E539 PDF

Water and Environmental Engineering
Department of Chemical Engineering

Modeling Anaerobic Digestion
-Validation and calibration of the Siegrist model
with uncertainty and sensitivity analysis

Master Thesis by

Oscar Lidholm & Elin Ossiansson

November 2008

Vattenfrsrjnings- och Avloppsteknik
Institutionen fr Kemiteknik
Lunds Universitet

Lund University, Sweden

Modeling Anaerobic Digestion

Master Thesis number: 2008-12 by
Oscar Lidholm & Elin Ossiansson

Lund University

November 2008

Supervisor: Professor J es la Cour J ansen
Examiner: Associate Professor Karin J nsson

Picture on front page:

Figure 26 Examples of scatter plots for the Aalborg household waste
experiment; gas production vs. TS (left) and gas production vs. kH (right)
from MC analysis on characterization and model parameters respectively
(Figure 26).

Postal address: Visiting address: Telephone:
P.O Box 124 Getingevgen 60 +46 46-222 82 85SE-
221 00 Lund +46 46-222 00 00
Sweden Telefax:
+46 46-222 45 26
Web address:
www.vateknik.lth.se
1

I

Summary
Anaerobic digestion is a complex system of biochemical and physical processes. Due to the
complexity of the process, it has traditionally been treated as a black box system, and
optimization has been based on experience or trial and error methods. As experiments of
anaerobic digestion processes are expensive and time consuming, modeling can provide a
useful tool for process understanding and optimization. Models have potentials for revealing
non-linear behaviors of the system and to quantify the performance of alternative operational
setups. The aim of this work was to evaluate the uncertainty, sensitivity and applicability of a
model for anaerobic digestion of mixed sludge published by Siegrist et al (2002) on pilot and
full scale processes.
Experimental data from two pilot scale experiments on household waste and one experiment
on mixed sludge published in Davidsson (2007) were used for validation of the model. The
household waste experiments were operated at meso- and thermophilic temperature
respectively, and the sludge experiment was conducted at mesophilic temperature. For the
household waste experiments, characterizations of the substrates were available and the
hydrolysis dynamics was determined from batch experiments. The results from the validation
showed that the steady state gas production could be predicted in the household waste
simulations, but not to the same extent in the sludge reactor simulation. The ammonium
correlated better for the sludge validation than for the household waste experiments, but VFA
concentrations needed to be calibrated for all simulations.
To assess the quality of the simulations, uncertainty analysis is required. The effect of
measurement uncertainty on the model predictions was evaluated using Monte Carlo methods.
It was concluded that measurement errors could explain some of the discrepancy between
simulations and data, but that further calibration was needed. It was also concluded that the
household waste characterization measurements reduced the prediction uncertainty
substantially, compared to using a general waste composition. The gas production and
ammonium proved to be highly affected by the quality of the input data while the alkalinity
and VFA were less influenced.
To deepen the understanding of the process, and to find the parameters best suited for
calibration, sensitivity analysis was conducted. The importance of individual input and model
parameters were quantified with a variance-based method with Monte Carlo sampling,
measuring the effect of individual parameter variations on the total variance of outputs. This
method needed to be supplemented with scatter plots, to enable visual evaluation of the
correlation between parameters and output. A drawback of the sensitivity analysis was the
lack of reliable distributions for the model parameters, which produced unreliable results. It
was concluded that better defined distributions would provide a better ground for finding
parameters for calibration.
The most important parameters to determine the gas production were the degradability and the
hydrolysis rate constant. The VFA and alkalinity, on the other hand, were dependent on the
model parameters connected to propionate degradation and acetoclastic methanogenesis. To
calibrate the VFA concentrations, the half saturation constants were changed substantially.
This implies different mass transfer conditions than suggested in the model implementation
by Siegrist et al. (2002). The precision for ammonium prediction could not be improved
without recalibration of the feed protein content. It was therefore concluded that a revision of
the model structure was needed for a successful validation. A slower hydrolysis rate for
protein than for e.g. sugar would be required. After calibration, the model precision was
II

improved significantly for the mixed sludge, while the precision for the household waste
simulations could not be improved to the same extent.
The full scale operation at Kppala waste water treatment plant on Liding was simulated to
validate the model and evaluate various process designs. Data of flows and VS measurements
were used as input to the model, complemented with literature values for the characterization
of the sludge. The simulation gave predictions of gas production and alkalinity with
acceptable accuracy. To evaluate different reactor setups and pretreatment options,
simulations were tested and evaluated with respect to gas production, sludge reduction and
economical viability. The simulations with different scenarios indicated that there are
economical incitements to operate reactors in series, at thermophilic temperature, and to use
enzymatic pretreatment of waste activated sludge.

III

Sammanfattning
Anaerob nedbrytning r ett komplext system av biokemiska och fysiska processer. P grund
av systemets komplexitet har det traditionellt blivit betraktat som ett s.k. black box system,
och optimering har baserats p erfarenhet och praktiska frsk. Eftersom experiment r bde
kostsamma och tidsdande kan modellering vara ett anvndbart verktyg fr kad frstelse
och processoptimering. Modeller har potentialen att exponera olinjrheter i systemet och att
kvantifiera prestandan fr alternativa utformningar av processer. Mlet med det hr arbetet
var att utvrdera oskerheten, knsligheten och anvndbarheten fr en modell av anaerob
nedbrytning av blandslam publicerad av Siegrist et al (2002). Modellen applicerades p
rtning av hushllsavfall i pilotskala och rtning av blandslam i pilot- och fullskala.
Data frn tv experiment i pilotskala med hushllsavfall och ett experiment med blandslam,
publicerade i Davidsson (2007), anvndes fr validering av modellen. Hushllsavfallet
rtades mesofilt och termofilt, och blandslammet rtades vid mesofil temperatur.
Karaktriseringar fr hushllsavfallen fanns att tillg, och hydrolyskonstanten fr ett av
hushllsavfallen och fr blandslammet bestmdes med data frn flaskfrsk. Resultaten frn
valideringen visade att gasproduktionen vid stabil drift kunde simuleras med god
trffskerhet, medan trffskerheten fr simuleringen med blandslam inte var lika god.
Mtningar och simuleringar fr ammonium korrelerade bttre fr valideringen med blandslam
n fr hushllsavfallen, medan koncentrationen av flyktiga organiska syror (FOS) behvde
kalibreras fr samtliga fall.
Oskerhetsanalys fordrades fr att bedma kvaliteten av simuleringarna. Effekten av
mtoskerheter p modellens prediktioner utvrderades med en Monte-Carlo metod.
Slutsatsen var att mtoskerhet kunde frklara en del av skillnaderna mellan simuleringar och
data, men att kalibrering av parametrar behvdes fr att ka korrelationen. Det kunde ven
observeras att en uppmtt karaktrisering vsentligen kunde minska oskerheten i utdata
jmfr med att anvnda en generell karaktrisering fr hushllsavfall. Gasproduktionen och
ammonium visade sig vara mycket pverkade av kvaliteten hos indata, medan alkaliniteten
och FOS var mindre pverkade.
Fr att ka frstelsen av processen och hitta de parametrar som r bst lmpade fr
kalibrering genomfrdes en knslighetsanalys. Vikten av infldes- och modellparametrarna
kvantifierades med en variansbaserad metod, och prover renderade med Monte-Carlo
metoden anvndes. Denna metod behvde kompletteras med punktdiagram fr att visualisera
korrelationen mellan parametrarna och utdata. En nackdel med knslighetsanalysen var
avsaknaden av trovrdiga frdelningar fr modellparametrarna, vilket ledde till otillfrlitliga
resultat. Bttre definierade intervall fr modellparametrarna skulle mjliggra en mer
tillfrlitlig knslighetsanalys och drmed ge bttre grund fr att hitta lmpliga parametrar att
kalibrera.
De viktigaste parametrarna fr att bestmma gasproduktionen var nedbrytbarheten och
hydrolyskonstanten. Alkaliniteten och FOS var andra sidan beroende av
modellparametrarna kopplade till acetogenes av propionsyra och acetoklastisk metanogenes.
Halvmttnadskonstanterna fr dessa processer kades avsevrt i kalibreringen, vilket visar p
frndrade betingelser fr masstransport n vid modellkalibreringen av Siegrist et al.
Trffskerheten fr ammoniumprediktionen kunde inte frbttras utan att kalibrera
proteininnehllet i substratet. Slutsatsen av detta var att en lgre hydrolyshastighet fr protein
n fr t.ex. kolhydrater skulle krvas fr en lyckad validering. Efter kalibreringen kade
korrelationen avsevrt fr simuleringen av blandslam, medan precisionen fr simuleringarna
med hushllsavfall inte kunde frbttras i samma utstrckning.
IV

Fullskaleprocessen p Kppala reningsverk p Liding simulerades fr att validera modellen
och jmfra olika processdesign. Data fr flden och VS anvndes som indata i modellen, och
kompletterades med karaktriseringar av primr och sekundrslam frn litteraturen.
Simuleringarna gav en godtagbar trffskerhet fr gasproduktionen och alkaliniteten. Olika
processkonstruktioner simulerades och utvrderades med avseende p gasproduktion, VS-
reduktion och ekonomisk nytta. Simuleringarna av de olika scenarierna pvisade att det finns
ekonomiska incitament att infra seriell drift, termofil rtning och att anvnda enzymer fr
frbehandling av sekundrslam.

V

Acknowledgements
This Master Thesis project would not have been possible without the valuable assistance from
our supervisors, Jes la Cour Jansen and Michael Recktenwald (Kemira Recycling
Competence Center). The cooperation has been a pleasure, and owing to your competence we
have gained a lot of insights to the field of anaerobic digestion. Mattias Alveteg has been very
helpful when we accoutered programming and modeling problems. The experimental data
received from sa Davidsson, was used extensively in this dissertation. Many thanks for
assisting us also with supplementary information. We would also like to recognize Andreas
Thunberg for providing us with information on the full scale process at Kppala WWTP.
Hansreudi Siegrist helped us with the implementation of the model, something we are very
grateful for. We would also like to thank our competent opponent, Niklas Borg, and the other
folks at the department for good company at the coffee breaks. Last, but not least, thank you
Emma and Fredric for your loving support.
VI

VII

Abbreviations and symbols

Waste Water Treatment Plant WWTP
Long-Chain Fatty Acids LCFA
Volatile Fatty Acids VFA
Gas-liquid transfer coefficient k
L
a d
-1
Continuously Stirred Tank Reactor CSTR
Upflow Anaerobic Sludge Blanket UASB
Total solids TS
Volatile solids VS
Person Equivalents p.e.
Chemical oxygen demand COD g/m
3

Solubilized COD SCOD g/m
3

Extracellular Polymeric Substances EPS
Waste Activated Sludge WAS
Anaerobic Digestion Model no. 1 ADM1
Sludge Retention Time SRT d
Hydraulic Retention Time HRT d
Ordinary Differential Equations ODE
Vector of liquid phase state variables x
i

Liquid concentrations
Protons
Hydrogen
Methane
Carbon dioxide
Hydrogen carbonate
Ammonium ions
Ammonia
Acetate
Acetic acid
Propionate
Propionic acid
Dissolved amino acids
Dissolved sugars
Dissolved long chain fatty acid
Dissolved inert COD
Particulate degradable COD
Amino acid fermenters
Sugar fermenters
Long chain fatty acid degraders
Propionate degraders
Acetotrophic methanogens
Hydrogenotrophic methanogens
Inert particulate matter
Partial pressures in gas bubbles
Hydrogen
Methane
Carbon dioxide

S
H

S
H2

S
CH4

S
CO2

S
HCO3

S
NH4

S
NH3

S
ac

S
hac

S
pro

S
hpro

S
aa

S
su

S
fa

S
in

X
S

X
aa

X
su

X
fa

X
pro

X
ac

X
H2

X
in

p
H2

p
CH4

p
CO2

mol/m
3

mg COD/m
3

g COD/m
3

mol/m
3

g COD/m
3

bar

Volumetric flow q m
3
/d
Volume V m
3
VIII

Vector of

inflow concentrations c
in
Vector of

outflow concentrations c
out
Vector of reaction rates r
Vector of reaction rates including
gas stripping reactions r
tot
Vector of gas stripping rates

r
stripping
Stoichiometric matrix T
Stoichiometric coefficient

Vector of reaction rates for all processes
Hydrolysis rate constant

k
H
d
-1
Maximum growth rate p
mux
d
-1
Half saturation constant for growth K
S
g COD/m
3

Decay rate k
d
d
-1
Inhibition of process I

Inhibition constant K
I

Chemical equilibrium rate constant k
cq

Acidity constant K
ucdbusc

Partial pressure p bar
Molecular weight MG g COD/mol
Gas constant R bar m
3
mol
-1
K
-1

Temperature T
Proportional constant K
p
Differential Algebraic Equations DAE
Uncertainty Analysis UA
Probability Density Function PDE
Monte Carlo MC
Stochastic Differential Equations SDE
Latin Hypercube Sampling LHS
Number of realizations N
Cumulative Distribution Function CDF
Sensitivity Analysis SA
Model coefficient of determination R
2

Model value for output parameter y
0

Simulated value for output parameter y
0

IX

Contents

1. Introduction ............................................................................................................................ 1
2. Theoretical background to anaerobic digestion ..................................................................... 3
3. Overview of existing models .................................................................................................. 7
4. Mathematical structure of the Siegrist model ........................................................................ 9
5. Numerical aspects ................................................................................................................ 13
6. System identification ............................................................................................................ 15
7. Model validation .................................................................................................................. 17
8. Uncertainty analysis ............................................................................................................. 33
9. Sensitivity analysis ............................................................................................................... 45
10. Sensitivity analysis of the Siegrist model .......................................................................... 49
11. Calibration of the model parameters .................................................................................. 59
12. Industrial scale application ................................................................................................. 71
13. Conclusions ........................................................................................................................ 77
14. Suggestions for further research ......................................................................................... 79
References ................................................................................................................................ 81
Appendices ...................................................................................................................................
Article ...........................................................................................................................................

X

1

1. Introduction
1.2. Background
Increasing awareness of global warming and rising energy prices have led to a growing
interest for renewable energy during recent years. Biogas, consisting mainly of methane, is an
energy carrier with many advantages. It can be produced from a variety of substrates,
preferably from different types of wastes, in the process of anaerobic digestion. The technique
is widely used at waste water treatment plants (WWTPs) for reducing the waste volume and
stabilizing the sludge produced from waste water treatment. The process can also be applied
to treat e.g. household and agricultural waste, turning them into valuable resources and
solving a waste handling problem. Greenhouse gas reduction is not only attained by the
replacement of fossil fuels, but digesting household waste, sludge and manure prevents direct
emission of the strong greenhouse gases methane and nitrous oxide to the atmosphere. The
residues from the digestion process can be used as fertilizer on farmland, and the carbon
dioxide produced from the methane combustion is thus taken up by plants. Plants which once
again can be used for biogas production, thus the carbon cycle is closed.
1.3. Why model anaerobic digestion?
To increase the biogas production capacity at WWTPs, different alternatives can be
considered. Serial digestion and thermophilic temperature are possible measures for improved
process efficiency. Another technique is to increase the degradation of the sludge with
different methods of sludge pretreatment. At many facilities, the possibilities of including
other wastes than sludge, e.g. household waste is considered as a way to increase the biogas
production. Before implementation of a new process design, it is essential to evaluate the
alternatives. Pilot plant and lab scale experiments are important tools, but costly and time
consuming.
Mathematical models can serve as useful tools to deepen the understanding of complex
systems, and to facilitate operation and design of the process. If the behavior of a system can
be predicted, the production can be optimized and process failure can be prevented. More
effective processes could lead to a better competitiveness for biogas as an energy carrier.
Despite of these motivations modeling has rarely been applied on anaerobic digestion
(Batstone, et al., 2003). The obstacles for introducing modeling to the industry are among
others that the models of anaerobic digestion are complex and require extensive input data,
and that the performance of the models on full scale processes has not yet been tested
(Batstone, 2006). It is therefore interesting to perform validations and uncertainty/sensitivity
analysis of the models to gain knowledge that will facilitate model application.
1.4. The aim of this work
The aim of this work was to evaluate the uncertainty, sensitivity and applicability of a
mathematical model on pilot scale digestion of sludge and household waste. The model for
anaerobic digestion of mixed sludge, published by Siegrist et al. (2002) was also used for
studying the potentials of simulating a full scale process with different operational settings
and pretreatments.
Some of the questions that needed to be answered were: What input data is required to use the
model with desirable accuracy? Which model parameters are most influential? Is the model
suitable for simulating household waste digestion? Can the results from a sensitivity analysis
be used for calibration of model parameters? Can the model predict the operation at a full
scale digester, and how can the model be used to evaluate different process designs?
2

Central methods for investigating these questions were uncertainty and sensitivity analysis.
1.5. Outline of the report
An introduction to anaerobic digestion is given in section 2, and an overview of models can
be found in section 3. A presentation of the Siegrist model (section 4) is then followed by a
discussion on the numerical aspects of the implementation (section 5). In section 6, the reader
will be introduced to system identification of models. The first validation of the model,
including description of the data, input parameters, results and reflections is presented in
section 7. The uncertainty analysis, section 8, includes introduction, results and reflections. In
section 9, a literature study on sensitivity analysis is provided as in introduction to section 10,
where the results from the sensitivity analysis on the studied examples are presented. These
results were used in the calibration of parameters (section 11). Examples of application on full
scale digesters, in section 12, give an insight on the practical benefits from model use. Last,
but not least, are the conclusions and proposals for further studies in section 13 and 14. The
last part of the report includes the appendices and the unpublished article Application,
uncertainty and sensitivity of the anaerobic digestion model by Siegrist et al. (2002) on
household waste digestion.

3

2. Theoretical background to anaerobic digestion
2.1. Biochemical reactions
Anaerobic degradation of organic material to methane and carbon dioxide is a complex
system of biochemical reactions. The reactions are commonly divided into four groups of
processes; hydrolysis, acidogenesis, acetogenesis and methanogenesis (Figure 1), each step
involving a specific group of microorganisms. To obtain a stable and efficient process all four
reaction steps need to function, as the processes are connected.

Figure 1 Overview of the biochemical reactions in anaerobic digestion with flows expressed as percent of
COD (from Siegrist et al 2002 based on Gujer et al 1983)
2.1.1. Hydrolysis
In this context hydrolysis refers to the group of reactions that degrades particulate organic
matter, consisting of complex carbohydrates, proteins and lipids to soluble monomers. The
end products are dissolved amino acids, sugars, long-chain fatty acids (LCFA) and dissolved
inert organic matter. The process can be subdivided into biological, chemical and physical
hydrolysis. Biological hydrolysis occurs when microorganisms excrete enzymes such as
lipase, protease and sucrase that attack the substrate. Chemical hydrolysis is often due to weak
acids or bases, which can be added to increase the hydrolysis rate. It can also be effective to
mechanically facilitate the disintegration of particulate matter by treatment with ultrasound or
with thermal treatment. These processes are examples of physical hydrolysis.
Hydrolysis is often rate limiting when the particulate matter is not readily degradable or in
systems with high loading rates (Davidsson, 2007). Even though the dynamics of hydrolysis
of some individual substrates are known, the process is often described as a simple first-order
process due to extensive variations in substrate composition (Batstone, 2006).

7) Hydrogenotrophic
methanogenesis

6) Acetotrophic
methanogenesis
14%
45%
5%
31%
7%
28% 67%
20%
3%
9%
29%
12%
50%
Amino acids (S
aa
), Sugars (S
su
)
Inert diss. COD (S
in
)
Long chain fatty acids (S
fa
)
Propionate (S
pro
)
Acetate (S
ac
) Hydrogen (S
H2
)
Methane (S
CH4
)
1) Hydrolysis
Fermentation of
2) amino acids
3) sugars
4) Anaerobic oxid. of S
fa
5) Anaerobic oxid. of S
pro
Degradable particulate organic material (X
S
)
Proteins Carbohydrates Lipids
30% 17% 5% 48%
4

2.1.2. Acidogenesis
Acidogenesis refers to the fermentation of dissolved sugars, amino acids and LCFA to volatile
fatty acids (VFA), carbon dioxide and hydrogen. These processes are energetically favorable
for the microorganisms, and are seldom rate limiting.
2.1.3. Acetogenesis
The VFA produced from acidogenesis consist of short chained fatty acids like acetate,
propionate and butyrate. During acetogenesis, a group of strict anaerobe Achaea oxidizes
VFA of higher order than acetate to acetate and hydrogen, which can be used to produce
methane in the methanogenesis. The energetic yields for these reactions are negative, and the
microorganisms need to co-metabolize with methane producing bacteria in order to make this
degradation step energetically feasible. Acetogens are easily inhibited by environmental
factors like pH, NH
3
, etc., and can thus be rate limiting and a source of instability to the
process. Accumulation of VFA is an indication of process failure. The dissolved hydrogen
concentration is a crucial parameter as it determines the energetic yield from the reaction
where VFA are degraded to acetate and hydrogen (Nilsson et al, 2007).
2.1.4. Methanogenesis
Methane producing bacteria can be divided into two groups; acetotrophic methanogens and
hydrogenotrophic methanogens. Acetotrophic methanogens produces methane and carbon
dioxide from acetic acid, and hydrogenotrophic methanogens produces methane from
hydrogen and carbon dioxide. Methanogenesis requires neutral pH, otherwise the gas
production is inhibited (Reith, et al., 2003).
2.1.5. Environmental requirements
The microbial processes are dependent on suitable environmental factors when it comes to
e.g. pH, temperature, nutrient levels and the levels of toxicants. The processes will be
inhibited if the requirements are not met. The reactor temperature is generally mesophilic
(355 C) or thermophilic (555C). These temperature ranges provide optimal conditions for
different populations of microorganisms. The reactions in a mesophilic reactor are not as fast
as in the thermophilic case, but mesophilic reactors are more stable and not as sensitive to
ammonia inhibition. Problems with instability are most likely to occur if easily degradable
substrates are digested in a thermophilic digester (Davidsson, 2007).
2.2. Physicochemical processes
2.2.1. Equilibrium processes
Important equilibrium reactions in anaerobic digestion systems are the acid-base reactions of
VFA, ammonium and the carbonate system. VFA affect the alkalinity and pH of the system,
which in turn can have inhibitory effects on the processes. Ammonia inhibition occurs mainly
at high pH, when the deprotonated ammonia is favored.
2.2.2. Gas stripping
The liquid-gas transfer is a crucial parameter for a bioreactor. A typical anaerobic digester
consists of a mixture of liquid phase, particles and gas bubbles. In a simplified manner, the
transport of dissolved gas to gas bubbles can be modeled as a two-phase system using
stationary liquid-film theory. All mass transport resistance is then assumed to be in the liquid
phase, and the kinetics is based on the overall transfer coefficient, k
L
a, which includes the
liquid-gas film area, the diffusivity of the specimen and the layer thickness. The k
L
a value
also depends e.g. on the degree of stirring and the temperature. The high concentrations of
dissolved gases in anaerobic digesters often led to supersaturation of the liquid. The effect of
5

particles that work as catalysts for the gas stripping process is neglected in the two-phase
system.
2.3. Reactor configuration
A key parameter for the reactor operation is the loading rate. Anaerobic stabilization ponds
are often used for low rate systems, which are often modeled as continuously stirred tank
reactors (CSTR) with liquid inflow and outflow streams and a separate gas collection system.
Examples of high-rate reactors are upflow anaerobic sludge blanket (UASB), expanded
granular sludge bed (EGSB) and anaerobic sequencing batch reactor (ASBR). The benefits of
these high rate reactors are for example less space requirements and odor problems, and a
more effective process (Batstone, 2006). The effectiveness of the reactors can be enhanced by
increasing the plug flow characteristics and by increasing the biomass retention. Two-step
digestion where the fermentation and methanogenesis occur in different reactors can also be
applied. The aim for this configuration is to make the process more stable, since the process
variables can be adapted for the fermenters and methanogens respectively. The first step is
usually thermophilic and the second step mesophilic (Blumensaat F, 2005). Thus, the
concentration of inhibitory ammonia can be kept lower in the second reactor where the
bacteria are more sensitive (Davidsson, 2007).
2.4. Substrate characterization
Total solids (TS), volatile solids (VS), total and solubilized chemical oxygen demand (COD,
SCOD) are common parameters to measure in the incoming substrate at digestion facilities. If
the fractions of fat, protein and carbohydrates are known, the theoretical methane yield can be
determined using the Buswell formula (eq. 2.1) (Buswell, et al., 1930).

C
n
E
u
0
b
N
c
+[n -
u
4
-
b
2
+
3c
4
E
2
0 - [
n
2
-
u
8
+
b
4
+
3c
8
C0
2
+[
n
2
+
u
8
-
b
4
-
3c
8
CE
4
+cNE
3
(2.1)

The methane production capacity can thus be expressed as [
n
2
+
u
8
-
b
4
-
3c
8
mole per mole
organic compound, (C
n
E
u
0
b
N
c
). If combining the Buswell formula with the oxygen demand of
the organic substance, the theoretical methane production potential per unit of COD can be
determined. The oxygen demand can be expressed as (eq. 2.2).

[n +
u
4
-
b
2
-
3c
4
mol O
2
per mole organic substance (2.2)

The average elemental compositions of the organic compounds used in this project are
presented in Table 1. These compositions were calculated for household waste samples in
Hjlund Christensen et al. (2003). Fat is a beneficial substrate with a much higher biogas
potential than both sugars and proteins; furthermore it gives a higher methane ratio. High
amounts of protein can on the other hand cause disturbances in the process due to a higher
ammonia concentration (Davidsson, 2007).

Table 1 Average composition of organic compounds (Hjlund Christensen, et al., 2003)
Compound Elemental composition
Fat
Protein
Carbohydrates
C
57
H
104
O
6
C
5
H
7
NO
2
C
6
H
10
O
5

The theoretical methane yield can be used as an indicator of the quality of the substrate, but
digestion tests needs to be carried out to determine the practical biogas potential. The ratio of
degradable COD to total COD (inert + degradable) will in this text be denoted as the
6

degradability of the substrate. The inert fraction can consist of complex structures of the
mentioned organics, fibers, extracellular polymeric substances (EPS) etc.
The processes covered in this project include digestion of sludge from wastewater treatment
plants and household waste. The sewage sludge from a waste water plant can be subdivided
into primary, secondary and tertiary sludge. Primary sludge is the residue after the mechanical
treatment stage at the plant, often including a chemical precipitation process. The secondary
sludge, or waste activated sludge (WAS) usually contains more inert organic material from
dead bacteria and is more difficult to degrade than the primary sludge (Davidsson, 2007).
High sludge age at the activated sludge process leads to more inert material and hence lower
degradation potential in the anaerobic digestion process. Household waste has a large
potential for anaerobic digestion. With improved collection systems, urban waste could be
added to the existing digesters at the wastewater plants, and increase the methane production.
Other commonly used substrates are grease trap waste, manure, food processing residuals or
crops (Davidsson, 2007).
2.5. Sludge pre-treatment
Pre-treatment can be applied to increase the degradability and dewatering properties of the
WAS. This can be done by physical treatments like thermal treatment or ultrasonication or
with chemical treatment, e.g. by acid or base addition or addition of enzymes (Davidsson,
2007). During these treatments, the flocs are disintegrated and both intracellular and
extracellular materials are extracted (Eskicioglu C, 2006). After thermal pre-treatment
experiments carried out by Eskicioglu (2006) the soluble COD increased with over 350 %,
and the digestibility with over 450 %. The rate of the digestion was however lower than for
the control sample, possibly due to denaturation of enzymes during the pre-treatment
(Wawrzynczyk, 2007). It has been shown that thermal treatment in combination with enzyme
addition gives a high methane yield (Davidsson, 2007).

7

3. Overview of models
Anaerobic digestion has traditionally been treated as a black box system due to the
complexity of the process. To facilitate design, system analysis, operational analysis and
control, a mathematical model describing the processes is required (Batstone, 2006). The
different purposes require different ranges of accuracy and model complexity. A complex,
non-linear model with focus on the biochemical reactions is well suited when the
understanding of the process is important, e.g. for operational analysis or for research
purposes. These models can facilitate optimization of operational stability and efficiency.
When implementing model-based control on a system, a linear and well parameterized model
is needed with measurable key parameters as input signals. For design purposes, the model
should focus on hydraulics and particle structure (Batstone, 2006). An example of such a
model is presented in Elmitwalli et al. (2003).
3.1. ADM1
Researchers have strived for a standardization of model structure and parameterization over
the last couple of years (Batstone, 2006). The most widely used model for anaerobic digestion
is the Anaerobic Digestion Model no. 1 (ADM1), developed by a task group for the
International Water Association (IWA) and published in 2002 (Batstone, et al., 2002). This
model has often been used as a framework model enabling researchers to focus on
modifications for specific purposes. The model includes kinetics for disintegration of
homogenous particles to carbohydrates, proteins and lipids, and hydrolysis of these particles
to sugars, amino acids and LCFA. The disintegration process is often rate limiting, and
modeled with first order kinetics with respect to the concentration of particulate matter.
Acidogenesis and acetogenesis are also accounted for, including the dynamics of the VFA
acetate, propionate, valerate and butyrate. Methanogenesis from acetate and hydrogen/CO
2
to
biogas is included. Inhibition functions of the metabolic activity by ammonia, pH, acetate and
hydrogen, and nitrogen limitation are included. The model also describes gas-liquid transfer
and ion association and -dissociation. Implemented as a system of differential equations, there
are 32 dynamic concentration state variables, and as a set of differential and algebraic
equations, 26 state variables and 8 implicit algebraic equations are used (Batstone, et al.,
2002). Examples of known excluded processes in ADM1 are lactate formation, sulphate
reduction, nitrate reduction, LCFA inhibition, competitive uptake of H
2
and CO
2
and chemical
and biological precipitation (Blumensaat F, 2005).

The benefits from using a model to evaluate and optimize anaerobic processes are not fully
recognized by the industry. ADM1 is not widely used due to the lack of implementation in a
commercial program, and due to the low number of case studies (Batstone, et al., 2003). In
Batstone, et al. (2003), two examples of ADM1 application were demonstrated to clarify the
advantages from modeling. Process modifications in the industry were evaluated, and the
model predictions were considered to be of high accuracy.
3.2. The Siegrist model
3.2.1. Comparison with ADM1
The ADM1 model has been regarded as too complex for practical applications (Batstone, et
al., 2003). In parallel to the ADM1 model, Siegrist, et al., (2002) published a slightly more
simplified model oriented towards mixed sludge treatment. The main differences are the
exclusion of valerate and butyrate as state variables. The hydrolysis rate is modeled as a
single step process with first order kinetics with respect to the concentration of particulate
matter, furthermore, the uptake and decay rates are higher than in the ADM1 model. The two
8

models were constructed with different approaches, the Siegrist model parameters are based
on experiments, whereas the ADM1 uses review consensus (Batstone, 2006). The Siegrist
model was calibrated with lab scale experiments and validated with full-scale experiments.
The incoming sewage sludge used to calibrate the Siegrist model was a mixture of primary,
secondary and tertiary sludge from a municipal treatment plant (Siegrist, et al., 2002). The
particulate matter, which is lumped together in the Siegrist model, is divided in sugars, fatty
acids, proteins and inert material with different hydrolysis constants in ADM1. The
particulate substrate composition can however be defined in the stoichiometry of the
hydrolysis.
3.2.2. Model assumptions and limitations of the Siegrist model
Anaerobic digestion is a complex process, and could not be modeled without many
simplifications and without neglecting many processes. Apart from the assumptions regarding
the processes already mentioned in 3.2.1, there are several others to declare, some of them are
stated below:
The fixed stoichiometry in the microbial processes is problematic, especially the
fermentation of sugars (Batstone, et al., 2006). Increased temperature can increase
maintenance and thus alter the stoichiometry.
According to the liquid film theory, mass transfer resistance is assumed to lie only on
the liquid side (Nielsen, et al., 2004). The k
L
a value is only dependent on temperature
in the model, but in reality it is affected by several other parameters.
Although the mass has been reduced in the digestion, the liquid phase is considered to
be dilute and the volume is assumed to be constant. This assumption is however
widely used (Batstone, 2006).
The reactor is assumed to be completely mixed and the sludge retention time (SRT) is
equal to the hydraulic retention time (HRT).
Due to the many assumptions in the model, there are limitations to what can be modeled,
some of them are:
A rapid change in reactor temperature from mesophilic to thermophilic cannot be
predicted due to the change in the microbial population (Siegrist, et al., 2002).
Since all parameters are calibrated for sewage sludge digestion, a change in substrate
may not be modeled well. The particle size of the substrate is not included in the
model.

