, P. D., Determination of the Area and Mass
Distribution of Orbital Debris Fragments, Earth, Moon and Planets 45, pp 29-51, 1989.
23. Bess, T. D., Mass Distribution of Man made Space Debris, NASA TN D-8108,
Langley Research Centre, 1975.
24. Bouttier F. and Courtier P., Data Assimilation Concepts and Methods. Meteorological
Training Course Lecture Series, 1999.
25. Brown, R.G., and Hwang,P.Y.C., Introduction to Random Signals and Applied Kalman
Filtering, Third Edn, John Wiley and Sons, 1997.
26. Candy J.V., Signal Processing; the Model Based Approach, McGraw-Hill International
Edition, 1986.
27. Chobotov, V. A., Herman, D. E., and Johnson, C. G., Collision and Debris Hazard
Assessment for a Low-Earth-Orbit Space Constellation,Journal of Spacecraft and
Rockets, Volume 34, 1997.
28. Cour-Palais, B. G., Meteoroid Environment Model-1969, NASA SP-8013, 1969.
29. Culp, R. D., and McKnight, D. S., Simulation of Satellite Breakups, AIAA/AAS
Astrodynamics Conference, AIAA 86-2220; Williamsburg, Virginia, 1986.
30. Culp, R.D., and McKnight, D.S., Distinguishing Between Collision Induced and
Explosion Induced Satellite Breakup Through Debris Analysis, AAS/AIAA
Astrodynamics Conference, 1985.
31. Culp, R. D. et al, Estimating and Modeling of the Background Orbital Debris
Population, Advances in Astronautical Sciences, Volume 71, Part I, 1989
objects

286
32. COSPAR International Reference Atmosphere, Compiled by COSPAR Working
Group 4, Akademia-Verlag, GmbH, Berlin, 1972.
33. Deb, K., Optimization for engineering design: Algorithms and examples , Prentice
Hall of India, 1995.
34. Divine, N., Five Populations of Inter Planetary Meteoroids, Journal of Geophysical
Research, Volume 98, 1993.
35. Debris Assessment Software Operators Manual Version 1.5, NASA, April, 2001.
36. Feller, W., "An Introduction to Probability theory and its Applications", Volume 1 Third
Edition, Wiley Eastern Limited, 1950.
37. Flury, W, Preparing For The Future, Volume 4 , No. 4, Space Debris, ESA, 1996.
38. Frye, J. W., "Collision Probability Estimate Method for Impact Generated Low Earth
Orbit Space Debris Clouds", AAS/AIAA Astrodynamics Conference, Durango,
Colarado, 1991.
39. Fucke, W., Fragmentation Experiments for the Evaluation of the Small Size Debris
Population, Proceedings of the First European Conference on Space Debris, ESA Sd
01, Darmstadt, Germany, 1993.
40. Ganeshan, A. S., and Ananthasayanam, M. R., Simulation and Modeling of Orbital
Debris Environment by Equivalent Breakups, Advances in Space Research, Volume 19,
No.2, 1997.
41. Gauss, K.G., Theoria Motus. (in Latin) Theory of Motion of the Heavenly Bodies about
Sun in a Conic Section, New York, Dover 1963.
42. Gelb, A.,(ed.), Applied Optimal Estimation, M.I.T. Press. Cambridge, Massachusetts,
1970.
43. Gemson, R. M. O., and Ananthasayanam, M. R., Importance Of Initial State Covariance
Matrix For The Parameter Estimation Using Adaptive Extended Kalman Filter, AIAA-
98-4153. 1998.
44. Gemson, R. M. O., Estimation Of Aircraft Aerodynamic Derivatives Accounting For
Measurement And Process Noise By EKF Through Adaptive Filter Tuning, Ph. D.
Thesis, Dept. of Aerospace Engineering IISc. Bangalore. 1991.
45. Gnedenko, B. V., The theory of Probability, MIR Publishers, Moscow, 1973
References
287
46. GoldBerg, D. E., "Genetic Algorithms in Search, Optimization & Machine Learning",
Addison-Wesley, 2000.
47. Gregg, W. R., Standard Atmosphere, NACA Report 147, 1922.
48. Guidelines and Assessment Procedures for Limiting Orbital Debris, NSS 1740.14,
1995.
49. Hamel,P.G.,,and Jategaonkar,R.V., Evolution of Flight Vehicle System Identification.
Jour. of Aircraft, Vol, 33, No 1, pp 9-28, 1996.
50. Jehn, R., Nazarenko, A., Ihringer, C., and Walker, R., Comparison of Space Debris
Models in the Centimeter Size Range, Proceedings of the Second European Conference
on Space Debris, ESOC, Germany, 1997.
51. Jehn, R., Modelling Debris Clouds, Shaker Verlag, 1996.
52. Jenniskens, P., Meteor Stream Activity, I. The Annual Streams, J. Astronomy and
Astrophysics 287, 990-1013, 1994.
53. Johnson, N. L., Krisko, P. H., Liou, J. C., and Anz-Meador, P. D., The New Breakup
Model of EVOLVE 4.0, COSPAR Meeting, Warsaw, 2000
54. Johnson, N. L., Major Satellite Breakup in June, Orbital Debris Quarterly News,
NASA JSC, September 1996.
55. Johnson, N. L., et al, History of Soviet/Russian Satellite Fragmentations-A Joint U.S.-
Russian Investigation, Kaman Sciences Corporation, October 1995.
56. Johnson, N. L., and McKnight, D.S., Artificial Space Debris, Orbit Book Company,
1987 and also in 1991.
57. Johnson, N. L., A Preliminary Analysis of the Fragmentation of the Spot 1 Ariane Third
Stage, Technical Report CS87-LKD-003, Teledyne Brown Engineering, Colorado
Springs, March 1987.
58. Kailath T., An Innovations Approach to Least-Squares Estimation, Part 1: Linear
Filtering in Additive White Noise, IEEE Trans. Automatic Control, Volume AC-13, pp.
646-655, 1969.
59. Kalman R.E., and Bucy R.S., New Results in Linear Filtering and Prediction Theory,
Jl. Basic Eng., Volume 83 D, pp. 95-108, 1961.
objects

