Você está na página 1de 12

DOI: 10.

1007/s00267-003-0027-0

Decision Tree Algorithm for Detection of Spatial


Processes in Landscape Transformation
JAN BOGAERT* cesses responsible for pattern change: aggregation, attrition,
Ecole Interfacultaire de Bioingenieurs creation, deformation, dissection, enlargement, fragmentation,
Universitie Libre de Bruxelles perforation, shift, and shrinkage. A novelty in this contribution
50 Av. F.D. Roosevelt, C.P. 165/05 is the inclusion of transformation processes causing expan-
B-1050 Bruxelles, Belgique sion of the land cover of interest. Consequently, we propose a
decision tree algorithm that enables detection of these pro-
REINHART CEULEMANS
cesses, based on three parameters that have to be deter-
DAVID SALVADOR-VAN EYSENRODE
mined before and after the transformation of the landscape:
Department of Biology
area, perimeter length, and number of patches of the focal
Research Group of Plant and Vegetation Ecology
University of Antwerp landscape class. As an example, the decision tree algorithm is
Universiteitsplein 1 applied to determine the transformation processes of three
B-2610 Wilrijk, Belgium divergent land cover change scenarios: deciduous woodland
degradation in Cadiz Township (Wisconsin, USA) 18311950,
ABSTRACT / The conversion of landscapes by human activi- canopy gap formation in a terra firme rain forest at the Tiputini
ties results in widespread changes in landscape spatial struc- Biodiversity Station (Amazonian Ecuador) 19971998, and
ture. Regardless of the type of land conversion, there appears forest regrowth in Petersham Township (Massachusetts, USA)
to be a limited number of common spatial configurations that 1830 1985. The examples signal the importance of the tem-
result from such land transformation processes. Some of poral resolution of the data, since long-term pattern conver-
these configurations are considered optimal or more desirable sions can be subdivided in stadia in which particular pattern
than others. Based on pattern geometry, we define ten pro- components are altered by specific transformation processes.

Human influence on the landscape has increased land conversion for human use involve ecosystem trans-
dramatically over the course of the 20th century. There- formation, e.g., from native grassland into agricultural
fore, methods are needed to properly monitor and field, from primary forest into clear-cut, and from
evaluate changes in land use and land cover (Herzog coastal shrub into housing development (Collinge and
and others 2001). The conversion of native landscapes Forman 1998). Consequently, the subject of land trans-
for human activities results in widespread changes in formation is significant to all human issues that involve
landscape spatial structure (Collinge 1998). These land. Wise forestry, economics, biodiversity conserva-
changes in land use and land cover are recognized as tion, agriculture, landscape architecture, sociology,
major components of the large-scale perturbations wildlife biology, soil science, and so forth explicitely
known as global change (Vitousek 1994, Walker and recognize and deal with dynamic land (Forman 1995a).
Steffen 1997). Due to the diffuse presence of people, An underlying assumption of many environmental
most changes that occur are produced by changes in decisions is that some patterns or combinations of land
the use of a territory. Two opposite processes can be cover are optimal or more desirable than other types.
observed in todays landscapes that have profound con- Management policies frequently seek to change the
sequences on their structure and functioning: (1) in- structure of a landscape toward particular goals, be-
tensification of agriculture and urbanization and (2) cause it is assumed that the spatial arrangement of
land abandonment (Farina 2000). Other processes of elements in a land-cover mosaic controls the ecological
processes that operate within it (Gustafson and Parker,
1992, Haines-Young and Chopping 1996, Gustafson
KEY WORDS: Decision tree algorithm; Fragmentation; Land cover
1998). Characteristic patterns of landscapes presum-
change; Landscape transformation; Spatial pattern
ably are a result of the operation of ecological pro-
Published online January 8, 2004. cesses, that is, ecological processes generate patterns,
*Author to whom correspondence should be addressed, email: and by analyzing these patterns useful inferences about
jan.bogaert@ulb.ac.be the underlying process can be made (Coulson and

Environmental Management Vol. 33, No. 1, pp. 6273 2004 Springer-Verlag New York Inc.
Spatial landscape transformation processes 63

others 1999). This proposition is known as the pattern/ configurations that result from such land transforma-
process paradigm, and it forms a central hypothesis of tion processes (Franklin and Forman 1987, Collinge
landscape ecology (Bogaert 2000), a branch of science and Forman 1998). From a geometric point of view,
developed to study ecological processes in their spatial eight processes are described in detail in literature for
context (Antrop 2001, Stine and Hunsaker 2001). The binary landscapes (composed of only two classes, such
study of the relationships between structural and func- as forest and not-forested): attrition, bisection, dissec-
tional consequences of landscape transformation offers tion, dissipation, fragmentation, incision, perforation,
insight into the more general question of how land- and shrinkage (Forman 1995a,b, Collinge 1998, Col-
scape patterns and processes are correlated (Jaeger linge and Forman 1998, Jaeger 2000). These spatial
2000). Consequently, the analysis of spatial landscape processes describe land cover (habitat, vegetation) loss
patterns is one of the central interests of (landscape) or degradation and were developed to assess the impact
ecologists and can enlarge the understanding of dy- of pattern change on species diversity, e.g., by consid-
namic ecological patterns, the role of disturbances in ering edge effects. Regretfully, a uniform relationship
ecosystems, and the characteristics of spatial and tem- or hierarchy between the processes in an ecological
poral scales of ecological events (Bannerman 1997). perspective has not been described to date. Forman
Incorporation of an understanding of the ecological (1995a,b) and Jaeger (2000) consider a kind of hierar-
consequences of particular spatial characteristics in- chical relationship between the processes, while Col-
creases the environmental benefit of landscape archi- linge (1998), and Collinge and Forman (1998) assume
tecture and planning proposals. Landscape planners that all spatial processes are equivalent.
and architects are increasingly involved in projects im- The purpose of this paper is to combine the visions
plicitely aimed at using the principles of landscape described in the seminal papers of Forman (1995a,b),
ecology to preserve, restore, and enhance biological Collinge (1998), Collinge and Forman (1998), and Jae-
diversity (Collinge 1996). Because land conversion pat- ger (2000) into a suite of spatial processes that can be
terns differ in their spatial configuration, they may used further on to identify the spatial processes in-
differ significantly in their impact on ecological pro- volved in land transformation. Consequently, the devel-
cesses as well (Collinge and Forman 1998). Land trans- opment of a typology of pattern forms the core of this
formation may severely compromise the integrity of contribution. Some novel spatial processes are defined
ecological systems through loss of native species, inva- to complete the existing list with those landscape con-
sion of exotic species, pronounced soil erosion, and versions characterized by an increase of the land cover
decreased water quality (Collinge 1996, Jorge and Gar- of interest. Some of the formerly used processes are not
cia 1997). The remnants of native vegetation left after considered because of their similarity to other ones. All
such transformation are often reduced in size and dis- processes are assigned an unequivocal definition. Next,
connected from adjacent, continuous habitat (Collinge a decision tree is proposed based on three spatial at-
1996). At the landscape scale, the frequency and spatial tributes that are easy to calculate and that are recog-
distribution of critical habitats and resources will deter- nized as key elements of landscape pattern measures
mine species distribution patterns (Debinski and others (Giles and Trani 1999): area, perimeter, and number of
2001). patches of the focal landscape class. This decision tree
The study of spatial pattern changes involves pattern will aid environmental managers, landscape ecologists,
quantification, as well as identification of the spatial and landscape architects to quickly and objectively
process involved (ONeill and others 1988). Pattern identify which process is transforming their landscape
quantification attempts have lead to the development at study. As an example, the decision tree algorithm is
and application of a plenitude of landscape indices applied to determine the transformation processes of
(e.g., ONeill and others 1988, Gustafson and Parker three divergent land-cover change scenarios: woodland
1994, Schumaker 1996, Wickham and others 1996, Li degradation in Cadiz Township (Wisconsin, USA) from
and Archer 1997, He and others 2000, Bogaert and 1831 to 1950, canopy gap formation in a terra firme rain
others 1999b, 2000a,b, 2001b,c, 2002a,b,c), and to spe- forest at the Tiputini Biodiversity Station (Amazonian
cialized software such as SPAN (Turner and Ruscher Ecuador) 19971998, and forest regrowth in Petersham
1988), SPAN (Gustafson and Parker 1991), HISA Township (Massachusetts, USA) 1830 1985.
(Gustafson and Parker 1992), FRAGSTATS (McGarigal
and Marks 1995), HAMS (Roseberry and Hao 1996),
Land Transformation Processes
and APACK (Mladenoff and DeZonia 1999).
Regardless of the type of land conversion, there The spatial transformation processes are defined for
appears to be a limited number of common spatial a landscape composed of only two classes (binary land-
64 J. Bogaert and others