9

4. Mathematical structure of the Siegrist model
In this section the mathematical structure of the model is presented, as it was implemented in
Matlab. For further reading, see Siegrist et al. (2002).
4.1. Model structure
The model is based on a system of ordinary differential equations (ODE) for the vector of
state variables x, (eq. 4.1).

dx
dt
= F(t, x) (4.1)

The state variables (x
i
) of the system are presented in Table 2. The model includes 23 state
variables of liquid phase concentrations (with units g COD/m
3
, mol/m
3
and g N/m
3
), and three
state variables describing the pressures of gases in the rising gas bubbles (measured in bar).

Table 2 State variables of the system
State variable Notation Unit
Liquid concentrations
1. Protons
2. Hydrogen
3. Methane
4. Carbon dioxide
5. Hydrogen carbonate
6. Ammonium ions
7. Ammonia
8. Acetate
9. Acetic acid
10. Propionate
11. Propionic acid
12. Dissolved amino acids
13. Dissolved sugars
14. Dissolved long chain fatty acid
15. Dissolved inert COD
16. Particulate degradable COD
17. Amino acid fermenters
18. Sugar fermenters
19. Long chain fatty acid degraders
20. Propionate degraders
21. Acetotrophic methanogens
22. Hydrogenotrophic methanogens
23. Inert particulate matter
Partial pressures in gas bubbles
24. Hydrogen
25. Methane
26. Carbon dioxide

S
H

S
H2

S
CH4

S
CO2

S
HCO3

S
NH4

S
NH3

S
ac

S
hac

S
pro

S
hpro

S
aa

S
su

S
fa

S
in

X
S

X
aa

X
su

X
fa

X
pro

X
ac

X
H2

X
in

p
H2

p
CH4

p
CO2

mol/m
3

mg COD/m
3

g COD/m
3

mol/m
3

mol/m
3

mol/m
3

mol/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

g COD/m
3

bar
bar
bar
4.2. Liquid phase differentials

The liquid phase differentials are expressed as for a CSTR (eq. 4.2). In the case of a pulse-fed
digester, the digester is modeled as batch reactor with instant feeding and withdrawal.

dx
dt
=
q
v
(c
n
-c
out
) +r
tut
(4.2)

10

The liquid phase system includes 17 reactions j (gas stripping not included) with individual
reaction rates
j
. The processes and related state variables are listed in Table 3 (compare with
Figure 1). The kinetics will be presented in the following sections.

Table 3 Liquid phase processes included in the model with process number (j) and reactants (i)
j Process i
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Hydrolysis
Fermentation of amino acids
Fermentation of sugars
Anaerobic oxidation of LCFA
Propionate oxidation
Acetotrophic methanogenesis
Hydrogenotrophic methanogenesis
Decay of amino acid fermenters
Decay of sugar fermenters
Decay of LCFA oxidizers
Decay of propionate oxidizers
Decay of acetotrophic methanogens
Decay of hydrogenotrophic methanogens
Equilibrium of CO
2
/HCO
3

Equilibrium of NH
3
/NH
4

Equilibrium of Ac/HAc
Equilibrium of Pro/HPro
X
S

S
aa

S
su

S
fa

S
pro

S
ac

S
H2

X
aa

X
su

X
fa

X
pro

X
ac

X
H2
S
CO2
, S
HCO2

S
NH3
, S
NH4

S
ac
, S
hac

S
pro
, S
hpro

The stoichiometric coefficients
j,i
of the reactions j with respect to the state variables i are
sorted as rows in a stoichiometric matrix, T, (see Appendix I). The stoichiometry is based on
the conservation of COD, nitrogen and electrons, and that the N/COD ratio in e.g. proteins
should remain unchanged (Appendix I). The 23 columns in T each represent one of the liquid
phase state variables (x
i
). The state variable reaction rates r
reaction
for the liquid phase reactions
could thus be written as in equation 4.3.

r
cucton
= T
1
p
1
(4.3)

In equation 4.3 r
reaction
is a column vector expressing the reaction rates of the 23 state
variables, and is a vector with the reaction rates for all processes.
For the stripping of the dissolved gases H
2
, CO
2
and CH
4
(state variables 2-4), the rates
(r
stripping
) were calculated separately (section 4.3.) and added to the vector r
reaction
to get the
total reaction rate r
tot
.
The concentration in the feed was expressed by the vector c
in
and the outflow concentration
vector c
out
also represented the state variables.
4.2.1. Kinetics of microbial reactions
The hydrolysis rate is described with a first order expression with respect to the concentration
of particulate organic matter (eq. 4.4).

p
1
= k
H
X
s
(4.4)

Processes 2-7 (Table 3) are functions of the reactants, inhibitory substances and model
parameters. These processes are modeled with first order kinetics with respect to the biomass
11

concentration X
i
(state variables 17-22 in Table 2) and with Monod kinetics with respect to
the substrate concentration (eq. 4.5).

p
]
= p
mux,]
S
i
K
S,i
+S
i
I
x,]
X
(4.5)

The inhibition factors, modeled with Monod kinetics, are described with equation 4.6.

I
x,]
=
K
I,x,]
z
K
I,x,]
z+S
x
z
(4.6)

x is here the inhibiting variables pH, acetate, ammonia or hydrogen. z is 2 for pH and NH
3

inhibition but otherwise 1. All processes calculated from (eq. 4.5) are slowed down by low pH
values. The anaerobic oxidation of LCFA and propionate are the most sensitive processes,
inhibited by high acetate and hydrogen concentrations and by low pH values. Propionate
oxidation is also inhibited by a high ammonia concentration, as is the acetotrophic
methanogenesis. The hydrogen inhibition of the hydrogen producing processes is due to the
thermodynamics of the reactions.
The biomass decay (Table 3) is modeled as first order reactions with respect to the biomass
concentrations (eq. 4.7). 80 % of the decayed biomass is assumed to become X
S
and the rest
X
in
.

p
]
= k
d,]
X
(4.7)
4.2.2. Protonation- deprotonation
The equilibriums of CO
2
/HCO
3
and NH
4
/NH
3
and protonations of acetate and propionate
were modeled as pseudo-equilibrium processes as in eq. 4.7-4.10. To achieve a more stable
model, the ammonia and VFA were protonated with CO
2
instead of a proton (Siegrist, et al.,
2002).

p
14
= k
cq1
(S
HC0
3
S
H
-S
C0
2
K
C0
2
HC0
3
) (4.7)
p
15
= k
cq2
(S
NH
3
S
C0
2
-S
NH
4
S
HC0
3
K
NH
4
NH
3
K
C0
2
HC0
3
) (4.8)
p
16
= k
cq3
(S
uc
S
C0
2
-S
huc
S
HC0
3
K
hucuc
K
C0
2
HC0
3
) (4.9)
p
17
= k
cq4
(S
po
S
C0
2
-S
hpo
S
HC0
3
K
hpopo
K
C0
2
HC0
3
) (4.10)
4.3. Gas-stripping model
The kinetics of the mass transfer of dissolved gases from liquid to gas phase was based on
stationary film theory (eq. 4.11). These reaction rates were added to the liquid reaction rate
vector r (eq. 4.3).

r
= k
L
o
(S
-S
S,
) (4.11)

The interfacial concentration S
S,i
was expressed with Henrys law (eq. 4.12).

S
S,
=
p
i
Mu
i
H
i
R1
(4.12)

To calculate the pressure differentials in the gas bubbles leaving the liquid, the common gas
law was used. As the molecules leave the liquid phase a pressure is induced in the gas
bubbles. The outflow of gas from the bubbles was modeled with a pressure control loop,
which kept the pressure close to the ambient pressure using the proportional constant K
p
. The
12

differentials for the pressures in the bubbles were calculated with equation 4.13. It is
important that K
p
is high enough to keep the pressure offset low and at the same time not too
high to avoid numerical instability (Siegrist, et al., 2002).

dP
i
dt
=
r
|
v
rccctcr
R1
Mu
i
v
bubblcs
-K
p
[Xp
-(p
tot
-p
H
2
0
) (4.13)

The biogas flow for 25 C and atmospheric pressure for the whole reactor was calculated with
equation 4.15.

q
bogus
=
K
p
[lp
i
-(p
tct
-p
H
2
O
)
p
tct
I
bubbIcs
[
298
1
cbs
(4.14)

4.4. Temperature dependency
The temperature dependencies of all constants were described with exponential expressions
like in equation 4.15.

F(I) = c
0(1-1
0
)
(4.15)

All constants were listed as values at standard temperature (T
0
= 35C), and then multiplied
with the factor from equation 4.15 to get the actual values. are constants determined from
experiments with varying temperature.

13

5. Numerical aspects
5.1. Modeling methods for avoiding numerical problems
ADM1 is a very stiff model, with time constants varying from seconds to months (Rosen C,
2006). The pH calculations are the fastest processes, e.g. the protonation/deprotonation of
VFA and equilibrium of CO
2
/HCO
3
and NH
3
/NH
4
. The problem of stiffness in the model can
be reduced if the fastest processes are assumed to be at equilibrium, (dx/dt=0). This means
that a system of Differential Algebraic Equations (DAE) has been created, a system which is
readily solved with a DAE solver, e.g. ode15s in Matlab. Rosen C, (2006) showed that the
differences between the results from the ODE system and the DAE system were minor.
Another benchmarking made by (Blumensaat F, 2005) proved that the ODE solvers in
Aquasim (ASIM) and Matlab (ode15s) gave very similar results for their adapted ADM1
model. The results from the implementation of Siegrist model in Matlab made in this report
and the same model implemented in ASIM are therefore not expected to differ significantly.
In the model by Siegrist et al, the problem of stiffness has been addressed differently from the
DAE system; the fastest processes have been slowed down instead. The reaction rate
constants of the equilibrium processes are chosen to be large enough to make the processes
faster than the other reactions, but they are very slow compared to the real process rates. The
hydrogen conversions are also very fast and the retention time is lower than for the other
substrates, which leads to a very low concentration of hydrogen. To avoid long simulation
times, the hydrogen concentration is artificially increased 1000 times, thus slowing down the
response time. The authors claim that this does not affect the result, since the response time
still is much shorter than for the other state variables. Another trick that has been used to
increase model stability and create a fast interaction between the acids and bases is to assume
that NH
3
is protonated with CO
2
.
5.2. Numerical methods in the Matlab implementation
Since the model is stiff, the preferred integration method is ode15s which is the stiff ODE-
solver in Matlab. Numerical problems are not to be expected when performing the simulation
and the solver is usually fast. When a pulse-fed reactor is fed at several times, the simulation
time is slower, but still within seconds and not minutes.

14

15

6. System identification
The goal of system identifiability analysis is to check that the model structure and
parameterization is adequate from a mathematical point of view. For nonlinear models,
system identification is complicated, and an analysis lies beyond the scope of this work.
Nonetheless, it is important to be aware of the problems that non-identifiable parameters may
cause when calibrating the model and conducting sensitivity analysis.
6.1. Parameter identifiability
If a parameter is identifiable, it can be uniquely determined. If a parameter is globally
identifiable only one unique solution exists for the parameter value, whereas an infinite
number of solutions exist for an unidentifiable parameter. A model with unidentifiable
parameters can produce the exact same result with many combinations of parameters, e.g. can
the product of two unidentifiable parameters be identifiable. Calibrating the parameter values
from measurements is not possible for a model with unidentifiable parameters (Jeppsson,
1996). Practical identifiability, on the other hand, depends on the quality of the measured
data. Measurements from similar conditions produce many different parameter results which
gives a poor parameter precision. Practical identifiability can be assessed by studying the joint
probability distributions with Monte Carlo analysis (see 8.1.1). Functional relationships may
exist between starting values of dynamical variables and parameters, or between model
parameters.
6.2. Examples of identifiability analysis
Identifiability analysis for dynamic models of anaerobic digestion was carried out by Mller,
et al., (2002). Two mathematical models were tested, one with simple Monod structure, and
one with inhibition and decay coefficients included. For nonlinear models, there are not many
available techniques for structural identifiability analysis; Taylor series expansion and a
differential-algebraic method are among the few. The problem is that the methods work well
for simpler models, but for more complex model they tend to fail. Even a simple Monod
model could not be evaluated with the methods, so the authors developed a new method based
on the covariance matrix of the probability distributions. If there is a functional relationship
between some of the parameters, the joint probability distributions have a correlation and the
covariance matrix will not be of full rank. This method assumes a white noise, which may not
always be the case. These methods will however not be employed in this report.

16

17

7. Model validation
To enable model validation and calibration of the Siegrist model, data from pilot-scale
experiments and a full-scale process were collected. The aim was to evaluate the usability of
the model for different applications, e.g. simulation of the full-scale operation at a WWTP or
pilot plant reactors with varying substrates and temperatures.
The input parameters that needed to be determined before running the simulations were
operational settings, such as the times of feeding, feed volumes, temperature of the reactor
and the properties of the feed. The constituents of the feed needed to be transformed into units
used in the model (mainly based on COD), and the characteristics of the particulate
degradable organic matter (X
S
) needed to be determined. As the available data differed for the
data sets, the methods for computing these calculations varied. Where data was missing,
values from the literature or standard values from the implementation by Siegrist, et al.,
(2002) were used instead.
7.1. Process descriptions
7.1.1. Methods for pilot and lab scale experiments
Pilot and lab scale data was collected from experiments where the biogas potentials of
household wastes were studied. The experiments were carried out by sa Davidsson within
the frame of a Danish-Swedish project (Hjlund Christensen, et al., 2003) and a project for
the city of Malm (la Cour Jansen, et al., 2004). Two sets of continuous data from
experiments on household wastes were used for validation, with waste from Aalborg,
Denmark, and from the city district Vstra Hamnen in Malm. Continuous data from a pilot
scale reactor fed with a mixture of primary sludge and WAS from Sjlunda WWTP in Malm
was used as well. For the experiments on the Sjlunda sludge and on the household waste
from Vstra Hamnen, lab scale batch tests were also available. Here a short description
follows of the continuous and batch experiments, for more detailed information, see
Davidsson, 2007.
The continuous experiments were performed in stirred and heated 35 liter reactors (20 liters
effective volume) with a connected gas collection tank of 77 liters. The reactors were operated
at mesophilic conditions (Sjlunda sludge, Vstra Hamnen household waste) or thermophilic
conditions (Aalborg household waste). The experiments were running over 2-3 months and
the operation was divided into three phases, start-up, continuous operation and post digestion.
During the start-up phase 10 liters of inoculum was added to the reactor and the volume was
gradually increased by increasing the daily feeding until a volume of 20 liters was reached.
Then the continuous phase followed, where the reactors were operated with daily feeding and
effluent withdrawal at constant HRT (15 days for household waste and 13.3 days for sludge).
After achieving a stable process, the feeding and effluent withdrawals were eventually
stopped and the post-digestion phase started. For the Sjlunda sludge process no start-up
phase was needed, since the inoculum was collected at a full scale digester fed with the same
sludge. No post digestion phase data was available for the reactor with household waste from
Vstra Hamnen.
The lab analyses of the continuous processes consisted of daily measurements of the gas
production, methane content, temperature and pH. Weekly analyses of the residue were
HCO
3
, VFA, TS, VS, COD, N-tot and NH
4
-N. For the household waste experiments, the
substrate was stored as frozen samples, and was considered to have constant composition. For
the Sjlunda sludge experiment, the same batch of substrate was used for 4-20 days, and a
new analysis of TS and VS was made for each batch. Characterization of the household
18

wastes was conducted within the projects, and included data of TS, VS, fat, protein, fiber,
sugar, ash and ammonium content. Calculations were needed for these values to be converted
to input parameters for the model, and these will be presented separately for each substrate.
For the Sjlunda sludge, there was no feed characterization made specifically for the
experiment.
The batch experiments were conducted in 2 liter glass bottles filled to about 30 % with a
mixture of inoculum and substrate (Davidsson, 2007). No withdrawal of liquid was conducted
during the test period, but gas was collected and analyzed for methane content a few times per
week. The bottles were stored for 30-50 days in 35C to assure that the maximum total
methane production for the waste was reached. The resulting methane potentials, however,
varied more than acceptable for double samples, and were not used in the validation. The
uncertainty can be due to difficulties in taking homogenous samples, described in (la Cour
Jansen, et al., 2004), or insecurities in the lab method.
7.1.2. The process at Kppala WWTP
Kppala WWTP in Liding receives 500 000 person equivalents (p.e.) and has two treatment
lines, one old and one new. The old line has chemical phosphorus removal while the new line
is equipped with biological phosphorus removal. Both lines include an activated sludge
process with a sludge age of 8-10 days. The amount of primary sludge is stable throughout the
year, but the WAS volume increases during the winter due to low temperature which leads to
slow degradation. After the primary sludge is dewatered, it flows into the first digester, called
R100. This digester is mesophilic (35 C) and the HRT is kept at 15 d. The outflow from
R100 is further digested in R200, together with the dewatered WAS. R200 is also mesophilic,
but with a HRT of only 10 days (Figure 2). Both reactors are stirred, but there are problems
with clogging of the impellers and the stirring speed is only about 5-6 rpm.

Figure 2 The digestion process at Kppala WWTP

7.2. Determination of the hydrolysis constants
The hydrolysis constants for the household waste from Vstra Hamnen and the Sjlunda
sludge were calibrated from the batch experiments described in 7.1.1. The dynamics of the
biogas production during these batch tests was used to determine the hydrolysis constants. It
was assumed that the hydrolysis step was rate limiting, and that there were no inhibition
effects on the microbial reactions. Under this assumption, a first order differential model for
the hydrolysis could be set up. After transferring the values to logarithm scale and applying
linear regression, k
H
was represented by the first order derivative. k
H
was determined to 0.33
19

d
-1
for the Sjlunda sludge and 0.20 d
-1
for the Vstra Hamnen household waste (Figure 3). As
no batch experiment was conducted for the Aalborg household waste, the hydrolysis constant
from Vstra Hamnen household waste was used.
Figure 3 Calibration of k
H
with batch experiment with Sjlunda sludge (left) and household waste from Vstra
Hamnen household waste (right)

7.3. Household waste experiment 1 - Aalborg household waste
The Aalborg household waste substrate originated from source sorted household waste from
the city of Aalborg in Denmark. The waste had been pretreated with a waste separator prior to
the experiments (Hjlund Christensen, et al., 2003). To be able to use the data for model
validation, the experimental data needed to be translated into units used in the model. Manual
calibration was made for the initial value of ammonium and VFA, as no data was available of
the inoculum used in the experiment. The hydrolysis constant was given from the batch
experiments (see 7.2.), but for the other model parameters default values from Siegrist et al.
(2002) was used.
7.3.1. Feed composition
The feed properties used as inputs to the model were the concentrations of particulate organic
material (X
S
and X
in
), dissolved organics (S
su
, S
fa
, S
aa
, S
in
) and other dissolved components (S
H
,
S
NH4
, S
HCO3
, etc).
The composition of the particulate organic matter was also calculated. The degradable
particulate organic material, X
S
, degrades into fractions of dissolved amino acids, sugars, fatty
acids and inert material during the hydrolysis process. The reaction also produces small
amounts of dissolved CO
2
and consumes small amounts of dissolved HCO
3
. The reaction with
stoichiometric coefficients can be written as in equation 7.1, and the default stoichiometric
coefficients from (Siegrist, et al., 2002) are presented in Appendix I (T-matrix).

-X
s
+
1,12
S
uu
+
1,13
S
su
+
1,14
S
]u
+
1,15
S
n
+
1,4
S
C0
2
+
1,5
S
HC0
3
= u (7.1)
Omitting the production of CO
2
and consumption of HCO
3
and introducing new notation
resulted in the simplified hydrolysis reaction described by equation 7.2.
X
s
-
uu
S
uu
+
su
S
su
+
]u
S
]u
+
n
S
n
(7.2)
0 10 20 30 40
-1
0
1
2
3
4
5
Time, [ Days]
G
a
s

p
r
o
d
u
c
t
i
o
n
,

[
N
m
l
/
D
a
y
]

Duplicate 1
Duplicate 2
kh=0.33
0 10 20 30 40 50
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
4
Time, [Days]
G
a
s

p
r
o
d
u
c
t
i
o
n
,

[
N
m
l
/
D
a
y
]

Tr iplicate 1
Tr iplicate 2
Tr iplicate 3
kh=0.2
20

The content of sugar, fat and proteins in X
S
could hence be described with the stoichiometric
coefficients
i
for the hydrolysis process. All species were given in COD-fractions, and due to
the conservation of COD, the hydrolysis could be written as in equation 7.3.
su
+
aa
+
fa
+
in
= 1 (7.3)

Characterization data for fat, protein, fiber, sugar, starch, VFA and ammonium content for the
waste is presented in Table 4 as percentages of TS. The soluble and particulate fractions of TS
were determined by centrifugation (Hjlund Christensen, et al., 2003).

Table 4 Crude data used for validation of the Aalborg household waste experiment (Hjlund Christensen, et al., 2003)
Characterization parameter Value Unit
TS
Protein
Sugar
Starch
Fat
Fibers
Ammonium
VFA
Soluble TS
5
17
7
18
18
11
0.32
0.14
23
%
% of TS
% of TS
% of TS
% of TS
% of TS
% of TS
% of TS
% of TS

To convert the data into units used in the model (g COD/m
3
, g N/m
3
) the Buswell formula
was used (eq.2.2). The average chemical compositions were needed (Table 1), for fibers it
was assumed to be C
5
H
10
O
5
.
To enable validation to the VFA levels in the reactor, some processing of the data were
needed. The analysis method used measured the total VFA level expressed as acetate. Under
the assumption that the measured VFA consisted of 2/3 acetate and 1/3 propionate on a COD
basis, the input variables, S
ac
and S
pro
, could be calculated from the data.
The degradable organics were assumed to be the sum of protein, fat, sugar, starch and fibers
(cellulose). The sugar, starch and fibers were lumped into a sugar fraction to comply with the
model structure. 23 % of the degradable organics X
S
were dissolved components (S
su
, S
fa
, S
aa
,
S
in
) with the same distribution as X
S
. The stoichiometric coefficients for the hydrolysis were
calculated as ratios of protein, sugar and fat to the total COD in X
S
(
in
was set to zero). These
ratios were also assumed to determine the composition of the soluble degradable organics in
the feed. For remaining input parameters, standard values from Siegrist et al. (2002) were
used. The calculated input parameters are presented in Table 5.

Table 5 Input values from the feed characterization for the Aalborg household waste simulation
Characterization parameter Notation
Value Unit
Total degradable organic matter S
aa
+ S
su
+ S
fa
+ Xs
58 193 g COD/m
3

Degradable particulate organic matter Xs
45 071 g COD/m
3

Protein fraction X
S

aa
0.21 -
Sugar fraction X
S

su
0.35 -
Fat fraction X
S

fa

0.45 -
Inert fraction of hydrolyzed X
S

in
0 -
Ammonium S
NH4

160 g N/m
3

21

7.3.2. Initial state variables
The inoculum was collected from a digester at a WWTP in Kalmar. As there were no analysis
data for the inoculum, standard values were applied for the first simulation. Then comparing
with data, it became evident that the concentrations of ammonium and VFA were higher than
first assumed, and these levels were iteratively increased to fit to data. The concentration of
ammonium in the inoculum was set to 3500 g N/m
3
and the VFA content to 720 g COD/m
3
.
7.3.3. Results
The simulation of the experiment included all feeding and withdrawals, simulated as new
initial states for the ODE-solver on these occasions. The feeding was hence made in pulses,
which was closer to reality than if the reactor would be modeled as a continuous reactor. The
frequent oscillations of some outputs were due to this fed-batch mode of operation. The
measured biogas was considered to be dry since it was measured at around 10 C, which
corresponded to a partial pressure of less than 15 mbar for water.
The simulated gas production (Figure 4) was overestimated compared to the measurements
during the first 50 days, but correlated better during the stable period and the post-digestion
phase. The sharp decrease in biogas production after 10 days was due to lower feeding
volumes; it was seen during the experiment that the pH dropped (Figure 6), and the feeding
rate was therefore decreased to prevent overload. The methane content was also overestimated
during the first part of the time series, and the drop in methane fraction when the feeding was
stopped can not be seen in the simulation (Figure 4). For the steady state phase, the simulated
result was well correlated to data.

Figure 4 Measured and simulated biogas production (left) and methane content (right) from continuous experiment
with Aalborg household waste
The totals VFA, however, were undervalued in the simulation and are 300-400 gCOD/m
3
below the

measured values (Figure 5). The fluctuations in the measured values of HCO
3
were
not captured in the simulated result, but the predictions were in the same range as the
measurements (Figure 5).
0 20 40 60 80
0
0.01
0.02
0.03
0.04
0.05
Time [days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

Simulated dr y biogas pr oduct ion
Measur ed biogas pr oduct ion
0 20 40 60 80
0
20
40
60
80
100
Time [days]
P
e
r
c
e
n
t

[
%
]

Simulated CH4
Measur ed CH4
22

Figure 5 Measured and simulated VFA (left) and HCO
3
concentrations (right) from continuous experiment with
Aalborg household waste
The high initial ammonium content was washed out during the start-up of the reactor, but
seemed to reach a stable concentration during the steady state phase (Figure 6). It was
observed that the model estimation of the final value was poor, while the dynamics of the start
up phase is reflected in the simulation. The measured pH was fluctuating (Figure 6), and
differed from the more stable simulated values. The values were however in the same range.

Figure 6 Measured and simulated ammonium concentration and pH from continuous experiment with Aalborg
household waste
7.3.4. Discussion
The inoculum used in the experiment was taken from a full scale sludge fed digester. Thus,
the microorganisms were not used to degrading food waste, and all processes could be
assumed to be slower in the beginning of the experiment. As the state of the biomass is not a
state variable in the model, the start-up phase can probably not be modeled very well with
constant parameters during the whole experiment. It would however be possible to lower the
hydrolysis constant during the start-up phase to improve the validation. The problem with
such measures is that the value of k
H
would need calibration to data and that the theoretical
background is not fully clear.
Another explanation for the poor correlation to the measured results can be that inhibition
from the high ammonium concentration in the reactor was underestimated in the model. A
higher inhibition would cause accumulation of VFA, due to reduced methanogenesis rate and
0 20 40 60 80
0
200
400
600
800
1000
1200
1400
1600
Time [days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

C
O
D
/
m
3
]

Simulated VFAs
Measur ed VFAs
0 20 40 60 80
0
10
20
30
40
50
60
70
Time [ days]
C
o
n
c
e
n
t
r
a
i
o
n

[
m
o
l
/
m
3
]

Simulated HCO3
Measur ed HCO3
0 20 40 60 80
0
500
1000
1500
2000
2500
3000
3500
Time [days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

Simulated NH4 modelled
Measur ed NH4
0 20 40 60 80
0
1
2
3
4
5
6
7
8
Time [days]
p
H

Simulat ed pH
Measur ed pH
23

a drop in pH. These trends could be seen in the data, as well as a lower methane fraction of
the biogas. The model parameters were not calibrated for household waste, thus it is likely
that the validation would be more successful with recalibrated parameters. Calibration of the
model parameters is the subject of section 11.

7.4. Household waste experiment 2 Vstra Hamnen household waste
The operation of the pilot scale experiment on substrate from Vstra Hamnen was similar to
the previously described experiment. In this case the reactor was kept at mesophilic
conditions, and batch experiments were also available. At the 20
th
and 21
st
day of the
experiment, 50 g of HCO
3
was added to the reactor to increase pH; this is also done for the
simulated reactor.
The same methodology as for the Aalborg household waste characterization was applied. The
available data is presented in Table 6, and the calculated input values in
Table 7. Values for the soluble fraction of TS and the VFA concentration of the feed were not
available, so the same values were used as in the Aalborg household waste simulation.

Table 6 Crude data used for validation of the Vstra Hamnen household waste experiment (Hjlund Christensen, et
al., 2003)
Characterization parameter Value Unit
TS
Protein
Sugar
Starch
Fat
Fibers
Ammonium
5
17
4.3
7
11.6
14
0.027
%
% of TS
% of TS
% of TS
% of TS
% of TS
% of TS
Table 7 Input values from the feed characterization for the Vstra Hamnen household waste simulation
Characterization parameter Notation Value Unit
Total degradable organic matter S
aa
+ S
su
+ S
fat
+ Xs
42 248 g COD/m
3

Degradable particulate organic matter Xs
32 721 g COD/m
3

Protein fraction of X
S

aa
0.28 -
Sugar fraction of X
S

su
0.32 -
Fat fraction of X
S

fat

0.40 -
S

in

0 -
Ammonium S
NH4

13.5 g N/m
3

The same methodology as for the Aalborg household waste simulation was used, the initial
ammonium concentration was set to 835 g N/m
3
and the total VFA to 300 g COD/m
3
.
7.4.3. Results
Both the simulated biogas production and the simulated methane content follow the
experimental values more closely during the start up phase than for the Aalborg household
waste experiments (Figure 7). After day 15 there is a major drop in methane content from 60
to 40 %, and a rise in VFA concentration (Figure 8). The measured alkalinity was declining
during this period, but increased after the addition of HCO
3
after 20 days (Figure 8). The
model simulation did not predict this minor reactor breakdown, and even though the VFA
concentration increased, the methane content in the gas, alkalinity and pH remained
24

practically unchanged. The measured ammonium concentration was declining during the
whole experiment, while the simulated ammonium remained fairly constant (Figure 9).

with Vstra Hamnen household waste

3
Vstra Hamnen household waste

Figure 9 Measured and simulated ammonium concentration (left) and pH (right) from continuous experiment with
0 10 20 30 40 50
0
0.01
0.02
0.03
0.04
0.05
Time [days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

0 10 20 30 40 50
0
20
40
60
80
100
Time [days]
P
e
r
c
e
n
t

[
%
]

Simulated CH4
Measur ed CH4
0 10 20 30 40 50 60
0
50
100
150
200
250
300
350
Time [ days]
V
F
A
s

a
s

a
c
e
t
a
t
e

e
q
u
i
v
a
l
e
n
t
s

[
g

C
O
D
/
m
3
]

Simulated VFAs
Measur ed VFAs
0 10 20 30 40 50 60
0
20
40
60
80
100
120
140
Time [ days]
C
o
n
c
e
n
t
r
a
i
o
n

[
m
o
l
/
m
3
]

Simulat ed HCO3
Measur ed HCO3
0 10 20 30 40 50 60
0
200
400
600
800
1000
Time [days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

Measur ed NH4
0 10 20 30 40 50
0
1
2
3
4
5
6
7
8
Time [days]
p
H

Simulated pH
Measur ed pH
25

7.4.6. Discussion
The validation of the gas production was acceptable, whereas the validations of the other
parameters were less successful. One of the purposes of modeling is to be able to prevent
reactor failures, and it can be concluded that the result was poor in this case. The experiment
was designed to keep the feed characteristics constant by freezing daily portions from the
same batch of feed, so it is unlikely that the problems in the reactor around day 20 was caused
by changes in the input. It is possible that the reason was a too high loading rate during the
start up phase, which led to consumption of alkalinity in the fermentation. The poor
correlation for ammonium indicates that the characterization of the waste might not be
reliable; the protein content or the hydrolysis rate seems to have been overestimated which led
to a higher ammonium concentration in the simulated result than for the measurements. VFA
were quickly produced and consumed, which gave large daily variations and made it difficult
to measure and validate the model. For the inhibition constants and reaction rate of the
methanogenesis, it is important that the VFA are in the right range. Methods for calibrating
the VFA concentration will be developed in the following sections.