288
60. Kalman, R.E., A New Approach to linear Filtering and Prediction Problems,
Transactions of the American Society of Mechanical Engineers. Series D, Journal of
Basic Engineering, Volume 82, pp 35-45, 1960.
61. Kessler, D. J., Zhang, J., Matney, M. J., Eichler, P., Reynolds, R. C., Anz-Meador, P. D.,
and Stansbery, E. G., A Computer-Based Orbital Debris Environment Model for
Spacecraft Design and Observations in Low-Earth Orbit, Online version of NASA
Technical Memorandum 104825 , 1997.
62. Kessler, D. J., and Cour-Palais, B. G., Collision Frequency of Artificial Satellites: The
Creation of a Debris Belt, The Journal of Geophysical Research, Volume 83, No. 46,
June 1978.
63. King-Hele, D., Satellite Orbits in an Atmosphere: Theory and Applications, Blackie,
Glasgow and London, 1987.
64. Klein, V., Estimation of Aircraft Parameters from Flight Data, Progress in Aerospace
Sciences, Volume 26, pp 1-77, Pergamon, Oxford, England, UK, 1989.
65. Kling, R. L., and Dowdy, J. S., The Fragmentation of Kosmos 1813, Technical Report
CS87-LKD-004, Teledyne Brown Engineering, Colorado Springs, 8 May 1987.
66. Klinkrad, H., Bendisch, J., Sdunnus, H., Wegener, P., and Westerkamp, R., "An
Introduction to the 1997 ESA MASTER Model", Proc. of the Second European Conf. on
Space Debris, pp. 217-224, ESA SP-393, May 1997.
67. Kozdron, D., Responsibility Regarding Space Debris, Proposal submitted to Prof.
Harvey, 1999.
68. Liou, J. C., Matney, M. J., Anz-Meador, P. D., Kessler, D., Jansen, M., and Theall, J. R.,
The New NASA Orbital Debris Engineering Model ORDEM2000, NASA/TP-2002-
210780, 2001.
69. Loftus, J. P., Jr (Ed), Orbital Debris from Upper Stage Breakup, Volume 121, Progress
in Astronautics and Aeronautics, AIAA, 1989.
70. Maine, R.E., and Iliff, K.W., Use of Cramer-Rao Bounds on Flight Data with Colored
Residuals, Jour. of Guidance, and Control. Vol 4. No 2 , 1981.
71. Makridakis, S., and Wheelwright, S. C., Forecasting Methods for Management, 5
th

