Landscape and Urban Planning 95 (2010) 181191

Contents lists available at ScienceDirect

Landscape and Urban Planning
j our nal homepage: www. el sevi er . com/ l ocat e/ l andur bpl an
Making graph theory operational for landscape ecological assessments,
planning, and design
Andreas Zetterberg

, Ulla M. Mrtberg, Berit Balfors

Environmental Management and Assessment Research Group, Department of Land and Water Resources Engineering, Royal Institute of Technology,
Teknikringen 76, SE-100 44 Stockholm, Sweden
a r t i c l e i n f o
Article history:
Received 16 January 2009
Received in revised form 21 January 2010
Accepted 21 January 2010
Available online 1 March 2010
Keywords:
Least-cost modeling
Functional connectivity
Environmental planning
European common toad
Metapatch
Spatial redundancy
a b s t r a c t
Graphtheory and network analysis have become established as promising ways to efciently explore and
analyze landscape or habitat connectivity. However, little attention has been paid to making these graph-
theoretic approaches operational within landscape ecological assessments, planning, and design. In this
paper, a set of both theoretical and practical methodological developments are presented to address
this issue. In highly fragmented landscapes, many species are restricted to moving among small, scat-
tered patches of different resources, instead of one, large patch. A life-cycle based approach is therefore
introduced, inwhichametapatchis constructed, spanningover theseresources, scatteredacross theland-
scape. The importance of spatially explicit and geographically dened representations of the network in
urban and regional planning and design is stressed, and appropriate, context-dependent visualizations
of these are suggested based on experience from real-world planning cases. The study moves beyond
the issue of conservation of currently important structures, and seeks to identify suitable redesigns
of the landscape to improve its socialecological qualities, or increase resilience. By introducing both
a system-centric and a site-centric analysis, two conicting perspectives can be addressed. The rst
answers the question what can I do for the network, and the second, what can the network do for
me. A method for typical planning strategies within each of these perspectives is presented. To illustrate
the basic principles of the proposed methods, an ecological study on the European common toad (Bufo
bufo) in Stockholm, Sweden is presented, using the betweenness centrality index to capture important
stepping-stone structures.
2010 Elsevier B.V. All rights reserved.
1. Introduction
Land use change represents the primary driving force in the loss
of biodiversity world wide, and negative effects reach far beyond
the directly impacted areas (Vitousek et al., 1997). To preserve and
develop biodiversity and other ecosystem services, planning and
management activities must recognize the dynamics and complex
interactions withinsocialecological systems, wherephysical plan-
ning activities are an integral part, and the physical landscape is
the common point of reference. Network analysis and graph the-
ory provide powerful tools andmethods for the analysis of complex
systems. The network is often represented by a graph, G(N,L), con-
sisting of a set of nodes, N(G) and a set of links, L(G). The link l
ij
connects nodes i and j. When using this model in landscape ecolog-
ical applications, a node typically represents a habitat patch and a
link typically represents dispersal.
Recently several papers have explored graph-based models of
specieshabitat interactions from a landscape perspective (for a

Corresponding author.
E-mail address: aze@kth.se (A. Zetterberg).
review, see Urban et al., 2009). Many of these feature analysis
and visualization techniques useful in landscape ecological assess-
ments, planning, and design. Graph theory can be used as an
initial, heuristic framework for management, driven in an iterative
and exploratory manner, and with very little data requirements
(Bunn et al., 2000; Calabrese and Fagan, 2004). It does not require
long-term population data, making it an important tool for rapid
landscape-scale assessments (Urban and Keitt, 2001), but graph
theory is at the same time dynamic, allowing additional knowledge
to be incorporated. Despite its simplicity, a graph model based only
on habitat and dispersal distance, has been shown to make predic-
tions very similar to a spatially explicit population model (SEPM),
which had nine additional life-history and behavioral parameters
(Minor and Urban, 2007).
Another attractive property of network analysis is its long tradi-
tion, well developedandtestedtools, as well as efcient algorithms,
used in a wide variety of disciplines (e.g. Ahuja et al., 1993), many
of which are used in planning. Several graph-theoretic metrics
related to classical network analysis problems, such as maximum
ow, connectivity, and shortest paths, have been developed over
decades, and Bunn et al. (2000) as well as Urban and Keitt (2001)
have proposedecological interpretations for some of these. Some of
0169-2046/$ see front matter 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.landurbplan.2010.01.002
182 A. Zetterberg et al. / Landscape and Urban Planning 95 (2010) 181191
the proposed graph-based metrics of functional connectivity have
also been summarized and evaluated (Pascual-Hortal and Saura,
2006; Saura and Pascual-Hortal, 2007).
In addition to comparing metrics for the overall network for
alternative planning scenarios, graph-based methods can be used
to explore important internal structures. For example, Estrada
and Bodin (2008) have used centrality indices to explore well-
connected backbone structures within the network. Another
approach is to evaluate the importance of each patch with respect
to a particular landscape connectivity index (I) by removing one
patch (i.e. one node) at a time, and recording the corresponding
change (I) of the connectivity index (I) (Keitt et al., 1997). This
has become a central technique for nding important patches, and
has also been used by Urban and Keitt (2001) to nd important
links within the network. The patches and links contributing the
most to overall connectivity can thus be found. Similar techniques
have also been used to explore trade-offs between the total pro-
tected area and the overall connectivity (Rae et al., 2007; Rothley
and Rae, 2005). Analyzing trade-offs between required area and
some desired property is an effective technique when assessing
alternative scenarios within planning.