7.5. Sludge experiment: Sjlunda mixed sludge
The model was originally calibrated with a mixture of primary and secondary sludge by
Siegrist et al. (2002). Hence, default values from the original implementation were assumed to
apply for this experiment when no better guess could be made.
The feeding of the Sjlunda sludge reactor was made from batches of sludge that were used
up after 4-15 days. These batches were analyzed for TS and VS but no analyses for protein,
fat or carbohydrate content were made. Therefore, the alterative characterization method
proposed by Siegrist et al. (2002) was used. The only parameters needed for the
characterization was the ammonium concentration in the reactor at steady state, the COD
reduction and the ratio of CH
4
to CO
2
in the produced biogas. The degraded particulate
organic matter, X
S,red
, was calculated from the experimental values for influent VS and mean
effluent VS at steady state (29 g/l and 14 g/l respectively). To calculate the COD content of
the degraded VS, the default characterization from Siegrist et al. (2002) was used, and the
resulting conversion factor was calculated to 1.9 g COD/g VS.
The mean ammonium concentration in the digester (S
NH4
= 860 g N/m
3
) was used for
calculating the stoichiometric coefficient for protein in the hydrolysis (
aa
) with equation 7.4.
The N-content of protein, i
NH4
, was given the value 0.1 (Siegrist, et al., 2002), for results from
the characterization, see Table 8.

u
po
=
S
NH4
NH4
X
S,rcd
(7.4)

The fat content (
fa
) was calculated from the relation between the real methane fraction in the
biogas and the theoretical methane content rendered from fat, protein and sugar (eq. 7.5). The
real methane fraction in the biogas at steady state (x
CH4,XS
= 0.65), could be calculated from
the known CO
2
pressure and HCO
3
concentration in the reactor with the equations in the
carbonate system (see Appendix II). The fractions of methane from anaerobic digestion of
sugar, protein and fat (x
CH4,su
= 0.5, x
CH4,aa
= 0.68 and x
CH4,fa
= 0.7) were found in VAV
(1981).

uu
x
CH
4,uu
+
]u
x
CH
4,]u
+(1 -
uu
-
]u
)x
CH
4,su
= x
CH
4,X
S
(7.5)
26

The stoichiometric parameter for sugar, (
su
) was calculated from the COD balance, in
equation 7.6. A standard value of 0.05 for the inert fraction (
in
) was used. The results of the
characterization (Table 8) does not differ substantially from the characterization by Siegrist, et
al., (2002) shown in Figure 1.

su
= 1 -
uu

]u

n
(7.6)

The total COD in the inflow was determined using the COD/VS ratio calculated from the
characterization and the Buswell formula (equation 2.2). The batch experiments resulted in a
degradability of 23 % and 54 %, indicating an unreliable method. The degradability based on
data for the continuous process, varied between 40-55 %, the ultimate degradability, however,
was assumed to be 60 % in the simulation. The soluble part of incoming COD was assumed to
be 30 %, and the fractions of the dissolved components were assumed to be the same as for
the hydrolysis stoichiometry. The ammonium concentration was suggested by the
experimenter, sa Davidsson, (2008).

Table 8 Input values from characterization of the Sjlunda sludge
Characterization parameter Notation Value Unit
Total degradable organic matter (mean value) S
aa
+ S
su
+ S
fa
+ Xs
30 785 g COD/m
3

Degradable particulate organic matter (mean value) Xs
21 549 g COD/m
3

S

aa
0.34 -
Sugar fraction of X
S

su
0.16 -
Fat fraction of X
S

fa

0.45 -
Inert fraction of X
S

in
0.05 -
Ammonium S
NH4

45 g N/m
3

The inoculum was collected from a full scale continuous reactor, fed with the sludge used for
the experiment. Steady state values after running the model could hence be used as initial
values.
7.5.3. Results
There was no start up phase as for the household waste experiments, since the inoculum was
taken from a reactor with the same feed and with similar operational settings. This resulted in
a gas production that was rather stable during the whole experimental period, both for the
measurements and simulation. The variations of the measured gas production were however
more pronounced than for the simulation, and the simulated values were also significantly
lower than the measurements. The modeled methane content in the gas seemed to be slightly
overrated (Figure 10). Both the measured and simulated VFA concentrations vary over a wide
range, but the measured values were generally higher (Figure 11). The validation of alkalinity
and ammonium were more successful (Figure 11 and Figure 12), while the simulated values
of pH were slightly higher than the measured values (Figure 12).
27

with Sjlunda sludge

3
Sjlunda sludge

Figure 12 Measured and simulated ammonium concentration (left) and pH (right) from continuous experiment on
Sjlunda sludge
7.5.4. Discussion
The degradability of the Sjlunda sludge used for the simulation was higher than measured,
and the hydrolysis constant was calibrated from the batch experiment. Despite this, the gas
0 20 40 60 80
0
0.01
0.02
0.03
0.04
0.05
Time [days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

0 20 40 60 80
0
20
40
60
80
100
Time [days]
P
e
r
c
e
n
t

[
%
]

Simulated CH4
Measur ed CH4
0 20 40 60 80 100
0
50
100
150
200
250
300
350
400
Time [ days]
V
F
A
s

a
s

a
c
e
t
a
t
e

e
q
u
i
v
a
l
e
n
t
s

[
g

C
O
D
/
m
3
]

Simulat ed VFAs
Measur ed VFAs
0 20 40 60 80 100
0
10
20
30
40
50
60
70
80
Time [days]
C
o
n
c
e
n
t
r
a
i
o
n

[
m
o
l
/
m
3
]

Simulat ed HCO3
Measur ed HCO3
0 20 40 60 80
0
500
1000
1500
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

Measur ed NH4
0 20 40 60 80
0
1
2
3
4
5
6
7
8
Time [days]
p
H

Simulat ed pH
Measur ed pH
28

production was lower for the simulation than for the measured data. It is possible that the
assumption of a non-inhibitory condition during the batch test was wrong. In that case, the
hydrolysis constant, k
H,
should have been higher for the simulation, which would have led to
more biogas and probably lower methane content in the gas. Another important parameter is
the ultimate degradability, which could have been higher than assumed. The uncertainty of the
characterization is higher for this simulation than for the household wastes since it is not
based on direct measurements. It would have been interesting to compare the results from the
characterization with measured data to evaluate the credibility of the method. On the other
hand, the default model parameters are probably more suitable for this validation than for the
household wastes given that they are calibrated for a similar substrate.

7.6. Full scale digester at Kppala WWTP (Liding)
Validation of the model to a full scale digester was made with data kindly provided by
Kemira Recycling Competence Center (Helsingborg) to evaluate the practical applicability of
the model. The time-scale is different from the pilot and lab scale experiments, and thus the
initial values are of less interest. The reactors were modeled as CSTRs and the stiffness of the
model was reduced compared to the previous simulations of pulse-fed reactors. For the model
validation, data for R100 and R200 from January to June 2006, and from October 2006 to
December 2007 was used. For each week, data of mean values for flows and biogas
production were available, as well as measurements of pH, VFA and VS/TS. The reactors
were modeled with weekly variations of inflow parameters, as well as of flows and reactor
volumes. Typical values for sludge variables can be found in Table 9. The mean
concentrations for each week from the simulation of R100 were used as input values to R200.
This method does not fully reflect the dynamics of R100, and the mass balance is not fully
closed but the accuracy of the simulation is still considered to be good enough considering the
quality of input data. A simulation with new input values at each time step would be very time
consuming, and it was seen in the simulations for R100 that the changes in concentration for
each week appeared to be linear.

Table 9 Typical values for sludge variables at Kppala WWTP
Variables Unit Primary
sludge
Digested
primary sludge
WAS Digested
primary sludge
and WAS
TS % 4.5-6.5 2-2.5 4-6 2.5-3
VS % of TS 80-85 65-70 65-70 2.5-3
pH - 7.1 - 7.2
VFA mg/l - 50-80 - 80-120
Alkalinity mg CaCO
3
/l - 3000 - 3500-4500
7.6.1 Feed composition
Since no measurements of the protein, fat and carbohydrate content in the sludge was made,
typical values were therefore sought after in the literature. A well defined characterization for
a primary sludge from the Netherlands which was published by Miron et al. (2000) was
chosen for input values to the model. The fractions expressed as percentages of VS (Table 10)
gave all inflow characteristics when converted into g COD/m
3
(with the Buswell formula,
equation 2.2) or g N/m
3
(Table 11). Alkalinity and pH were set to 5 mol/m
3
and 7
respectively. The measured degradability in R100 was 60-70 % with a fairly high HRT, hence
the degradable fraction in the inflow was assumed to be 70 %.
29

No data on the properties of the WAS was available from Kppala, thus a characterization
from the literature (Chen, et al., 2007) was used for the substrate for R200. This
characterization was obtained from a waste water treatment plant in Shanghai with a sludge
age of 7 days (Table 10) and was converted to g COD/m
3
as for the primary sludge (Table
11). The ammonium concentration in the inflow was assumed to be the same as for the
Sjlunda sludge, and the VFA concentration were assumed to be low (100 g/m
3
) with a
distribution of 2/3 acetate and 1/3 propionate on a COD basis. The alkalinity was set to 5
mol/m
3
, which is the value proposed in Siegrist et al. (2002), and the pH of the sludge was
found in Chen et al. (2007) (Table 11).

Table 10 Characterization of primary sludge and WAS
Characterization parameter Primary sludge WAS Unit
VS (mean value)
COD to VS ratio
Protein
Amino acids
Sugar
Dissolved carbohydrates
Fat
LCFA
4.7
1.64
18
1.7
45
0.65
21
13
3.5
1.24
83
0
15
0
1.3
0
%
g COD/g VS
% of COD
% of COD
% of COD
% of COD
% of COD
% of COD
Table 11 Input parameters for primary sludge and WAS

Characterization parameter Notation R100 R200 Unit
Mean total degradable organic matter S
aa
+ S
su
+ S
fa
+ Xs
52 000 21 000 g COD/m
3

Mean degradable particulate organic matter Xs
45 000 70 g COD/m
3

S

aa
0.22 0.83 -
Sugar fraction of X
S

su
0.54 0.15 -
Fat fraction of X
S

fa

0.25 0.013 -
S

in

0 0 -
Mean ammonium S
NH4

410 45 g N/m
3

VFA S
hac
+S
hpro
5000 100 g COD/m
3

HCO
3

pH
S
HCO3

-log(S
H
)
10
7
5
6.8
mol/m
3

7.6.3. Hydrolysis constant
A hydrolysis constant of 0.4 d
-1
for primary sludge was found in a review by Vavilin et al.
(2008). A lower value of 0.15 d
-1
was assumed for the WAS. Only primary sludge is
hydrolyzed in R100, and in R200 it was assumed that only the WAS was hydrolyzed.
Owing to the long time scale for the simulation, the initial values for R100 were of little
importance for this simulation, and were therefore set to steady state values to avoid stiffness
in the simulation. The initial values for R100 were also used for R200.
7.6.5. Results
The biogas production was integrated for each time step and divided with the time difference
to enable a comparison with the measured data. The result from the simulation of the biogas
production in R100 shows that the correlation is fair, except for some outliers (Figure 13).
The variations in the modeled gas production are due to changes in TS and flow rate, and it
seems that these variations can explain the variations in measured values to some extent. The
residual analysis of the gas production in R100 in Figure 13 shows that the model
30

underestimates the gas production in the beginning of the time series. It is also shown in the
normal probability plot that the residuals seem to be normal distributed with a mean slightly
lower than 0.

Figure 13 Simulated and measured biogas production for R100 and R200 (left) with residuals and normal probability
plot for R100 (right)
The simulation results of R200 depend on the effluent from R100 and the quality of the
characterization of the WAS. Despite the underestimation of biogas production in R100, the
biogas production in R200 correlates well during the last part of the time series (Figure 13)
while it is underestimated during the first part. As for the biogas production in R100, the
trends and variations seem to be reflected in the modeled results for R200. The simulated
VFA are underestimated in the simulations, but are in the same order of magnitude as the
measured values (Figure 14). It has been observed previously in validations of ADM1 that the
VFA correlate worse than the other parameters (Tartakovsky et al. (2008), Jeong et al.
(2005)). The simulated alkalinity, on the other hand, correlates well with the measured values
(Figure 14). The pH is stable around 7.1 for both the simulation and the measured values,
while the ammonium is variable and higher in R200 (Figure 15).

Figure 14 Simulated and measured VFA (left) and HCO
3
(right) for R100 and R200
0 100 200 300 400 500
0
0.5
1
1.5
2
x 10
4
Time [days]
W
e
t

b
i
o
g
a
s

p
r
o
d
u
c
t
i
o
n

[
m
3
/
d
a
y
]

Simulated, R100
Measur ed, R100
Simulated, R200
Measur ed, R200
0 200 400 600 800
-2
-1
0
1
x 10
4
Time [ days]
[
m
3
/
d
a
y
]
-10000 -5000 0 5000
0.10
0.25
0.50
0.75
Data
P
r
o
b
a
b
i
l
i
t
y
Nor mal Pr obability Plot
0 200 400 600
0
50
100
150
200
250
Time [days]
T
o
t
a
l

V
F
A
s

a
s

a
c
e
t
a
t
e

e
q
u
i
v
a
l
e
n
t
s
,
[
g
C
O
D
/
m
3
]

Measur ed, R100
Simulated, R100
Measur ed, R200
Simulated, R200
0 100 200 300 400 500 600 700
0
20
40
60
80
100
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
m
o
l
/
m
3
]

Simulated HCO3
Mesur ed HCO3
Simulated HCO3R200
Measur ed HCO3
31

Figure 15 Simulated and measured pH (left) and ammonium (right) for R100 and R200
7.6.6. Discussion
Although the characterization was obtained from the literature and not from direct
measurements, the simulations correlated fairly well with the experimental data. A major
problem with the validation is the lack of data for ammonium in the reactor, which means that
the real inhibitory effect of free ammonia is unknown. It is possible that the VFA
concentration in R200 was underestimated in the simulation due to the underestimation of
ammonia, which led to a higher rate of acetotrophic methanogenesis and thus a lower acetate
concentration.

The fluctuations in the biogas production could not be entirely explained by the model. This
was not expected, however, since the model did not include the daily and seasonal variations
in sludge composition. These variations are due to changes in the waste water treatment
processes and to variability in the inflow to the WWTP. Despite this lack of information, the
model could predict some of the observed trends. The result could hopefully be improved if a
better characterization and more data were available. The model parameters were calibrated
for a mixed sludge reactor, and the microbial population in the reactor may be different for the
reactors with digestion of household waste. Therefore, a recalibration of the model parameters
could possibly improve the model fit.

0 100 200 300 400 500 600 700
0
1
2
3
4
5
6
7
8
Time [days]
p
H

Simulated, R100
Measur ed, R100
Simulated, R200
Measur ed, R200
0 200 400 600 800
700
800
900
1000
1100
1200
1300
1400
1500
1600
Time [days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

Simulat ed NH4, R100
Simulat ed NH4 R200
32

33

8. Uncertainty analysis
In this section, uncertainty analysis (UA) of the model is conducted to enable evaluation of
the accuracy and credibility of the model output. The uncertainty of the model is evaluated
using Monte Carlo methods, with known or assumed distributions of the input parameters to
the model.
In section 7, measured values were used as inputs for the simulations. A good fit to the
measured output values were considered as a successful validation, and if the correlation was
poor the validation was considered to be a failure. In this section, the measurement errors will
be taken into consideration, to evaluate the quality of the model predictions. If the uncertainty
of the model output is high, then better analysis methods may be required to produce a
satisfying fit to data.
Another topic that will be addressed in this section is the benefit from extensive
characterization of household waste. The variabilities of 40 different household wastes were
used to describe a general characterization. The quality of model predictions with or without
measurements could then be compared, in order to evaluate the importance of a detailed
characterization.
8.1. Introduction to uncertainty analysis
A complex natural system like anaerobic digestion is impossible to perfectly predict, thus the
uncertainty is an important measure of the reliability of the prediction. The prediction
uncertainty will depend on the uncertainty of the model structure, parameters and
measurements and on the mathematical uncertainty as in Figure 16. The mathematical
uncertainty arises from the processes where the equations are solved numerically. Model
structure uncertainty can have different sources, e.g. excluded processes or a false
representation of the system. Uncertainty from model structure can change with time if the
excluded processes change with time. An example in the anaerobic digestion model is the
inhibition from H
2
S, which most likely varies with time but is excluded in the model (Kops,
et al., 1996). The model structure can be mathematically verified only if the parameters can be
identified (see section 6). The model structure is assumed to be valid for this case, even if
system identification is not likely to be successful. Indications of the lack of identifiable
parameters for the model were discussed in section 6.2.

Figure 16 Correlation between uncertainties (adapted from Kops et al, 1996)
Real
Prediction
Uncertainty
Model
Uncertainty
Parameter
Uncertainty
Mathematical
Uncertainty
Determined
Prediction
Uncertainty
Measurement
Uncertainty
34

The parameter and measurement uncertainties are in focus in this analysis. Uncertainty in
input parameters propagates through the model and affects the model output. The statistical
probability of parameters and measurements can often be described with probability density
functions (PDEs), with e.g. normal, uniform, or triangular distribution. To be able to
determine the uncertainty in the model prediction, the PDEs are used as input in the Monte
Carlo analysis.
8.1.1 Monte Carlo analysis
If the PDEs of the parameters are known, it is possible to generate random numbers from
these and use as input to the model. The procedure where a set of generated random numbers
are used as input to a model is called Monte Carlo (MC) analysis. The distribution of the
output from a MC analysis can be used to quantify the uncertainty of the prediction. This
method has many benefits and is widely used; one of the most important advantages is that it
can be applied for non-linear models. The method enables analysis of the model for a span of
input parameters, i.e. global sensitivity analysis, which will be discussed in section 9.
The sampling of input values is commonly made from the entire spectrum of the PDEs
independently for each input variable, called simple random sampling, or brute force
sampling. If an infinite number of samples are generated, the resulting set of outputs will give
the PDE of the model output. In reality, however, a number of runs are chosen that is large
enough to ensure that the result is not changed notably by increasing the number of runs. This
sampling method is commonly used, since it is relatively easy to conduct and to understand.
One drawback with the method is that a large number of simulations often are required, which
can be very time consuming.
The ordinary MC method does not include variation of the parameters with time, which is a
disadvantage (Kops, et al., 1996). In wastewater treatment plants, input parameters are known
to vary with time, for example flow rates and concentrations of phosphorous and nitrogen. An
alternative to MC simulation is therefore produce stochastic MC input variables at each time
step, a method called MC with stochastic parameters. Another method is the stochastic
differential equations method (SDE) where a noise term is added to the ODE system which
ads the uncertainty for each time step (Kops, et al., 1996). These methods are rarely used, they
have the drawback of being slower than MC and they require further assumptions regarding
the quantity of the noise.
8.1.2. Latin Hypercube Sampling (LHS)
To reduce the number of simulations needed in the MC analysis, Latin hypercube sampling
(LHS) can be applied. This method divides each parameter PDE into equiprobable intervals,
and combinations of input values are chosen from these intervals so that as much as possible
of the parameter space is covered (Figure 17). An underlying assumption in this method is
that all parameters are independent and could be paired as such, but in real life it is fairly
common with correlated parameters. If the parameters are correlated the LHS sampling can
produce unlikely input combinations.
35

0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
ProbabilityX
2
P
r
o
b
a
b
i
l
i
t
y
X
1
0
0.2
0.4
0.6
0.8
1
-2 -1 0 1 2
X
2
C
u
m
u
l
a
t
i
v
e

P
r
o
b
a
b
i
l
i
t
y
0
0.2
0.4
0.6
0.8
1
-2 -1 0 1 2
X
1
C
u
m
u
l
a
t
i
v
e

P
r
o
b
a
b
i
li
t
y

Figure 17 Visualization of the LHS sampling method (Minasny, 2004)
When generating the MC samples, it is important to verify that the number of realizations, N,
is large enough not to affect to result. If the total mean or variance of the output is changed
when increasing N, it is not sufficiently large. To evaluate the benefits of the LHS sampling
methods, a comparison between the simple random method and LHS sampling was conducted
(Figure 18 and Figure 19). The LHS method produced results converged faster to the final
value, and fluctuated less. This method was hence chosen for the analyses in this work. The
choice of N is a trade-off between desired accuracy of the result and the time available for the
simulations. N was chosen to be 5000 for all simulations, a value large enough to produce
sufficiently consistent results but still low enough to keep the required simulation time within
hours and not days.

Figure 18 Mean (left) and variance (right) for steady state gas production as a function of N for a continuous process
when using simple random MC sampling
0 2000 4000 6000 8000 10000
0.0494
0.0495
0.0496
0.0497
0.0498
0.0499
0.05
Number of r uns (N)
M
e
a
n

g
a
s
p
r
o
d
u
c
t
i
o
n
0 2000 4000 6000 8000 10000
2
2.2
2.4
2.6
2.8
3
x 10
-3
N
S
t
a
n
d
a
r
d

d
e
v
i
a
t
i
o
n
36

Figure 19 Mean (left) and variance (right) for steady state gas production as a function of N for a continuous process
when using LHS sampling
8.2 Measurement uncertainty of the household waste simulations
8.2.1. Methods for the uncertainty analysis
The uncertainties of the simulations of household waste experiments from section 7 were
analyzed with MC method. The uncertainty from the model structure and parameters is hard
to quantify as stated in 6.1., and was omitted in this work. Focus of the analysis was instead
on the impact of the measurement quality on the model predictions. The standard deviations
of the measurements for the characterization of the waste were found in the literature, and are
presented in Table 12. The measurement errors were assumed to be normally distributed.
Values for some parameters were not found, either because they were measured with
unknown instruments, as in the case of pH and pressure, or because they were impossible to
measure, as in the case of HCO
3
. For these cases, high values for uncertainty were assumed.
Table 12 Standard deviations for measurements of input parameters
Input parameter Aalborg
household waste
Vstra Hamnen
household waste
Unit (%)
TS
Soluble COD
Protein
Fat
Starch
Sugar
Fibers
pH
NH
4
-N
Acetate
Propionate
Temperature
Pressure
Feed flow rate
5.0
23
17
18
18
7
11
7.4
0.32
48.7
34.3
53.3
1.02
1.33
5.0
23
N
17
12
7
4
14
7.2
0.027
48.7
N
34.3
N

35.4
1.03
1.33
%
%
% of TS
% of TS
% of TS
% of TS
% of TS

% of TS
g COD/m
3
g COD/m
3
C
bar
l/d
2.5
S

20
U
0.97
J
3.5
J

3.4
J

18
J

4.7
J

2
U
3
L

10
U

10
U

2
U

2
U

2
U

N
No available data, the same values as for Aalborg household waste were used in the simulations,
U
Estimated value,
S
(SIS, 1981),
J
(la Cour Jansen, et al., 2004),
L
(Dr Lange)

The uncertainty of the measurements for the target output parameters gas production, VFA,
alkalinity and ammonium are presented in Table 13. The gas production was the key
parameter of interest and the measurements were considered to be very reliable (a relative
standard deviation of 2 % was estimated for the measurements). Measurement variability for
0 2000 4000 6000 8000 10000
0.0494
0.0495
0.0496
0.0497
0.0498
0.0499
0.05
Number of r uns (N)
M
e
a
n

g
a
s
p
r
o
d
u
c
t
i
o
n
0 2000 4000 6000 8000 10000
2
2.2
2.4
2.6
2.8
3
x 10
-3
N
S
t
a
n
d
a
r
d

d
e
v
i
a
t
i
o
n
37

VFA was unknown, and the method for determination was considered less unreliable.
Uncertainties for the ammonium and alkalinity measurements were found in laboratory
documentation.

Table 13 Standard deviations for measurements of output parameters
Output parameter (%)
Gas production
VFA
NH
4
-N
HCO
3

2
U

10
U
3
L
3.5
S2
U
Estimated value,
L
(Dr Lange),
S2
(SIS, 1994)

The experiments were modeled as pulse-fed reactors in section 7, but the stiffness of these
simulations led to long simulation times. MC simulations of these models would be very time
consuming for a sufficient number of realizations. Instead the reactors are approximated to
CSTRs with flows and volumes as in the steady state phase (see section 7.1.1.). One drawback
for this method is that the results from the uncertainty analysis cannot be fully compared with
the experimental data. The advantage is, apart from the short simulation time, that the results
will be more useful for real processes which are fed with shorter intervals. To ensure that the
samples are well distributed, the LHS sampling technique was used in Matlab (Minasny,
2004). With normally distributed samples, it was possible to get negative values for the
concentrations; in that case they were set to 0.
8.2.2. Results of the measurement uncertainty analysis
The resulting distributions of the model outputs from the MC analysis on the Aalborg
household waste simulation are presented as cumulative distribution functions (CDFs) in
Figure 20. The median for the simulated output is found on the x axis for F(x)=0.5, and the
95% confidence interval between F(x)=0.025 and F(x)=0.975. Although the inputs to the
model are normally distributed, it can be seen that the output are not, since the CFDs are not
fully symmetrical. Especially for the alkalinity, where it can be seen that the higher output
values are more spread out than the lower values (F(x)=0.5 is not in the middle of the curve).
The biogas production, however seem to be normally distributed.

38

Figure 20 CDF of the model outputs for the Aalborg household waste simulation, gas production (upper left), VFA
(upper right), NH
4
-N (lower left) and HCO
3
(lower right)
Intervals of the modeled outputs from the Aalborg household waste simulation with a
confidence level of 95% are presented in Figure 21. These are results from Monte Carlo
simulations, and are compared to data with measurement errors also presented with
confidence intervals. The confidence intervals for the gas production and alkalinity are the
broadest, which means that the measurement errors are most influential on these parameters.
When the confidence intervals overlap, the validation error can be explained by measurement
error. If not, the explanation for the lack of fit is to be found elsewhere since it is highly
unlikely that measurement error is the cause. Most of the data points for the gas production
are within the interval, except for the first time period. The alkalinity has a broad interval and
all data points overlap the simulated confidence intervals. For the VFA and the ammonium;
none of the measurements are included in the interval.

0.035 0.04 0.045 0.05
0
0.2
0.4
0.6
0.8
1
x
F
(
x
)
Empir ical CDF
100 150 200 250
0
0.2
0.4
0.6
0.8
1
x
F
(
x
)
Empir ical CDF
900 950 1000
0
0.2
0.4
0.6
0.8
1
x
F
(
x
)
Empir ical CDF
50 60 70 80 90 100
0
0.2
0.4
0.6
0.8
1
x
F
(
x
)
Empir ical CDF
39

Figure 21 Plots of the confidence intervals of the model outputs ammonium and alkalinity with respect to
measurement errors for the Aalborg household waste simulation
The same method for uncertainty analysis was applied on the Vstra Hamnen household
waste simulation from section 7.4., and the results from the simulations are presented in
Figure 22. There are many measurements for the gas production which are far from the
confidence interval, while the VFA measurements are all within the interval. The lack of fit
for ammonium and alkalinity cannot be explained by measurement error.

40 45 50 55 60 65 70
0
0.01
0.02
0.03
0.04
0.05
0.06
Time [ days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

Biogas pr oduct ion modelled
95 % conf idence int er val
Biogas pr oduct ion measur ed
40 45 50 55 60 65 70
0
200
400
600
800
1000
1200
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

C
O
D
/
m
3
]

VFAs modelled
VFAs measur ed
40 45 50 55 60 65 70
0
500
1000
1500
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

NH4 modelled
NH4 measur ed
40 45 50 55 60 65 70
0
20
40
60
80
100
Time [ days]
C
o
n
c
e
n
t
r
a
i
o
n

[
m
o
l
/
m
3
]

HCO3 modelled
HCO3 measur ed
30 35 40 45 50
0
0.01
0.02
0.03
0.04
0.05
0.06
Time [ days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

30 35 40 45 50
0
200
400
600
800
1000
1200
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

C
O
D
/
m
3
]

VFAs modelled
VFAs measur ed
40

Figure 22 Plots of the confidence intervals of the model outputs with respect to measurement errors for the Vstra
Hamnen household waste simulation
8.2.3. Discussion
A considerable measurement uncertainty is naturally not desirable since it means that the
model cannot predict the parameters better than the confidence interval, even if the model
parameters were improved by calibration. The only way to reduce measurement uncertainly
is, of course, to improve the analytical method for determination of these inputs. If the
parameters that influence the uncertainty the most are determined, the measurements that need
improvement can be found. Even so, there is also the uncertainty of sampling, which is harder
to quantify but further broadens the confidence interval of the result. It can seem unpromising
that so many measurements are outside of the confidence interval for the household waste
simulations, but it is in fact promising in one way; it allows the correlation to be improved by
calibration of the model parameters. Sensitivity analysis and model calibration is discussed in
section 9 and 10.
8.3. Variability of feed composition
8.3.1 Method
To enable comparison of the prediction accuracy when using the characterization from
measurements and when using a general composition of household waste, the variation of
data from 40 samples was analyzed. The general characterization was calculated from the
literature and based on household waste collected in Danish cities (
Table 14). This comparison is interesting since measurements of proteins, fibers, fat and other
constituents in the substrate are costly and thus seldom carried out. If the uncertainty of the
model results mainly depends on accurate measurement of these parameters, it means that
these measurements are needed for a satisfactory simulation. The 40 measurements were
regarded as normally distributed, which was a well-grounded assumption e.g. for the fat
content (compare with Figure 23). The standard deviations of the parameters can be found in
Table 14. It is worth noting that the fractions of the different constituents are assumed not to
be correlated, although it is likely that they are correlated in reality. The total amount of the
constituents makes up the degradable part of TS.