Edition, John Wiley & Sons, New York, 1989.
References
289
72. Maybeck, P.S., Stochastic Models, Estimation, and Control, Academic Press, Inc.
Volume 1, 1979 and Volume 2, 1982.
73. McBride, N., Modelling The Meteoroid Environment, A Method for Modelling
Meteoroid Fluxes for Inclusion in an Upgraded ESABASE, 1998.
74. McKnight, D. S., Determination of Breakup Initial Conditions, Jl. of Spacecraft;
Volume 28, No.4, 1991.
75. McKnight, D. S., and Brechin, C. B., "Debris Creation Via Hypervelocity Impact", 28
th

Aerospace Science Meeting, AIAA-90-0084, Reno, 1990.
76. McKnight, D. S., Refined Algorithms for Structural Breakup due to Hyper Velocity
Impact, Hyper Velocity Impact Symposium, Santa Fe, 1994.
77. McLaughlin, M. P., "A Tutorial on Mathematical Modeling", Free downloadable at
http://www.geocities.com/~mikemclaughlin/software/Regress_plus.html, 1999.
78. Mclay, T. D., et al.,"On the Subject of Mass Distribution from Catastrophic
Hypervelocity Impacts, or MAD CHIMP's", Astrodynamics Conference, AIAA-92-4438,
Hilton Head Island, 1992.
79. Meirovitch, L., Methods of Analytical Dynamics, McGraw Hill Company, New York,
1970.
80. Meshishnek, M. J., Overview of the Space Debris Environment, 1995.
81. Minsky., M. The Society of the Mind, Picador Edition , 1988.
82. Myers, K. A., and Tapley, B. D., Adaptive Sequential Estimation With Unknown Noise
Statistics, IEEE Transactions On Automatic Control, Volume AC 21, pp 520-525,
1976.
83. Myers, K.A., and Tapley, B.D., Dynamic Model Compensation for near-earth Satellite
Orbit Determination, AIAA Jl., Vol 13, No.3. , 1975.
84. Myers K.A., Filtering Theory Methods and Applications to the Orbit Determination
Problem for Near-Earth Satellites, Ph. D. Thesis, University of Texas, January 1974.
85. Narasimha, R. and Ananthasayanam, M. R., On Standard Atmospheres, Report No.
1981 FM 12, Department of Aeronautics, Indian Institute of Science. Bangalore, India,
1981.
objects