Although graph theory has been proposed as an efcient way
to explore and analyze landscape connectivity, little attention has
been paid to making graph-theoretic approaches operational (i.e.
being in operation) within ecological assessments, planning, and
design. In this paper a set of methodological developments, both
theoretical and practical, are presented to address this issue:

In highly fragmented landscapes, such as urbanizing regions,

many species are restricted to moving among small, scattered
patches of different resources instead of one, large, contiguous
habitat patch. These patches are not capturedwiththe traditional
ways of constructing patches by nding contiguous regions of
some selected land use classes. Here, a life-cycle based approach
is therefore introduced, in which a metapatch can be constructed
spanning over different kinds of life-cycle resources scattered
across the fragmented landscape. This metapatch also has the
potential to capture issues related to scaling in that it always has
to be dened with respect to a temporal scale related to a spec-
ied part of the life-cycle. Shorter time frames are then nested
spatially within longer time frames.

The importance of spatially explicit and geographically dened

representations of the network in urban and regional planning
and design are stressed, and appropriate, context-dependent
visualizations of these are suggested. This, in combination with
network analysis and graph theory, allows the planner to study
the details of a local site in a relevant planning context while
bringing systemic or regional properties into the planning and
decision making process.

The study moves one step beyond the issue of conservation of

currently important structures, and seeks to identify suitable
redesigns of the landscape withimprovement potential. Inorder to
systematically address conicts between local and regional per-
spectives on improvement, a system-centric and a site-centric
analyses are introduced. The rst answers the question what
can I do for the network, and the second, what can the network
do for me. A method for a typical planning issue within each of
these two perspectives is presented. The system-centric analy-
sis illustrates how the spatial redundancy in the network can be
increased, which is argued to improve the resilience of the entire
system with respect to node/link removal. The site-centric per-
spective aims at nding areas with the potential of mitigating the
exposure or isolation of locally important sites.
To illustrate the basic principles of working with the proposed
methods, the paper is framed around an ecological study on the
Europeancommontoad (Bufo bufo) inStockholm, Sweden. The net-
work analysis in the study uses the betweenness centrality index
(Freeman, 1979) in order to capture important stepping-stone
structures (Bodin and Norberg, 2007; Minor and Urban, 2007). By
onlycalculatingleast-cost-distances of thelinks withintheselected
threshold distances, a fairly large network could be analyzed with
22 428 potential reproduction patches grouped into 1361 sepa-
rate annual home-range patches connectedby 4372 dispersal links.
Despite there being very few habitat patches in the highly urban-
ized municipality of Stockholm, the regionally most important
network path with respect to betweenness centrality, consisting
of several small but critical stepping-stones, turned out to pass
through this region. A redesign of the landscape also managed to
restore a dispersal link into a locally important site.
2. Methods
The methods presented in this paper are of two main types.
A set of landscape ecological, theoretical methods and concepts
are introduced which have been developed to better meet the
requirements in fragmented landscapes. The other type pertains to
process related methods and visual representations that are con-
sidered important in order to make graph-theoretic approaches
operational in practice.
2.1. Terminology
In this paper, the term network is used as short for ecolog-
ical network, and dened in a general manner as a complex,
multi-species network representation of a set of ecosystems, inter-
acting with the landscape and linked through functional relations
between the organisms of the ecosystem (sensu Opdam et al.,
2006). In the ecological study of the European common toad, the
network is reduced to a single-species habitat network. The terms
patch and node are conceptually the same within this network but
node is here used when referring to general network theory and
patch is used when referring to a spatially explicit, geographically
dened part of the network, with a physical realization in the land-
scape. The same reasoning holds for the terms connectivity zone
and link. Connectivity zone is introduced as a general term den-
ing a spatially explicit area corresponding to the functional link
between two patches, and is used instead of the often used term
corridor in order to avoid confusion with the structural corridors
used in greenway planning (Fig. 1).
2.2. The spatial aspects of urban or regional planning, and design
Urban and regional planning are commonly referred to as phys-
ical planning (or spatial planning). An important aspect of physical
planning with respect to this paper is that it refers to planning with
a spatially explicit, geographically dened component. The general
idea is to plan the spatial conguration of land use and activities in
Fig. 1. Three representations of networks. (a) A traditional representation of a net-
work, but with suitable spatially explicit locations of the nodes and links. (b) A
patch-link representationshowingthephysical extent of thepatches andtheleast-
cost path (LCP). (c) The physical extent of both the patches and connectivity zones.
Note the multiple connectivity zones between patches.
A. Zetterberg et al. / Landscape and Urban Planning 95 (2010) 181191 183
such a way as to achieve the planning objectives. The following cri-
teria are based on the results fromseven real-world planning cases
involving stakeholders from many disciplines (Zetterberg, 2009).
In order to be successfully operational within planning, design,
or assessment activities, the graph-based approaches need to be
spatially explicit, and geographically dened. Another important
aspect is the ability to deploy these theoretic concepts in a case-
dependent context anddaily working environment already inplace
by planners and designers. The operational maps of the network
should therefore be implemented in such a way that they can be
used seamlessly with other maps, tools and constructs used within
the planning and design process, which is often achieved using a
GIS. As an example, a mapping of each site together with a quick
overview of potential impacts and concerns has been shown to
be an important communicative tool between different types of
stakeholders (Theobald et al., 2000).