30 35 40 45 50
0
500
1000
1500
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

NH4 modelled
NH4 measur ed
30 35 40 45 50
0
50
100
150
200
Time [ days]
C
o
n
c
e
n
t
r
a
i
o
n

[
m
o
l
/
m
3
]

HCO3 modelled
HCO3 measur ed
41

Table 14 Means and standard deviations for household waste
Characterization
parameter
General
household
waste values
Unit (%
of TS)
Rel std (%)
TS
Protein
Sugar
Starch
Fat
Fibers
Soluble COD
26.4
A
14.9
6
M
15
M
13.8
18.5
13.9
%
% of TS
% of TS
% of TS
% of TS
% of TS
%
6.5
1.37
3
5
1.65
3.69
8.3
24.6
B

9.2
H

50
H

33.3
H

12.0
H

20.0
H

59.7
B

A
Before dilution, uncertainty is equal to measurement uncertainty after dilution,
M
Mean values from (Hansen, et
al., 2007) and from assuming that 95 % of measurements are within the range of 2,
B
(Hjlund Christensen, et
al., 2003),
H
(Hansen, et al., 2007)

Figure 23 Normal probability plot and histogram for fat fractions in data from (Hansen, et al., 2007)
8.3.2. Results and discussion
The uncertainties of the model predictions when using the general characterization instead of
measurements were determined for the Aalborg and Vstra Hamnen household waste
simulations. This analysis was conducted simply by changing the means and distributions of
the characterizations from 8.2. to the values in
Table 14. A comparison between the resulting variances and the variances due to
measurement errors is presented in Table 15. This table shows that the variance increases
almost 10-fold when it comes to gas production and ammonium. The predictions of VFA and
HCO
3
are not as influenced of the quality of the input data.

Table 15 Variance of model predictions from analysis with varying characterization parameters
Parameter Unit Aalborg
household waste
with measured
characterization
Aalborg
household waste
with general
characterization
Vstra
Hamnen
household waste
with measured
characterization
Vstra
Hamnen
household waste
with general
characterization
Gas production
VFA
NH
4

HCO
3

m
3
/d
gCOD/m
3

gN/m
3

mol/m
3

4.1x10
-6
422
428
52.9
2.3x10
-5
506
5.8x10
3
82
1.2x10
-6

22.4
493
59.5
1.5x10
-5

13.3
6.1x10
3

92.7

11 12 13 14 15 16 17
0.01
0.02
0.05
0.10
0.25
0.50
0.75
0.90
0.95
0.98
0.99
P
r
o
b
a
b
i
l
i
t
y
10 12 14 16 18
0
5
10
42

The confidence intervals of the model outputs presented in Figure 24 and Figure 25, show that
the confidence interval for biogas production is significantly broader when the variability of
the characterization parameters are used instead of the measurement uncertainty. This implies
that the measurements of the substrate are important to reduce the uncertainty of the results
for household waste. One can hence expect that the difference in gas production from
different household wastes can be significant if the fractions of the feed are non-correlated.
Figure 24 and Figure 25 also show that the levels of ammonium and alkalinity in the reactor
are affected by the content in the waste, while the VFA seem to be largely unaffected in the
same range. Measurements of the specific household waste would therefore not be expected
to improve the correlation of the simulation for VFA.
The confidence intervals for the general composition also reveal that there are values that
cannot be explained by varying the waste composition. This implies that the model
parameterization or structure needs to be revised. To be able to calibrate the model, the
influence of individual parameter on outputs needs to be investigated. This is the subject of
section 9 and 10.

Figure 24 Aalborg household waste simulation with general waste composition
40 45 50 55 60 65 70
0
0.01
0.02
0.03
0.04
0.05
Time [ days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

40 45 50 55 60 65 70
0
200
400
600
800
1000
1200
Time [days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

C
O
D
/
m
3
]

VFAs modelled
95 % conf idence inter val
VFAs measur ed
40 45 50 55 60 65 70
0
500
1000
1500
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

NH4 modelled
NH4 measur ed
40 45 50 55 60 65 70
0
20
40
60
80
100
Time [ days]
C
o
n
c
e
n
t
r
a
i
o
n

[
m
o
l
/
m
3
]

HCO3 modelled
HCO3 measur ed
43

Figure 25 Vstra Hamnen household waste simulation with general waste composition

30 35 40 45 50
0
0.01
0.02
0.03
0.04
0.05
0.06
Time [ days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

95 % confidence int er val
30 35 40 45 50
0
200
400
600
800
1000
1200
Time [days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

C
O
D
/
m
3
]

VFAs modelled
VFAs measur ed
30 35 40 45 50
0
500
1000
1500
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

NH4 modelled
NH4 measur ed
30 35 40 45 50
0
50
100
150
200
Time [ days]
C
o
n
c
e
n
t
r
a
i
o
n

[
m
o
l
/
m
3
]

HCO3 modelled
HCO3 measur ed
44

45

9. Sensitivity analysis
In the uncertainty analysis from the previous section the overall uncertainty of the model
output due to uncertainties in input was studied. In this section the sources of the variations in
output is investigated to find the input parameters which the model is sensitive to. Sensitivity
analysis (SA) can be conducted to find the input parameters that are most important to
measure to minimize the prediction uncertainty. SA can also be used to test the model
stability and to find the model parameters that are most suitable for model calibration. If a
parameter has a negligible impact on the output, then the efforts to determine that parameter
might be reconsidered. For some applications, like control, it is desirable to simplify the
model as much as possible, and if the result is insensitive to a parameter, it can be excluded
from the model. The parameter for which the output is most sensitive to is probably the best
control signal. Another obvious purpose of SA is to improve the understanding of the system.
Large ODE systems are not easily understood, and the SA can help to quantify the
contribution of different parts of the system to the output result and give hints on what parts of
the process that needs special consideration. A specific goal of the SA in this case was to find
key parameters that could be used to calibrate the pilot scale simulations from section 7.
First, a general presentation of different methods for sensitivity analysis is made, in section
9.1. and 9.2., then previous SA on anaerobic digestion models from the literature are
presented in 9.3. The sensitivity analysis on the pilot scale simulations from section 7 are
found in section 10.

9.1. Local SA methods
Local methods evaluate linear perturbations for the output for a specific set of parameters.
Thus, the result applies for this set of parameters, but may not be applicable for other values.
For linear models without correlation between parameters, these methods are useful.
9.1.1. Sensitivity functions
A common SA method is to study the partial derivative of an output variable with respect to a
certain input parameter. This derivative is called the sensitivity function , and can be
interpreted as the linear change in output due to the change in the input parameter. The
sensitivity function can be computed continuously during the simulation time, to study the
dynamics of the sensitivity of the output to the parameter. The absolute change in an input
variable x per unit change in the parameter is presented in (eq. 9.1) (Jeppsson, 1996).

o =
x
(9.1)

The definition in equation 9.1 gives a result that strongly depends on the units of the
parameter, and often the relative change in parameter is more interesting to compute (eq. 9.2).
The derivative of the variable x with respect to the parameter is here multiplied with the
parameter value. Here the sensitivity function depends on the parameter on a relative basis,
and sensitivity functions for different parameters can easily be compared.

o =
x
(9.2)

When analyzing the sensitivity functions for different parameters dependencies between
parameters can be detected. If parameters have similar but inverse sensitivity functions they
can be non-identifiable, i.e. a change in one parameter can be compensated by a change in
another. The best identifiability is achieved when sensitivity functions have different behavior
46

and when they differ significantly from zero. As sensitivity functions give a linear sensitivity
analysis they are valid only locally for the specific parameter values that were used.

9.2. Global methods
Models of complex systems such as the anaerobic digestion process are usually nonlinear.
Using local SA methods to analyze these models can be questionable, since the results only
are valid for the set of parameters that were analyzed. For other parameter values, the
sensitivity analysis may not be accurate. For nonlinear models, global methods are preferred,
that evaluate the sensitivity for a broader spectrum of input parameters. This analysis will not
be as straightforward as the linear methods, though. Here three examples of useful methods
for nonlinear SA are presented.
9.2.1. Scatter plots
A natural first step when analyzing a model is to study plots of outputs vs. input for a
qualitative analysis of the model behavior. Dependencies and nonlinearities can be detected,
and the understanding of the system can be improved. A MC generated sample of inputs and
the corresponding outputs can be used for this analysis. Figure 26 shows an example of how
scatter plots can be used for SA.

Figure 26 Examples of scatter plots for the Aalborg household waste experiment; gas production vs. TS (left) and gas
production vs. k
H
(right) from MC analysis on characterization and model parameters respectively
The qualitative sensitivity analysis from these plots indicates that the TS content in the feed
has an impact on the biogas flow from the reactor, and a nonlinear correlation between k
H
and
gas production was detected.
9.2.2. Standardized regression coefficients
Another common SA method is to use the output from a set of MC sampled parameters for
linear regression with respect to the different parameters. The standardized regression
coefficients are a measure of the sensitivity of the output to the specific parameters. The
model coefficient of determination, R
2
, can be used to determine the precision of the model.
For R
2
values close to 1, the regression model is valid and the regression coefficients are a
good measure of the sensitivity. If R
2
is low, on the other hand, it indicates a nonlinear model,
with a poor correlation to the linear regression. In these cases, a nonlinear approach to SA is
0.044 0.046 0.048 0.05 0.052 0.054 0.056
0
0.01
0.02
0.03
0.04
0.05
0.06
TS [ %]
G
a
s

p
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
]
0.1 0.15 0.2 0.25 0.3
0
0.01
0.02
0.03
0.04
0.05
0.06
k
H
[d
-1
]
G
a
s

p
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
]
47

preferred. This analysis hence can quantify the degree of nonlinearity, and if the linear
behavior is dominant, give a measure of the degree of sensitivity.

9.2.3. Variance-based methods
Scatter plots represent a qualitative method, and is a very useful for revealing nonlinearities
and finding correlations between inputs and outputs. To be able to quantify the sensitivity of
input parameters, and to compare sensitivities for different parameters, more sophisticated
methods are required. The method that was chosen for the main part of the SA in this project
is based on the correlation between the variance of input parameters and the variance of
outputs for steady state simulations. The variance in the parameters will lead to a variance in
output, and an important parameter will have a bigger impact on the output variance than a
parameter that the model is less sensitive to.
If, for example, the inputs are generated with MC methods from PDEs as in section 8.2., the
outputs will be determined with a total variance. This is the same variance that was used for
determining the uncertainties of the predictions. In the case of sensitivity analysis, the source
of the variance is of interest. The variance that each parameter adds to the total output
variance is the measure of sensitivity that is sought. For a linear model, without correlation
effects, the total uncertainty could be decomposed into fractions of variances assigned to each
parameter. The contributions of variance for one parameter at a time could simply be
calculated, and the total output variance would be the sum of these. For a nonlinear model, the
total variance cannot be decomposed in this simple manner due to correlation effects. Instead,
the difference between the total output variance and the output variance when one parameter
was held constant was chosen as an estimator of the correlation between variance in input and
output. This value also represents the possible reduction in uncertainty that would be gained if
a parameter would be determined without uncertainty.
9.3. Sensitivity analysis of anaerobic digestion models in the literature
Linear sensitivity analysis is the most commonly used method for analyzing the parameter
sensitivity in anaerobic digestion models, in fact no examples of SA based on MC simulations
were found in the literature. Tartakovsky et al. (2008) developed a distributed version of
ADM1 for simulation of an UASB reactor and performed a sensitivity analysis on the model
parameters. They used sensitivity functions to find the most sensitive parameters, which were
used to calibrate the model. The sensitivity analysis showed that specific uptake rates and half
saturation constants for acetate and propionate and butyrate/valerate had the strongest impact
on reactor COD concentration, acetate and propionate, while k
L
a affected the methane
percentage. The measurements had high variability, but model correlation with reasonable
accuracy was achieved. COD predictions were better than VFA predictions; the authors
suggested that the model underestimated inhibition of fermentation from free acid, thus
overestimating the VFA.
Another method for SA was used by Jeong et al (2005) as they analyzed the sensitivity of
ADM1 using bottle tests with glucose and acetate as substrate. ADM1 was simplified as the
decay processes for microorganisms, inhibition from pH and free ammonia and the gas-liquid
transfer for methane were ignored. The sensitivity analysis was carried out by varying one
parameter at a time over a chosen interval, and then calculating the mean of the absolute
differences between the standard parameter setting result and the model result with altered
parameter value for each time step. The benefit from this method compared to sensitivity
functions was that different intervals of variance could be chosen for the parameters, and that
the non-linearity of the output could be revealed in that range. Yield of product on substrate
and Monod maximum specific uptake rates proved to be significant parameters, while yield of
48

biomass on substrate were less important. Correlation with the model for methane
concentration was good after calibration, but worse for acetate which the authors explain with
adsorption and storage by the microorganisms. The question which arose from the results of
this SA was whether the chosen intervals of variance are reasonable, for example the yield of
product on substrate varied with 30 %, a figure which was not supported in the article.
An example of SA for the purpose of parameter reduction can be found in Noykova (2000)
where an SA is made on an older and simpler model than ADM1. The only output in the SA
was the biogas flow, and logarithmic or relative sensitivity functions were used (eq. 9.3)

I
]
=
InX
]
Inp
(9.3)

The analytical equation to calculate T
ij
could not be adapted for non-linear systems, therefore
simulation results were required.
max
for methanogens proved to be the most significant
parameter, as were the yield coefficient for methane and inhibition coefficient for
methanogens. Decay rates, on the other hand, had little impact on the results and could be
neglected. The simplified model with four optimized parameters had a reasonable fit. When
reducing a model, it is important to be aware that the conditions under which the model could
be used become narrower. The process needed to be very stable for the reduced model to be
useful.
9.4. Discussion
If a model prediction is sensitive to a specific parameter, this parameter will also contribute
more to the uncertainty of the prediction. Uncertainty and sensitivity analysis therefore go
hand in hand.
All sensitivity analyses on anaerobic digestion found in the literature were based on local
methods. These analyses were easy to perform and to grasp. Sensitivity functions, for
example, give a general sensitivity analysis in the sense that the parameters are always
changed 100 %. The result should however be treated with caution. Some parameters are not
at all likely to vary as much as 100 %, while others could vary more than that. The non-linear
behavior and correlation effects of the model are also excluded using this method. It is known
that both ADM1 and the Siegrist model are non-linear and it is doubtable whether sensitivity
functions are a suitable method.
All sensitivity analyses found in the literature only included model parameters and not inflow
parameters, although they are interesting as well. It is for example not useful to put much
effort info calibrating the model parameters if the level of detail is not supported by the data.
Furthermore, it could be interesting to compare the relative importance of input parameters
for various operational settings. Increase knowledge of the model sensitivity for all
parameters would enhance model application and ensure that measurements are made for the
most influential input parameters.
Global MC methods are preferred over local methods when it comes to analyzing nonlinear
models. It is however important to stress that sensitivity or uncertainty analysis using MC
methods are performed for a certain case, with chosen ranges for the parameters, and that the
results only are valid for that particular parameter space. The analysis will not be reliable if
the probability density functions of the parameters are chosen poorly. One of the drawbacks
of MC-based methods is of course the extensive computational force that is required.
49

10. Sensitivity analysis of the Siegrist model
As discussed in chapter 9, a global and variance-based method was considered most suitable
for SA of the validation simulations of section 7. The chosen method quantified the reduction
of variance in output when one parameter was kept constant compared to when all parameters
were varying. The background to the sampling method and distributions for characterization
parameters can be found in chapter 8. The sensitivity analysis was conducted by first
rendering a LHS sample of 5000 parameter sets. Steady state simulations were run for the
LHS sample, giving the total variance, and also for the same parameter sets but where one
parameter at a time was kept at a constant value (at the mean value). The sensitivity measure
for each parameter was expressed as the relation between the variance when the parameter
was held constant to the total variance. A low value for this relative sensitivity means that the
output is very sensitive to the parameter.
SA was conducted for the validation simulations of the pilot scale experiments presented in
section 7. Each of these experiments were now simulated as steady state simulations, and
analyzed for three different categories of parameters. One of the purposes of the SA was to
find the parameters that were important to measure in order to reduce the output uncertainty.
This was achieved by calculating the sensitivity to the measured parameters, using known and
estimated measurement errors as parameter distributions. Another aim for the SA was to find
the parameters that would be most suitable for calibration. In this case, the sensitivity to the
model parameters was studied, using rectangular distributions for these parameters with a
range of 50 % to 150 % of the default values. This means that all values in this interval were
assumed to be equally probable. This use of general intervals was applied due to the lack of
information, and does not always reveal what parameters that was best suited for calibration.
The reason for this was partly that some parameters are more variable by nature than others,
and may be changed to a higher extent. The sensitivity measure for some parameters could
also be explained with a high rate of unstable simulations, due to e.g. washout. These
parameters will not always be good for calibration purposes. The SA therefore needed to be
supplemented with scatter plots to study relationships between parameter value and output
more closely.
The sensitivity of the measured parameters and model parameters were also studied when all
input and model parameters were allowed to vary simultaneously. This could display the
relative importance of the measured parameters compared with the model parameter.
The values used in the model were given for 35C, and recalculated for the actual reactor
temperature with the equations for temperature dependency (see equation 4.16).
10.1. Aalborg household waste experiment
For this first example, all results are presented in the text, but for the other cases only the most
interesting results are shown, and the rest are found in Appendix III.
10.1.1. Sensitivity of input parameters
The first step of the sensitivity analysis for this thermophilic digester was to find
measurement uncertainties for the input parameters, described as standard deviations. The
standard deviations and rank of the uncertainties were presented in Table 16. The resulting
relative sensitivity for each parameter with respect to different output parameters is also
presented here. Rank 1 is given to the parameter that lowers output variance the most when
kept at a constant value.
50

Table 16 Results of SA, UA and variability of measured parameters for the Aalborg household waste simulation
Parameter Mean Unit Uncertainty

rank (%)
Sensitivity
(Gas)

rank relative
Sensitivity
(VFA)

rank relative
Sensitivity
(NH
4
-N)

rank relative
Sensitivity
(HCO
3
)

rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
0.05
0.23
0.17
0.18
0.07
0.18
0.11
7.4
10
160
48.7
34.3
55.3
1.33
1.02
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
12
2
14
9
3
10
8
13
1
11
4
4
13
6
6
2.5
S
20
U
0.97
J
3.5
J
18
J
3.4
J
4.7
J
2
U

100
U

3
L
10
U

10
U

2
U

2
U

2
U
1
8
10
6
4
7
9
14
11
13
15
12
5
3
2
0.74
0.98
1.00
0.95
0.88
0.97
0.99
1.00
1.00
1.00
1.00
1.00
0.90
0.87
0.78
4
3
7
11
5
8
10
14
2
9
15
13
1
12
6
0.95
0.95
0.99
1.00
0.98
1.00
1.00
1.00
0.59
1.00
1.00
1.00
0.49
1.00
0.98
1
3
2
6
7
9
11
14
4
5
13
15
12
8
10
0.42
0.93
0.84
0.97
0.97
0.99
1.00
1.00
0.93
0.95
1.00
1.00
1.00
0.98
1.00
3
2
4
8
5
13
7
11
1
12
9
10
15
6
14
0.97
0.95
0.99
1.00
1.00
1.00
1.00
1.00
0.080
1.00
1.00
1.00
1.00
1.00
1.00
S
(SIS, 1981),
U
(Unknown),
J
L
(Dr Lange),
B
(Hjlund Christensen, et al., 2003),
H

The results from the sensitivity analysis show that the gas production is most sensitive to the
TS and flow of the feed and to the pressure in the reactor. The temperature and starch content
are also influential (rank 5 and 4), but the other parameters seem to be measured well enough
or have an insignificant effect on the variability for the gas production. The alkalinity, which
was given the highest uncertainty, is still irrelevant for the gas production. It can also be
concluded that the hydrolysis is not limiting, since the solubility of the waste also is
insignificant for the variance in gas production. The VFA, on the other hand, are very much
affected by the temperature and by the unknown inflow of carbonate alkalinity.
The measurement uncertainty for protein that was used in this analysis was very low. This
resulted in a higher contribution of the TS in the feed to the variance of the ammonium levels
in the reactor than for the protein content (compare 0.42 and 0.84). The carbonate alkalinity in
the reactor was mainly affected by the inflow concentration of HCO
3
, and not by the other
constituents of the feed.
10.1.2. Model parameter sensitivity
The results from the sensitivity analysis on model parameters are presented in Table 17. In the
first column the parameters are listed. The sensitivity analysis did not include the chemical
equilibrium constants, since these were considered to be well defined. The parameters
included maximum growth rate constants for the processes (
max,j
), death rate constants (k
d,j
),
half saturation constants (K
Si
), inhibition constants (KI
ac
), k
L
a for CO
2
(which decides the k
L
a
for H
2
and CH
4
as well) and hydrolysis constant (k
H
).

51

Table 17 Results of SA of model parameters for the Aalborg household waste simulation
Parameter
at 35 C
Mean Unit Sensitivity
(Gas)
rank relative
Sensitivity
(VFA)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO
3
)
rank relative
max3

max4

max5

max6

max7

max8

k
d9

k
d10

k
d11

k
d12

k
d13

k
d14

K
Saa

K
Ssu

K
Sfa

K
Spro

K
Sac

K
SH2

KI
ac56

KI
H2,5

KI
H2,6

KI
H,34

KI
H,58

KI
NH3,6

KI
NH3,7

k
L
a
CO2

k
H

4
4
0.6
0.6
0.37
2
0.8
0.8
0.06
0.06
0.05
0.3
50
50
1000
20
40
1
1500
3
1
0.01
5e-4
25
17
200
0.2
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

mgCOD/m
3

gCOD/m
3

mgCOD/m
3

mgCOD/m
3

mol/m
3

mol/m
3

gN/m
3

gN/m
3

d
-1

d
-1

14
16
2
24
7
5
17
18
4
13
23
11
15
21
3
22
6
8
25
10
12
19
9
27
26
20
1
1.00
1.00
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.019
27
17
24
1
4
6
23
22
26
3
8
13
21
18
16
15
11
9
14
20
5
19
10
2
7
25
12
1.01
1.00
1.00
0.35
0.72
0.80
1.00
1.00
1.00
0.64
0.86
0.96
1.00
1.00
1.00
0.98
0.95
0.90
0.98
1.00
0.80
1.00
0.91
0.47
0.85
1.00
0.96
2
21
17
5
8
6
11
27
10
12
3
7
13
22
26
20
25
15
18
24
14
23
16
4
9
19
1
0.99
1.00
0.99
0.99
0.99
0.99
1.00
1.00
0.99
1.00
0.99
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
0.99
1.00
0.051
19
22
16
3
2
8
20
25
27
7
6
13
18
23
17
15
10
12
14
21
9
24
11
4
5
26
1
1.00
1.00
1.00
0.80
0.75
0.93
1.00
1.00
1.00
0.88
0.88
0.99
1.00
1.00
1.00
0.99
0.96
0.97
0.99
1.00
0.94
1.00
0.96
0.85
0.86
1.00
0.57

The gas production seemed to be virtually insensitive to all model parameters except the
hydrolysis constant, which indicates that the hydrolysis could get rate-limiting for certain
values of k
H
(Table 17). The non-linear correlation between k
H
and gas production is
illustrated in Figure 26 (in section 9.2.1).
It can be noted that the hydrolysis was irrelevant for the VFA concentrations, as were the rates
of fermentation (processes 3, 4 and 5). The VFA were instead linked to the constants for
propionate degradation (process 6; rank 1, 2 and 3) and the acetoclastic methanogenesis
(process 7, rank 4). The high sensitivity for the maximum growth rates could be explained
with a high rate of washout of propionate degraders when the parameters reached critical
values. Figure 27 shows an example where reactor instability can be detected for low values
for
max6
. These parameters would not be suited for calibration, since it would result in
unstable reactors. Instead the half saturation constants for processes that determine the VFA
levels should be considered for calibration of VFA. This result could not be found in the
sensitivity analysis, and a better approach for the SA would hence have been to either use
realistic variations for the parameters, or to sort out values when bacteria were washed out.
This analysis was not performed in this project due to lack of time.

52

Figure 27 Scatter plot with VFA levels as function of
max6

The ammonium is primarily produced in the fermentation of proteins, a process which is fast
and limited by the rate of the hydrolysis. Therefore, the hydrolysis constant was far more
important for the ammonium concentration than the other model parameters.
The carbonate alkalinity is mostly produced in the methanogenesis (process 7 and 8),
consequently the maximum rates of these reactions proved to be important. The most
significant parameter, however, was the hydrolysis constant. It affects alkalinity by
consumption in the hydrolysis, but moreover it decides the rates of the other processes, of
which many include uptake or production of alkalinity.
A relative sensitivity higher than 1 is unlikely, but it can be seen for the VFA and
max3
. It can
happen if more extreme values are produced for the run where one variable is kept constant
than for the run where all parameters vary. If the distributions for the model parameters were
known and the number of runs was increased, this problem would probably not occur to the
same extent.
It could be shown that TS contributed to a great deal to the uncertainty for the gas production,
although it was measured with a standard deviation of 2.5 %. k
H
stands out as an important
parameter for gas production, ammonium and alkalinity when it varies between 0.1 and 0.3. It
is possible that the real variability is even higher which makes it clear that k
H
is essential to
measure or approximate with sufficient accuracy. This is also practical; k
H
is easier to
measure than many of the other constants. The uncertainty of the biogas production can
namely be reduced by measuring three easy variables: k
H
, TS and the gas pressure. Moreover,
measuring the TS and pressure is inexpensive and already implemented at many plants.
Reducing the uncertainty for the alkalinity is more complicated; the most significant variable
HCO
3
in the feed is immeasurable. Numerical calibration of the feed alkalinity is therefore
required to find a suitable inflow concentration (section 11).
10.2. Vstra Hamnen household waste experiment
This reactor was run with the same operational settings as for the Aalborg household waste
experiment, but with mesophilic temperature instead of thermophilic. The SA could hence
reveal if the importance of certain parameters change with the temperature.
The results from the SA on measured parameters are presented in Table 18. It was shown that
the sensitive parameters for the biogas production rate were similar to the Aalborg household
0.2 0.4 0.6 0.8 1
0
1000
2000
3000
4000
5000
max6
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

C
O
D
/
m
3
]
53

waste simulation, while the results for the VFA were more interesting. In this case, the
carbonate alkalinity was the most important parameter, followed by the TS and solubilisation
of TS. The temperature, which was the most important parameter for the Aalborg household
waste reactor was rather unimportant for the mesophilic reactor. This is probably due to the
exponential expressions for temperature dependence (compare eq. 4.15), which cause the
parameters to change more in the thermophilic interval than in the mesophilic. It is clear that
these nonlinearities in the model were exposed by the SA. The sensitivities for ammonium
and carbonate alkalinity are virtually the same as for the thermophilic reactor.

Table 18 Results of SA, UA and variability of measured parameters for the Vstra Hamnen household waste
simulation
Parameter
at 35 C
Mean Unit Uncertainty

rank (%)
Sensitivity
(Gas)

rank relative
Sensitivity
(VFA)

rank relative
Sensitivity
(NH
4
-N)

rank relative
Sensitivity
(HCO
3
)

rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
0.05
0.23
0.17
0.12
0.04
0.07
0.14
7.21
10
13.5
48.7
34.3
35.4
1.33
1.03
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
12
2
14
9
3
10
8
13
1
11
4
4
13
6
6
2.5
S
20
U
0.97
J
3.5
J
18
J
3.4
J
4.7
J
2
U

100
U

3
L
10
U

10
U

2
U

2
U

2
U
1
5
10
7
4
9
6
12
15
13
14
11
8
3
2
0.69
0.93
1.00
0.95
0.90
0.99
0.95
1.00
1.00
1.00
1.00
1.00
0.99
0.87
0.79
3
2
6
10
7
11
9
14
1
12
15
13
4
8
5
0.88
0.88
0.97
1.00
0.98
1.00
0.99
1.00
0.29
1.00
1.00
1.00
0.93
0.99
0.96
1
2
3
6
7
9
8
13
15
10
12
14
4
5
11
0.46
0.76
0.86
0.98
0.98
1.00
0.99
1.00
1.00
1.00
1.00
1.00
0.97
0.97
1.00
3
2
4
13
7
14
8
11
1
12
10
9
6
5
15
0.97
0.95
0.99
1.00
1.00
1.00
1.00
1.00
0.074
1.00
1.00
1.00
1.00
1.00
1.00
S
(SIS, 1981),
U
(Unknown),
J
L
(Dr Lange),
B
H
The most influential parameters from the SA with model parameters for the Vstra Hamnen
household waste simulation are presented in Table 19. The sensitivity for the gas production
was different for mesophilic conditions; k
H
is still the most influential parameter, but there
were also other parameters that affected the output variance. The maximum growth rate of the
acetoclastic methanogens (ranked 2) and the inhibition of the same process (rank 3) were
ranked higher for this reactor. This was probably due to the fact that this process was slower
under mesophilic conditions, and washout of biomass occurred for more parameter settings.
The maximum growth rate of the acetoclastic methanogens was therefore ranked high for the
VFA and carbonate concentrations, which was not the case in the thermophilic reactor. The
acetoclastic methanogenesis is important for the reactor stability; these results therefore
indicate that the stability for the mesophilic Vstra Hamnen household waste digester should
be lower than for the thermophilic Aalborg household waste digester. However, this finding
could not be reasonable since the opposite is known by experience. The drawback of this
method for a poor precision of the parameter distributions is exposed again, and it is clear that
the results need to be evaluated before conclusions are drawn.

Table 19 Results from SA of model parameters for the Vstra Hamnen household waste simulation
Parameter
at 35 C
(Gas)
rank relative
Sensitivity
(VFA)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO3)
rank relative
max7

KI
H,58

KI
NH3,7

k
L
a
CO2

k
H

0.37
5e-4
17
200
0.2
d
-1
mol/m
3

gN/m
3

d
-1

d
-1

2
4
3
17
1
0.78
0.86
0.78
1.00
0.21
1
3
2
20
6
0.09
0.52
0.14
1.00
0.88
2
5
3
24
1
0.98
0.99
0.98
1.00
0.024
1
5
2
21
3
0.41
0.81
0.47
1.00
0.65
54

10.3. Sjlunda sludge experiment
For the Sjlunda sludge, the uncertainty of the feed characterization method was unknown,
and instead guessed standard deviations of 20 % were used (Table 20). The results from this
SA were in other words less reliable than for the household wastes.