scape), representing the land cover of interest gener-


ally a particular vegetation or habitat typeand its
surroundings, which correspond respectively to the
patches and the matrix as described in the patch
corridormatrix landscape model (Forman and Go-
dron 1981, 1986, Forman 1995a). The definitions of
the processes are combinations of statements by For-
man (1995a,b), Heywood and Watson (1995), Fahrig
(1997), Collinge (1998), Collinge and Forman (1998),
Bogaert (2000), Farina (2000), and Jaeger (2000). Web-
sters Third New International Dictionary (Anonymous
1971), American Heritage Dictionary (Anonymous 1982),
Longman Dictionary of the English Language (Anonymous
1991), and Websters New Universal Unabridged Dictionary
(Anonymous 1996) were used to verify and improve the
existing definitions. The following ten spatial processes
for landscape transformation are considered to reflect
pattern changes in landscapes (Figure 1):

1 aggregation: the action or process of collecting


units or parts into a whole; to bring or gather
together into a whole; to fill gaps or open space;
2 attrition: the reduction or decrease in the number
of patches; dissappearance of patches;
3 creation: the formation of new patches, which re-
sults in an increase of the total number of patches;
the act of causing to exist of patches; patch genesis;
4 deformation: the change of patch shape, without
patch size change; patch disfigurement;
5 dissection: the carving up or subdividing of an area
or patch using equal-width lines; sectioning of an
area or patch; area or patch (sub)division;
6 enlargement: the increase of patch size; patch size
expansion;
7 fragmentation: the breaking up of an area into
smaller parcels, resulting in unevenly separated Figure 1. Diagrammetric representation of the ten spatial
patches; the breaking up of extensive landscape fea- processes in land transformation. Black areas refer to the land
tures into disjunt, isolated, or semi-isolated patches; cover of interest, white areas to the landscape matrix: (a)
8 perforation: the process of making holes in an area aggregation: merging of patches; (b) attrition: dissappearance
or patch; gap formation; interruption of land cover of one of the four patches present in the original landscape;
continuity by formation of openings; (c) creation: landscape transformation by formation of a new
9 shift: patch repositioning; patch translocation; patch, increasing the number of patches from two to three;
(d) deformation: transformation characterized by shape
10 shrinkage: the decrease or reduction in size of
change of both patches from a square to rectangular shape;
patches, without attrition; progressive reduction (e) dissection: subdivision of a continuous area by equal-width
of the initial land cover patch, ideally maintaining lines into six parts; (f) enlargement: transformation by size
its original shape. increase of both patches; (g) fragmentation: landscape con-
version by breaking up the continuous land cover into five
Note that attrition, dissection, fragmentation, perfo- disjunct patches of unequal size and shape; (h) perforation:
ration, and shrinkage involve land-cover degradation. transformation by the formation of four gaps; (i) shift: trans-
Aggregation, creation, and enlargement imply the ap- location of one of both patches; (j) shrinkage: patches un-
pearance of new land-cover units. Deformation and dergo a decrease in size.
shift are neutral towards land cover area. The incorpo-
ration of the latter five processes is novel compared to
Spatial landscape transformation processes 65

Figure 2. Decision tree to identify


transformation processes that alter
the spatial pattern of landscapes.
The parameters a0, p0, and n0 refer
to the habitat area, perimeter, and
number of patches before transfor-
mation, while a1, p1, and n1 are the
reciprocal values after pattern
change.