Graph models of networks can be presented in many different
ways depending onthe context (Fig. 1). Asimple visualizationtech-
nique used within landscape ecology is to present geographically
explicit graphs, showing the full extent of the linked patches on
top of a map (e.g. Bodin and Norberg, 2007; Fall et al., 2007; Keitt
et al., 1997; OBrien et al., 2006; Urban and Keitt, 2001; Uy and
Nakagoshi, 2007; van Langevelde, 2000; Zhang and Wang, 2006).
This corresponds to Fig. 1b.
However, the extent of the area corresponding to the links,
such as the migration or juvenile dispersal zones, must also be
considered (Fig. 1c). This is just as important from the mapping
perspective as the geographic extent of the patches. In addition,
there are of course several other plausible mapping possibilities
and the users engaged in planning, assessments or design activi-
ties needtobe able toswitchbetweendifferent visualizations of the
graph depending on the context and situation. The ability to visual-
ize different aspects of the network using a varying degree of detail
depending on the context and scale, as for example when zooming
between a regional overview and a detailed local perspective, is
a crucial part of making the graph-based approaches operational,
placing the local planning and design in a regional network con-
text. One interesting approach has been presented by Theobald et
al. (2006) where several different visual andgeographically dened
representations of nodes, links, patches, and linkages can be mixed.
Several of these visual representations are adopted and used in
this paper and suggestions on when to use which representation
are presented throughout the case study. The tools used to create
the different representations span fromstandard software through
third-party software to our own developments and are presented
in their respective methods section.
2.3. Study area and geospatial data
An ecological study was performed within the county of Stock-
holm, the capital of Sweden (Fig. 2a), using the European common
toad(Bufobufo) as afocal species. First, aregional studyof thewhole
county was performed (Fig. 2b), after which a local study (Fig. 2c),
followed by a more detailed study (Fig. 2d), both within the munic-
ipality of Stockholm, were carried out. The regional study aimed
at nding important ecological structures through the region. The
local study illustrated examples of nding areas with improvement
potential both from a system-centric, and site-centric perspec-
tive. The detailed study showed how to liberate this improvement
potential through the construction of three new spawning ponds.
Fig. 2. Study areas. (a) The location of the study area for the regional study, encompassing Stockholm County. (b) The entire study area in detail and points out the location
of (c), showing the study area for nding improvement potential in the municipality of Stockholm. (d) The study area for designing a link through the National Urban Park.
184 A. Zetterberg et al. / Landscape and Urban Planning 95 (2010) 181191
The geospatial data used as input for modeling were GSD-
Landcover andGSD-Propertymap(National Landsurveyof Sweden,
2006), a biotope map (Lfvenhaft et al., 2002; Stockholm
Municipality, 1999), and a vegetation map (Sollentuna kommun,
1993). GSD-Landcover contains about 60 landcover and vegetation
classes andis basedonMultispectral satellite data fromLandsat TM
with 30m30m geometric resolution. The biotope map and the
vegetation map are based on interpretation of aerial photographs,
and together with the GSD-Property map, they were used to com-
plement GSD-Landcover with smaller waterbodies and wetlands.
The biotope map has previously been used to assess the amphibian
distribution in Stockholm Municipality (Lfvenhaft et al., 2004).
2.4. A life-cycle based approach and the metapatch concept
In highly fragmented landscapes, such as urbanizing regions,
many species are restricted to moving among small, scattered
patches of different resources instead of one, large, contiguous
habitat patch. In these cases, the traditional approach of construct-
ing a patch by selecting contiguous cells of some land use class
will fail. If instead modeling the accessibility of different resources
needed, a habitat patch can be constructed that spans over scat-
tered resources and includes the regions needed to move between
them. Such a patch is in this sense a metapatch resembling the idea
of smaller sub-populations forming a metapopulation.
Another seldom emphasized fact is that the size and type of a
patch vary with the temporal scale. Theobald (2006) argues that
the functional denition of a patch depends on the movement type
considered, and that these occur at a range of temporal scales. For
example, for a certain organismwith temporal dynamics similar to
human, themovement typeonadailybasis couldtypicallybeforag-
ing, andthe patches reecting the corresponding time frame would
therefore contain foraging resources. On a yearly basis, the move-
ment types could typically be natal dispersal and genetic exchange,
and the patch types would be home-range patches (either annual
or lifetime). On a centurial basis, the relation could typically be
long-term genetic exchange and the patches would correspond
to (meta-) population patches. The same reasoning can be used
for the traditional matrix and corridor concepts; as the temporal
scale increases and the patches change type, parts of what was
previously considered matrix or connectivity zone are successively
incorporatedintothepatch, andnewconnectivityzones areformed
reecting the processes relevant at the new time scale (Fig. 3).
In order to generalize this view and devise a method more suit-
able for fragmented landscapes, a life-cycle based approach and
the general metapatch were introduced. Metapatches and links
need to be dened with respect to a temporal scale related to a
speciedpart of the life-cycle. These were here constructedby nd-
ing contiguous areas containing all resources within reach, needed
throughout this selected part of the life-cycle. In this study, the
metapatches were set to represent annual home-ranges, contain-
ing all the resources needed throughout the year (e.g. spawning,
foraging, overwintering), including the migration zones between
these resources. The links and corresponding connectivity zones
were accordingly set to represent juvenile dispersal.