Table 20 Results of SA, UA and variability of measured parameters for the Sjlunda sludge simulation
Parameter
at 35 C

rank (%)
Variability

rank (%)
Sensitivity
(Gas)

rank relative
Sensitivity
(VFA)

rank relative
Sensitivity
(NH
4
-N)

rank relative
Sensitivity
(HCO
3
)

rank relative
TS
VS
Soluble TS
Protein
Fat
Inert
pH
HCO
3

NH
4
-N
Temperature
Flow
Pressure
Degradability
0.039
0.75
0.3
0.34
0.45
0.05
7.13
12
45
34
1.5
1.02
0.6
kg/kg
g/TS

gCOD/gX
S
gCOD/gX
S
gCOD/gX
S

mol/m
3

gN/m
3

C
l/d
bar
gCOD/gCOD
8
13
2
2
2
2
9
1
7
9
9
9
2
2.5
S
0.53
J

20
U

20
U

20
U

20
U

2
U

100
U

3
L

2
U

2
U

2
U
20
U

12
10
-
-
-
-
-
-
-
0.6
0
0.9
-
4
10
8
2
13
3
11
5
12
9
6
7
1
0.99
1.00
1.00
0.97
1.01
0.98
1.00
0.99
1.00
1.00
0.99
0.99
0.13
5
8
12
1
4
6
10
3
9
13
11
7
2
0.99
1.00
1.00
0.46
0.97
0.99
1.00
0.90
1.00
1.00
1.00
1.00
0.48
3
5
6
1
13
12
11
4
8
10
7
9
2
0.99
1.00
1.00
0.43
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.60
7
12
5
2
4
6
9
1
10
11
8
13
3
1.00
1.00
0.99
0.78
0.98
1.00
1.00
0.69
1.00
1.00
1.00
1.00
0.81
S
(SIS, 1981),
U
(Unknown),
J
L
(Dr Lange),
B
H

The most influential input parameters for the sludge simulation were the degradability and
protein content (Table 20), which indicated that ammonia inhibition became important for
high protein levels, and that the degradability of the substrate is fundamental to measure. The
composition of the waste did not affect the gas flow as much as the degradability, which
indicated that the characterization was less important to measure. The VFA are affected by the
degradability, which could not be seen for the household wastes, probably because of the
broadened intervals for the inputs.
The results of interest from the model parameter SA are presented in Table 21. The gas
production was mostly dependent on the rate of the acetoclastic methanogenesis which
implied that it was the rate-limiting process. One could suspect that the lower bound for the
growth rate was too low and that this led to washout of bacteria. The degradability was so low
that the biogas production could not be increased much by a faster hydrolysis in this range.
For shorter HRT or higher degradability, the hydrolysis rate would be rate limiting and k
H
would have contributed more to the uncertainty of the result. The instability of the reactor led
to a high sensitivity for all outputs to KI
NH3,7
and
max7
, parameters determining the ammonia
inhibition and growth rate of acetoclastic methanogens. The scatter plot in Figure 28 shows
that the frequency of reactor failure was higher for low values on
max,7
.

Table 21 Results of SA of model parameters for the Sjlunda sludge simulation
Parameter
at 35 C
(Gas)
rank relative
Sensitivity
(VFA)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO
3
)
rank relative
max7

KI
NH3,7

k
H

0.37
17
0.2
d
-1
gN/m
3

d
-1

1
2
6
0.19
0.53
0.94
1
2
26
0.25
0.54
1.03
1
3
2
0.38
0.60
0.60
1
2
27
0.24
0.51
1.03

55

Figure 28 Gas production rate vs.
max
from MC simulation with varying model parameters
10.4. Aalborg household waste experiment with general characterization
In this case the same reactor settings were used as for the previous simulation on Aalborg
household waste, but with the general feed characterization from section 8.3. The distributions
for the characterization variables hence reflect the variability of the parameters when
collected from different sites. The previous SA investigated the relative decrease in output
uncertainty that would be achieved if the characterization variables were determined without
measurement errors. In this case, the SA is conducted to analyze the output uncertainties if no
characterization measurements were made at all. The SA methodology is the same as in
previous sections, but with other means and standard deviations for the characterization. It
should be noted that the variability of the fractions are caused both by real variability in the
waste, but also by measurement errors.
The results from the SA with varying input parameters are presented in Table 22. When
throwing a glance at the results, it may first seem like the starch was a more important
fraction of the substrate than the other constituents. But it can also be seen that the starch
content in the waste had a high variability, and consequently the high values for starch would
increase the degradability of the waste. This was a consequence of the determination of
degradable fraction from the sum of protein, carbohydrates, fibers and fat. The increased
uncertainty for the biogas production seen in 8.3. for the unmeasured substrate can in fact be
explained by the varying degradability of the waste. When plotting the gas production as a
function of degradablility, this relationship is clearly shown (Figure 29). The protein content
of the feed was not very variable and does not affect the biogas production as much, but had
an impact on the concentrations of VFA, ammonium and alkalinity, three parameters that are
of major importance for the reactor stability.

0.2 0.3 0.4 0.5
0
0.005
0.01
0.015
0.02
max7
[ d
-1
]
G
a
s

p
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
]
56

Table 22 Results of SA, UA and variability of measured parameters for the general thermophilic household waste
simulation
Parameter
at 35 C

rank (%)
Variability

rank (%)
Sensitivity
(Gas)

rank relative
Sensitivity
(VFA)

rank relative
Sensitivity
(NH
4
-N)

rank relative
Sensitivity
(HCO
3
)

rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
0.05
0.14
0.15
0.14
0.06
0.15
0.19
7.4
10
160
48.7
34.3
55.3
1.33
1.02
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
5
-
-
-
-
-
-
6
1
4
2
2
6
6
6
2.5
S
-
-
-
-
-
-
2
U

100
U

3
L
10
U

10
U

2
U

2
U

2
U
-
1
7
6
3
2
4
-
-
-
-
-
-
-
-
-
59.7
B
9.2
H
12.0
H
33.3
H
50
H
20.0
H
-
-
-
-
-
-

-
-
5
10
7
4
3
1
2
14
11
13
15
12
9
8
6
0.96
1.00
0.98
0.92
0.85
0.55
0.83
1.00
1.00
1.00
1.00
1.00
0.99
0.98
0.97
8
5
1
10
6
4
7
14
3
11
15
13
2
12
9
0.98
0.95
0.60
1.00
0.96
0.88
0.96
1.00
0.77
1.00
1.00
1.00
0.66
1.00
0.99
2
7
1
5
4
3
6
12
10
8
11
13
15
9
14
0.96
0.99
0.13
0.99
0.98
0.96
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
4
3
2
8
6
5
7
11
1
13
12
10
15
9
14
0.98
0.94
0.70
1.00
0.99
0.99
0.99
1.00
0.39
1.00
1.00
1.00
1.00
1.00
1.00
S
(SIS, 1981),
U
(Unknown),
J
L
(Dr Lange),
B
H

Figure 29 The gas production vs. the degradability of the substrate from the simulation of the Aalborg household
waste with general characterization
k
H
is still the highest ranked model parameter, and in general it seems that the broadened
distributions of the feed fractions has no significant impact on the sensitivity for the
parameters (Table 23). It could be expected that the broader span for the protein could
increase the importance of the ammonia inhibition constants for the VFA, but this was not the
case. Instead, the decay rate for the propionate degraders turned out to be more important. The
amount of propionate depends on the amount of sugars and proteins; fat does not lead to
propionate production in the model. When the span of sugars and proteins are broadened, the
propionate production is increased, and the slow process of propionate degradation becomes
even more important for the total VFA concentration than it already was for the measured
characterization. Another interesting difference is the increased sensitivity for KI
H2,6
(the half
saturation concentration for H
2
inhibition of propionate degradation) which means that the
thermodynamics of the process is affected by the variable substrate. The hydrogenotrophic
methanogenesis (process 8) is affected if the fat content is too high and the hydrogen level is
increased, or if the growth rate of the bacteria is too low. This can also be seen in the SA, as
the sensitivity for
max8
and

KI
H,58
are increased as well.

0 0.1 0.2 0.3 0.4 0.5 0.6
0
0.01
0.02
0.03
0.04
0.05
0.06
Non-degradable fraction of TS
G
a
s

p
r
o
d
u
c
t
i
o
n

[
m
3
/
d
]
57

Table 23 Results of SA of model parameters for the general thermophilic household waste simulation
Parameter
at 35 C
(Gas)
rank relative
Sensitivity
(VFA)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO
3
)
rank relative
max3

max5

max6

max7

KI
H2,6

k
H

4
0.6
0.6
0.37
1
0.34
d
-1
d
-1
d
-1
d
-1
mgCOD/m
3

d
-1

19
2
4
8
9
1
1.00
0.98
0.98
0.98
0.98
0.04
26
16
1
6
5
14
1.01
1.00
0.34
0.68
0.65
0.98
2
17
9
12
14
1
0.98
1.00
0.99
1.00
1.00
0.08
16
14
3
2
8
1
0.99
0.98
0.78
0.72
0.88
0.58

10.5. Vstra Hamnen household waste experiment with general
characterization
This SA was conducted using the same characterization as in the previous example, but with
operational settings as in the Vstra Hamnen household waste simulation (mesophilic). The
results (Table 24) show that the ranking of the parameters remains mainly unchanged
compared to the Aalborg household waste reactor. The same pattern can be observed for the
gas production, but the sensitivity for the VFA is slightly altered and the importance for the
protein is higher than in the thermophilic reactor.

Table 24 Results of SA, UA and variability of measured parameters for the general mesophilic household waste
simulation

rank (%)
Variability

rank (%)
Sensitivity
(Gas)

rank relative
Sensitivity
(VFA)

rank relative
Sensitivity
(NH
4
-N)

rank relative
Sensitivity
(HCO
3
)

rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
0.05
0.14
0.15
0.14
0.06
0.15
0.11
7.2
10
13.5
48.7
34.3
35.4
1.33
1.03
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
5
-
-
-
-
-
-
6
1
4
2
2
6
6
6
2.5
S
-
-
-
-
-
-
2
U

100
U

3
L
10
U

10
U

2
U

2
U

2
U
-
1
6
5
3
2
4
-
-
-
-
-
8
-
7
-
59.7
B
9.2
H
12.0
H
33.3
H
50
H
20.0
H
-
-
-
-
-
1.4
A

-
1.6
A

5
9
7
4
3
1
2
14
11
13
15
12
10
8
6
0.96
0.99
0.98
0.92
0.84
0.54
0.83
1.00
1.00
1.00
1.00
1.00
1.00
0.98
0.97
8
4
1
11
6
3
7
15
2
12
14
13
5
9
10
0.96
0.90
0.38
1.00
0.94
0.81
0.94
1.00
0.70
1.00
1.00
1.00
0.91
0.98
1.00
3
5
1
7
4
2
6
12
15
10
11
13
9
8
14
0.96
0.98
0.17
0.98
0.98
0.94
0.98
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
5
3
2
8
6
4
7
12
1
13
14
11
10
9
15
0.99
0.93
0.72
1.00
0.99
0.98
0.99
1.00
0.39
1.00
1.00
1.00
1.00
1.00
1.00
S
(SIS, 1981),
U
(Unknown),
J
L
(Dr Lange),
B (Hjlund Christensen, et al., 2003)
,
H
10.5.1. Model parameters
As for most other cases, k
H
stands out as the first parameter to be calibrated (Table 25). This
can be seen by the much lowered variances in the gas production, ammonium and alkalinity.
If k
H
is varying, the VFA will have more extreme values than if it is kept constant (relative
sensitivity >1), the same applies for the K
Sac
. The reason for these odd values could be the
extreme values for VFA that are produced by some combinations. It is also possible that the
VFA have not converged to the steady state values during the simulation time.

58

Table 25 Results of SA of model parameters for the general mesophilic household waste simulation
Parameter Mean Unit Sensitivity
(Gas)
rank relative
Sensitivity
(VFA)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO
3
)
rank relative
max7

k
d13

KI
H,58

KI
NH3,7

k
H

0.37
0.05
5e-4
17
0.34
d
-1
d
-1
mol/m
3

gN/m
3

d
-1

4
3
2
5
1
0.87
0.87
0.86
0.87
0.19
1
4
2
3
26
0.44
0.52
0.45
0.52
1.12
3
2
5
4
1
0.98
0.97
0.98
0.98
0.06
2
5
3
4
1
0.72
0.77
0.76
0.76
0.52
10.5. Discussion on SA
As discussed earlier, using spans of 50 % when varying model parameters produces
misleading results. The ranges of variability for the model parameters were not available, and
a realistic uncertainty or sensitivity analysis was not feasible to conduct within this project.
Although the quality of the SA results may be disputable, it could still give interesting
information on the system. For example: even if the relation between the importance of input
and model parameters may be incorrect, the difference in this relationship can still be
compared for different modes of operation, for example mesophilic and thermophilic
conditions. In the results from the uncertainty analysis it was revealed that the outputs were
affected differently by the changes in input data. The SA could provide some clues on what
one should measure to reduce uncertainty but also give a deeper understanding of the process
and of the mathematics of the model. Nonetheless, it was clear that the method was useful to
find interesting parameters but it needed to be supplemented with scatter plots to avoid false
conclusions from the results.
10.6. Summary of the SA
The most important findings from the SA were:
The hydrolysis rate constant k
H
and the degradability are important input parameters
for the biogas production for the household waste digesters.

In the Sjlunda sludge digester, the degradability of the substrate was so low that the
hydrolysis constant was unimportant for the studied HRT.

The VFA concentration is much more affected by the model parameters linked to
propionate degradation and acetoclastic methanogenesis than by the input parameters.

The sensitivities for the parameters differ between a mesophilic and thermophilic
digester; the model parameters are more important for the gas production in a
mesophilic digester.

If the protein content in a household waste is not measured, this contributes to the
uncertainty for the ammonium, VFA and alkalinity.

The SA method can give some information on which parameters are suitable for
calibration, but it is important to consider reasonable distributions of the model
parameters and to study scatter plots before performing a calibration based in the SA
results.

59

11. Calibration of the model parameters
In this part of the thesis, some parameters are calibrated to improve the fit of model
predictions to data. The calibration is made for the pilot scale simulations for which
sensitivity analysis was conducted in the previous section. The results from the SA will be
used for choosing suitable parameters for calibration of different output variables. As far as
possible, changes in the parameterization are motivated with a physical explanation.
11.1. Method for calibration
Most of the calibrations were conducted by minimizing the sum of absolute errors for model
predictions,
i
, compared to measured data (eq. 11.1). The sums of absolute errors were
simply plotted as functions of the calibration variable, and the calibration value for which the
error function was minimized was chosen. Compared to using optimizers in Matlab such as
fmincon and fminsearch this method has lower precision, but it is faster, and gives a graphical
presentation of the error which makes it easier to avoid local minimums as solutions. The
magnitude of the error and non linear behaviors are also revealed with this method. The start-
up phases of the pilot scale reactors on household waste were not stable, and were therefore
excluded when performing the calibration.
E = |y
0,
-y
0,i
|
n
=1
(11.1)

The first step in the calibration procedure for the Vstra Hamnen household waste and
Sjlunda sludge simulations was to calibrate the alkalinity of the feed. This was done first of
all, because the sensitivity analysis showed that the reactor alkalinity was sensitive to this
parameter, and because it is immeasurable in these types of wastes. Many processes in the
digestion involve uptake and production of alkalinity, and it would not be desirable that the
lack of calibration for inflow alkalinity affected the calibration of other parameters. For the
Aalborg household waste, this method could not be applied since the dependency between the
alkalinity and the VFA needed to be considered. The VFA were sensitive to the inflow
alkalinity while the alkalinity in the digester was sensitive to the parameters in propionate
degradation and acetoclastic methanogenesis which are important for the VFA concentration.
It was thus impossible to calibrate the parameters apart from each other; a numerical method
for minimizing the errors for several parameters at a time was required here. An optimization
with fminsearch was therefore conducted. The same relation existed for the Vstra Hamnen
household waste simulation but in that case it did not cause a problem in the calibration
because the alkalinity in the reactor was much too high compared to the measured values. If
the calibrations were conducted simultaneously, the alkalinity in the inflow would still need to
be reduced substantially.
The other model parameters to be calibrated differed from case to case, but for all cases the
biogas production was an interesting output parameter to include in the calibration process,
e.g. for calibration of k
H
. In the sensitivity analysis, it was concluded that the VFA
concentration mostly depended on the parameters connected to the propionate degradation
and the acetoclastic methanogenesis. These parameters were calibrated to minimize the sum
of errors for VFA predictions.

60

11.2. Aalborg household waste experiment
11.2.1. Hydrolysis constant k
H

The SA showed that k
H
was more important for determining the gas production, alkalinity and
ammonium concentration than any of the other model parameters. It was not calibrated from
batch experiment, but assumed to be the same as for the Vstra Hamnen household waste and
was therefore an interesting candidate for calibration. Due to the high relative error for
ammonium, it would have had a major impact on the optimization if it was included in the
objective function for the calibration. A trial to calibrate k
H
from the ammonium
measurements only revealed that the k
H
would have to be so low that the hydrolysis almost
stopped to lower the error for ammonium. It was therefore concluded that the relative error for
ammonium cannot be reduced by calibration of k
H
. The gas production, on the other hand, the
most economically important output, could probably be used successfully to calibrate k
H
. It
was therefore chosen as output for the calibration. The result (Table 26) showed that the
former value which was determined by experiments should be reduced slightly to improve the
model fit to the data.
11.2.2. Protein content
The deviation from the measured values for the simulation of ammonium is problematic since
ammonia is important for inhibition of propionate degradation and acetotrophic
methanogenesis. k
H
could not be used in the calibration, since it was calibrated from the gas
production rate, although it was the most important parameter for the ammonium
concentration. This means that either the protein measurement was unsuccessful, or the
hydrolysis model could not describe the degradation of protein properly. For the ammonium
to correlate with the measured values, the protein content in the feed would have to be 9 % on
a weight basis instead of the measured value 17 % (Table 26). It is unlikely that the
measurement of protein would deviate so much from the real value. A more plausible
explanation is that the protein was hydrolyzed slower than the other fractions in the substrate,
but to include this in the model, an alteration of the hydrolysis model would be required.
Throughout this work, however, a fixed model structure was used, and in the further
calibration the calibrated value for protein content was used instead. This was used because it
is important to have reasonable values of ammonium to calibrate the processes that are
affected by ammonia inhibition. To compensate for the error in inflow COD as the protein
content is reduced, the weight was added to the sugar content in order to retain the biogas
production.
11.2.3. Parameters for propionate degradation, acetotrophic methanogenesis
and inflow alkalinity
Both the VFA and the alkalinity were sensitive to
max7
,
max6
(growth rates of propionate
degraders and acetotrophic methanogens) and K
INH3,6
, (inhibition constant for propionate
degradation). The SA thus indicated that these parameters could be used for calibration of
these outputs. However, as discussed in 10.1.2 the high sensitivity to these parameters was
mainly due to a frequent washout of bacteria when these parameters reached extreme levels.
When trying to calibrate the VFA using growth rates, the result showed that the fit to the
measured values only could be improved slightly, and that instability occurred when changing
the values further. A better result was achieved when calibration with half saturation constants
instead, which were calibrated by Siegrist et al. by measuring propionate and acetate for
different HRTs in a mixed sludge digester. The fit to data could be improved more in this
case, but the changes in parameter values were high for optimal fit. In Figure 30, a
comparison between a calibration of VFA with
max7
and K
S,ac
is shown. When increasing the
61

half saturation constant, the Monod kinetics produces the same reaction rate at a higher
substrate concentration. Thus, the gas production should not be affected from the calibration,
which made the calibration process less complicated.

Figure 30 Example of calibration of VFA using
max7
(left) and K
S,ac
(right)
VFA needed most calibration for the household waste experiments. It is possible that the
accessibility of the substrate for the propionate degraders and acetotrophic methanogens was
lower for the household waste digester, which could give a physical explanation for
calibrating K
S,pro
and K
S,ac
.
For the Aalborg household waste simulation both parameters were first calibrated
independently with the graphical method, and it was seen that they should be increased many
times to lower the error for the VFA, to 220 and 240 g COD/m
3
respectively. The shape of the
curve for the total VFA was more similar to the measured values for the calibration of K
S,ac

than for K
S,pro
, which means that K
S,ac
probably is altered more than K
S,pro
for the substrate. It
was thus decided that K
S,pro
should be set to 100 g COD/m
3
and that

calibration of K
S,ac
should
account for the rest of the output error. As was discussed in 11.1., the inflow alkalinity and
VFA are correlated and must be calibrated simultaneously. With the start guesses 180 g COD
/m
3
for K
S,ac
and 30 mol/m
3
for S
HCO3
, the optimizer fminsearch converged well, but the results
(Table 26) differed much from the values suggested by Siegrist et al. (2002). The half
saturation concentration for acetate would have to be increased more than six fold to
minimize the error. On the other hand, it has been seen in calibration of ADM1 that the half
saturation constants for propionate and acetate uptake were 297 and 582 g/m
3
respectively
(Jeong, et al., 2005). The suggested values in ADM, 100 and 150 (Jeong, et al., 2005) were
also much higher than the values in the Siegrist model. The substrate uptake rate in ADM1,
which is equivalent to the maximum growth rate parameter in the Siegrist model has been
shown to be important and was calibrated to much lower values by Tartakovsky et al. ( 2008).
This indicates that the uptake rate of VFA is variable, Tartakovsky, et al., (2008) suggested
that the content of intert material in the substrate could affect the the mass transfer. The
resulting VFA levels after calibration (Figure 32) shows that the steady state concentration
could be modeled, but that the fit to data in the start-up phase was poor.

0 0.1 0.2 0.3 0.4 0.5 0.6
0
1000
2000
3000
4000
5000
6000
7000
max7
S
u
m

o
f

a
b
s
o
l
u
t
e

e
r
r
o
r
s
0 50 100 150 200 250
0
1000
2000
3000
4000
5000
6000
7000
S
u
m

o
f

a
b
s
o
l
u
t
e

e
r
r
o
r
s
K
Sac
62

Table 26 Values for former and calibrated parameters for the Aalborg household waste simulation
Parameter Unit Former
value
Calibrated
value
Output
k
H
d
-1
0.20 0.15 Dry gas production
Protein content g/g TS 0.17 0.09 Ammonium in reactor
K
S,pro
for 35 C g COD/m3 20 100
*
VFA in reactor
K
S,ac
for 35 C
g COD/m3 40 260 VFA in reactor+ Reactor alkalinity
Inflow S
HCO3
mol/m
3
5 42 VFA in reactor+ Reactor alkalinity
*Assumed value

An important issue when calibrating the model was the problem with parameter identifiability
which was discussed in section 6. There are many combinations of for example
max6
, k
d,12
,
K
S,pro
and K
INH3,6
that give the same reaction rate for the propionate degradation which means
that a correct calibration is impossible. This problem could easily be dealt with by using one
parameter at a time for the calibration, and that strategy was employed in this work. The
alternatives would have been to make guesses for parameter combinations or simply to avoid
calibrating these parameters.
11.2.4. Simulation results with calibrated parameters
The gas production and methane content were not changed notably after the calibration
(compare Figure 31 with Figure 4), and the start up phase was still not correlating well.
Considering that the calibration was performed for the steady state phase, this was not
surprising. The lower biogas production rate in the start-up phase was probably due to that the
microorganisms were not adapted to the substrate, and lacked specific enzymes for the
hydrolysis. When modeling a slower hydrolysis in the beginning, the measured gas
production correlated better to measurements. An example of how this could be modeled
iteratively is shown in Figure 34. The VFA were over predicted during the start-up with
calibrated parameters, and the correlation for alkalinity was poorer (Figure 32). The
calibration improved the correlation for ammonium and pH (Figure 33).

with Aalborg household waste after model calibration
0 20 40 60 80
0
0.01
0.02
0.03
0.04
0.05
0.06
Time [days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

Biogas production modelled
Biogas production measured
0 20 40 60 80
0
20
40
60
80
100
Time [days]
P
e
r
c
e
n
t

[
%
]

CH4 modelled
CH4 measur ed
63

3
Aalborg household waste after model calibration

household waste after model calibration

Figure 34 Measured and simulated biogas production (left) with k
H
increased stepwise as in the figure to the right

0 20 40 60 80
0
500
1000
1500
2000
2500
3000
3500
Time [days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

C
O
D
/
m
3
]

VFAs modelled
VFAs measur ed
0 20 40 60 80
0
10
20
30
40
50
60
70
80
Time [days]
C
o
n
c
e
n
t
r
a
i
o
n

[
m
o
l
/
m
3
]

HCO3 modelled
HCO3 measur ed
0 20 40 60 80
0
500
1000
1500
2000
2500
3000
3500
Time [days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

NH4 modelled
NH4 measur ed
0 20 40 60 80
0
1
2
3
4
5
6
7
8
Time [days]
p
H

pH modelled
pH measur ed
0 20 40 60 80
0
0.01
0.02
0.03
0.04
0.05
0.06
Time [ days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

0 20 40 60 80
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Time [ days]
k
H

[
d
-
1
]
64

11.3. Vstra Hamnen household waste experiment
11.3.1. Inflow alkalinity
As seen in section 7.4.3., the simulated alkalinity during the steady state operation was too
high. The calibration of inflow alkalinity showed that it should be as low as possible, but even
if it was set to zero, large errors remained. The calibration was hence not successful, and other
possible calibration parameters were examined. The calibration of the reactor alkalinity was
however considered good enough since the deviations were small enough to have only minor
impact on the result. It was clear that the variations in alkalinity could not be modeled very
well; the UA showed that the uncertainty of alkalinity is high which means that detailed
calibration is impossible.
11.3.2. Hydrolysis constant k
H
It was revealed in the SA that k
H
had a significant impact on both the gas production and on
the ammonium concentration in the reactor. Only the error in gas production was used for the
calibration, however, since this parameter was prioritized because of the stronger correlation
to the other processes in the system. If both the gas production and ammonium levels in the
reactor were to be calibrated, the model would have to be expanded to include an individual
hydrolysis constant for particulate protein, as discussed for the calibration of the Aalborg
household waste simulation. It turned out that the value used previously in the simulations
gave the best fit to data (Table 27), which showed that the batch experiment produced a
reasonable value. The value for the hydrolysis rate constant when calibrating the ammonium
levels was low (k
H
= 0.02), which indicated that the degradation of the protein fraction of the
particles was ten times slower than the other constituents.
11.3.3. Protein content of feed
The ammonium concentration in the reactor was also dependent on the protein content of the
feed, and could be calibrated with this value. The value for protein content that gave the
optimal fit to data was 10 % instead of 17 % of TS (Table 27). The measurements are not
expected to be this poor for protein, thus indicating that it rather is the hydrolysis of protein
that is slower. The result, however, had the same effect on the amount of degraded protein in
the reactor, and hence the ammonia inhibition will be better modeled with this new value. The
seven percent of particles that was withdrawn from the total degradable TS was added to the
sugar fraction to compensate for the loss of degradable matter.
11.3.4. Half saturation constant for acetotrophic methanogenesis, K
S,ac

For calibration the VFA in the reactor, the sensitivity analysis indicated that the growth rate of
aceticlastic methanogens and the inhibition of this process were suitable as calibration
parameters. The high sensitivity could however be explained by the high rate of washout of
bacteria, as discussed in the previous section. Using the half saturation constant K
s,ac
turned
out to be more successful, and is easier to motivate from a biological point of view, see
11.2.3. The required value for best fit to data K
S,ac
=190 (Table 27), was lower than the value
calibrated for the Aalborg household waste. It could probably have been useful to calibrate
more parameters than only one in this case, but this was not prioritized.
11.3.5. Simulation results with calibrated values
The resulting simulation results are shown in Figure 35 - Figure 37. Compared to the first
crude validation in 8.4., the gas production was more stable but otherwise not much changed.
The correlation for ammonium was much better after the calibration of the protein content,
but it can also be noted that the initial value probably was higher than the average in the
Sjlunda sludge digester. The modeled VFA were variable, but the range of variation was
65

now closer to the measured values. It seemed that the drop in methane content in the gas
could and pH not be reflected in the results after calibration.

Table 27 Values for former and calibrated parameters for the Vstra Hamnen household waste simulation
Parameter Unit Former value Calibrated value Output
Inflow S
HCO3
mol/m
3
10 0 Reactor alkalinity
k
H
d
-1
0.20 0.20 Dry gas production
Protein content of feed g/g TS 0.17 0.10 Ammonium in reactor
K
s,ac
d
-1
40 190 VFA in reactor

with Vstra Hamnen household waste after model calibration

3
Vstra Hamnen household waste after model calibration
0 10 20 30 40 50
0
0.01
0.02
0.03
0.04
0.05
Time [ days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

Simulat ed dr y biogas pr oduct ion
0 10 20 30 40 50
0
20
40
60
80
100
Time [ days]
P
e
r
c
e
n
t

[
%
]

Simulat ed CH4
Measur ed CH4
0 10 20 30 40 50
0
50
100
150
200
250
300
350
400
Time [ days]
V
F
A
s

a
s

a
c
e
t
a
t
e

e
q
u
i
v
a
l
e
n
t
s

[
g

C
O
D
/
m
3
]

Simulated VFAs
Measur ed VFAs
0 10 20 30 40 50
0
20
40
60
80
100
120
Concent r at ion of HCO3
Time [days]
C
o
n
c
e
n
t
r
a
i
o
n

[
m
o
l
/
m
3
]

Simulat ed HCO3
Measur ed HCO3
66

Figure 37 Measured and simulated ammonium concentration and pH from continuous experiment with Vstra
Hamnen household waste after model calibration
11.4. Sjlunda sludge
11.4.1. Inflow alkalinity
The inflow alkalinity was calibrated with the graphical method, and it was shown that the
calibrated value was the same as the former used assumed value (Table 28).
11.4.2. Degradability
In the sensitivity analysis, it was concluded that the gas production was more sensitive to the
degradability of the substrate than to the model parameters connected to the acetoclastic
methanogenesis and propionate degradation. Furthermore, the model parameters had already
been calibrated while the ultimate degradability of the sludge was more or less assumed. It
was thus decided to use the gas production to calibrate the degradability of the substrate. The
result, (Table 28) is reasonable for mixed sludge. As seen in section 8, the measurement
uncertainty of VFA is low, but the sampling uncertainty is probably more significant. The
correlation for VFA concentration (Figure 39) was therefore considered to be good enough
after the calibration of degradability; no further calibrations were needed for the parameters in
propionate degradation and acetate methanogenesis.

Table 28 Values for former and calibrated parameters for the Aalborg household waste simulation
Parameter Unit Former value Calibrated
value
Output
Inflow S
HCO3
mol/m
3
10 10 Reactor alkalinity
Degradability gCOD/gCOD 0.60 0.72 Dry gas production

0 10 20 30 40 50
0
200
400
600
800
1000
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

Simulat ed NH4 modelled
Measur ed NH4
0 10 20 30 40 50
6
6.5
7
7.5
8
pH
Time [ days]
p
H

Simulat ed pH
Measur ed pH
67

11.4.3. Simulation results calibrated parameters
The resulting simulation results from the calibration of the Sjlunda sludge experiments are
presented in Figure 38 - Figure 40. It can be seen that the calibration of degradability
significantly improved the correlation for the gas production. Overall, the quality of the
predictions is higher for the sludge than for the household wastes, which further emphasize
the importance of parameter calibration for household wastes. For mixed sludge, however,
these simulation results show that results with high accuracy can be achieved with only minor
calibration if the characterization method of Siegrist, et al., (2002) is used.

with Sjlunda sludge digestion after model calibration

3
Aalborg household waste after model calibration
0 20 40 60 80
0
0.01
0.02
0.03
0.04
0.05
Time [days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

0 20 40 60 80
0
20
40
60
80
100
Time [days]
P
e
r
c
e
n
t

[
%
]

Simulated CH4
Measur ed CH4
0 20 40 60 80 100
0
100
200
300
400
500
Time [ days]
V
F
A
s

a
s

a
c
e
t
a
t
e

e
q
u
i
v
a
l
e
n
t
s

[
g

C
O
D
/
m
3
]

Simulat ed VFAs
Measur ed VFAs
0 20 40 60 80
0
20
40
60
80
100
Time [days]
C
o
n
c
e
n
t
r
a
i
o
n

[
m
o
l
/
m
3
]

Simulated HCO3
Measur ed HCO3
68

household waste after model calibration
11.5. Discussion on the calibration
From the SA in section 10 it was concluded that the most important output variable, the gas
production rate, could be used to calibrate k
H
. In section 11, it was shown that the values of k
H

which were calibrated from batch experiments were good approximations. It was however
impossible to get a good fit to data in the start-up phase for the Aalborg household waste
simulation with a constant k
H
, indicating that the hydrolysis was slower before the
microorganisms had adapted to the new substrate.