Forman (1995a,b), Collinge (1998), Collinge and For- equality, increase, or decrease of land cover area, pe-
man (1998), and Jaeger (2000). rimeter length, and number of patches.
The number of patches is the first parameter to be
evaluated. This pattern feature allows partition of the
Decision Tree Algorithm to Identify Spatial processes in three groups. If the number of patches
Processes shows a decrease during transformation (n1 n0), two
processes can be observed. If this decrease of the num-
Description of Decision Tree Algorithm ber of patches goes together with a decrease of the total
To identify the spatial process responsible for land- area of the class of interest (a1 a0), attrition (Figure
scape pattern changes during a well-defined time pe- 1b) is the process of transformation. Note that this type
riod, a decision tree was developed (Figure 2). The aim of transformation is expected to occur in a patchy
was to provide environmental scientists with a quick landscape, which can already be the result of another
protocol to determine the transformation process(es) pattern conversion (Forman 1995a, Jaeger 2000). On
present in the study area of interest. the other hand, if the total area does not decrease (a1
Input data for the decision tree consists of the area a0), this indicates the mergence of existing patches
or size (a), the perimeter or edge length (p), and the by physical connection. This process is denoted as ag-
number of patches (n) observed for the land-cover class gregation (Figure 1a).
of interest. Calculation of a, p, and n is discussed in the An increase of the number of patches (n1 n0)
next section. The data referring to the landscape pat- requires, analogously as for attrition and aggregation,
tern before transformation are denoted as a0, p0, and the consideration of a second parameter. An increase
n0. At the end of the considered time period in which of the area (a1 a0) combined with an increase of the
the pattern transformation is analyzed, a1, p1, and n1 number of patches implies that new patches are
are recorded. The data used in the decision tree are formed, and that these patches contribute to the land
considered as key elements of landscape pattern mea- cover expansion observed. This formation of new
sures that will encompass most of the phenomena of patches is denoted as creation (Figure 1c). Creation
observed patterns in a landscape (Giles and Trani can be anthropogenic (e.g., afforestation as a conse-
1999). Since a binary landscape model is assumed, quence of landscape architectural measures) or can be
which is suitable for most applications, other key pa- caused by natural vegetation dynamics (colonization of
rameters mentioned by Giles and Trani (1999), such as bare soil, succession to a pioneer forest). When the
proportion of the dominant class or as the number of vegetated area decreases (a1 a0), this indicates that
classes present, are not considered. The decision tree large, continuous land-cover units are substituted by
algorithm leads to the identification of the spatial pro- smaller, disjunct parcels. If this breaking up coincides
cess involved by comparing the values of a0, p0, and n0 with minimal area loss associated with linear structures
with, respectively, a1, p1, and n1, and is thus based on of disruption (a1 a0), the pattern change sequence is
66 J. Bogaert and others

processes. The total perimeter can increase (p1 p0)


due to the formation of gaps inside the land cover of
interest. The gap area can be considered as belonging
to the matrix, and the perimeter segments that demar-
cate the gaps make the overall perimeter increase. This
process is known as perforation (Figure 1h). Gap for-
mation is a widespread form of forest disturbance, with
a key role in regeneration dynamics, since gaps provide
spatiotemporal opportunities for plant species to de-
velop (Salvador-Van Eysenrode and others 1998b,
2000a, 2002). Gap formation can be induced by nature
(e.g., treefall gaps after windthrow) or by human action
(e.g., logging for timber). If the total perimeter does
not increase (p1 p0), one or more of the patches are
characterized by shrinkage (Figure 1j). This involves
loss of the peripheral parts of the patches. Shrinkage
could be a consequence of the edge effect, which refers
Figure 3. Relationship between creation, dissection, and frag- to those microclimatic changes observed after the ex-
mentation based on the area before (a0) and after (a1) trans- posure of an isolated patch to the altered physical
formation. The choice between fragmentation and dissection environment of the adjacent cleared area (Collinge
is based on a predefined reference value t, which describes the
1996). When a forest is cleared, the floor in the edge is
extent to which the area can decrease (a0 a1 0) before the
exposed to direct sunlight, the ground becomes hotter
habitat is considered as fragmented. tobs is the a1/a0 ratio
observed. If tobs t, the pattern change is considered as a during the day, and without the canopy to reduce heat
consequence of fragmentation. and moisture loss, the forest floor is also much colder at
night and generally less humid. These effects will de-
crease towards the center of the vegetation (Lovejoy
denoted as dissection (Figure 1e) (Forman 1995a). and others 1986, Chen 1991, Saunders and others 1991,
Typical examples of linear dissecting structures are rail- Groom and Schumaker 1993). Associated with the abi-
roads. Fragmentation (Figure 1g) combines the in- otic changes (Lovejoy and others 1986), changes in the
creasing number of patches with considerable area loss structure and composition of the floral and faunal com-
(a1 a0). The difference between a1 a0 and a1 a0 munities are observed (Collinge 1996). Profound abi-
can be subject to debate. A predefined threshold value otic changes could therefore lead towards land cover
t, used as a reference for the ratio of a1 to a0, is change. Note that for p1 p0, a change of patch shape
therefore suggested (t 1). If tobs is calculated by is observed together with a shrinkage, because the patch
form should become more elongated to prevent perime-
a1
t obs ter decrease. This type of shrinkage could therefore be
a0 (1)
denoted as anisotropic, while the type with the concomi-
then dissection is present if tobs t. Fragmentation will tant decrease in perimeter can be denoted as isotropic.
then be observed for t obs t. The relationship between Two types of conversion are possible for landscapes
creation, dissection, and fragmentation, as determined by in which both the number of patches and the total
the values for a0, a1, tobs, and t is illustrated in Figure 3. land-cover area do not change during conversion (n1
If the number of patches is not influenced by land- n0; a1 a0). If the total perimeter length also is con-
scape conversion (n1 n0), three clusters of transfor- stant (p1 p0), the pattern can only change by reloca-
mation processes can be distinguished. An increase of tion of the patches; this is known as shift (Figure 1i).
the land-cover area (a1 a0) indicates that one or more This is, in fact, a combination of creation and attrition
of the already present patches expands: this process is occurring at a distinct location but is considered here
known as enlargement (Figure 1f). Succession or com- as a separate process. Vegetation dynamics and land-
petition processes or anthropogenic action at the patch scape design can generate this pattern change type. If
perimeter can induce this kind of pattern change. To the total perimeter changes (p1 p0), deformation
avoid creation, the newly formed land-cover area (Figure 1d) is observed. Note that deformation implies
should appear contiguous to patches present before that only the shape of the patch changes, by removal
the transformation occurred. In case of area decrease and addition of land-cover area at the periphery of the
(a1 a0), the patchy landscape can be subject to two patches. Shape change is generally quantified by an
Spatial landscape transformation processes 67