2.5. Cost-distance modeling of the common European toad (Bufo
bufo)
Cost-distance analysis (for details, see for example Adriaensen
et al., 2003) was used to dene the annual home-range patches
according to our life-cycle based approach. This enabled the con-
struction of a metapatch to be based on the accessible necessary
life-cycle resources scattered across the landscape. The migra-
tion zones between the resources were automatically integrated
into the patches. For the analysis, ArcGIS cost-distance was used
Fig. 3. Schematic overview of how clusters of patches and links gradually build
up larger metapatches, connected by new links as the temporal domain increases.
Clusters of resource patches and links make up home-range patches, which are con-
nected by dispersal links. On a longer time scale, these home-range clusters make
up (meta-)population patches, connected by genetic links.
(ESRI, 2006). This tool needs two GIS-layers, a source and a friction
layer, as input. In the proposed method, the source layer contains
patches of the resources inthe landscape, andthe frictionlayer con-
tains the different costs of movement across each raster cell. The
annual home-range patches were based around potential breeding
sites, and the source layer was therefore constructed by selecting,
from the geospatial datasets, all pixels corresponding with suit-
able parts of lakes, ponds, or wetlands (Table 1). A range for each
input parameter was found through semi-structured interviews
withfour experts inNovember 2005. Anassessment of these ranges
was conductedby the authors inanattempt tonda representative
parameter set.
The output is an accessibility map, containing the total least-cost
to move from each cell to its closest source across the friction sur-
face. By setting the friction for optimal habitat to 1, the least-cost
throughoptimal habitat is equal to the Euclideandistance along the
same path. The geographic extents (i.e. the borders) of the annual
home-range patches were nally foundby setting a thresholdvalue
for the maximum effective distance from a resource, in effect cor-
responding to a probability threshold of reaching that far within a
year. The probability of migrating a certain distance from a source
is often modeled using a negative exponential decay kernel (e.g.
Bunn et al., 2000; Urban and Keitt, 2001):
p
ij
= e
(d
ij
)
,
where p
ij
is the probability of migrating from point i to point j,
d
ij
, is the distance from point i to j, and is a species-specic
parameter. The parameter assessment suggested =2.3010
3
,
corresponding to 90.0% of the individuals migrating a distance
less than 1km, and 99.0% less than 2km within the annual home-
range patch. The probability threshold for the metapatches was
selected to cover 90% of the annual migration events, correspond-
ingtoaneffectivedistanceof 1km. Anymetapatches not containing
all the required resources, in this case, reproduction sites, sum-
mer habitat and winter habitat (Table 1), within reach (i.e. within
A. Zetterberg et al. / Landscape and Urban Planning 95 (2010) 181191 185
Table 1
Parameter values for the different kinds of vegetation classes found through expert solicitation.
Biotope Resource Friction
(S) summer habitat
(W) winter habitat
(R) reproduction habitat
Forest
(deciduous, mixed, coniferous)
S, W 25
Wetland
(smaller open, semi-open, wooded)
S, W, R 1
Larger water bodies 2
Open grassland
(natural or cultivated)
S, W 25
Semi-open land
(gardens, sparsely developed land w
trees/bushes)
S, W 15
No or sparse vegetation 1050
the effective distance and therefore within its boundaries) were
eliminated.
Juvenile dispersal was also modeled using cost-distance meth-
ods. Male toads are sexually mature in their 3rd year (Hemelaar,
1983), and therefore the probability of effective juvenile dispersal
distances was modeled in the same way as the annual migration
distance, only this time using =7.6710
4
, corresponding to a
three times longer distance than the annual migration distance (to
be reached within 3 years). Potential dispersal zones were mod-
eledusingtheConditional MinimumTransit Cost (CMTC) (Pintoand
Keitt, 2009), where the CMTC of each raster cell is the effective dis-
tance between two given patches along the least-cost path passing
through that particular cell. This was done using ArcGIS least-cost
corridor (ESRI, 2006), whichuses twocost-distance rasters as input,
one for each of the two patches between which the corridor is to
be found. The output is a raster showing, for each cell, the CMTC
between the two sources used to form the input rasters. By using
the threshold level for juvenile dispersal given above, a potential
dispersal zone is found between the two patches, containing all the
cells along all least-cost paths shorter than the threshold distance.
2.6. Finding important structures within the network
After having found the annual home-range patches and the dis-
persal links between these, graph theory was used to lter out
important structures withinthe network, i.e. whichof these patches
and links could be considered more important from a certain per-
spective. For the purposes of this study, the chosen perspective of
importance was nding stepping-stones through the highly frag-
mented and urbanized parts of the landscape that are critical for
keeping the network connected. These stepping-stones are typi-
cally not important for example as major sources of recruitment,
but rather for the long-term genetic ow through barriers such
as built-up areas, infrastructure, or large water bodies. Without
these stepping-stones, the network would fall apart into at least
two smaller components instead of one larger component.
The betweenness centrality index (Freeman, 1979) has been
proposed as a suitable measure of stepping-stone importance
(Bodin and Norberg, 2007; Minor and Urban, 2007). For a graph,
G=(N,L), the betweenness centrality C
B
(n) of node n is calculated
as:
C
B
(n) =

i / = n / = j N
i / = j

ij
(n)

ij
where
ij
is the total number of least-cost paths from node i to j,
and
ij
(n) is the number of least-cost paths from i to j that actually
pass through node n. Hence, the index for node n corresponds to the
proportion of all possible least-cost paths of the network that are
routed through node n.