When it came to the ammonium levels in the reactor, however, the only way to improve the fit
to data with a fixed k
H
was to drastically change the protein content in the feed. It is unlikely
that the measurements would be of such low quality. Furthermore, the mean protein content in
the tested household wasted was 15 % (Table 22 in 10.4.), which is much higher than the
values that were needed to achieve an acceptable correlation to the measurement. If the model
is to be used successfully for household waste digestion, it is essential to calibrate the protein
content with the ammonium concentration in the reactor or to extend the hydrolysis model.
Both methods produce reasonable results, but the second option has a theoretical background
and is therefore preferable. The hydrolysis description applied in ADM1 (Batstone, et al.,
2002) includes different hydrolysis rate for all constituents, and could be used in further
studies. The hydrolysis rate for protein would in that case be approximately ten times slower
that the average particular organic component.
For the Sjlunda sludge, the correlation of modeled ammonium to data was much better, but
this was not surprising since the protein was calculated from the ammonium in the digester. It
is possible that the real protein content in the mixed sludge was higher, which would mean
that the hydrolysis of protein is slower in this case as well. A measurement of the protein in
the Sjlunda sludge would confirm if this theory is likely. Another option is that the
hydrolysis description in the model is less suitable for household waste than for mixed sludge.
Household waste includes meat and other components with high protein content and may in
other words be less available for hydrolysis than the protein in the mixed sludge. This can
depend on the particle size which is affected by the pretreatment method, and on the protein
characteristics. Protein in the primary sludge is readily available, and the protein in the WAS
becomes available after cell lysis, together with the other substrates.
As suspected in the first validation of the model (7.4.6. and 7.6.6.), the model parameters for
the household wastes needed to be recalibrated to improve the correlation for VFA. The lack
0 20 40 60 80
0
500
1000
1500
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

Measur ed NH4
0 20 40 60 80
0
1
2
3
4
5
6
7
8
Time [days]
p
H

Simulat ed pH
Measur ed pH
69

of identifiability and the correlation between the parameters were however problematic for
calibration of model and inflow parameters. It was concluded that the SA did not give the best
calibration values, but that the half saturation constants was better suited for calibration
purposes. As discussed in 11.2.3., half saturation constants of VFA seem to vary considerably
for different substrates and it was not surprising that these parameters needed recalibration for
household wastes. The VFA are thus less likely to be successfully modeled without
calibration than other outputs, which was also indicated in the crude validation (7.3.3. and
7.4.3.). Since the sampling uncertainty is suspected to be high for VFA, it is also possible that
the model results are more credible than the measurements if the measurements are slightly
lower than the simulated values. However, as long as simulated VFA levels do not exceed
levels that are inhibitory to the methanogenesis, the correlation for biogas will still be
acceptable.

70

71

12. Industrial scale application
In this section, the model will be used for evaluation of a full scale process at Kppala
WWTP; various process designs will be assessed and discussed. A simple economical
analysis is made to compare the economical benefits for the process designs.
12.1. Example with Kppala digesters
Kppala WWTP which receives 500 000 p.e. has two mesophilic digestion chambers, R100
and R200 (Figure 2, section 7.2.1.). The primary sludge is digested separately in R100 to
avoid filamentous bacteria clogging the upper part of the digester and causing reactor
overflow. The digested primary sludge is mixed with the WAS in R200, for typical
characteristics of the inflow and validation to data, see section 7.2.1. With the current setup
the biogas production from R200 is low and it could possibly be improved with thermophilic
temperature, different process design or pretreatment of the WAS. Experimental trials to
compare different solutions would be both expensive and time consuming. Nonetheless,
different alternatives for reactor setup and process variables can be tested on the full scale
process, which has been done on some occasions. The drawback of such full scale
experiments is that they are can be costly e.g. if the VS reduction is poor or if practical
problems arise. Simulating different alternatives for process setups can be a useful tool for
discovering possibilities and problems quickly and inexpensively. Several possibilities for the
process are explored in this section, including thermophilic digestion and pretreatment of the
WAS. Some parameters are not included in the model, like the dewaterability of the sludge,
which will raise the costs for handling of the remaining sludge, or the amount of filamentous
bacteria. The results can however be used to evaluate the potentials of the different setups.
12.1.1. Simulated cases

A) Original design described in 7.1.2., mesophilic (35 C)
The retention time for the primary sludge is very long for this design, 15+10 days
while the WAS has a shorter retention time of 10 days.

B) Original setup, thermophilic (55 C)
The degradability in thermophilic digestion has proved to increase with 7.6 %
compared to mesophilic digestion (Song, et al., 2004). A representative retention time
for mesophilic digestion is 25 days, while it is about 15 days for thermophilic
digestion. Consequently, the temperature dependency of k
H
(see 4.4) resulted in an
increase for the WAS k
H
from 0.15 to 0.24 d
-1
, and 0.4 to 0.65 d
-1
for the primary
sludge k
H
. The digester volumes and thus retention times were kept constant.

C) The digesters in series, mesophilic (35 C)
For this case, the sludge is mixed and flows first into R100 and then R200 (Figure 41).
The total retention time for the primary sludge is reduced in this case, from 25 to 20
days total but it is increased for the WAS from 10 to 20 days total.

D) The digesters in series, thermophilic (55 C)
The same reactor design as in case C is used, but with thermophilic temperature 55 C.
The degradability and k
H
are the same as for case B.

E) The digesters in series, enzyme addition mesophilic (35 C)
Sludge pretreatments have been proved to increase the gas production significantly as
discussed in section 2.5. Many treatment methods require expensive investments in
72

form of vessels, heat exchangers, pumps etc. Enzymatic treatment has the advantage
of low investment costs; the enzymes are simply poured into the digester. The
solubilisation and degradability of the primary sludge are not notably improved by
enzyme addition; the substrate is already highly degradable and hydrolysable. The
WAS, on the other hand, is not readily degradable unless the EPS are degraded
(Recktenwald, 2008). Solubilisation of WAS with a dose of 60 mg/g TS increased to
0.41 g SCOD/g TS (Wawrzynczyk, 2007); recalculated for the characterization of the
WAS this means a solubilisation of 49 % on a COD basis. The degradability of WAS
with enzyme pretreatment was calculated from the increased methane production from
batch tests, which increased with 60 % (Wawrzynczyk, 2007). It should be noted that
the enzymatic dose of 60 mg/L in the laboratory scale tests is significantly higher than
the dose of 0.25 mg/L which is used for full scale processes (Recktenwald, 2008). The
laboratory scale data was used due to lack of known input variables for full scale
digesters with enzymatic pretreatment.

F) One reactor with the volume V
R100
+ V
R200

In many WWTPs, there is only one digester in which both the primary sludge and
WAS are digested (Figure 42). The total retention time is 20 days for both flows, the
same as for case D.

Figure 41 Reactors in series, cases C, D and E

R100+R200
Primary sludge
Waste activated sludge
To dewatering

Figure 42 The digester in case F
73

12.1.2. Method
Data from 2007 for flows and VS content of the primary sludge and WAS were used. All
initial values and parameters for case A can be found in section 7.6. The allocation for
calculating the characterization of the substrate and k
H
was made on a basis of mass VS of
primary sludge and WAS into the reactor.
With thermophilic temperature, (cases B and D) the methane content in the gas is 3
percentage points lower than for the mesophilic cases (A and C), but it can also be seen that
the methane production is increased with 9 percentage points compared to mesophilic
digestion (Table 29). Enzyme addition is also an effective measure to improve the methane
production, an increase of 11 % was reached compared to the first case. The methane
production in the single digester (case F) is the lowest; 5 % lower than for case A.
The increase of ammonium for the thermophilic digesters compared to the mesophilic is
small, only 100 and 200 g COD/m
3
for case B-A and D-C (Table 30). The risk for ammonia
inhibition, however, increases more compared to the mesophilic cases and it was seen that the
inhibition was more important in case D than case C.
The alkalinity was mainly unaffected by the temperature (compare case A-B and C-D) but
varied for different process designs (compare A-B and C-D). The enzymatic treatment in case
E increased solubilization which led to a fast acidogenesis and high concentration of VFA
(Table 30). Although the concentration reached over 3000 g COD/m
3
, the gas production was
still high, this indicated that VFA inhibition was not a substantial problem for the gas
production (Table 29). Furthermore, it is worth noting that the VFA concentrations for cases
A and B are similar in R100 and R200, while the VFA level is higher in R100 when the
digesters are run in series (cases C and D). The thermophilic cases B and D showed higher
VFA concentration than the mesophilic cases A and C.
The total COD reduction could be increased 3 percentage points compared to case A by
running the process in series, and 7 percentage points by introducing thermophilic digestion
(Table 30). Enzymatic treatment and thermophilic digestion in series give the highest VS
reduction, 0.64 and 0.65 respectively.

Table 29 Biogas production for R100 and R200 with different process designs for 2007
Case Total
dry gas
R100

[10
6
m
3
]
Total
dry gas
R200

[10
6
m
3
]
Total
methane
R100

[10
6
m
3
]
Total
methane
R200

[10
6
m
3
]
Mean
methane
content
in dry gas
Total dry gas
production in
R100 and
R200
[10
6
m
3
]
Total
methane
production

[10
6
m
3
]
Relative
change in
total
methane
production
Original setup, A
5.1

1.1

3.2 0.79 0.64 6.3 4.0 0 %
B, 55 C
5.9 1.3 3.5 0.83 0.61 7.2 4.3 9%
C, series mesophilic
5.6 1.6 3.2 0.97 0.58 7.2 4.2 5%
D, series thermophilic
6.8 1.4 3.7 0.78 0.55 8.2 4.5 14 %
E, Enzyme treatment, series
5.9 1.8 3.3 1.1 0.57 7.7 4.4 11 %
F, one digester
6.6 - 3.8 - 0.58 6.6 3.8 -5%

74

Table 30 Process variables for R100 and R200 with different process designs
Case Mean
NH4-N
R100

[g N/m
3
]
Mean
NH4-N
R200

[g N/m
3
]
Mean
HCO3
R100

[mol/m
3
]
Mean
HCO3
R200

[mol/m
3
]
Mean VFA
R100

[g COD/m
3
]
Mean VFA
R200

[g COD/m
3
]
Total
COD
reduction
after
R100
Total
COD
reduction
after
R200
Original setup, A
1100 1300 68 87 66 49 0.30 0.56
B, 55 C
1200 1400 71 88 200 220 0.38 0.63
1400 1600 80 110 780 170 0.45 0.59
1600 1800 92 110 870 330 0.54 0.65
1600 1800 8 200 3500 1800 0.50 0.64
F, one digester
1500 - 94 - 580 - 0.53 -

It is common in WWTPs to run the digesters in parallel instead of series to facilitate the
control of the process, but it can be seen that this is not the best process design if increased
gas production, VS degradation and stabilization is the goal (Table 29 and Table 30).
Apparently, the process rates are increased with serial digestion. This is due to higher
concentrations of the substrates in R100 than in a single digester where the feed is more
diluted. The substrates are more easily available for the biomass, and the substrate uptake rate
is increased. Although the retention time for R100 in the case of serial operation (C) is half of
the retention time with one big digester (F), it can be seen in Table 29 that the biogas
production in R100 for case C is much higher than half than for case F. It is understandable
that many WWTP prefer parallel digestion, but it could be preferable to implement serial
digestion with improved process control.
Enzyme addition was beneficial for increasing the VS degradation and methane production,
but resulted in a high VFA concentration which could lead to inhibition of the
methanogenesis and propionate degradation (case E). The VFA increased during the winter
due to the higher production of WAS, and the total effect from the enzyme addition became
more pronounced. Problems with high VFA levels could be avoided if the dose of enzymes
was adapted to the amount of WAS, or if the active volume was increased. However, the dose
of enzymes used in the simulation was, as mentioned, too high to be economically defendable.
If enzyme addition was introduced, the dose would be significantly lower and the VFA
concentrations would not reach the levels shown in (Table 30). The drawback of a lower dose
could be that the methane production would not be increased as much as the 11 % for case E.
The effect from enzyme dosing is also dependent on the sludge characteristics, e.g. ratio
municipal/industrial waste water, sludge genesis, season and sludge age. In order to evaluate
the process with enzyme dosing with less uncertainty, experimental results of the sludge from
Kppala would be needed. The literature values used for case E in this simulation produce
results that are less credible than for the other cases.
12.2. Economical evaluation
12.2.1. Method
A simple economical analysis of the results from section 12.1. was conducted for the
quantification of the economical benefits for the process designs. The cost for investments
and running costs for the processes would be dependent on the individual factors for the
WWTPs and were not included here. A switch from mesophilic to thermophilic digestion may
for example require new heat exchangers in some cases, and the cost depends on the reactor
configuration. The fee for the handling of the dewatered sludge and the income from selling
the methane gas are similar for all WWTP in Sweden and were therefore included in the
economical analysis.
75

The fee for handling of the sludge is 300 SEK per m
3
, and considering the amounts of sludge
produced each year this is a significant cost for the WWTPs. After dewatering, the TS content
is stable at around 25 % and the total amount sludge can therefore be calculated from the mass
of TS out from digester R200. In this case the VS reduction is assumed to be equal to the
COD reduction shown in (Table 30), and the outgoing VS is simply calculated from the total
influent mass of VS. Total amount of ashes out from digester R200 during the year is
calculated from the data of TS and VS in incoming flows (equation 12.1), and summarized
with outgoing VS for each case. The total outgoing TS is transformed to a volume assuming
that the densities of the dewatered sludge and water are 1 ton/m
3
. The WWTP sells the
methane gas for 5 SEK /m
3
in these calculations.

Iotol osbcs = IS
n
-IS
n
= 1.61 1u
7
-1.2S 1u
7

kg
yr
= S.S8 1u
6

kg
yr
(12.1)
For cases B-E, it is possible to earn and save money if the yearly costs for investments and
continuous expenses compared to case A are lower than the total change shown in the outmost
right column (Table 31). For case F, it is only possible if the expenses are lower than the total
change compared to case A. It is less expensive to run a single digester with the double
volume, but it is also associated with poorer mixing and heating due to the upscaling. For
most WWTPs, it would probably be economically favorable to choose serial digestion. If the
WWTP is equipped with several digesters run in parallel, it is both simple and inexpensive to
convert the process. The economical benefit would be around 3 million SEK per year for a
WWTP similar to Kppala (compare case C and case F).
At Kppala WWTP, the problems with filamentous bacteria means that case A is chosen
instead of case C. With improved control of the activated sludge process, Kppala could have
applied serial digestion and saved 1.5 million SEK 2007 from reduced fees and increased
income from biogas. The cost for changing from case A to C is practically zero.
The most economically favorable option is thermophilic digestion in series, which would have
spared Kppala 4 million SEK 2007, enzymatic treatment is next with 3.4 million SEK. A
comparison between these measures would require a detailed analysis of the process at
Kppala and of the costs for investments and maintenance. As mentioned before, the effect of
enzyme addition on the Kppala sludge would need to be studied closer, the degradability and
solubilization could be both lower and higher than the literature values. A lower enzyme dose
would mean less effect, but the sludge at Kppala could be more or less suitable for enzymatic
treatment than the sludge studied in the literature. It is however likely that the investment
costs for enzymatic treatment are lower than for conversion to thermophilic digestion; the
equipment for enzyme dosing is about 50 000 - 100 000 SEK depending on the reactor
configuration (Recktenwald, 2008). Improved heat exchangers, revetment of the reactor and
new pumps which could have been required for thermophilic digestion are expensive.
Furthermore, thermophilic digestion has the drawback of being less stable than mesophilic
digestion, which means that the risk for expensive process failures is higher.

76

Table 31 Change in cost for sludge handling and income from methane production for the cases
Case Outgoing
mass of TS

[10
6
kg /yr]
Outgoing
volume of
sludge

[10
3
m
3
/yr]
Fee for
sludge
handling
compared to
case A

[10
6
SEK/yr]
Income from
methane
production
compared to
case A

[10
6
SEK /yr]
Total change
compared to
case A

[10
6
SEK /yr]
A Original setup,
9.09 36 0 0 0
B, 55 C
8.22 33 1.1 1.8 2.9
8.72 35 0.45 1.0 1.5
7.97 32 1.4 2.7 4.0
8.09 32 1.2 2.2 3.4
F, one digester
9.47 38 -0.45 -1.1 -1.5

The increase in income from selling the methane gas is higher than the savings from the
sludge handling fee for all cases. The price for the gas is however variable and less
predictable compared to the sludge handling fee which could make it hazardous to introduce
expensive equipment with high continuous costs. Political decisions and public awareness of
climate change are important to make the methane gas competitive compared to other
alternatives and to enhance investments to increase biogas production.
The economical analysis showed that there are many options for process design that could be
economically beneficial if the investment and maintenance costs do not exceed the profit from
decreased sludge handling fees and increased income. The model can be a useful tool to
evaluate different process designs and choose interesting options for further investigations. It
is of course still essential to perform pilot plant experiments, but the number of experiments
could be reduced if the least appealing alternatives are sorted out after simulations. The
process variables VFA and alkalinity give indications on the stability of the processes and a
comparison between those can be useful when assessing the stability of the process.
Unwanted side-effects of a process configuration can thus be found by simulations and not by
expensive pilot plant experiments.

Important variables, like the formation of mercaptans which causes odor problem and the
dewaterability of the remaining sludge, are not included in the model. These neglected
factors, and the problem with filamentous bacteria described above, could be affected by the
alterations of the process. This needs to be taken into account when evaluating the simulation
results.
77

13. Conclusions
The first part of this dissertation was aimed at giving the reader a background of the process
of anaerobic digestion, and to summarize the prior research in the field of modeling anaerobic
digestion. It was stated that models have not been widely used in this field, and due to the
complex microbiologic and physiochemical processes, they are difficult to parameterize and
impossible to identify mathematically.
For household waste experiments, it was possible to get accurate predictions of steady state
gas production, but the correlation for ammonia was less satisfying. Simulations of the sludge
experiment produced the opposite result. VFA concentrations were difficult to model in all
three cases. In the household waste experiments, the inoculums were not adapted to the
substrates. The startup phase before adaptation proved to be difficult to simulate with the
current model structure. Validation of the model to the full scale process at Kppala proved to
be successful, considering the lack of data for the feed characterization.
It was shown in the uncertainty analysis that measurement errors could not be used to explain
the discrepancy between the simulations and data. The output uncertainties when using a
general characterization of household waste were significantly higher than when conducting
measurements for gas production and ammonia, while VFA and alkalinity were less affected
by input data.
The global, variance based method chosen for the sensitivity analysis was useful for the
inflow variables, but as the distributions of model parameters were unknown, the analysis
needed to be supplemented with scatter plots to study the correlations between parameters and
output. The results from the sensitivity analysis indicated that to determine the gas production,
information of degradability and hydrolysis rate of the sludge were needed, in addition to the
actual flows and sludge concentrations. The batch experiments used for determination of the
hydrolysis rate constant proved to give reasonable values. For better precision for VFA
predictions, the parameters connected to propionate degradation and acetoclastic
methanogenesis were essential.
A revision of the model structure is needed for a successful calibration of ammonium. In the
model by Siegrist, particulate organic matter is lumped. To be able to calibrate the
experiments on household waste digestion, a slower degradation of particulate protein would
be required. To calibrate the VFA concentrations, the half saturation constants were changed
substantially. This implies different mass transfer conditions than suggested in the model
implementation by Siegrist et al.
The model proved to be a promising tool for testing different process designs for a full scale
process. The productivity could be improved by running the reactors in series, at thermophilic
temperature and by implementing enzymatic pretreatment.

78

79

14. Suggestions for further research
It was concluded that the degradability and hydrolysis rate are more important to measure
than the individual constituents of a substrate, when it comes to prediction of biogas
production rate. This could be due to the relative stability of the simulated systems, without
significant problems with inhibited processes. It would be interesting to validate the model at
more critical states, to evaluate the ability of the model to simulate reactor instability.
The Siegrist model was primarily developed for modeling mixed sludge digestion. In this
dissertation the model was applied on household waste digestion, but not without
complications. A changed hydrolysis model would be required for this task, as in the more
complex model ADM1. Further research is however needed to determine the hydrolysis
kinetics and other process parameters for different wastes, as there is scarce information
available in the literature.
It was seen in the validation and SA of the household wastes that the precision for VFA
correlation was poor and that a calibration of parameters connected to propionate degradation
and acetoclastic methanogenesis was necessary to improve the model fit. It would be
beneficial to increase the knowledge about the uptake rates for VFA; is it dependent of the
fraction of inert material or particle size in the reactor or is it a related to the microbial
species? If the relations between the parameters
max
, K
S
, and K
I
for propionate degradation
and acetoclastic methanogenesis could be expressed as functions of the substrate
characteristics, this would allow application of the model to various substrates without
recalibration of these parameters. More research about the theoretical background for these
parameters could therefore be valuable.
It was shown in 11.2.4. that an upstart phase cannot be modeled well with the Siegrist model.
Siegrist, et al., (2002) also stressed that a simulation of a transition between mesophilic and
thermophilic digestion could not be performed with accuracy. An inclusion of a state variable
describing the physiological state of the biomass may solve this problem. More research on
this subject would be interesting, although it must be admitted that such a variable would be
difficult and expensive to verify with measurements.
More studies are required to specify the distributions of the model parameters. This would
enable a more useful SA and it would be possible to get a better ground for allocation of
resources to determine model parameters, and to deepen the understanding of the processes.
In some cases, it could be used for simplification of the model if several parameters are
proven to be irrelevant.

80

81

References
Batstone, D. J. (2006). Mathematical modelling of anaerobic reactors treating domestic. Reviews in
Environmental Science and Bio/Technology , 5, 5771.
Batstone, D. J. (2006). Mathematical modelling of anaerobic reactors treating domestic wastewater:
rational criteria for model use. Reviews in Environmental Science and Bio/Technology 5:5771 , 5:57
71.
Batstone, D. J., Keller, J., Angelidaki, I., Kalyuzhnyi, S. V., Pavlostathis, S. G., Rozzi, A., et al.
(2002). The IWA Anaerobic Digestion Model No 1 (ADM1). 45, 65-73.
Batstone, D., & Keller, J. (2003). Industrial applications of the IWA anaerobic digestion model No. 1
(ADM1). Water science and technology , 47, 199-206.
Batstone, J., Keller, J., & Steyer, J. P. (2006). A review of ADM1 extensions, applications, and
analysis: 2002-2005. Water Science and Technology , 54, 1-10.
Bernard, O., Polit, M., Hadj-Sadok, Z., Pengov, M., Dochain, D., & Estaben, M. (2001). Advanced
monitoring and control of anaerobic wastewater treatment plants: software sensors and controllers for
an anaerobic digester. Water Science and Technology , 43(7), 175-182.
Blumensaat F, K. J. (2005). Modelling of two-stage anaerobic digestion using the IWA Anaerobic
Digestion Model No. 1 (ADM1). Water Research , 39, 171183.
Buswell, E. G., & Neave, S. L. (1930). Laboratory studies of sludge digestion. Illinois Division of
State Water Survey , Bulletin no. 30.
Chen, Y., Jiang, S., Yuan, H., & Guowei, G. (2007). Hydrolysis and acidification of waste activated
sludge at different pHs. Water Research , 41, 683-689.
Davidsson, . (2008) Interview, August 6
Davidsson, . (2007). Increase of biogas production at wastewater treatment plants. Doctoral
Disseration, ISBN 978-91-7422-143-5, Lund University.
Elmitwalli, T., Sayed, S., Groendijk, L., van Lier, J., Zeeman, G., & Lettinga, G. (2003). Decentralised
treatment of concentrated sewage at low temperature in a two-step anaerobic system: two upflow-
hybrid septic tanks. Water Science and Technology , 48(6), 219-226.
Eskicioglu C, K. J. (2006). Characerization of soluble organic matter of wast activated sludge before
and after thermal pretreatment . 40 (3725-3736).
Gujer, W., & Zehnder, A. J. (1983). Conversion processes in anaerobic digestion. Water Science and
Technology , 15 (8/9); 127-167.
Hansen, T. L., la Cour Jansen, J., Spliid, H., Davidsson, ., & Christensen, T. H. (2007). Composition
of source-sorted municipal organic waste collected in Danish cities. Waste Management , 27; 510
518.
Hansen, T. L., Schmidt, J. E., Angelidaki, I., Marca, E., la Cour Jansen, J., Mosback, H., et al. (2003).
Method for determination of methane potentials of solid organic waste. 24:4; 393-400.
Hjlund Christensen, T., la Cour Jansen, J., & Jrgensen, O. (2003). Datarapport om sammenstning
og biogaspotentiale i organisk dagrenovation. Miljstyrelsen (Danish Environmental Board).
Hjlund Christensen, T., la Cour Jansen, J., Kjems Toudal, J., & Gruvberger, C. (2003).
Basisdokumentation for biogaspotentialet i organisk dagrenovation. Miljprojekt Nr. 815,
Miljstyrelsen (Danish Environmental Board).
Jeong, H.-S., Suh, C.-W., Lim, J.-L., Lee, S.-H., & Shin, H.-S. (2005). Analysis and application of
ADM1 for anaerobic methane production. Bioprocess Biosyst Eng , 27; 8189.
Jeppsson, U. (1996). Modelling aspects of waste water treatment processes. Doctoral Disseration,
ISBN 91-88934-00-4, Lund University.
82

Kops, S. G., & Vanrolleghem, P. A. (1996). An evaluation of Methodologies for uncertainty analysis
in biological waste water treatment. Unknown publisher.
la Cour Jansen, J., Gruvberger, C., Hanner, N., Aspegren, H., & Svrd, . (2004). Digestion of sludge
and organic waste in the sustainability concept for Malm, Sweden. Water Science and Technology ,
49:10 163-169.
la Cour Jansen, J., Spliid, H., Lund Hansen, T., & Svrd, . (2004). Assessment of sampling and
chemical analysis of source-separated organic household waste. Waste Management , 24; 541-549.
Minasny, B. (2004, 1 11). (reference from internet 2009-07-16 9.00 page creator Budiman Minasny,
University of sidney). (The Mathworks) Retrieved July 16, 2008, from
www.mathworks.com/matlabcentral:
http://www.mathworks.com/matlabcentral/fileexchange/loadFile.do?objectId=4352&objectType=file
Miron, Y., Zeeman, G., van Lier, J. B., & Lettinga, G. (2000). The role of sludge retention time in the
hydrolysis and acidification of lipids, carbohydrates and protein during digestion of primary sludge in
CSTR systems. Water Reserch , 34:5 1705-1713.
Muller, T. G., Noykova, N., Gyllenberg, M., & Timmer, J. (2002). Parameter identification in
dynamical models of anaerobic waste water treatment. Mathematical biosciences , 177; 147-160.
Nielsen, J., Villadsen, J., & Lidn, G. (2004). Bioreaction Engineering Principles. Springer.
Nilsson et al, I. (2007). Environmental biotechnology. Department of biotechnology, Lund University.
P54, V. (1984). Enkla analysmetoder fr driftkontroll vid avloppsreningsverk. Svenska Vatten- och
avloppsverksfreningen.
Recktenwald, M. (2008, November 3). Ingeneer, Kemira Recycling Competence Center in
Helsingborg. Lund.
Reith, J. H., Wijffels, R. H., & Barten, H. (2003). Status and perspectives of biological methane and
hydrogen production. The Hague: Smiet offset.
Rosen C, V. D. (2006). Implementing ADM1 for plant-wide benchmark simulations in
Matlab/Simulink. Water Science & Technology , 54, 1119.
Siegrist, H., Vogt, D., Garcia-Heras, J. L., & Gujer, W. (2002). Mathematical model for meso- and
thermophilic anaerobic sewage sludge digestion. Environmental Science and Technology , 36, 1113-
1123.
SIS. (1994). Bestmning av karbonatalkalinitet SS-EN ISO 9963-2. Standardiseringskommisionen i
Sverige.
SIS. (1981). Svensk Standard SS 02 81 13. SIS Standardiseringskommisionen i Sverige.
Song, Y.-C., Kwon, S.-J., & Woo, J.-H. (2004). Mesophilic and thermophilic temperature co-phase
anaerobicdigestion compared with single-stage mesophilic- and thermophilic digestion of sewage
sludge. Water Research , 38, 16531662.
VAV. (1981). Rtning av kommunalt slam, P42. Vatten- och avloppsverksfreningen (VAV) AB.
Vavilin, V. A., Fernandez, B., Palatsi, J., & Flotats, X. (2008). Hydrolysis kinetics in anaerobic
degradation of particulate organic material: An overwiev. Waste Management , 28; 939-951.
Wawrzynczyk, J. (2007). Enzymatic treatment of wastewater sludge. Doctoral Disseration, ISBN 978-
91-628-7246-5, Lund University.
I

Appendix I

Matrix of stoichiometric coefficients T (with default values for
j,i
):

State variable (see table 2)

S
H
S
H2
S
CH4
S
CO2
S
HCO3
S
NH4
S
NH3
S
ac
S
hac
S
pro
S
hpro
S
aa
S
su
S
fa
S
in
X
s
X
aa
X
su
X
fa
X
pro
X
ac
X
H2
X
in
P
r
o
c
e
s
s

n
u
m
b
e
r

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
1 0 0 0 0.0004 -0.0005 0 0 0 0 0 0 0.3 0.2 0.45 0.05 -1 0 0 0 0 0 0 0
2 0 0.96 0 0.043 -0.022 0.587 0 3.29 0 1.42 0 -6.67 0 0 0 0 1 0 0 0 0 0 0
3 0 0.96 0 0.91 -0.07 -0.08 0 3.29 0 1.42 0 0 -6.67 0 0 0 0 1 0 0 0 0 0
4 0 6.7 0 0.199 -0.202 -0.08 0 14.3 0 0 0 0 0 -22 0 0 0 0 1 0 0 0 0
5 0 8.2 0 0.162 0.004 -0.08 0 10.8 0 -20 0 0 0 0 0 0 0 0 0 1 0 0 0
6 0 0 0 -0.006 0.618 -0.08 0 -40 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
7 0 -22 39 -0.353 -0.006 -0.08 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
8 0 0 21 0 0.003 0.045 0 0 0 0 0 0 0 0 0 0.8 -1 0 0 0 0 0 0.2
9 0 0 0 0 0.003 0.045 0 0 0 0 0 0 0 0 0 0.8 0 -1 0 0 0 0 0.2
10 0 0 0 0 0.003 0.045 0 0 0 0 0 0 0 0 0 0.8 0 0 -1 0 0 0 0.2
11 0 0 0 0 0.003 0.045 0 0 0 0 0 0 0 0 0 0.8 0 0 0 -1 0 0 0.2
12 0 0 0 0 0.003 0.045 0 0 0 0 0 0 0 0 0 0.8 0 0 0 0 -1 0 0.2
13 0 0 0 0 0.003 0.045 0 0 0 0 0 0 0 0 0 0.8 0 0 0 0 0 -1 0.2
14 -1 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
15 0 0 0 -1 1 14 -14 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
16 0 0 0 -1 1 0 0 -64 64 0 0 0 0 0 0 0 0 0 0 0 0 0 0
17 0 0 0 -1 1 0 0 0 0 -112 112 0 0 0 0 0 0 0 0 0 0 0 0