area-to-perimeter ratio (Bogaert and Impens 1998, Bo- because its information is already incorporated by the
gaert and others 2000a, 2002a), or by patch diameter separate parameters a and n. A separate measure for
measurement (Feret diameters) (Glasbey and Horgan interior (or edge) is avoided because a correct definition
1995, Cruz and Matos 1998). of edge width is not feasible due to variation in distur-
bance penetration, location of impact, and periodicity of
Discussion of Decision Tree Algorithm the disturbance (Chen 1991, Bogaert and others 1998,
The decision tree algorithm uses patch area, patch 1999a, 2000b, 2001a). Moreover, by using perimeter and
perimeter, and number of patches data. A raster data- area, interior presence is well described (Bogaert and
base, e.g., cartographic data digitized by scanning, or Impens 1998, Bogaert and others 2000a,b), since large
satellite imagery portrays features as a matrix of equal- isodiametric patches will be characterized by high interi-
area cells that are usually squares (Johnston 1998, Bo- or-to-edge ratios and small, elongated patches by the ab-
gaert and Rousseau 2001). The smallest nondivisible sence of an interior. Connectivity (sensu Forman 1995a
element in the image is then a grid cell or pixel. Spatial and Farina 1998, equivalent to the connectedness of an
primitives such as points, lines, and polygons are then area regarding a particular process, e.g., species migra-
represented by pixel configurations: points by isolated tion), or its reciprocal isolation, is not included in the
pixels, lines by strings of pixels, and polygons by pixel decision tree, because connectivity is different according
clusters. The determination of patch area, perimeter, to the landscape process considered. For example, con-
and number of patches to be distinguished is deter- nectivity will be clearly different for water fluxes and ani-
mined by the neighborship function used. If 4-connec- mal dispersal. An abundant number of landscape param-
tivity applies, four neighboring positions (nearest eters need to be considered, such as the 3D vegetation
neighbors) are assumed for every pixel not located in structure, soil data, and hydrological data (Bogaert and
the map border. If the diagonal pixels are also consid- others 2000b), when this type of landscape process are
ered as neighboring pixels (next-nearest neighbors), incorporated in the decision tree. The issue of process
the neighborship function is denoted as 8-connectivity scale and the set of processes to be assessed could then
(Stauffer 1985). Pixels that are connected belong to the consequently also be subject to debate. Therefore, we
same patch, and its area is then represented by the restricted the parameters to a, p, and n, which describe
number of pixels (a measured as the number of equal- key pattern features (Giles and Trani 1999), and are
area pixels), or this number multiplied by the actual concepts that are unambiguous and easily to determine.
area or resolution of the pixel (e.g., a measured in The decision tree is designed to identify one single
km2). The use of 8-connectivity will generally lead to process using land-cover area, perimeter, and number
less patches with a larger patch area. Patch perimeter is of patches. Some processes in the decision tree algo-
based upon the presence of nearest neighbors in a rithm are defined by fewer parameters than others.
patch and is calculated as Aggregation, attrition, creation, dissection, enlarge-
ment, and fragmentation are defined by two parame-
p(a) 4a 2 a (2)
ters (number of patches and area); for deformation,
with p(a) the perimeter for a patch with area equal to a perforation, shift, and shrinkage, all three parameters
square pixels, and ra the number of nearest neighbor- are required. Introduction of a third parameter for all
ships (1 0; 2 1) (Bribiesca 1997, Bogaert and processes could refine the process definition, but also
Impens 1998, Bogaert and others 2000a). For a patch introduces complexity. An example for this case is en-
composed of a 16 square pixels organized as a 4 4 largement. It can be expected that perimeter increase
square, 16 24 and p(16) (4 16) (2 24) 16. coincides with enlargement. However, a decrease of the
Note that this perimeter definition equals the prag- perimeter can occur if a patch shape change from
matic approach of Lawrence and Ripple (1996), but is elongated (large perimeter) to isodiametric (small pe-
clearly different from Rosenfelds (1974). rimeter) occurs at the same time. The absence of a
The algorithm presented does not incorporate other parameter can also indicate redundancy of the param-
land cover characteristics, such as average patch size, in- eter; this is the case for the transformation processes
terior (habitat), connectivity, and patch isolation (For- with n1 n0: these processes are all associated with
man 1995a). These characteristics are directly related to perimeter increase (Forman 1995a), so that incorpora-
fragmentation (Bogaert and others 2000b) and are less tion of this parameter does not provide supplementary
important for the other transformation processes consid- information. It should be noted that patch attrition can
ered. Although average patch size is often used to quantify dominate the analysis, due to the design of the decision
spatial pattern (e.g., Groom and Schumaker 1993, Bender tree algorithm. In the case of attrition of a single or a
and others 1998, Davidson 1998), it is not considered here few patches, we suggest a methodological fix: redo the
68 J. Bogaert and others

analysis with omission of the decrease in number of age. Fragmentation in Fahrig (1997) equals the dissection
patches (consider n1 n0 instead of n1 n0). In this process in the current paper. Finally, Hansen and others
way, other conversion scenarios may be found. This (1992) considernext to forest fragmentationtwo
latter case is illustrated in the example of Cadiz Town- other processes: forest conversion (transition of natural to
ship in the following section. regulated forest), and deprise agricole, which implies the
The proposed definitions and the decision tree are transition of agricultural land into seminatural forest.
partly a synthesis of other attempts to classify land Since these processes are bound to particular vegetation
transformation processes, but are significantly different or land-cover types (forest, agriculture), and since our aim
through the addition of new processes (aggregation, was to develop a methodology applicable to multiple land-
shift, creation, deformation, enlargement) and by omis- cover types, these processes are not considered here.
sion of others. In Forman (1995a,b) only five processes Moreover, the actual spatial or geometrical component of
are listed: perforation, dissection, fragmentation, these processes is not explicitely defined, as it is for the
shrinkage, and attrition. This classification is designed current nine processes.
to assess landscape fragmentation in particular. A time
sequence is assumed: a landscape is firstly transformed
Examples
by dissection and/or perforation. Dissection is consid-
ered a special case of fragmentation, and the two spatial Three examples are elaborated to exemplify the use
processes are differentiated by Forman (1995a,b), in of the decision tree algorithm for detecting landscape
part because the separating elements typically are so transformation processes. It is demonstrated how com-
different and widespread. Moreover, the ecological plex pattern sequences are the result of a series of
consequences may be similar or highly dissimilar, de- different processes. The examples underline the impor-
pending on the function of the cleared zone (Forman tance of data availability, measured as the temporal
1995a). Shrinkage and attrition are processes assumed resolution of the data. This resolution is a component
to happen in already fragmented landscapes. The time of the temporal scale, which also includes the extent,
sequence as defined by Forman (1995a,b) can also be equal to the space of time over which records are
found in Eng (1997). One study (anonymous 1999) available. Data sets covering a long time period, with
only considers dissection, perforation, and fragmenta- many intervals, are preferred for the analysis of land-
tion. Jaeger (2000) distinguishes six processes, equal to scape transformation processes.
six steps of fragmentation: perforation, incision, dissec-
tion, dissipation, shrinkage, and attrition. Incision is to Cadiz Township, Wisconsin, USA
be interpreted as incomplete dissection, and dissipa- Land-cover change has been documented over a
tion is a combination of dissection and shrinkage. Ini- long period in fine resolution in only a few locations
tial landscape transformation (sensu fragmentation) is (Shafer 1990). One example is the rate and areal extent
expected by perforation, incision, and dissection. After of the destruction of original cover by agricultural
dissipation, patch shrinkage and attrition can occur. clearing in Cadiz Township, Green County, Wisconsin,
For their microlandscape experiment with grassland USA (8954W, 4330N), which took place during the
insects, Collinge (1998) and Collinge and Forman period of European settlement (Curtis 1956). Before
(1998) considered four scenarios of landscape conver- agricultural settlement began, its vegetation was mostly
sion, which are based on Formans (1995a,b) concept: upland deciduous forest, characterized by one single
shrinkage, bisection, fragmentation, and perforation. patch of a 87.24 km2 (Burgess and Sharpe 1981,
Pattern change was based on clearings which become Dunn and others 1991). Five records are reported for a,
more extended. In Kolasa and Waltho (1998), patch p, and n: 1831, 1882, 1902, 1935, and 1950 (Curtis 1956,
size reduction (in the current paper denoted as shrink- Burgess and Sharpe 1981, Dunn and others 1991). In
age), isolation (a particular type of shift) and fragmen- Figure 4, the evolution of a, p, and n is illustrated. All
tation are considered, the latter more in the sense of values are given relative to the highest value present in
dissection. Farina (2000) only considers fragmentation the period 18311950 to show the relative changes of
and aggregation, and both processes are considered as the parameters. Considering the entire time period
opposites, with aggregation defined as regaining the 18311950, thus comparing the 1831 pattern features
open space maintained by the disturbance process ( with the 1950 features, fragmentation is observed (Fig-
fragmentation). Farina (2000) puts emphasis on the ure 2), due to the strong area decrease (exceeding
gap-filling aspect of this type of pattern dynamics. Fahrig 95%, Figure 4a) and the increase of both perimeter
(1997) puts focus on habitat loss as the main process. Two (Figure 4b) and number of patches (Figure 4c). This
types are considered, equivalent to attrition and shrink- result confirms the recognition of the Cadiz Township
Spatial landscape transformation processes 69