In order to nd important stepping-stone structures through
the network, betweenness centrality was therefore calculated, and
the corresponding important landscape structure was visualized
by emphasizing the resulting important patches in the maps. These
were found using ArcGIS natural breaks (ESRI, 2006), which creates
data classes according to clusters by maximizing their differences.
The classes are ranked, and when increasing the number of desired
classes, additional break points do not affect the previously found,
higher ranked break points. The top two classes ltered out less
than 1% of the patches while still forming a connected structure
through the bottleneck in the center, and were therefore selected
as important (Fig. 5).
2.7. Exploring the improvement potential from two perspectives
In order to move from an assessment of the current situation
and illustrate howto look for improvement potential, two different
perspectives, the site-centric and system-centric, were introduced.
The site-centric perspective focuses on nding parts of the sys-
tem that can be modied to improve the situation at a particular
geographic site, while the system-centric perspective focuses on
nding parts that can improve some studied property for the entire
system. These perspectives can often be conicting and trade-offs
between them need to be considered.
A method for nding improvement potential fromeach of these
twoperspectives is presentedbelowandinFig. 4. Thesystem-centric
perspective, aimedat ndingareas withthe potential for improving
the resilience of the important internal structure of the network by
increasing its spatial redundancy. The increase inresilience was not
based on a comparison between quantied measures, but rather on
the argument that increasing the link (or node) redundancy in the
network makes the network more resilient to the removal of links
(or nodes) (Janssen et al., 2006).
First the important structure needs to be identied. In the case
study, this was made up of the nodes with high betweenness
centrality and their interconnecting links as previously explained.
Regions within this structure with little redundancy (i.e. important
regions but with very few alternative routes through the network)
were consideredcritical. The critical regions facing highthreats (e.g.
urbanizing areas), received the highest priority (Fig. 4a) in which
improvement potential should be looked for.
The attention then turned to the existing non-important struc-
ture in this high-priority region. The non-important structure
contains nodes and links that could be considered as poten-
tial building-blocks. Areas where redundancy could effectively
186 A. Zetterberg et al. / Landscape and Urban Planning 95 (2010) 181191
Fig. 4. Schematic overview of the methods for nding improvement potential. (a) System-centric perspective. Regionally important nodes (solid) are presented together
with the other nodes, the building-blocks (hollow). A vulnerable area with lowredundancy is indicated and an area suitable for creating redundancy is found. (b) Site-centric
perspective. Three locally important nodes (solid), with different network-related properties, are shown: (i) well-connected, (ii) isolated, (iii) exposed (due to the single link).
Areas with improvement potential for the latter two are highlighted.
be created by restoring nodes or links were considered to have
improvement potential from the system-centric perspective.
The site-centric perspective aimed at nding areas with the
potential of mitigating the exposure or isolation of locally impor-
tant sites. A generalization of the concept of exposure (Jenelius et
al., 2006) is that a site is exposed if small changes in the network
have large consequences for the local site, which can ultimately
become isolated. For this analysis, important sites, not with respect
to the system but rather for some other reason such as high recre-
ational values, were identied (Fig. 4b). Among these sites, those
highly exposed or isolated from the network were identied. In
the case study, a highly exposed site was simply a site with one or
very few links. Areas in the network where the exposure or isola-
tion could be mitigated, for example by restoring a link or adding
redundancy, were considered to have improvement potential from
the site-centric perspective.
Asuitable representation of the network corresponding to Fig. 4
was created using the third-party ArcGIS tool FunnConn (Theobald
et al., 2006). By using the previously created home-range patches
and the friction values as input and setting the thresholds for dis-
persal previously given, this tool can create an attractive visual
representationof thenetworkthat wesuggest is suitableat this par-
ticular scale. This representation shows the patches together with
simpliedmultiple linkages (Theobald, 2006; Theobaldet al., 2006)
between these (Fig. 6). The linkages are simple representations
showing the approximate locations of the different connectivity
zones between the patches.
3. Results
3.1. Cost-distance analysis
Aggregation of the pixels selected as suitable for reproduction
resulted in 22 428 potential reproduction patches within the study
area of whichmany were incorporated into the same annual home-
range patches during cost-distance analysis. The cost-distance
analysis resulted in 1361 separate annual home-range patches
(Fig. 5a) and4372 links withaneffective distance belowthe thresh-
old level of 6km, covering 99.0% of the potential dispersal events.
3.2. Finding important structures within the network
The betweenness centrality index managed to clearly highlight
the important small stepping-stones connecting the northern with
the southern parts of the county. Two threshold levels, 1.7410
2
(class 2) and 8.3810
2
(class 1), were found using ArcGIS natural
breaks (ESRI, 2006). The99out of all 1361metapatches correspond-
ingtoavaluewithinanyof theseclasses wereconsideredimportant
with respect to this index, since the removal of any of these patches
would signicantly increase the average least-cost dispersal dis-
tance betweentworandomlychosennodes of the network(Fig. 5b).
Two major structures emerged within the study area: a stronger
western path, with several small but critical stepping-stones
through parts of the municipality of Stockholm, and a weaker east-
ern path through parts of the Stockholm archipelago. The patches
witha value belowthese thresholds were not consideredregionally
important with respect to betweenness centrality, and therefore
only visualized in Fig. 5a.