Conservation matrix with respect to mass conservation of theoretical COD (1), conservation of N/COD ratio or N/N ratio (2), charge per g of
COD, N or mol (3) and mol of C/g of COD or mol:
State variable
S
H
S
H2
S
CH4
S
CO2
S
HCO3
S
NH4
S
NH3
S
ac
S
hac
S
pro
S
hpro
S
aa
S
su
S
fa
S
in
X
s
X
aa
X
su
X
fa
X
pro
X
ac
X
H2
X
in
c
x
z
x
c
z

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
1 0 1 1 0 0 0 0 1 1 1 1 1 1 1 1 1 0 0 1 0 0 0 1
2 0 0 0 0 0 1 1 0 0 0 0 0.1 0 0 0.02 0.031 1 0 0.08 0 0 0 0.05
3 1 0 0 0 -1 -1/14 0 -1/64 0 -1/112 0 0 0 -0.0012 0 0 0 0 0 0 0 0 0
4 0 0 0.0156 1 1 0 0 0.0313 0 0.0268 0 0.029 0.031 0.0216 0.03 0.026 0.03 0 0 0 0 0 0.03

II

III

Appendix II
Calculation of the Sjlunda sludge characterization

%cal cul at i ng t he amount of CO2 i n t he car bonat e syst em
pH=7. 14;
H=10^- pH; %kmol / m3
HCO3m=4000/ 61000; %kmol / m3, t he mean val ue i n t he di gest er at st abl e
oper at i on
T=35+273; %K
kHCO3=10^( 6. 53- 2906/ T- 0. 0238*T) ; %equi l i br i umconst ant CO2 t o HCO3
kH2CO3=10^( 14. 82- 3401/ T- 0. 0327*T) ; %equi l i br i umconst ant HCO3 t o H2CO3
kH=10^( - 12. 59+2198/ T+0. 0126*T) ; %henr y' s const ant f or CO2
pCH4=0. 65; %bar , measur ed i n r eact or
pCO2=1- pCH4;

H2CO3=kH*pCO2
HCO3=kH2CO3*kH*pCO2/ %The measur ed and cal cul at ed val ues f or
HCO3 ar e al i ke
CO3=HCO3*kHCO3/ H

mol es_i n_l i qui d=1. 5*( H2CO3+HCO3+CO3) %mol es of car bon i n t he
car bonat e syst eml eavi ng t he r eact or 1. 5 i s t he out t ake vol ume
mol es_i n_gas=( 21. 8e- 3*pCO2) / ( 0. 082e- 3*( 273+18) ) %mol es l eavi ng as gas,
cal ucl at ed f r ommean f l ow of gas and gener al gas l aw
CO2t ot =mol es_i n_gas+mol es_i n_l i qui d %mol es l eavi ng as gas and
l i qui d
CH4=( 21. 8e- 3*pCH4) / ( 0. 082e- 3*( 273+18) ) %mol es l eavi ng as gas,
cal ucl at ed f r ommean f l ow of gas and gener al gas l aw

par t _i n_l i qui d=mol es_i n_l i qui d/ ( mol es_i n_l i qui d+mol es_i n_gas)

r eal CO2f r act i on=CO2t ot / ( CH4+CO2t ot ) % The f r act i on of CO2
pr oduced by t he bi omass f r omt he subst r at e
r eal CH4f r act i on=CH4/ ( CH4+CO2t ot ) % The f r act i on of CH4
pr oduced by t he bi omass f r omt he subst r at e

%Cal cul at i on of t her or et i cal pr ot ei n f r act i on f r omhansr eudi met hod
i N=0. 1; %g N/ g COD t aken f r omt he st oi chi omet r i c mat r i x
SNH4=834; %mean val ue i n r eact or f or st abl e oper at i on

%COD- r educt i on
ds. VSr ed=ds. i ndat a. ch. VS*0. 45; %45 %r educt i on
ds. CODr ed=1e6*ds. VSr ed*1. 9; % amount of COD r educed, 1. 9 gCOD/ g VS and
1e6 t o get g/ m3
XST=mean( ds. CODr ed) ; %mean val ue f or r educed COD
vaa=SNH4/ ( i N*XST) %cal cul at i on of pr ot ei n cont ent i n f eed

%val ues f r omRoedi ger et al . 1967 f r omt he book by VAV ar e used f or t he
%met hane f r act i on f r omdi f f er ent subst r at es.
pr oCH4_f r act i on=0. 68;
f at CH4_f r act i on=0. 7;
suCH4_f r act i on=0. 5;

%cal cul at i on of f at cont ent
vf a=( - vaa*pr oCH4_f r act i on+suCH4_f r act i on*vaa-
suCH4_f r act i on+( pCH4/ 1) ) / ( f at CH4_f r act i on- suCH4_f r act i on)

vsu=1- vf a- vaa

IV

V

Appendix III
Results from the sensitivity analysis
Aalborg household waste

Table III, 1 Results of SA, UA and variability on measured parameters for the Aalborg household waste simulation

rank (%)
Variability

rank (%)
Sensitivity
(Gas)

rank relative
Sensitivity
(VFAs)

rank relative
Sensitivity
(NH
4
-N)

rank relative
Sensitivity
(HCO
3
)

rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
0.05
0.23
0.17
0.18
0.07
0.18
0.11
7.4
10
160
48.7
34.3
55.3
1.33
1.02
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
12
2
14
9
3
10
8
13
1
11
4
4
13
6
6
2.5
S
20
U
0.97
J
3.5
J
18
J
3.4
J
4.7
J
2
U

100
U

3
L
10
U

10
U

2
U

2
U

2
U
5
1
7
6
3
2
4
-
-
-
-
-
9
-
8
24.6
B
59.7
B
9.2
H
12.0
H
33.3
H
50
H
20.0
H
-
-
-
-
-
1.4
A

-
1.6
A

1
8
10
6
4
7
9
14
11
13
15
12
5
3
2
0.74
0.98
1.00
0.95
0.88
0.97
0.99
1.00
1.00
1.00
1.00
1.00
0.90
0.87
0.78
4
3
7
11
5
8
10
14
2
9
15
13
1
12
6
0.95
0.95
0.99
1.00
0.98
1.00
1.00
1.00
0.59
1.00
1.00
1.00
0.49
1.00
0.98
1
3
2
6
7
9
11
14
4
5
13
15
12
8
10
0.42
0.93
0.84
0.97
0.97
0.99
1.00
1.00
0.93
0.95
1.00
1.00
1.00
0.98
1.00
3
2
4
8
5
13
7
11
1
12
9
10
15
6
14
0.97
0.95
0.99
1.00
1.00
1.00
1.00
1.00
0.080
1.00
1.00
1.00
1.00
1.00
1.00
S
(SIS, 1981),
U
(Unknown),
J
L
(Dr Lange),
B
(Hjlund Christensen, et al., 200
3)
,
H

Table III, 2 Results from SA on model parameters for the Aalborg household waste simulation
(Gas)
rank relative
Sensitivity
(VFAs)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO
3
)
rank relative
max3

max4

max5

max6

max7

max8

k
d9

k
d10

k
d11

k
d12

k
d13

k
d14

K
Saa

K
Ssu

K
Sfa

K
Spro

K
Sac

K
SH2

KI
ac56

KI
H2,5

KI
H2,6

KI
H,34

KI
H,58

KI
NH3,6

KI
NH3,7

k
L
a
CO2

k
H

4
4
0.6
0.6
0.37
2
0.8
0.8
0.06
0.06
0.05
0.3
50
50
1000
20
40
1
1500
3
1
0.01
5e-4
25
17
200
0.2
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

mgCOD/m
3

gCOD/m
3

mgCOD/m
3

mgCOD/m
3

mol/m
3

mol/m
3

gN/m
3

gN/m
3

d
-1

d
-1

14
16
2
24
7
5
17
18
4
13
23
11
15
21
3
22
6
8
25
10
12
19
9
27
26
20
1
1.00
1.00
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.019
27
17
24
1
4
6
23
22
26
3
8
13
21
18
16
15
11
9
14
20
5
19
10
2
7
25
12
1.01
1.00
1.00
0.35
0.72
0.80
1.00
1.00
1.00
0.64
0.86
0.96
1.00
1.00
1.00
0.98
0.95
0.90
0.98
1.00
0.80
1.00
0.91
0.47
0.85
1.00
0.96
2
21
17
5
8
6
11
27
10
12
3
7
13
22
26
20
25
15
18
24
14
23
16
4
9
19
1
0.99
1.00
0.99
0.99
0.99
0.99
1.00
1.00
0.99
1.00
0.99
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
0.99
1.00
0.051
19
22
16
3
2
8
20
25
27
7
6
13
18
23
17
15
10
12
14
21
9
24
11
4
5
26
1
1.00
1.00
1.00
0.80
0.75
0.93
1.00
1.00
1.00
0.88
0.88
0.99
1.00
1.00
1.00
0.99
0.96
0.97
0.99
1.00
0.94
1.00
0.96
0.85
0.86
1.00
0.57

VI

Table III, 3 Results of SA of all parameters for the Aalborg household waste simulation
(Gas)
rank relative
Sensitivity
(VFAs)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO3)
rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
max3

max4

max5

max6

max7

max8

k
d9

k
d10

k
d11

k
d12

k
d13

k
d14

K
Saa

K
Ssu

K
Sfa

K
Spro

K
Sac

K
SH2

KI
ac56

KI
H2,5

KI
H2,6

KI
H,34

KI
H,58

KI
NH3,6

KI
NH3,7

k
L
a
CO2

k
H

0.05
0.23
0.17
0.18
0.07
0.18
0.11
7.4
10
160
48.7
34.3
55.3
1.33
1.02
4
4
0.6
0.6
0.37
2
0.8
0.8
0.06
0.06
0.05
0.3
50
50
1000
20
40
1
1500
3
1
0.01
5e-4
25
17
200
0.2
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

mgCOD/m
3

gCOD/m
3

mgCOD/m
3

mgCOD/m
3

mol/m
3

mol/m
3

gN/m
3

gN/m
3

d
-1

d
-1

2
22
24
20
4
15
23
36
9
31
30
34
6
5
3
39
40
21
17
14
8
32
38
29
13
16
10
37
35
26
25
18
11
42
27
12
33
7
19
41
28
1
0.88
0.99
0.99
0.98
0.91
0.97
0.99
1.00
0.97
1.00
1.00
1.00
0.96
0.93
0.90
1.00
1.00
0.98
0.97
0.97
0.97
1.00
1.00
1.00
0.97
0.97
0.97
1.00
1.00
0.99
0.99
0.98
0.97
1.01
1.00
0.97
1.00
0.97
0.98
1.00
1.00
0.52
15
40
20
29
14
18
25
31
7
27
28
30
23
42
17
41
35
19
1
4
5
39
36
38
3
8
12
37
34
21
16
13
11
26
22
6
32
10
2
9
24
33
0.97
1.01
0.99
1.00
0.95
0.98
1.00
1.00
0.82
1.00
1.00
1.00
0.99
1.02
0.98
1.01
1.00
0.98
0.33
0.69
0.78
1.00
1.00
1.00
0.66
0.83
0.91
1.00
1.00
0.99
0.97
0.92
0.88
1.00
0.99
0.80
1.00
0.88
0.43
0.85
0.99
1.00
2
4
3
8
14
26
29
39
5
6
37
40
33
31
24
9
34
25
10
13
12
17
42
19
18
7
15
23
36
41
32
28
21
27
35
20
38
22
11
16
30
1
0.86
0.98
0.97
0.99
0.99
1.00
1.00
1.00
0.98
0.99
1.00
1.00
1.00
1.00
1.00
0.99
1.00
1.00
0.99
0.99
0.99
1.00
1.00
1.00
1.00
0.99
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
0.99
1.00
0.30
9
3
18
32
17
20
26
35
1
34
30
33
24
41
22
27
38
19
5
4
8
31
40
42
6
7
14
39
37
21
29
16
13
25
23
10
36
12
11
15
28
2
0.98
0.96
0.99
1.00
0.99
1.00
1.00
1.00
0.24
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.97
0.96
0.98
1.00
1.00
1.00
0.97
0.98
0.99
1.00
1.00
1.00
1.00
0.99
0.99
1.00
1.00
0.98
1.00
0.99
0.98
0.99
1.00
0.93

VII

Table III, 4 Results of SA, UA and variability on measured parameters for the Vstra Hamnen household waste
simulation

rank (%)
Variability

rank (%)
Sensitivity
(Gas)

rank relative
Sensitivity
(VFAs)

rank relative
Sensitivity
(NH
4
-N)

rank relative
Sensitivity
(HCO
3
)

rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
0.05
0.23
0.17
0.12
0.04
0.07
0.14
7.21
10
13.5
48.7
34.3
35.4
1.33
1.03
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
12
2
14
9
3
10
8
13
1
11
4
4
13
6
6
2.5
S
20
U
0.97
J
3.5
J
18
J
3.4
J
4.7
J
2
U

100
U

3
L
10
U

10
U

2
U

2
U

2
U
5
1
7
6
3
2
4
-
-
-
-
-
9
-
8
24.6
B
59.7
B
9.2
H
12.0
H
33.3
H
50
H
20.0
H
-
-
-
-
-
1.4
A

-
1.6
A

1
5
10
7
4
9
6
12
15
13
14
11
8
3
2
0.69
0.93
1.00
0.95
0.90
0.99
0.95
1.00
1.00
1.00
1.00
1.00
0.99
0.87
0.79
3
2
6
10
7
11
9
14
1
12
15
13
4
8
5
0.88
0.88
0.97
1.00
0.98
1.00
0.99
1.00
0.29
1.00
1.00
1.00
0.93
0.99
0.96
1
2
3
6
7
9
8
13
15
10
12
14
4
5
11
0.46
0.76
0.86
0.98
0.98
1.00
0.99
1.00
1.00
1.00
1.00
1.00
0.97
0.97
1.00
3
2
4
13
7
14
8
11
1
12
10
9
6
5
15
0.97
0.95
0.99
1.00
1.00
1.00
1.00
1.00
0.074
1.00
1.00
1.00
1.00
1.00
1.00
S
(SIS, 1981),
U
(Unknown),
J
L
(Dr Lange), B (Hjlund Christensen, et al., 2003),
H

Table III, 5 Results of SA on model parameters for the Vstra Hamnen household waste simulation
(Gas)
rank relative
Sensitivity
(VFAs)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO3)
rank relative
max3

max4

max5

max6

max7

max8

k
d9

k
d10

k
d11

k
d12

k
d13

k
d14

K
Saa

K
Ssu

K
Sfa

K
Spro

K
Sac

K
SH2

KI
ac56

KI
H2,5

KI
H2,6

KI
H,34

KI
H,58

KI
NH3,6

KI
NH3,7

k
L
a
CO2

k
H

4
4
0.6
0.6
0.37
2
0.8
0.8
0.06
0.06
0.05
0.3
50
50
1000
20
40
1
1500
3
1
0.01
5e-4
25
17
200
0.2
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

mgCOD/m
3

gCOD/m
3

mgCOD/m
3

mgCOD/m
3

mol/m
3

mol/m
3

gN/m
3

gN/m
3

d
-1

d
-1

12
19
26
7
2
23
24
21
6
10
5
22
18
16
9
15
20
11
27
13
8
25
4
14
3
17
1
1.00
1.00
1.02
0.98
0.78
1.01
1.01
1.00
0.98
0.99
0.91
1.00
1.00
1.00
0.99
0.99
1.00
1.00
1.02
1.00
1.00
1.01
0.86
1.00
0.78
1.00
0.21
26
15
27
5
1
12
9
24
11
14
4
17
21
22
16
18
19
13
25
23
8
7
3
10
2
20
6
1.02
0.98
1.09
0.84
0.09
0.95
0.95
1.00
0.95
0.97
0.70
0.99
1.00
1.00
0.99
1.00
1.00
0.95
1.02
1.00
0.93
0.90
0.52
0.95
0.14
1.00
0.88
6
26
7
8
2
13
9
27
10
12
4
21
16
19
23
20
25
22
15
17
14
18
5
11
3
24
1
1.00
1.00
1.00
1.00
0.98
1.00
1.00
1.00
1.00
1.00
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
1.00
0.98
1.00
0.024
19
14
27
6
1
12
22
24
8
10
4
23
17
16
13
18
25
9
26
20
7
15
5
11
2
21
3

1.00
1.00
1.02
0.96
0.41
0.99
1.00
1.00
0.99
0.99
0.78
1.00
1.00
1.00
0.99
1.00
1.01
0.99
1.02
1.00
0.97
1.00
0.81
0.99
0.47
1.00
0.65

VIII

Table III, 6 Results of SA on all parameters for the Vstra Hamnen household waste simulation
(Gas)
rank relative
Sensitivity
(VFAs)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO3)
rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
max3

max4

max5

max6

max7

max8

k
d9

k
d10

k
d11

k
d12

k
d13

k
d14

K
Saa

K
Ssu

K
Sfa

K
Spro

K
Sac

K
SH2

KI
ac56

KI
H2,5

KI
H2,6

KI
H,34

KI
H,58

KI
NH3,6

KI
NH3,7

k
L
a
CO2

k
H

0.05
0.23
0.17
0.12
0.04
0.07
0.14
7.21
10
13.5
48.7
34.3
35.4
1.33
1.03
4
4
0.6
0.6
0.37
2
0.8
0.8
0.06
0.06
0.05
0.3
50
50
1000
20
40
1
1500
3
1
0.01
5e-4
25
17
200
0.34
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

mgCOD/m
3

gCOD/m
3

mgCOD/m
3

mgCOD/m
3

mol/m
3

mol/m
3

gN/m
3

gN/m
3

d
-1

d
-1

6
11
35
14
9
17
12
29
42
27
25
28
10
7
8
33
34
40
13
2
21
37
36
15
41
5
30
24
26
20
23
18
19
38
22
16
39
4
32
3
31
1
0.94
0.98
1.00
0.99
0.96
0.99
0.98
1.00
1.03
1.00
1.00
1.00
0.97
0.95
0.95
1.00
1.00
1.00
0.99
0.83
1.00
1.00
1.00
0.99
1.00
0.92
1.00
1.00
1.00
1.00
1.00
0.99
1.00
1.00
1.00
0.99
1.00
0.91
1.00
0.83
1.00
0.45
39
20
27
25
15
33
23
31
40
28
32
30
7
14
24
37
19
42
5
1
11
18
29
13
16
4
22
36
34
17
26
21
12
41
38
9
8
3
10
2
35
6
1.01
0.99
1.00
1.00
0.98
1.00
0.99
1.00
1.01
1.00
1.00
1.00
0.91
0.97
1.00
1.01
0.98
1.10
0.81
0.10
0.95
0.98
1.00
0.96
0.98
0.69
0.99
1.00
1.00
0.98
1.00
0.99
0.96
1.04
1.01
0.94
0.94
0.58
0.95
0.16
1.00
0.87
2
3
4
9
17
40
27
34
13
22
33
35
11
24
31
10
39
12
16
5
15
28
42
20
21
7
38
26
29
37
32
41
25
19
30
23
14
8
18
6
36
1
0.92
0.96
0.98
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.98
1.00
1.00
1.00
1.00
1.00
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
1.00
0.98
1.00
0.19
8
7
12
38
14
22
19
32
1
31
25
27
9
15
20
37
28
42
10
2
13
39
40
17
18
5
21
34
29
23
35
26
16
41
33
11
30
6
24
3
36
4
0.98
0.97
0.99
1.00
1.00
1.00
1.00
1.00
0.42
1.00
1.00
1.00
0.98
1.00
1.00
1.00
1.00
1.01
0.99
0.80
0.99
1.00
1.00
1.00
1.00
0.91
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.01
1.00
0.99
1.00
0.93
1.00
0.82
1.00
0.89

IX

Sjlunda sludge

Table III, 7 Results of SA, UA and variability on measured parameters for the sludge simulation

rank (%)
Sensitivity
(Gas)

rank relative
Sensitivity
(VFAs)

rank relative
Sensitivity
(NH
4
-N)

rank
relative
Sensitivity
(HCO
3
)

rank relative
TS
VS
Soluble TS
Protein
Fat
Sugar
Inert
pH
HCO
3

NH
4
-N
Temperature
Flow
Pressure
0.039
0.75
0.3
0.34
0.45
0.16
0.05
7.13
12
45
34
1.5
1.02
kg/kg
g/TS

gCOD/gX
S
gCOD/gX
S
gCOD/gX
S
gCOD/gX
S

mol/m
3

gN/m
3

C
l/d
bar
8
13
2
2
2
2
2
9
1
7
9
9
9
2.5
S
0.53
J

20
U

20
U

20
U

20
U

20
U

2
U

100
U

3
L

2
U

2
U

2
U

4
10
8
2
13
3
11
5
12
9
6
7
1
0.99
1.00
1.00
0.97
1.01
0.98
1.00
0.99
1.00
1.00
0.99
0.99
0.13
5
8
12
1
4
6
10
3
9
13
11
7
2
0.99
1.00
1.00
0.46
0.97
0.99
1.00
0.90
1.00
1.00
1.00
1.00
0.48
3
5
6
1
13
12
11
4
8
10
7
9
2
0.99
1.00
1.00
0.43
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.60
7
12
5
2
4
6
9
1
10
11
8
13
3
1.00
1.00
0.99
0.78
0.98
1.00
1.00
0.69
1.00
1.00
1.00
1.00
0.81
S
(SIS, 1981),
U
(Unknown),
J
L
(Dr Lange), B (Hjlund Christensen, et al., 2003),
H

Table III, 8 Results of SA on model parameters for the sludge simulation
(Gas)
rank relative
Sensitivity
(VFAs)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO
3
)
rank relative
Zero
gas-
prod.
max3

max4

max5

max6

max7

max8

k
d9

k
d10

k
d11

k
d12

k
d13

k
d14

K
Saa

K
Ssu

K
Sfa

K
Spro

K
Sac

K
SH2

KI
ac56

KI
H2,5

KI
H2,6

KI
H,34

KI
H,58

KI
NH3,6

KI
NH3,7

k
L
a
CO2

k
H

4
4
0.6
0.6
0.37
2
0.8
0.8
0.06
0.06
0.05
0.3
50
50
1000
20
40
1
1500
3
1
0.01
5e-4
25
17
200
0.2
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

mgCOD/m
3

gCOD/m
3

mgCOD/m
3

mgCOD/m
3

mol/m
3

mol/m
3

gN/m
3

gN/m
3

d
-1

d
-1

25
23
5
17
1
7
26
24
14
10
4
9
21
19
16
18
13
12
8
15
11
22
3
27
2
20
6
1.00
1.00
0.91
1.00
0.19
0.96
1.00
1.00
0.99
0.98
0.68
0.97
1.00
1.00
1.00
1.00
0.99
0.99
0.97
0.99
0.98
1.00
0.63
1.00
0.53
1.00
0.94
24
12
27
5
1
6
13
25
15
11
4
9
23
20
17
18
14
21
10
22
8
16
3
7
2
19
26
1.01
0.96
1.03
0.85
0.25
0.91
0.98
1.02
0.99
0.96
0.79
0.95
1.00
1.00
1.00
1.00
0.99
1.00
0.96
1.00
0.95
0.99
0.74
0.94
0.54
1.00
1.03
7
27
6
25
1
13
26
8
14
12
5
18
23
16
17
21
10
19
9
11
24
22
4
15
3
20
2
0.95
1.09
0.95
1.01
0.38
1.00
1.04
0.97
1.00
1.00
0.65
1.00
1.00
1.00
1.00
1.00
0.99
1.00
0.99
0.99
1.01
1.00
0.62
1.00
0.60
1.00
0.60
22
18
26
5
1
6
25
16
21
10
4
8
20
23
19
24
11
13
12
17
7
15
3
9
2
14
27
1.00
1.00
1.01
0.96
0.24
0.97
1.01
1.00
1.00
0.99
0.80
0.99
1.00
1.00
1.00
1.00
0.99
1.00
0.99
1.00
0.98
1.00
0.80
0.99
0.51
1.00
1.03
20
27
21
23
0
22
23
21
23
22
12
22
21
22
22
22
21
21
21
23
22
21
11
22
9
22
21
(23)

X

Table III, 9 Results of SA of all parameters for the sludge simulation
(Gas)
rank relative
Sensitivity
(VFAs)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO3)
rank relative
TS
VS
Soluble TS
Protein
Fat
Sugar
Inert
pH
HCO3
NH4-N
Temperature
Flow
Pressure
max3

max4

max5

max6

max7

max8

k
d9

k
d10

k
d11

k
d12

k
d13

k
d14

K
Saa

K
Ssu

K
Sfa

K
Spro

K
Sac

K
SH2

KI
ac56

KI
H2,5

KI
H2,6

KI
H,34

KI
H,58

KI
NH3,6

KI
NH3,7

k
L
a
CO2

k
H

0.039
0.75
0.3
0.34
0.45
0.16
0.05
7.13
12
45
34
1.5
1.02
4
4
0.6
0.6
0.37
2
0.8
0.8
0.06
0.06
0.05
0.3
50
50
1000
20
40
1
1500
3
1
0.01
5e-4
25
17
200
0.34
kg/kg
g/TS

gCOD/gXS
gCOD/gXS
gCOD/gXS
gCOD/gXS

mol/m3
gN/m3
C
l/d
bar
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

mgCOD/m
3

gCOD/m
3

mgCOD/m
3

mgCOD/m
3

mol/m
3

mol/m
3

gN/m
3

gN/m
3

d
-1

d
-1

19
33
23
3
8
11
35
6
31
24
20
21
1
28
30
7
16
2
10
39
27
12
22
9
13
34
32
36
38
15
14
17
25
26
37
4
40
5
29
18
0.99
1.00
0.99
0.88
0.94
0.98
1.00
0.92
1.00
1.00
0.99
0.99
0.48
1.00
1.00
0.93
0.98
0.79
0.96
1.00
1.00
0.98
0.99
0.96
0.98
1.00
1.00
1.00
1.00
0.98
0.98
0.99
1.00
1.00
1.00
0.90
1.01
0.90
1.00
0.99
24
34
14
2
9
13
27
6
28
26
22
23
1
33
17
39
10
3
15
32
37
12
25
7
31
18
30
36
29
38
19
21
20
16
11
5
8
4
35
40
1.00
1.00
0.99
0.67
0.98
0.99
1.00
0.90
1.00
1.00
1.00
1.00
0.66
1.00
0.99
1.01
0.98
0.72
0.99
1.00
1.00
0.98
1.00
0.96
1.00
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
0.98
0.89
0.97
0.82
1.00
1.01
12
33
34
1
14
29
32
6
35
23
24
37
2
17
40
11
10
3
8
39
18
26
20
7
13
38
30
31
36
15
22
16
25
21
19
5
27
4
28
9
0.99
1.00
1.00
0.46
1.00
1.00
1.00
0.98
1.00
1.00
1.00
1.00
0.60
1.00
1.00
0.99
0.99
0.96
0.99
1.00
1.00
1.00
1.00
0.99
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.98
1.00
0.98
1.00
0.99
22
36
16
4
6
14
33
1
35
19
18
24
5
29
37
10
17
2
9
34
26
20
15
8
11
38
30
28
31
12
13
21
25
23
32
7
40
3
27
39
1.00
1.00
0.99
0.86
0.91
0.99
1.00
0.77
1.00
1.00
0.99
1.00
0.88
1.00
1.00
0.98
0.99
0.77
0.98
1.00
1.00
1.00
0.99
0.96
0.99
1.00
1.00
1.00
1.00
0.99
0.99
1.00
1.00
1.00
1.00
0.92
1.02
0.85
1.00
1.00

XI

Aalborg household waste with general characterization
Table III, 10 Results of SA, UA and variability on measured parameters for the general thermophilic household waste
simulation

rank (%)
Variability

rank (%)
Sensitivity
(Gas)

rank relative
Sensitivity
(VFAs)

rank relative
Sensitivity
(NH
4
-N)

rank relative
Sensitivity
(HCO
3
)

rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
0.05
0.14
0.15
0.14
0.06
0.15
0.19
7.4
10
160
48.7
34.3
55.3
1.33
1.02
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
5
-
-
-
-
-
-
6
1
4
2
2
6
6
6
2.5
S
-
-
-
-
-
-
2
U

100
U

3
L
10
U

10
U

2
U

2
U

2
U
5
1
7
6
3
2
4
-
-
-
-
-
9
-
8
24.6
B
59.7
B
9.2
H
12.0
H
33.3
H
50
H
20.0
H
-
-
-
-
-
1.4
A

-
1.6
A

5
10
7
4
3
1
2
14
11
13
15
12
9
8
6
0.96
1.00
0.98
0.92
0.85
0.55
0.83
1.00
1.00
1.00
1.00
1.00
0.99
0.98
0.97
8
5
1
10
6
4
7
14
3
11
15
13
2
12
9
0.98
0.95
0.60
1.00
0.96
0.88
0.96
1.00
0.77
1.00
1.00
1.00
0.66
1.00
0.99
2
7
1
5
4
3
6
12
10
8
11
13
15
9
14
0.96
0.99
0.13
0.99
0.98
0.96
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
4
3
2
8
6
5
7
11
1
13
12
10
15
9
14
0.98
0.94
0.70
1.00
0.99
0.99
0.99
1.00
0.39
1.00
1.00
1.00
1.00
1.00
1.00
S
(SIS, 1981),
U
(Unknown),
J
L
(Dr Lange),
,
H

Table III, 11 Results of SA on model parameters for the general thermophilic household waste simulation
(Gas)
rank relative
Sensitivity
(VFAs)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO
3
)
rank relative
max3

max4

max5

max6

max7

max8

k
d9

k
d10

k
d11

k
d12

k
d13

k
d14

K
Saa

K
Ssu

K
Sfa

K
Spro

K
Sac

K
SH2

KI
ac56

KI
H2,5

KI
H2,6

KI
H,34

KI
H,58

KI
NH3,6

KI
NH3,7

k
L
a
CO2

k
H

4
4
0.6
0.6
0.37
2
0.8
0.8
0.06
0.06
0.05
0.3
50
50
1000
20
40
1
1500
3
1
0.01
5e-4
25
17
200
0.34
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

mgCOD/m
3

gCOD/m
3

mgCOD/m
3

mgCOD/m
3

mol/m
3

mol/m
3

gN/m
3

gN/m
3

d
-1

d
-1

19
25
2
4
8
3
21
23
15
5
13
6
20
24
12
18
14
10
27
16
9
22
7
11
26
17
1
1.00
1.00
0.98
0.98
0.98
0.98
1.00
1.00
0.99
0.98
0.99
0.98
1.00
1.00
0.99
1.00
0.99
0.99
1.03
1.00
0.98
1.00
0.98
0.99
1.00
1.00
0.04
26
22
16
1
6
4
25
24
23
2
9
10
20
21
17
13
12
8
27
18
5
19
7
3
11
15
14
1.01
1.00
1.00
0.34
0.68
0.58
1.00
1.00
1.00
0.53
0.80
0.82
1.00
1.00
1.00
0.96
0.92
0.77
1.01
1.00
0.65
1.00
0.76
0.54
0.91
0.99
0.98
2
23
17
9
12
4
6
27
7
10
3
5
8
24
26
19
20
13
21
22
14
25
15
11
16
18
1
0.98
1.00
1.00
0.99
1.00
0.99
0.99
1.00
0.99
0.99
0.99
0.99
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
1.00
1.00
0.08
16
25
14
3
2
5
19
26
22
4
7
12
18
24
17
15
13
11
27
20
8
23
9
6
10
21
1
0.99
1.00
0.98
0.78
0.72
0.85
1.00
1.00
1.00
0.84
0.86
0.93
1.00
1.00
0.99
0.99
0.93
0.91
1.01
1.00
0.88
1.00
0.90
0.85
0.91
1.00
0.58