mentation; the landscape is in that period converted


from a continuous (n0 1) forest patch into dispersed
patches (n1 70), with concomitant area loss (tobs
0.296) and perimeter increase (p1 156.9 km). Note
that for 19351950 the decrease in number of patches
does not exceed 5%, hence n1 n0 can be substituted
by n1 n0, which leads to shrinkage (isotropic) as the
dominant process and which is consistent with the vi-
sual interpretation of the cartographic data of Curtis
(1956) . The sequence of transformations could then
be modified in fragmentation (18311882), attrition
(18821935), and shrinkage (19351950).

Tiputini Biodiversity Station, Amazonian Ecuador


A gap survey (19971998) was conducted at the
Tiputini Biodiversity Station (TBS), 76 10'15W,
003705S, altitude c. 300 m, in the province of Orel-
lana in Ecuador. For details on the survey, the reader is
referred to Salvador-Van Eysenrode and others
(1998a,b, 1999a,b, 2000a,b) and to Bogaert and others
(2001c). TBS occupies a 650-ha tract of virgin lowland
terra firme rain forest on the north bank of the Tiputini
river. The plot in which the gaps were monitored (size:
135000 m2) consists of a platform surrounded by
Figure 4. Evolution of woodland area (a), perimeter (b), and
number of patches (c) in Cadiz Township, Green County, swamps and an oxbow lake. Canopy gapsformed
Wisconsin, USA, from 1831 to 1950. All parameters are ex- since October 1996 were located and measured dur-
pressed relative to the maximum value of the parameter for ing three field trips: June 1997, March 1998, and De-
the time period considered. cember 1998. The minimum area to be recognized as
gap was set to 4 m2. A metallic marking pole was put in
the approximate center (centroid) of each gap, and the
land-cover change sequence as the main example in distances from the pole to the gap edge (ground pro-
ecological literature of landscape fragmentation by jection of the first vegetation) were measured along
man (Burgess and Sharpe 1981, Shafer 1990, Dunn and eight radii, at intervals of 45 starting from the north.
others 1991, Forman 1995a). Identical conclusions can Gap area was estimated by adding the areas of the
be drawn from the time periods 18311882, 18311902, octants. Hemispherical pictures were taken at 2 m
and 18311935. Because data for 1882, 1902, and 1935 above the ground at the centroid, oriented to the mag-
are available, analysis of the different steps of the 1831 netic north, using a CID-100 Digital Plant Canopy Im-
1950 transformation process is appealing. Because the ager (CID 1997). Images were enhanced for continuity
number of patches and perimeter data peak in 1882, all and contrast. Gap area and perimeter were calculated
subsequent periods are characterized by a1 a0, p1 using a Fortran-77 program. An equal ratio of area to
p0, n1 n0, which points towards attrition of landscape perimeter squared was assumed for the digital images
patches (Figure 2). The initial phase of landscape and the field observations. At the first field trip, 14 gaps
change, before the end of the 19th century, is therefore were observed. Figure 5 shows the gap area, perimeter,
the only period characterized by fragmentation. After- and number of gaps, expressed as fractions of the larg-
wards, the patchy landscape is dominated by continu- est values considered in the entire time period. An
ous forest patch removal for agricultural practice. This increase in the number of gaps is observed with time;
example shows that the temporal resolution of the data, for gap area and boundary length, first an increase and
here 30 years, influences the final conclusion consid- then secondly a decrease is observed. This decrease can
erably. Finer resolutions, i.e., shorter time periods be- be explained by the effect of horizontal closure, caused
tween the records, enable a more accurate determina- by ingrowth of the surrounding vegetation (Salvador-
tion of the active processes. It can be concluded for Van Eysenrode and others 1998a). The data presented
Cadiz Township that the major influences on the pat- in Figure 5 refer to gap characteristics and should be
tern have occurred before 1882 with profound frag- converted to know the parameters for the forest in
70 J. Bogaert and others

Figure 5. Evolution of total gap area (a), total gap perimeter Figure 6. Evolution of forested area (a), perimeter (b), and
(b), and number of gaps (c) present in the experimental plot number of patches (c) in Petersham Town- ship, Worcester
at the Tiputini Biodiversity Station (Amazonian Ecuador) County, Massachusetts, USA from 1830 to 1985. All parame-
from June 1997 (first survey date) to December 1998. All ters are expressed relative to the maximum value of the
parameters are expressed relative to the maximum value of parameter for the time period considered.
the parameter for the time period considered. Using the
presented gap data, the changes in forest area, perimeter, and
number of patches can be assessed. Petersham Township, Massachusetts, USA
Foster (1992) investigated the histories of changing
which they are formed. Gap area increase corresponds land use and vegetation in central New England (Mas-
to forest area decrease (a1 a0), whereas an increase of sachusetts, USA) to assess environmental controls over
the total gap perimeter implies an increase of the total land ownership patterns, agricultural practice and log-
forest perimeter (p1 p0). The total number of gaps ging activity, and vegetation response to these land use
does not provide any information regarding the num- factors. Petersham Township is located in north west-
ber of forest patches. The field survey showed n0 n1 ern Worcester County (central Massachusetts, 7130W,
1, so that only deformation, shift, perforation, shrink- 4130N) and was settled in 1733. In the subsequent
age, or enlargement could possibly be observed (Figure period, due to economic development from a small-
2). Both time intervals represent different transforma- scale rural economy with farming to one characterized
tion processes. From June 1997 to March 1998, forest by small-scale industry, better transportation, and com-
area decreases (a1 a0), while its perimeter increases mercial farming, forest clearance to provide pasture
(p1 p0). This points to perforation, with formation of land for beef cattle and sheep resulted in an increase of
16 supplementary gaps in this period. From March open land from 50% in 1800 to 85% in 1850 (Foster
1998 to December 1998, 18 extra gaps were formed, but 1992, Dunn and others 1991). Commencing in the
the increase in forest area (a1 a0) due to ingrowth of mid-1800s, the population of Petersham declined:
the older gaps prevails: enlargement of the forest is farms were abandoned and neglected fields were refor-
here the type of transformation. Thus, two distinct ested. Since the 1930s, there has been an increase in
phases are distinguished, both featuring gap formation, homes and a reversal in the downward population
but also characterized by opposite trends in forest area, trend, but a decrease in the intensity of direct human
which signals the presence of different conversion ge- impact on the land. Forest cover for Petersham Town-
ometries (Figure 2). ship was recorded using cartographic data and aerial
Spatial landscape transformation processes 71