3.3. Identication of improvement potential
In the regional study, the graph was based on 22 428 potential
reproduction sites, which is too many to be effectively visualized
as any of the graph examples in Fig. 1, and the visualizations in
Fig. 5 were therefore considered more appropriate. When zoom-
ing in, however, and studying the role of the part of the network
situated within Stockholm Municipality (Fig. 2c), fewer nodes and
links need to be visualized. This opens up for an intermediate and
more detailedrepresentation(Fig. 6, similar to Fig. 1b), showing the
same information as Fig. 5, but in addition the links are shown. All
patches and links are included in this representation to show the
potential building-blocks (striped) of the network, and at the same
time the regionally important patches withrespect to betweenness
centrality (both class 1 and class 2) are highlighted (solid).
A. Zetterberg et al. / Landscape and Urban Planning 95 (2010) 181191 187
Fig. 5. All of the 1361 patches are shown to the left (a), and the patches considered important (class 1 and class 2) with respect to betweenness centrality are shown to the
right (b). These small stepping-stone patches run through, and on either side of the city of Stockholm.
This patch-link representation gives insights into the relations
between the local network structure and the regionally important
structure; it matches the method presented in Fig. 4, and can be
used to nd parts of the network with improvement potential both
from a system-centric, and a site-centric perspective. Fig. 6 shows
examples of both these perspectives. Restoring area (b) would have
virtually no effect on the rest of the network but rather enable
a stronger inux of propagules into the site itself, which is con-
sidered locally important due to its accessibility, biodiversity, and
recreational values for the inhabitants of Stockholm Municipality.
3.4. Connecting the National Urban Park
In order to nd the spatial extent of an area in which a new link
could be created into the southern part of the National Urban Park,
the least-cost corridor tool (ESRI, 2006) was run between the two
involved patches. By relaxing the threshold for the CMTC (which
would normally correspond to the maximum juvenile dispersal
distance), a potential restoration zone emerged (Fig. 7). This area
was studied in detail with the aim of adding new ponds, and for
this the potential restoration zone was visualized in a GIS together
with other contextually important information, such as the biotope
map, topography, an orthophoto, property boundaries, and a city
map for local reference. Three new ponds were considered to be
sufcient to direct the ow of amphibians through the area and to
restore the connection, and within this new connection there was
sufcient summer and winter habitat. Fig. 7 shows the suggested
locations for these three newponds, and also illustrates the impor-
tance of visualizing this ina spatiallyexplicit manner, together with
relevant GIS-data.
The effect of the redesign was evaluated by re-running the cost-
distance analysis for the new network containing the new ponds,
andtheir correspondinglinks usingthesethreenewsources (Fig. 8).
The longest of the least-cost effective distances between the two
initial home-range patches decreased dramatically from 9702m
to 1096m (effective), improving the probability of dispersal for
a propagule through the link. As a result of the additional three
breeding ponds, the productive area, and thus the potential net
production, also increased. The combined effect of the increased
dispersal probability and potential net production is a signicant
potential increase of the dispersal ux in the region. The potential
inux to the previously isolated area therefore increased.
4. Discussion
4.1. Operational aspects
One of the major advantages of the network approach was the
abilitytozoominandout betweenalocal areaandtheentireregion,
while efciently incorporating important system properties into
the respective planning context at each scale. Knowledge from the
network analysis in combination with the corresponding spatial
extents of the nodes and links, can thus provide input to both the
188 A. Zetterberg et al. / Landscape and Urban Planning 95 (2010) 181191
Fig. 6. Patch-link representation of a network revealing the relationship between important patches, other available building-block patches and improvement potential
gaps. Solid patches are regionally important with respect to betweenness centrality (from Fig. 5b). Striped patches are those with lower betweenness centrality (potential
building-blocks). Two example sites are highlighted. Site (a) in the upper left corner has the potential of restoring a link to improve resilience of the regional network through
increased spatial redundancy. Site (b), has the potential of connecting the locally important, but isolated southern part of the National Urban Park, to the rest of the network
in the north.
regional and the local planning process, such as when developing
comprehensive plans. Understanding which nodes and links are
important, or even critical, and their spatial extents, facilitates the
identication of areas with high ecological values that are in need
of protection, as well as areas with low ecological and social val-
ues that could have an improvement potential. For example, there
may be a potential for improving the ecological properties such as
connectivity or resilience in some cases, and social properties such
as housing or transportation in others.
Results from seven real-world planning and assessment cases,
involving different kinds of stakeholders (Zetterberg, 2009),
demonstrate the need to be able to move from the overall systems
analysis level downto the planning and assessment maps, under-
standing not only howthe overall systemis affected, but also where
critical areas are situated. It is also valuable to understand why cer-
tain impacts arise, and the geographic extent of critical regions. The
results also call for the inclusion of an assessment of the improve-
ment potential inadditionto the assessment of the current situation,
and nally how to design the landscape so as to mitigate negative
impacts, or evenimprove desirable properties. The ability to switch
to a close-upof network, showing its spatial extent, andinthe same
GIS add other important information such as topography, property
boundaries, roads, and vegetation turned out to be a major quality
in the process of landscape design.
4.2. Betweenness centrality
The betweenness centrality index (Freeman, 1979) managed to
clearly highlight the importance of the smaller stepping-stones.
This result is in agreement with Bodin and Norberg (2007) stating
that this index manages to emphasize areas thought to be impor-
tant to the connectivity of the network even when the risk for
habitat isolation is low. Minor and Urban (2007) also showed that
betweenness centrality could be used to identify stepping-stone
patches that were not easily identied with an SEPM.