XII

Table III, 12 Results of SA of all parameters for the general household waste simulation for thermophilic conditions
(Gas)
rank relative
Sensitivity
(VFAs)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO
3
)
rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
max3

max4

max5

max6

max7

max8

k
d9

k
d10

k
d11

k
d12

k
d13

k
d14

K
Saa

K
Ssu

K
Sfa

K
Spro

K
Sac

K
SH2

KI
ac56

KI
H2,5

KI
H2,6

KI
H,34

KI
H,58

KI
NH3,6

KI
NH3,7

k
L
a
CO2

k
H

0.05
0.14
0.15
0.14
0.06
0.15
0.19
7.4
10
160
48.7
34.3
55.3
1.33
1.02
4
4
0.6
0.6
0.37
2
0.8
0.8
0.06
0.06
0.05
0.3
50
50
1000
20
40
1
1500
3
1
0.01
5e-4
25
17
200
0.34
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

mgCOD/m
3

gCOD/m
3

mgCOD/m
3

mgCOD/m
3

mol/m
3

mol/m
3

gN/m
3

gN/m
3

d
-1

d
-1

6
20
9
5
3
1
2
34
42
33
30
31
14
8
7
38
41
22
11
25
10
27
32
26
13
18
12
36
29
24
28
40
19
21
35
16
39
17
15
23
37
4
0.97
1.00
0.98
0.93
0.86
0.58
0.84
1.00
1.00
1.00
1.00
1.00
0.99
0.98
0.97
1.00
1.00
1.00
0.98
1.00
0.98
1.00
1.00
1.00
0.98
0.99
0.98
1.00
1.00
1.00
1.00
1.00
0.99
1.00
1.00
0.99
1.00
0.99
0.99
1.00
1.00
0.92
17
20
11
22
27
9
13
34
16
29
28
33
23
41
21
26
18
39
1
12
4
31
40
36
3
7
5
32
30
38
19
25
10
14
35
6
24
8
2
15
37
42
0.93
0.97
0.80
0.99
1.00
0.75
0.83
1.00
0.93
1.00
1.00
1.00
1.00
1.01
0.98
1.00
0.96
1.01
0.26
0.83
0.51
1.00
1.01
1.00
0.47
0.73
0.63
1.00
1.00
1.00
0.97
1.00
0.76
0.85
1.00
0.69
0.99
0.73
0.46
0.90
1.00
1.04
3
8
1
5
6
4
7
33
17
9
32
34
39
18
28
15
25
31
14
12
16
26
42
41
20
10
19
38
29
40
27
37
22
24
36
21
35
30
13
11
23
2
0.96
0.99
0.22
0.99
0.99
0.97
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.93
9
3
2
19
14
7
13
29
1
34
31
27
33
22
37
23
32
20
8
16
6
35
36
38
5
15
10
39
28
21
40
18
17
25
26
12
30
11
41
42
24
4
0.99
0.95
0.73
1.00
0.99
0.99
0.99
1.00
0.43
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
0.99
0.98
1.00
1.00
1.00
0.98
0.99
0.99
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
1.00
0.99
1.00
1.01
1.00
0.98

XIII

Vstra Hamnen household waste with general characterization

Table III, 13 Results of SA, UA and variability on measured parameters for the general thermophilic household waste
simulation

rank (%)
Variability

rank (%)
Sensitivity
(Gas)

rank relative
Sensitivity
(VFAs)

rank relative
Sensitivity
(NH
4
-N)

rank relative
Sensitivity
(HCO
3
)

rank relative
TS
Soluble TS
Protein
Fat
Sugar
Starch
Fibers
pH
HCO
3

NH
4
-N
Acetate
Propionate
Temperature
Flow
Pressure
0.05
0.14
0.15
0.14
0.06
0.15
0.11
7.2
10
13.5
48.7
34.3
35.4
1.33
1.03
kg/kg

g/TS
g/TS
g/TS
g/TS
g/TS

mol/m
3

gN/m
3

gCOD/m
3

gCOD/m
3

C
l/d
bar
5
-
-
-
-
-
-
6
1
4
2
2
6
6
6
2.5
S
-
-
-
-
-
-
2
U

100
U

3
L
10
U

10
U

2
U

2
U

2
U
5
1
7
6
3
2
4
-
-
-
-
-
9
-
8
24.6
B
59.7
B
9.2
H
12.0
H
33.3
H
50
H
20.0
H
-
-
-
-
-
1.4
A

-
1.6
A

5
9
7
4
3
1
2
14
11
13
15
12
10
8
6
0.96
0.99
0.98
0.92
0.84
0.54
0.83
1.00
1.00
1.00
1.00
1.00
1.00
0.98
0.97
8
4
1
11
6
3
7
15
2
12
14
13
5
9
10
0.96
0.90
0.38
1.00
0.94
0.81
0.94
1.00
0.70
1.00
1.00
1.00
0.91
0.98
1.00
3
5
1
7
4
2
6
12
15
10
11
13
9
8
14
0.96
0.98
0.17
0.98
0.98
0.94
0.98
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
5
3
2
8
6
4
7
12
1
13
14
11
10
9
15
0.99
0.93
0.72
1.00
0.99
0.98
0.99
1.00
0.39
1.00
1.00
1.00
1.00
1.00
1.00
S
(SIS, 1981),
U
(Unknown),
J
L
(Dr Lange),
,
H

Table III, 14 Results of SA on model parameters for the general thermophilic household waste simulation
(Gas)
rank relative
Sensitivity
(VFAs)
rank relative
Sensitivity
(NH
4
-N)
rank relative
Sensitivity
(HCO
3
)
rank relative
max3

max4

max5

max6

max7

max8

k
d9

k
d10

k
d11

k
d12

k
d13

k
d14

K
Saa

K
Ssu

K
Sfa

K
Spro

K
Sac

K
SH2

KI
ac56

KI
H2,5

KI
H2,6

KI
H,34

KI
H,58

KI
NH3,6

KI
NH3,7

k
L
a
CO2

k
H

4
4
0.6
0.6
0.37
2
0.8
0.8
0.06
0.06
0.05
0.3
50
50
1000
20
40
1
1500
3
1
0.01
5e-4
25
17
200
0.34
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
d
-1
gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

gCOD/m
3

mgCOD/m
3

gCOD/m
3

mgCOD/m
3

mgCOD/m
3

mol/m
3

mol/m
3

gN/m
3

gN/m
3

d
-1

d
-1

19
17
6
8
4
7
16
22
24
25
3
27
14
18
12
11
26
9
23
10
20
21
2
13
5
15
1
1.00
1.00
0.98
0.99
0.87
0.98
1.00
1.01
1.02
1.02
0.87
1.06
1.00
1.00
1.00
1.00
1.04
1.00
1.02
1.00
1.00
1.00
0.86
1.00
0.87
1.00
0.19
15
22
17
5
1
6
24
11
23
10
4
25
18
21
19
14
27
8
16
12
7
13
2
9
3
20
26
0.98
1.02
0.99
0.53
0.44
0.56
1.03
0.95
1.02
0.93
0.52
1.10
0.99
1.00
1.00
0.98
1.23
0.82
0.98
0.96
0.70
0.98
0.45
0.92
0.52
1.00
1.12
7
26
10
14
3
11
9
8
6
23
2
13
15
18
25
20
19
17
27
24
12
22
5
16
4
21
1
0.99
1.00
1.00
1.00
0.98
1.00
1.00
1.00
0.99
1.00
0.97
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.98
1.00
0.98
1.00
0.06
16
22
11
6
2
7
21
23
25
14
5
27
17
20
15
12
26
9
24
13
8
19
3
10
4
18
1
1.00
1.00
0.98
0.84
0.72
0.85
1.00
1.00
1.03
1.00
0.77
1.06
1.00
1.00
1.00
0.99
1.03
0.93
1.02
0.99
0.90
1.00
0.76
0.97
0.76
1.00
0.52

XIV

1

Application, uncertainty and sensitivity analysis of the anaerobic digestion
model by Siegrist et al. (2002) on household waste digestion
E.Ossiansson*andO.Lidholm**
Water and Environmental Engineering, Department of Chemical Engineering
E-mail: * w03eno@student.lth.se, ** w03orl@student.lth.se
Abstract
In this paper the anaerobic digestion model published by Siegrist et al. (2002) is applied on household waste digestion. Data
from two pilot scale experiments with measured characterizations were used, one with mesophilic and one with thermophilic
temperature. Validation simulations with focus on the gas production rate, ammonium and VFA were first performed, using
default values for model parameters. The steady state gas production could be predicted with acceptable accuracy, while the
simulated ammonium and VFA concentrations were overestimated by the model. An uncertainty analysis was conducted to
analyze the quality of the predictions with respect to measurements of input data. The measurement errors influenced the gas
production the most, but that the VFA and ammonia predictions could not be improved with more accurate measurements. To
find suitable parameters for calibration of model parameters, a Monte Carlo based sensitivity analysis was conducted,
supplemented with scatter plots. The sensitivity analysis showed that the hydrolysis constant was the most important
parameter to determine for both ammonium and gas production. This was problematic for the calibration, as the particulate
organic matter is lumped and calibrating the ammonium from the hydrolysis constant therefore led to unrealistically low gas
production. To model the protein degradation accurately, without changing the overall hydrolysis constant or the model
structure, the protein content in the substrate was calibrated instead. The Siegrist model was hence not suitable for simulation
of characterized household waste, unless a hydrolysis model with slower protein hydrolysis was introduced or the
stoichiometry for the hydrolysis was calibrated. The VFA concentrations in the reactor were dependent on model parameters
related to the aceticlastic methanogenesis and the propionate degradation. Calibration of half saturation constants for these
processes was successful, but the values were increased considerably. A possible explanation for the considerable change in
these parameter values could be that the mass transfer of substrate was more limiting than for the default Siegrist model.
Keywords anaerobic digestion, model, household waste, uncertainty analysis, sensitivity analysis, calibration, Monte Carlo,
hydrolysis constant

Introduction
Anaerobic digestion is a complex system of
biochemical and physical processes. Due to the
complexity of the process, it has traditionally been
treated as a black box system, and optimization has
been based on experience or trial and error
methods. As experiments of anaerobic digestion
processes are expensive and time consuming,
modeling can provide a useful tool for process
understanding and optimization. Models have
potentials for revealing non-linear behaviors of the
system and to quantify the performance of
alternative operational setups.
The aim of this report is to evaluate the
applicability of a model for anaerobic digestion
published by Siegrist, et al (2002), here referred to
as the Siegrist model. The model was primarily
developed for simulating the digestion of mixed
sludge, but in this report the applicability for the
model on household waste is studied. Compared to
the more commonly used model by the
International Water Association (IWA), called
Anaerobic Digestion Model no 1 (ADM1), the
Siegrist model is less complex, with a lumped
hydrolysis constant and fewer Volatile Fatty Acids
(VFA) included. The two models were constructed
with different approaches, the Siegrist model
parameters are based on experiments, whereas the
ADM1 uses review consensus (Batstone, 2006).

ADM1 has been validated in the literature with
varying success (Parker, 2005; Batstone, et al.,
2003; Tartakovsky, et al., 2008). The Siegrist
model is less utilized, but interesting because of the
simpler structure with fewer input variables.
The aim of the report is also to evaluate the
required quality of input parameters, and to find the
most important parameters to measure when
modeling household waste digestion. Furthermore,
the model parameters most suitable for calibration
are evaluated and results with calibrated parameters
are shown.
Materials & Methods
The model was implemented in Matlab as a system
of differential equations with 23 liquid phase state
variables and 3 state variables describing gaseous
compounds. The stiff ODE-solver ode15s was used
for the numerical integration. The hydrolysis rate
was described with first order kinetics with respect
to the degradable particulate organic matter X
S
(process 1 in Figure 1). Monod kinetics, in
combination with inhibition expressions with
respect to ammonia, pH, acetate and dissolved
hydrogen, were used to simulate the microbial
processes (process 2-7 in Figure 1). The model also
included stripping of the gases methane, carbon
dioxide and hydrogen from the liquid with a
pressure control loop (Siegrist, et al., 2002).

2

Figure 1 Overview of the biochemical reactions in anaerobic digestion
with flows expressed as percent of COD (from Siegrist et al 2002
based on Gujer et al 1983)
Data from one mesophilic (35C) and one
thermophilic (55C) pilot scale experiment was
used for validation and model analysis (Davidsson,
et al. 2007). The household waste used in the
mesophilic experiment was collected in the city
district of Vstra Hamnen in Malm, Sweden, and
the thermophilic digester was fed with household
waste from Aalborg, Denmark. Characterization
data for fat, protein, fiber, sugar, starch, VFA and
ammonium content as fractions of TS was found in
Hjlund Christensen, et al. (2003). The
measurement errors of the characterization data
were found in an evaluation of the analysis methods
by la Cour Jansen, et al. (2004). Relative standard
deviation for the ammonium measurements were
provided by the analysis instrument manufacturer
(3.5 %, Dr Lange). The relative standard
deviation for VFA was assumed to be 10 %, and
2.5 % for gas production measurements.
The experiments were divided into three phases;
startup, steady state and post digestion. As the
inoculums were collected from digesters fed with
sludge, the startup phase was needed for the
microorganisms to get accustomed to the new type
of feed. Post digestion data was not available for
the mesophilic digestion experiment.
Validation simulations
Before the pilot scale experiments could be
simulated, some processing of data was needed,
including unit conversions and calculation of
stoichiometry and the rate constant for the
hydrolysis process. The elemental compositions for
protein, fat etc. were found in Davidsson, et al.
(2007) and converted to the unit gCOD/gTS with
the Buswell formula (Buswell, et al., 1930). The
default stoichiometry of mixed sludge hydrolysis in
Figure 1 was hence updated to apply for the
household wastes.
Data from batch experiments on the waste from
Vstra Hamnen was used to determine the first
order hydrolysis constant for degradation of
particulate degradable organic matter, X
S
. It was
assumed that the hydrolysis step was rate limiting
and that there were no inhibition effects on the
microbial reactions. The hydrolysis constant, k
H
,
was determined to 0.20 d
-1
(Figure 2).

Figure 2 Calibration of k
H
from batch experiment on household waste
from the mesophilic digestion
Except for the hydrolysis constant and the
composition of the waste, default values for model
parameters were used for the validation
simulations. The results show that the steady state
gas production rate could be predicted fairly well
for both household wastes, while the correlation
during the start-up phase was poorer (Figure a and
4a). The modeled gas production rate during the
startup phase was higher than measured, indicating
a slower degradation before adaptation to the
substrate. The fit to data for steady state ammonium
concentrations were unsatisfactory for both
simulations, especially for the mesophilic case
(Figure 3b and 4b). The good fit to data for the
startup of the thermophilic experiment was mainly
due to a washout effect. The VFA concentrations
were underestimated in the model predictions,
especially for the thermophilic household waste
digestion (Figure 3c and 4c).
Uncertainty analysis
The characterizations were based on measurements
with inherent measurement errors. A Monte Carlo
method with Latin Hypercube Sampling (LHS) was
used to study the effects of these measurement
errors on prediction uncertainty. The distributions
of the measurement errors were assumed to be
described by Gaussian functions, and the
distributions of outputs from 5000 simulations were
studied.
In Figure 5 the uncertainty of simulated outputs
presented as 95 % confidence intervals can be
compared with measurement uncertainties of
validation data (vertical intervals) for the
thermophilic household waste simulation. When the
confidence intervals overlap, the lack of fit could be
explained by measurement error. Figure 5a shows
that measurement errors can explain some of the
discrepancy between simulations and data for the
0 10 20 30 40 50
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
4
Time, [Days]
G
a
s

p
r
o
d
u
c
t
i
o
n
,

[
N
m
l
/
D
a
y
]

Tr iplicate 1
Tr iplicate 2
Tr iplicate 3
kh=0.2

3

Figure 3a Modeled and measured biogas production Figure 3b Modeled and measured ammonia Figure 3c Modeled and measured VFA
for the thermophilic experiment for the thermophilic experiment for the thermophilic experiment

Figure 4a Modeled and measured biogas production Figure 4b Modeled and measured ammonia Figure 4c Modeled and measured VFA
for the mesophilic experiment for the mesophilic experiment for the mesophilic experiment
gas production rate, but that further calibration was
needed for ammonia and VFA (Figure 5b and 5c).
The confidence intervals are broader for the gas
production than the ammonia and VFA, indicating
that the gas production is affected more by the
quality of input data than the other two outputs.
Sensitivity analysis
Sensitivity analysis was conducted with two main
purposes; to determine the importance of measuring
specific input parameters and to find model
parameters suited for calibration. Sensitivity
analyses of ADM1 found in the literature used
sensitivity functions or other local analysis methods
(Tartakovsky et al. 2008, Jeong et al. 2005, etc.). In
this report, a global sensitivity analysis method was
employed instead, to include the non-linear
behavior of the system.The sensitivity analysis was
carried out with the same methodology as in the
uncertainty analysis, e.g. Monte Carlo with LHS.
This time, however, instead of varying all
parameters, one parameter at a time was kept at a
constant value. The relative decrease in output
variance when keeping a parameter constant,
compared to if all parameters varied, was used as a
measure of the sensitivity for this parameter.
This variance-based method needed to be
supplemented with scatters plots to enable visual
evaluation of the correlation between parameters
and output.
The analysis of measured input parameters was
based on the same parameter distributions as in the
uncertainty analysis. As no information of realistic
distributions for model parameters was found, the
model parameters were assumed to be uniformly
distributed between 50 percent of the default
values.

Figure 5a Confidence intervals of the gas production Figure 5b Confidence intervals of the ammonium Figure 5c Confidence intervals of the VFA
for the thermophilic experiment for the thermophilic experiment for the thermophilic experiment

0 20 40 60 80
0
0.01
0.02
0.03
0.04
0.05
Time [days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

Simulated dr y biogas pr oduction
Measur ed biogas pr oduction
0 20 40 60 80
0
500
1000
1500
2000
2500
3000
3500
Time [days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

Measur ed NH4
0 20 40 60 80
0
200
400
600
800
1000
1200
1400
1600
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

C
O
D
/
m
3
]

Simulat ed VFAs
Measur ed VFAs
0 10 20 30 40 50
0
0.01
0.02
0.03
0.04
0.05
Time [ days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

Simulat ed dr y biogas pr oduct ion
0 10 20 30 40 50 60
0
200
400
600
800
1000
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

Simulat ed NH4 modelled
Measur ed NH4
0 10 20 30 40 50 60
0
50
100
150
200
250
300
350
Time [ days]
V
F
A
s

a
s

a
c
e
t
a
t
e

e
q
u
i
v
a
l
e
n
t
s

[
g

C
O
D
/
m
3
]

Simulat ed VFAs
Measur ed VFAs
40 45 50 55 60 65 70
0
0.01
0.02
0.03
0.04
0.05
0.06
Time [days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

40 45 50 55 60 65 70
0
500
1000
1500
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

NH4 modelled
NH4 measur ed
40 45 50 55 60 65 70
0
200
400
600
800
1000
1200
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

C
O
D
/
m
3
]

VFAs modelled
VFAs measur ed

4

Table 1 Results from sensitivity analysis of measured input parameters

(Gas)

Sensitivity
(VFA)

Sensitivity
(NH
4
-N)

Thermophilic
digestion
rank relative
Mesophilic
digestion
rank relative
Thermophilic
digestion
Rank relative
Mesophilic
digestion
rank relative
Thermophilic
digestion
rank relative
Mesophilic
digestion
rank relative
TS
Soluble TS
Protein
HCO
3

0.05
0.23
0.17
10
kg/kg

g/TS
mol/m
3

1
5
10
15
0.69
0.93
1.00
1.00
3
2
6
1
0.88
0.88
0.97
0.29
4
3
7
2
0.95
0.95
0.99
0.59
3
2
6
1
0.88
0.88
0.97
0.29
1
3
2
4
0.42
0.93
0.84
0.93
1
2
3
15
0.46
0.76
0.86
1.00
The sensitivity analysis of measured input
parameters showed that the variance for prediction
of gas production rate decreased the most if the TS
measurements were made without errors (Table 1).
Reliable measurements for TS are thus important.
The most significant sources of uncertainty for the
ammonium predictions were the TS, solubilized TS
and protein content of the feed (Table 1). For VFA,
the uncertainty from distributed input parameters
was not significant, but the HCO
3
of the feed was
most influential (Table 1).
With the aid of scatter plots, the correlations
between the measured input parameters and output
could be visualized. From this analysis, the
correlation between the degradable fraction of TS
could be pointed out as more important for the gas
production rate than the total measured TS (Figure
6 and 7). The linear correlation to the gas
production is more pronounced for the degradable
TS and the output values are less dispersed. This
result is no surprise, however, since it is the
degradable fraction of TS that is used for gas
production. This result emphasizes the importance
of measuring the degradability of the substrate.

Figure 6 Gas production vs. total TS for the thermophilic experiment

Figure 7 Gas production vs. degradable TS for the thermophilic
experiment
The results from the variance-based sensitivity
analysis of model parameters showed that the
hydrolysis constant k
H
is the most important
parameter both for determining the gas production
rate and the ammonium levels in the reactor (Table
2). This indicates that the hydrolysis step is rate
limiting, and that the hydrolysis constant can be
calibrated from either the gas production rate or the
ammonia level in the reactors. Figure 8 shows the
non-linear correlation between the hydrolysis
constant and the gas production.

Figure 8 Gas production vs. k
H
for the thermophilic experiment

VFA levels in the reactor were sensitive to
parameters related to the acetotrophic
methanogenesis and the anaerobic degradation of
propionate (Table 2). The scatter plot analysis of
the correlations revealed that the high sensitivity to
these parameters was partly due to that bacteria
were washed out when these constants reached
extreme values. Figure 9 shows an example of a
frequent washout of bacteria for low maximum
growth rates for the aceticlastic methanogens.

Figure 9 Gas production vs. maximum growth rate of aceticlastic
methanogens
0.045 0.05 0.055
0
0.01
0.02
0.03
0.04
0.05
0.06
G
a
s

p
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
]
TS [ kg/kg]
0.03 0.032 0.034 0.036 0.038 0.04
0
0.01
0.02
0.03
0.04
0.05
0.06
G
a
s

p
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
]
TS [ kg/kg]
0.1 0.15 0.2 0.25 0.3
0
0.01
0.02
0.03
0.04
0.05
0.06
k
H
[ d
-1
]
G
a
s

p
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
]
0.2 0.3 0.4 0.5 0.6
0
0.005
0.01
0.015
0.02
0.025
0.03
max6
G
a
s

p
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
]

5

Table 2 Results from sensitivity analysis of model parameters

Calibration
In the sensitivity analysis, the most influential
model parameters for each output were ranked. This
result could then be used for calibration purposes,
where the most influential parameters were chosen,
and used to increase the fit to data. The calibration
was conducted by varying parameters so that the
sum of absolute errors for model predictions
compared to data during the steady state phase
could be minimized.
Calibration of the hydrolysis constant to the gas
production measurements resulted in an unchanged
value for the mesophilic experiment, while the best
fit to data for the thermophilic experiment was
slightly decreased (from 0.2 to 0.15). This indicated
a successful determination of k
H
from batch
experiment for the household waste from Vstra
Hamnen. To enable calibration during the startup
phase, the hydrolysis constant was iteratively
increased until a good fit to data was achieved. The
results from the calibration of the hydrolysis
constant to the gas production errors are presented
in Figure 10a.
The hydrolysis constant was also the most
influential parameter when it came to determining
the ammonia levels in the reactor. When calibrating
the hydrolysis constant with the ammonia
measurements, the resulting hydrolysis constants
were lower, indicating slower degradation of
protein than other particulate constituents. The fact
that the particulate organics were lumped into one
state variable, X
S,
complicated the calibration
process. As the hydrolysis constant determined both
the protein degradation and the gas production rate,
only one of the two processes could be calibrated at
the same time. The model structure hence needed to
be revised, to include a separate process of
hydrolysis of particulate protein. The model
structure was however not changed in this project,
but is suggested for further studies. The problem
was instead solved by decreasing the stoichiometric
coefficient of protein in the degradation of X
S
. The
result from this calibration is presented in Figure
10b.
The sensitivity analysis showed that the
maximum growth rate constants for the acetoclastic
methanogenesis and the propionate degradation
were most important for the VFA concentration. As
discussed in the sensitivity analysis section, this
result was partly due to the high rate of washout for
extreme values of these parameters. This indicated
that they may not be suited for calibration purposes.
When calibrating with these values, the fit to data
only improved slightly, and when changing the
value further, the errors increased rapidly. An
example when calibrating the maximum growth
rate constant for the aceticlastic methanogens is
presented in Figure 11.

Figure 10 Sum of absolute errors for VFA for the thermophilic
simulation when varying maximum growth rate constant for the
aceticlastic methanogenesis

Figure 11a Biogas production for the thermophilic Figure 11b Ammonium concentration for the Figure 11c VFA concentration for the
experiment after calibration thermophilic experiment after calibration thermophilic experiment after calibration

0 0.1 0.2 0.3 0.4 0.5 0.6
0
1000
2000
3000
4000
5000
6000
7000
max7
S
u
m

o
f

a
b
s
o
l
u
t
e

e
r
r
o
r
s
0 20 40 60 80
0
0.01
0.02
0.03
0.04
0.05
0.06
Time [ days]
P
r
o
d
u
c
t
i
o
n

r
a
t
e

[
m
3
/
d
a
y
]

Biogas pr oduction modelled
Biogas pr oduction measur ed
0 20 40 60 80
0
500
1000
1500
2000
2500
3000
3500
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

N
/
m
3
]

NH4 modelled
NH4 measur ed
0 20 40 60 80
0
500
1000
1500
2000
2500
3000
3500
Time [ days]
C
o
n
c
e
n
t
r
a
t
i
o
n

[
g

C
O
D
/
m
3
]

VFAs modelled
VFAs measur ed
(Gas)

Sensitivity
(VFA)

Sensitivity
(NH
4
-N)

Thermophilic
digestion
rank relative
Mesophilic
digestion
rank relative
Thermophilic
digestion
Rank relative
Mesophilic
digestion
rank relative
Thermophilic
digestion
rank relative
Mesophilic
digestion
rank relative
max6

max7

k
H

0.6
0.37
0.2
d
-1
d
-1
d
-1

24
7
1
1.00
1.00
0.019
7
2
1
0.98
0.78
0.21
1
4
12
0.35
0.72
0.96
5
1
6
0.84
0.09
0.88
5
8
1
0.99
0.99
0.051
2
8
1
0.98
1.00
0.024

6

Better calibration results were instead achieved
when using half saturation constants. A better fit to
data was obtained, and the process was stable for a
wide range of values (Figure 12). The minimum
error was acquired for a K
S
of 100 g COD/m
3
for
propionate degradation and 260 g COD/m
3
in the
thermophilic digester (compare with the default
values 20 and 40 g COD/m
3
respectively). For
mesophilic digestion, only K
S
for acetoclastic
methanogenesis was calibrated, to 190 g COD/m
3
.
The physical explanation for changing the half
saturation constants could be that the mass transfer
conditions was different than suggested in the
implementation by Siegrist et al. The resulting
simulation after calibration is showed in Figure 10c.

Figure 12 Sum of absolute errors for VFA for the thermophilic
simulation when varying half saturation constant for the aceticlastic
methanogenesis
Conclusions
Applying the Siegrist model on household waste
experiments gave acceptable predictions of the
steady state gas production. The simulation of the
startup phase, i.e. before the inoculum was adapted
to the substrate, was less successful. The correlation
of ammonia and VFA predictions to data was not
satisfying for simulations without calibrated values.
The uncertainty analysis showed that
measurement errors could explain some of the
discrepancy for gas production predictions to data,
but not to the same extent for ammonium and VFA.
From the sensitivity analysis it could be
concluded that the most important parameters to
determine for predictions of gas production are the
amount of degradable TS in the substrate, and the
hydrolysis rate constant. The VFA concentrations
of the reactor were mostly influenced by the
availability of VFA as substrate in propionate
degradation and acetoclastic methanogenesis. This
implied that the mass transfer in the digester was
more important for the VFA level than the input
variables.
The calibration showed that the hydrolysis rate
constant determined from batch experiments gave
the best fit to data for the gas production; this
method for determining the hydrolysis constant
could thus be used in this case.
The ammonium concentration in the reactor was
determined from the degradation of protein. As the
particulate organics are lumped in the Siegrist
model, the individual hydrolysis of particulate
protein could not be modeled. This was
problematic, since the protein degradation was
much slower in the simulated experiments. It was
therefore concluded that a hydrolysis model with
separate hydrolysis rates as in ADM1 is preferable
when modeling household waste degradation. This
applies particularly when characterization
measurements of the substrate are used; otherwise
calibrations of the stoichiometric coefficients are
needed.
References
Batstone D J Mathematical modelling of anaerobic reactors
treating domestic [Journal] // Reviews in Environmental Science
and Bio/Technology. - 2006. - pp. 5, 5771.
Batstone D.J. and Keller J. Industrial applications of the IWA
anaerobic digestion model No. 1 (ADM1) [Article] // Water
science and technology. - 2003. - Vols. 47, 199-206.
Buswell E G and Neave S L Laboratory studies of sludge
digestion [Article] // Illinois Division of State Water Survey. -
1930. - Vol. Bulletin no. 30.
Davidsson sa [et al.] Methane yield in source-sorted organic
fraction of municipal solid waste [Article] // Waste
Management. - 2007. - Vols. 27, p.406-414 .
Gujer W and Zehnder A J B Conversion processes in anaerobic
digestion [Article] // Water Science and Technology. - 1983. -
Vols. 15 (8/9); 127-167.
Hjlund Christensen T, la Cour Jansen J and Jrgensen O
Datarapport om sammenstning og biogaspotentiale i organisk
dagrenovation [Report]. - [s.l.] : Danska Miljstyrelsen, 2003.
Jeong Hyeong-Seok [et al.] Analysis and application of ADM1
for anaerobic methane production [Article] // Bioprocess
Biosyst Eng. - 2005. - Vols. 27; 8189.
la Cour Jansen J [et al.] Assessment of sampling and chemical
analysis of source-separated organic household waste
[Article] // Waste Management. - 2004. - Vols. 24; 541-549.
Parker Wayne J. Application of the ADM1 model to advanced
anaerobic digestion [Artikel] // Bioresource Technology. -
2005. - Vol. 96, p.1832-1842.
Siegrist H [et al.] Mathematical model for meso- and
thermophilic anaerobic sewage sludge digestion [Journal] //
Environmental Science and Technology. - 2002. - pp. 36, 1113-
1123.
Tartakovsky B [et al.] Anaerobic digestion model No 1- based
distributed prameter model of an anaerobic reactor: II. Model
validation [Article] // Bioresource technology. - 2008. - Vols. 99
3676-3684.

0 50 100 150 200 250
0
1000
2000
3000
4000
5000
6000
7000
S
u
m

o
f

a
b
s
o
l
u
t
e

e
r
r
o
r
s
K
Sac

E539 PDF

Enviado por

Dados do documento

Descrição original:

Título 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

E539 PDF

Enviado por

Direitos autorais:

Formatos disponíveis

Water and Environmental Engineering

Department of Chemical Engineering

4.2. Liquid phase differentials

Table 11 Input parameters for primary sludge and WAS

Você também pode gostar