photograph analysis for 1830, 1900, 1938, and 1985. Anonymous. 1982. The American heritage dictionarySec-
Figure 6 shows the evolution of forest area, perimeter, ond college edition. Houghton Mifflin Company, Boston,
1568 pp.
and number of patches for 1830 1985. The three time
periods considered (1830 1900, 1900 1938, and Anonymous. 1991. Longman dictionary of the English lan-
guageNew edition. Longman Group UK Ltd., Essex, 1890
1938 1985) are dominated by different land transfor-
pp.
mation processes. From 1830 to 1900, the number of
Anonymous. 1996. Websters new universal unabridged dic-
patches increased (n1 n0), and together an expan- tionary. Barnes & Noble Books, New York, 2230 pp.
sion of the forest was observed (a1 a0), as well as an
Anonymous. 1999. Assessing the ecological impacts of timber
increase of the perimeter (p1 p0). This conversion is, management: apparent impacts, actual impacts, and pre-
therefore, denoted as creation (Figure 2). After 1900, cautionary forest development. Silva Forest Foundation Lit-
the number of patches decreases (n1 n0), probably erature Review Series, Slocan Park, 16 pp.
due to patch mergence, together with an increase of Antrop, M. 2001. The language of landscape ecologists and
the forest area (a1 a0), so aggregation was dominat- plannersa comparative content analysis of concepts used in
ing the pattern conversion (Figure 2). From 1938 on, landscape ecology. Landscape and Urban Planning 55:163173.
only one connected forest is present (n1 n0 1). The Bannerman, S. 1997. Spatial patterns and landscape ecology:
third period is characterized by the increase in area (a1 implications for biodiversity. British Columbia Ministry of
Forests Research Program, Victoria, BC, 9 pp.
a0), so that Figure 2 identifies enlargement as the
Bender, D. J., T. A. Contreras, and L. Fahrig. 1998. Habitat
process of transformation.
loss and population decline: a meta-analysis of the patch
size effect. Ecology 79:517533.
Conclusion Bogaert, J., and I. Impens. 1998. An improvement on area-
perimeter ratios for interior-edge evaluation of habitats.
The subject of land transformation is significant to Pages 55 61 in F. Muge, R. C. Pinto, and M. Piedade. Eds,
all human issues that involve land (Forman 1995a). RecPad98: proceedings of the 10th Portuguese conference
Processes of landscape transformation are useful for on pattern recognition. Associac a o Portuguesa de Recon-
environmental managers to define geometrically the hecimento de Padro es, Lisbon.
processes responsible for pattern change in the land- Bogaert, J., P. Van Hecke, D. Salvador-Van Eysenrode, and I.
Impens. 1998. Quantifying habitat edge for nature reserve
scape. Ten processes are considered to alter the spatial
design. Coenoses 13:131136.
configuration of landscapes: aggregation, attrition, cre-
Bogaert, J., P. Van Hecke, and I. Impens. 1999a. A reference
ation, deformation, dissection, enlargement, fragmen-
value for the interior-to-edge ratio of isolated habitats. Acta
tation, perforation, shift, and shrinkage. A decision tree Biotheoretica 47:6777.
algorithm has been presented that determines the ac- Bogaert, J., P. Van Hecke, R. Moermans, and I. Impens.
tive process based on area, perimeter length, and num- 1999b. Twist number statistics as an additional measure of
ber of patches of the class of interest. Application of the habitat perimeter irregularity. Environmental and Ecological
decision tree algorithm for three study areas [woodland Statistics 6:275290.
conversion in Cadiz Township (USA), gap formation in Bogaert, J. 2000. Quantifying habitat fragmentation as a spa-
a rain forest in Tiputini (Ecuador), and forest regrowth tial process in a patch corridormatrix landscape model.
in Petersham Township (USA)] shows the practical use PhD dissertation. University of Antwerp, Wilrijk 209.
of the decision tree. The examples signal the impor- Bogaert, J., R. Rousseau, P. Van Hecke, and I. Impens. 2000a.
Alternative area-perimeter ratios for measurement of 2-D
tance of the temporal resolution of landscape transfor-
shape compactness of habitats. Applied Mathematics and Com-
mation data to detect the presence of short-term con- putation 111:71 85.
versions in a long-term transformation scenario. Bogaert, J., P. Van Hecke, D. Salvador-Van Eysenrode, and I.
Impens. 2000b. Landscape fragmentation assessment using
a single measure. Wildlife Society Bulletin 28:875 881.
Acknowledgments
Jan Bogaert is indebted to the Fund for Scientific Bogaert, J., and R. Rousseau. 2001. Spatial aggregation of
two-dimensional objects in raster data structures. Applied
ResearchFlanders (F.W.O. deren) for his postdoc- Mathematics and Computation 119:117126.
toral fellowship. The authors acknowledge the review-
Bogaert, J., D. Salvador-Van Eysenrode, I. Impens, and P. Van
ers for their significant input. Hecke. 2001a. The interior- to-edge breakpoint distance as
a guideline for nature conservation policy. Environmental
Management 27:493500.
References
Bogaert, J., D. Salvador-Van Eysenrode, P. Van Hecke, and I.
Anonymous. 1971. Websters third new international dictio- Impens. 2001b. Geometrical considerations for evaluation
nary of the English language unabridged. G. & C. Merriam of reserve design. Web Ecology 2:6570, (erratum Web
Company, Springfield, 2662 pp. Ecology 2:74).
72 J. Bogaert and others