An ecological interpretation of betweenness centrality is that
it could indicate areas with long-term genetic variety. The index
identies the patches routing the highest proportionof the shortest
effective dispersal paths within the network. Since the algorithm
is inuenced by all patches, including those that are far from
each other and thus probably more genetically different, patches
with the highest betweenness (Fig. 5b) may indicate the geneti-
cally most diverse paths through the network. These paths could
therefore be regarded as important for biodiversity at the genetic
level, and should then be considered in impact assessments, such
as Environmental Impact Assessment and Strategic Environmental
Assessment.
When increasing the number of patches within a region, the
relative importance of each of these decreases. Hence, an impor-
tant region with many alternative patches (i.e. high redundancy)
may not be identied as important until all but a few patches are
removed. This may result indifculties nding or designing regions
that are important and redundant at the same time.
The example in this paper has only dealt with importance with
respect to stepping-stone quality, using betweenness centrality.
However, even though most of the patches of Fig. 5a are not
important with respect to betweenness, several of them are most
probably important in other aspects, such as major sources for
A. Zetterberg et al. / Landscape and Urban Planning 95 (2010) 181191 189
Fig. 7. An area suitable for redesigning the network was found by calculating a potential dispersal zone (striped area) between two existing but unconnected annual
home-range patches, and presenting this area together with other important information for the design, such as topography and buildings.
Fig. 8. The effect of the redesign is illustrated by showing the new home-range patches (striped patches), formed around each of the three new ponds, and the new links
(dashed lines) between the patches. Note that two new links were found between the southernmost of the new patches and the two original patches in the very south.
190 A. Zetterberg et al. / Landscape and Urban Planning 95 (2010) 181191
recruitment, or as redundant back-up routes building resilience.
Within the planning and design process as well as in ecological
assessments, several different aspects of the network need to be
studied. Size, quality andconnectivity are three examples of impor-
tant attributes of the patches within a network (e.g. Minor and
Urban, 2007).
4.3. Important network structures
Notwithstanding its limitations due to uncertainties, the graph
is useful as a heuristic framework requiring little data (Bunn et
al., 2000; Urban and Keitt, 2001), and critical parts of the network
can still be identied, for example using patch importance indices
found through patch removal (Keitt et al., 1997), or using central-
ity indices (Estrada and Bodin, 2008). The critical structures can
then be emphasized using a ranking of these important patches
and selecting the top ranked patches (Urban and Keitt, 2001), or
as was done in this paper, nding thresholds using natural breaks
for the index and emphasizing all patches that are more important
than the threshold level. The number of important patches can be
increased by including more classes into the natural breaks tool.
The natural breaks correspond to threshold values where the index
changes rapidly, which could be used as an effective way of negoti-
ating the trade-off betweenprotected important structures and the
corresponding area required. Similar techniques have been used to
explore trade-offs between a patchs contribution to overall con-
nectivity and its corresponding increase in protected area (Rothley
andRae, 2005), or toanalyze critical thresholds inconnectivitywith
respect to dispersal distance (e.g. Keitt et al., 1997). In this kind of
analysis, the critical spatial structures of the links need also to be
considered in addition to those of the patches.
4.4. Cost-distance modeling and ecology
Traditionally, the friction values in cost-distance modeling are
related to a mixture of energy expenditure, behavioral aspects, and
mortality risks (e.g. Adriaensen et al., 2003; Joly et al., 2003; Ray et
al., 2002; Theobald, 2006). As a consequence, areas with high mor-
tality are assigned a higher friction value, which in effect results
in a reduction of the accessible patch area instead of for example
a reduction of the surviving number of propagules. In this study
we therefore separated mortality risks from energy expenditure,
only acknowledging energy expenditure as being part of the effec-
tive distance. This opens up for other methods for handling the
mortality risks, such as probability-related models, which in turn
can result in both a better geographic representation of the poten-
tially accessible patch, and a separate analysis of how to mitigate
mortality risks.
As an example, roads were assigned to the class no or sparse
vegetation (Table 1) with friction values only related to energy
expenditure. Traditionally however, roads would be considered a
barrier due to high mortality and assigned extremely high friction
values. This would give the false impression that the potentially
accessible patch is small, bordered by a road and with no informa-
tion about mortality. In this paper, the roads are considered just as
accessible as any land with no or sparse vegetation, but deadly. In
reality, they should be considered to be population sinks. By sepa-
rating mortality fromfriction, the road layer can be added on top of
the potentially accessible patch producing a map of conict areas
in need of mitigation or monitoring.
There are also a number of problems related to the ecological
assumption of least-cost. First of all, low cost (i.e. low friction)
of the landscape for an organism does not necessarily imply that it
is the chosen route through the landscape for that particular organ-
ism. Indeed, taking amphibians as anexample, there are indications
that juveniles of some species tend to migrate towards a spe-
cic habitat, such as distant forests, even though the local friction
may be much lower in other directions (e.g. Sjgren-Gulve, 1998;
Walston and Mullin, 2008). Second, even though some indices of
connectivity take the size and often even some quality-weighted
area of the patches into account, this is usually not the case for
the links. In essence, basing the connectivity index only on some
functionof the least-cost pathhas the unwanted side effect that the
entire dispersal zone between two patches could be removed, leav-
ingnothingmorethananarrowregionaroundthepathwithout this
affecting the value of the index. This may make an index calculated
in this way unsuitable as an indicator of connectivity, for example,
both within environmental assessments, planning, and design. One
of the major challenges will be to better model the probability of
dispersal or the dispersal ux between two patches as well as the
corresponding spatial extent of the dispersal zones.