290
86. Narasimha, R. and Ananthasayanam, M. R., Proposed Standards for Critical
Atmospheric Temperature distributions in India, Report No. 1980 FM 3, Dept. of
Aeronautical Engineering, Indian Institute of Science, Bangalore, India, 1980.
87. Narasimha. R., Performance Reliability of High Maintenance Systems, Journal of
Franklin Institute. Vol 303, pp 15-29. (1975).
88. Nazerenko, A. I., Prediction and Analysis of Orbital Debris Environment Evolution,
Proceedings of First European Conference on Space Debris, Darmstadt, Germany, 1993.
89. Nazarenko, A. I., and Menchikov, I. L., Engineering Model of Space Debris
Environment, Third European Conference on Space Debris, Darmstadt, Germany, 2001.
90. Ojakangas, G. W., et al., "Orbital Debris Environment", AIAA Space Programs and
Technologies Conference, AIAA-90-3863, Huntsville, AL, 1990.
91. On - Orbit Collision Hazard Analysis in Low Earth Orbit Using The Poisson
Probability Distribution, Kaman Sciences Corporation, August, 1992.
92. Orbital Debris Quarterly News, Volume 8, Issue 1, January 2004.
93. Orbital Debris Quarterly News, Volume 7, Issue-1, January 2002.
94. Orbital Debris: A Technical Assessment, Committee on Space Debris, National
Research Council, The National Academics Press, 1995.
95. Orbiting Debris: A Space Environmental Problem Background Paper, OTA-BP-ISC-
72, Washington DC, US Government Printing Office, 1990.
96. Papoulis A., Probability, Random Variables, and Stochastic Process McGraw-Hill,
1991.
97. Philip N.K., and Ananthasayanam M.R., The Method of Maximum Likelihood
Estimation: A Generalized Computer Program, Report No. 95 FM 3. Department of
Aerospace Engineering, Indian Institute of Science, Bangalore, 1995.
98. Philip, N. K., A Study of the Estimation Schemes with Controller for the Final Phase of
a Rendezvous and Docking Mission of Spacecraft, Ph. D. Thesis, Department of
Aerospace Engineering, Indian Institute of Science, Bangalore, July 2001.
99. Position Paper on Orbital Debris, Updated 2000 by the Space Debris Subcommittee of
the International Academy of Astronautics, France, 2000
References
291
100. Protecting the Space Shuttle from Meteoroids and Orbital Debris, National
Academies Press , 1997.
101. Rex, D., et al., "A Review of Orbital Debris Modeling in Europe", Orbital Debris
Conference, AIAA-90-1354, Baltimore, 1990.
102. Reynolds, R. C., Documentation of Program EVOLVE: A Numerical Model to
Compute Projections of the Man-Made Orbital Debris Environment, OD91-002-U-CSP,
1991.
103. Reynolds, R. C., A Review of Orbital Debris Environment Modeling at
NASA/JSC, Orbital Debris Conference; AIAA-90-1355, Baltimore, 1990.
104. Roes, D.,(Ed.)., The COSPAR International Reference Atmosphere 1986. Part 1,
The Thermosphere, Pergamon Press, New York, 1989.
105. Rossi, A., Anselmo, L., Cordelli, A., Farinella, P., and Pardini, C., Modelling the
evolution of the space debris population,
http://apollo.cnuce.cnr.it/~rossi/publications/aquila/aquila.html 1997.
106. Rossi, A. et al., Long Term Evolution of the Space Debris Population, Advances in
Space Research, 19, No.2, 331-340, 1997.
107. Rossi, A., Cordelli, A., Farinella, P., and Anselmo, L., Collisional evolution of the
Earth's orbital debris cloud, J. Geophys. Res., 99, No. E11, 23,195-23,210, 1994.
108. Sdunnus, H., Beltrami, P., Klinkrad, H., Matney, M., Nazarenko, A., and Wegener,
P., Comparison of Debris Flux Models, COSPAR02-A-00986, 2002.
109. Sdunnus, H., Meteoroid and Space Debris Terrestrial Environment Reference Model
MASTER , Final Report to ESOC Contract 10453/93/D/CS, 1995.
110. Sdunnus, H., and Klinkrad, H., An Introduction to the ESA Reference Model for Space
Debris and Meteoroids, Proceedings of the first European Conference on Space Debris,
ESA SD 01, Darmstadt, Germany, 1993.
111. Sehnal, L., "Probability of collisions with artificial bodies in the earth environment",
Advances in Space Research 5, 1985.
112. Sharma, R. K., and Anilkumar, A., K., An integrated approach for risk object re-entry
prediction: Development and validation of KSGEN software, VSSC Technical Report,
VSSC/APMD/TR/073/2003, 2003.
objects