Bogaert, J., D. Salvador-Van Eysenrode, P. Van Hecke, I. Im- movement and patch size affect diversity?. Biological Conser-
pens, and R. Ceulemans. 2001c. Land cover change: quan- vation 98:179 190.
tification metrics for perforation using 2-D gap features. Dunn, C. P., D. M. Sharpe, G. R. Guntenspergen, F. Stearns,
Acta Biotheoretica 49:161169. and Z. Yang. 1991. Methods for analyzing temporal changes
Bogaert, J., R. B. Myneni, and Y. Knyazikhin. 2002a. A math- in landscape pattern. Pages 173198 in M. G. Turner, and
ematical comment on the formulae for the aggregation R. H. Gardner. Eds, Quantitative methods in landscape
index and the shape index. Landscape Ecology 17:8790. ecology. Springer-Verlag, New York.
Bogaert, J., P. Van Hecke, and R. Ceulemans. 2002b. The Eng, M. 1997. Spatial patterns and landscape ecology: implica-
Euler number as an index of spatial integrity of landscapes: tions for biodiversity. Extension Note 14, British Columbia
evaluation and proposed improvement. Environmental Man- Ministry of Forests Research Program, Victoria, BC, 9 pp.
agement 29:673 682.
Fahrig, L. 1997. Relative effects of habitat loss and fragmen-
Bogaert, J., Zhou, L. Tucker, C. J. Myneni, R. B. Ceulemans. R. tation on population extinction. Journal of Wildlife Manage-
2002c. Evidence for a persistent and extensive greening ment 61:603 610.
trend in Eurasia inferred from satellite vegetation index
Farina, A. 1998. Principles and methods in landscape ecology.
data. Journal of Geophysical Research 107:D11-10.1029/
Chapman & Hall, London 235.
2001JD001075.
Farina, A. 2000. Landscape ecology in action. Kluwer Aca-
Bribiesca, E. 1997. Measuring 2-D shape compactness using
demic Publishers, Dordrecht 317.
the contact perimeter. Computers and Mathematics with Appli-
cations 33:19. Forman, R. T. T. 1995a. Land mosaicsthe ecology of land-
scapes and regions. Cambridge University Press, Cambridge
Burgess, R. L., and D. M. Sharpe. 1981. Introduction. Pages 15
632.
in R. L. Burgess, and D. M. Sharpe. Eds, Forest island dynamics
in man-dominated landscapes. Springer-Verlag, New York. Forman, R. T. T. 1995b. Some general principles of landscape
Chen J. 1991. Edge effects: microclimatic pattern and biological and regional ecology. Landscape Ecology 10:133142.
responses in old-growth Douglas-fir forests. PhD dissertation, Uni- Forman, R. T. T., and M. Godron. 1981. Patches and struc-
versity of Washington, Seattle, 174 pp. tural components for a landscape ecology. BioScience
CID. 1997. CID-100 Digital plant canopy imager. Instruction 31:733740.
manual. CID, Vancouver, BC, 18 pp. Forman, R. T. T., and M. Godron. 1986. Landscape ecology.
Collinge, S. K. 1996. Ecological consequences of habitat frag- John Wiley & Sons, New York 620.
mentation: implications for land- scape architecture and Foster, D. R. 1992. Land use history (1730-1990) and vegeta-
planning. Landscape and Urban Planning 36:59 77. tion dynamics in central New England, USA. Journal of
Collinge, S. K. 1998. Spatial arrangement of habitat patches Ecology 80:753772.
and corridors: clues from ecological field experiments. Franklin, J. F., and R. T. T. Forman. 1987. Creating landscape
Landscape and Urban Planning 42:157168. patterns by forest cutting: ecological consequences and
Collinge, S. K., and R. T. T. Forman. 1998. A conceptual principles. Landscape Ecology 1:518.
model of land conversion processes: predictions and evi- Giles, R. H., Jr., and M. K. Trani. 1999. Key elements of
dence from a microlandscape experiment with grassland landscape pattern measures. Environmental Management
insects. Oikos 82:66 84. 23:477 481.
Coulson, R. N., H. Saarenmaa, W. C. Daugherity, E. J. Jr. Glasbey, C. A., and G. W. Horgan. 1995. Image analysis for the
Rykiel, M. C. Saunders, and J. W. Fitzgerald. 1999. A knowl- biological sciences. John Wiley & Sons, Chichester, UK 218.
edge system environment for ecosystem management.
Groom, M. J., and N. H. Schumaker. 1993. Evaluating land-
Pages 5779 in J. M. Klopatek, and R. H. Gardner. Eds,
scape change: patterns of worldwide deforestation and local
Landscape ecological analysisissues and applications.
fragmentation. Pages 24 44 in P. M. Kareiva, J. G. King-
Springer-Verlag, New York.
solver, and R. B. Huey. Eds, Biotic interactions and global
Cruz, M. D., and R. P. Matos. 1998. Automatic detection, change. Sinauer Associates Inc., Sunderland, Massachuse.
characterization, and description of objects morphology.
Gustafson, E. J. 1998. Quantifying landscape spatial pattern:
Pages 63 67 in F. Muge, R. C. Pinto, and M. Piedade. Eds,
what is the state of the art?. Ecosystems 1:143156.
RecPad98: proceedings of the 10th Portuguese conference
on pattern recognition. Associac a o Portuguesa de Recon- Gustafson, E. J., and G. R. Parker. 1991. An automated
hecimento de Padro es, Lisbon. method to quantify habitat spatial pattern from satellite
images. Bulletin of the Ecological Society of America 72:132.
Curtis, J. T. 1956. The modification of mid-latitude grasslands
and forests by man. Pages 721736 in W. L. Thomas, C. O. Gustafson, E. J., and G. R. Parker. 1992. Relationships be-
Sauer, M. Bates, and L. Mumford. Eds, Mans role in chang- tween landcover proportion and indices of landscape spa-
ing the face of the earth. The University of Chicago Press, tial pattern. Landscape Ecology 7:101110.
Chicago. Gustafson, E. J., and G. R. Parker. 1994. Using an index of
Davidson, C. 1998. Issues in measuring landscape fragmenta- habitat patch proximity for landscape design. Landscape and
tion. Wildlife Society Bulletin 26:3237. Urban Planning 29:117130.
Debinski, D. M., C. Ray, and E. H. Saveraid. 2001. Species Haines-Young, R., and M. Chopping. 1996. Quantifying land-
diversity and the scale of the landscape mosaic: do scales of scape structure: a review of landscape indices and their
Spatial landscape transformation processes 73