292
113. Sharma, R. K., Anilkumar, A., K., and Xavier James Raj, M., IADC Re-entry
prediction campaign 2003 #1: Re-entry of COSMOS 389 satellite, VSSC Technical
report, VSSC/APMD/SP/082/2003, 2003.
114. Sorenson H.W., Least Squares Estimation from Gauss to Kalman, IEEE Spectrum,
Volume 7, pp. 63-68, 1970.
115. Sternberg, H. M., Fragment Weight Distributions from Naturally Fragmenting
Cylinders Loaded with Various Explosives, Naval Ordnance Laboratory Report AD-
722 480, White Oak, ML, 1973.
116. Strang G. and Borre K., Linear Algebra, Geodesy, and GPS, Wellesley-Cambridge
Press, 1997.
117. Su, S. Y., and Kessler, D. J., Contributions of Explosions and Future Collision
Fragments to the Orbital Debris Environment, Advances in Space Research, Volume
5, No. 2, 1985.
118. Suresh, I., and Anilkumar, A. K., A note on assessment of NASA guidelines for debris
in the context of PSLV-C3 mission, VSSC / APMD/ TM-PSLV/029/2001, 2001.
119. "Space Surveillance", Satellite Times, Volume 4 , Number 1, 1950.
120. Tapley B.D., and Hagar H. Estimation of Unmodelled Forces on a Low-Thrust Space
Vehicle, Journal of Spacecraft, Volume 12. , No.10, 1975.
121. Tribble, A. C., The Space Environment: Implications for Spacecraft Design, Princeton
University Press, 1995.
122. Technical Report on Space Debris, Text of the report adopted by the Scientific and
Technical Subcommittee of the United Nations Committee on the Peaceful uses of
Outer Space, United Nations, New York, 1999.
123. The Ethics Concerning Space Debris, http://cseserv.engr.scu.edu/nquinn/
ENGR019_299Fall2000/StudentWebSites/Leopold/ResearchPaper.htm, 1999.
124. "US Standard Atmosphere", NOAA, NASA and USAF, U.S. Government Printing
Offices, Washington. D.C., 1976.
125. Vedder, J. D., and Tabor, J. L., New Method for Estimating Low Orbit Collision
Probability, Journal of Spacecraft and Rockets, Volume 22, 1991.
References
293
126. Walker, R., Hauptmann, S., Crowther, R., Stokes, H., and Cant, A., Introducing
IDES: Characterising the Orbital Debris Environment in the Past, Present and Future,
Paper AAS 96-113, Advances in the Astronautical Sciences, Volume 93 Part I, 201-
220, 1996.
127. Wegener, P., Bendisch, J., Bunte, K. D., and Sdunnus, H., Upgrade of the ESA
MASTER Model; Final Report, ESOC/TOS-GMA contract 12318/97/D/IM, 2000.
128. Whitley, L. D. (ed.), Foundations of Genetic Algorithms. 2 ", Morgan Kauffmann
Publishers., 1993.
129. Whitten, R. C., and Vaughan, W. W., Guide to Reference and Standard Atmosphere
Models, AIAA-G-003-1988, American Institute of Aeronautics and Astronautics,
1990.
130. Wiesel, W. E., Space flight Dynamics, McGraw-Hill Book Company, New York,
1989.
131. Zang, W. X., and Liao, S. Y., Analyzing the Cause of LM-4 (A)s Upper Stages
Disintegration and the Countermeasures, 5 th International Conference of Pacific
Basin Societies, 6-9 Jun 1993, Shanghai.

objects

294

WEBSITES REFERENCES

1. http://www.geocities.com/iss25544/#TLEs
2. http://www.wingar.demon.co.uk
3. http://apollo.cnuce.cnr.it
4. http://www.wws.princeton.edu
5. http://www.esa.int/
6. http://www.spaceref.com/redirect.ref?url=sn-callisto.jsc.nasa.gov&id=1627
7. http://celestrak.com/
8. http://www.orbitaldebris.jsc.nasa.gov/library/EducationPackage.pdf
9. www.aero.org/cords/orbdebris.html
10. http://sn-callisto.jsc.nasa.gov/model/modeling.html
11. www.weibull.com

AKANIL

Enviado por

Dados do documento

Descrição original:

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

AKANIL

Enviado por

Direitos autorais:

Formatos disponíveis

NEW PERSPECTIVES FOR ANALYZING THE

BREAKUP, ENVIRONMENT, EVOLUTION,

g 1936 m for ), 05756 . 0 exp( 10 69 . 8

Appendix 3 Basics of orbital motion

Você também pode gostar