application to forested landscapes. Progress in Physical Geog- Salvador-Van Eysenrode, D., Bogaert, J. and Impens. I. 1998a.
raphy 20:418 445. Horizontal canopy gap occlusion at the Tiputini Biodiver-
Hansen, A. J., D. L. Urban, and B. Marks. 1992. Avian com- sity Station, Amazonian Ecuador. Pages 60 61 in Proceed-
munity dynamics: the interplay of landscape trajectories and ings of the second international canopy conferenceforest
species life histories. Pages 170195 in A. J. Hansen, and F. canopies: global perspectives. The Marie Selby Botanical
diCastri. Eds, Landscape boundaries consequences for bi- Gardens, Sarasota.
otic diversity and ecological flows. Springer-Verlag, New York. Salvador-Van Eysenrode, D., J. Bogaert, P. Van Hecke, and I.
He, H. S., B. E. DeZonia, and D. J. Mladenoff. 2000. An Impens. 1998b. Influence of tree-fall orientation on canopy
aggregation index (AI) to quantify spatial patterns of land- gap shape in an Ecuadorian rain forest. Journal of Tropical
scapes. Landscape Ecology 15:591 601. Ecology 14:865 869.
Herzog, F., A. Lausch, E. Muller, H. H. Thulke, U. Steinhardt, Salvador-Van Eysenrode, D., J. Bogaert, and I. Impens. 1999a.
and S. Lehmann. 2001. Landscape metrics for assessment of Canopy gap morphology determinants in an Amazonian
landscape destruction and rehabilitation. Environmental rain forest. Selbyana 20:339 344.
Management 27:91107.
Salvador-Van Eysenrode, D., J. Bogaert, P. Van Hecke, and I.
Heywood, V. H., and R. T. Watson. 1995. Global biodiversity Impens. 1999b. Canopy perforation in Amazonian Ecuador.
assessment. Cambridge University Press, Cambridge 1140. Coenoses 14:1321.
Jaeger, J. A. G. 2000. Landscape division, splitting index, and Salvador-Van Eysenrode, D., J. Bogaert, P. Van Hecke, and I.
effective mesh size: new measures of landscape fragmenta- Impens. 2000a. Forest canopy perforation in time and space
tion. Landscape Ecology 15:115130.
in Amazonian Ecuador. Acta Oecologica 21:285291.
Johnston, C. A. 1998. Geographic information systems in ecol-
Salvador-Van Eysenrode, D., V. Zak-Mnacek, J. Bogaert, and P.
ogy. Blackwell Science, Oxford 239.
Van Hecke. 2000b. Species diversity and community struc-
Jorge, L. A. B., and G. B. Garcia. 1997. A study of habitat ture of canopy gaps and mature forest patches in Amazo-
fragmentation in southeastern Brazil using remote sensing nian Ecuador. Pages 133141 in R. Ceulemans, J. Bogaert,
and geographic information systems (GIS). Forest Ecology G. Deckmyn, and I. Nijs. Eds, Topics in ecology: structure
and Management 98:35 47. and function in plants and ecosystems. University of Ant-
Kolasa, J., and N. Waltho. 1998. A hierarchical view of habitat werp, UIA, Wilrijk.
and its relationship for species abundance. Pages 5576 in
Salvador-Van Eysenrode, D., F. Kockelbergh, J. Bogaert, I.
D. L. Peterson, and V. T. Parker. Eds, Ecological scaletheory
Impens, and P. Van Hecke. 2002. Canopy gap edge deter-
and applications. Columbia University Press, New York.
mination and the importance of gap edges for plant diver-
Lawrence, R. L., and W. J. Ripple. 1996. Determining patch sity. Web Ecology 3:15.
perimeters in raster image processing and geographic in-
formation systems. International Journal of Remote Sensing Saunders, D. A., R. J. Hobbs, and C. R. Margules. 1991.
17:12551259. Biological consequences of ecosystem fragmentation: a re-
view. Conservation Biology 5:18 32.
Li, B.-L., and S. Archer. 1997. Weighted mean patch size: a
robust index for quantifying landscape structure. Ecological Shafer, C. L. 1990. Nature reserves. Smithsonian Institution
Modelling 102:353361. Press, Washington 189.
Lovejoy, T. E., R. O. Jr. Bierregaard, A. B. Rylands, J. R. Stauffer, D. 1985. Introduction to percolation theory. Taylor
Malcolm, C. E. Quintela, L. H. Harper, K. S., Jr. Brown, & Francis, London 124.
A. H. Powell, G. V. N. Powell, H. O. R. Schubart, and M. B. Schumaker, N. H. 1996. Using landscape indices to predict
Hays. 1986. Edge and other effects of isolation on Amazon habitat connectivity. Ecology 77:1210 1225.
forest fragments. Pages 257285 in M. E. Soule Eds, Con-
Stine, P.A., and C.T. Hunsaker. 2001. An introduction to
servation biology: the science of scarcity and diversity. Si-
nauer Associates, Sunderland, Massachusetts. uncertainty issues for spatial data used in ecological appli-
cations. Pages 91107 in C.T. Hunsaker, M. Goodchild,
McGarigal, K., and Marks. B. J. 1995. FRAGSTATS: spatial pat-
M.A. Friedl, and T.J. Case. Eds, Spatial uncertainty in ecol-
tern analysis program for quantifying landscape structure (PNW-
ogyimplications for remote sensing and GIS applications.
GTR-351). US Department of Agriculture, Forest Service, Pa-
Springer-Verlag, New York.
cific Northwest Research Station, Portland, OR 122 pp.
Turner, M.G., and C.L. Ruscher. 1988. Changes in landscape
Mladenoff, D. J., and DeZonia. B. 1999. APACK 2.11 Analysis
software users guide. Draft version. 43 pp. patterns in Georgia, USA. Landscape Ecology 1:241251.
ONeill, R. V., J. R. Krummel, R. H. Gardner, G. Sugihara, B. Vitousek, P.M. 1994. Beyond global warming: ecology and
Jackson, D. L. DeAngelis, B. T. Milne, M. G. Turner, B. global change. Ecology 75:18611876.
Zygmunt, S. W. Christensen, V. H. Dale, and R. L. Graham. Walker, B., and Steffen W. (eds.). 1997. The terrestrial biosphere
1988. Indices of landscape pattern. Landscape Ecology and global change: implications for natural and managed ecosys-
1:153162. tems. A synthesis of GCTE and related research. IGBP Science
Roseberry, J. L., and Q. Hao. 1996. Interactive computer No. 1. IGBP, Stockholm, 32 pp.
program for landscape-level habitat analysis. Wildlife Society Wickham, J.D., K.H. Riiters, R.V. ONeill, K.B. Jones, and T.G.
Bulletin 24:340 341. Wade. 1996. Landscape contagion in raster and vector
Rosenfeld, A. 1974. Compact figures in digital pictures. IEEE environments. International Journal of Geographical Informa-
Transactions on Systems, Man, and Cybernetics 4:221223. tion Systems 10:891 899.

Você também pode gostar