Theobald (2006) has raised parts of this issue by introducing
the concept of multiple-paths, also recognizing the problems asso-
ciated with the least-cost path. This has been further explored
by Pinto and Keitt (2009), using two different methods to nd
multiple-paths. One of them, CMTC, was used in this paper to con-
struct dispersal zones. However, eventhoughthese methods create
multiple-paths or connectivity zones, they are still based on the
concept of the least-cost path which again may not be ecologi-
cally relevant. Another promising approach, based on randomwalk
theory, is using circuit theory which also allows the modeling of
multiple-paths between nodes (Mcrae et al., 2008).
4.5. Redundancy, resilience, and planning
Within physical planning, it is of interest to know which areas
are suitable to develop without a large negative ecological impact.
One way of achieving this is to look for redundancies in the net-
work. However, one has to keep in mind that the resilience of
the network with respect to link (or patch) removal is degraded
when removing spatial redundancy in the network (Janssen et al.,
2006). Indeed, one of the results in this paper illustrated how to
increase resilience by nding areas suitable for creating redundancy
in important structures. Nevertheless, a deeper understanding of
the network structure helps to select areas where redundancy can
be increased as well as areas that are of less ecological importance
andwhere redundancy couldbe decreased, allowing for other func-
tional aspects of the landscape, such as housing.
Acknowledgements
The study was nanced by Formas, the Swedish Research Coun-
cil for Environment, Agricultural Sciences and Spatial Planning.
We thank Ebbe Adolfsson, Bjrn-Axel Beier, Margareta Ihse, Oskar
Kindvall, and Lars-Gran Mattsson for interesting discussions and
valuable input. We would also like to thank the reviewers for tak-
ing their time and giving constructive feedback. Finally, we would
like to thank Claes Andrn, Jon Loman, Jan Malmgren, and Per
Sjgren-Gulve for their valuable time and input during the expert
solicitation.
References
Adriaensen, F., Chardon, J.P., De Blust, G., Swinnen, E., Villalba, S., Gulinck, H., et al.,
2003. The application of least-cost modelling as a functional landscape model.
Landscape Urban Plan. 64 (4), 233247.
Ahuja, R.K., Magnanti, T.L., Orlin, J.B., 1993. Network Flows: Theory, Algorithms and
Applications. Prentice Hall, Englewood Cliffs, NJ.
Bodin, O., Norberg, J., 2007. A network approach for analyzing spatially structured
populations in fragmented landscape. Landscape Ecol. 22 (1), 3144.
Bunn, A.G., Urban, D.L., Keitt, T.H., 2000. Landscape connectivity: a conservation
application of graph theory. J. Environ. Manage. 59 (4), 265278.
Calabrese, J.M., Fagan, W.F., 2004. A comparison-shoppers guide to connectivity
metrics. Front. Ecol. Environ. 2 (10), 529536.
A. Zetterberg et al. / Landscape and Urban Planning 95 (2010) 181191 191
ESRI, 2006. ArcGIS (Version 9.2) [GIS Application]. Environmental Systems Research
Institute, Inc., Redlands, CA.
Estrada, E., Bodin, O., 2008. Using network centrality measures to manage landscape
connectivity. Ecol. Appl. 18 (7), 18101825.
Fall, A., Fortin, M.J., Manseau, M., OBrien, D., 2007. Spatial graphs: principles and
applications for habitat connectivity. Ecosystems 10 (3), 448461.
Freeman, L.C., 1979. Centralityinsocial networks conceptual clarication. Soc. Netw.
1 (3), 215239.
Hemelaar, A.S.M., 1983. Age of Bufo bufo in Amplexus over the spawning period.
Oikos 40 (1), 15.
Janssen, M.A., Bodin, O., Anderies, J.M., Elmqvist, T., Ernstson, H., McAllister, R.R.J.,
et al., 2006. Toward a network perspective of the study of resilience in
socialecological systems. Ecol. Soc. 11 (1), Artn 15.
Jenelius, E., Petersen, T., Mattsson, L.G., 2006. Importance and exposure in road
network vulnerability analysis. Transportation Res. Part A: Policy Pract. 40 (7),
537560.
Joly, P., Morand, C., Cohas, A., 2003. Habitat fragmentation and amphibian conserva-
tion: building a tool for assessing landscape matrix connectivity. C. R. Biol. 326,
S132S139.
Keitt, T.H., Urban, D.L., Milne, B.T., 1997. Detecting critical scales in fragmented
landscapes. Conserv. Ecol. 1 (1), pp. XVXVI.
Lfvenhaft, K., Bjrn, C., Ihse, M., 2002. Biotope patterns in urban areas: a conceptual
model integrating biodiversity issues in spatial planning. Landscape Urban Plan.
58 (24), 223240.
Lfvenhaft, K., Runborg, S., Sjgren-Gulve, P., 2004. Biotope patterns and amphibian
distribution as assessment tools in urban landscape planning. Landscape Urban
Plan. 68 (4), 403427.
Mcrae, B.H., Dickson, B.G., Keitt, T.H., Shah, V.B., 2008. Using circuit theory to
model connectivity in ecology, evolution, and conservation. Ecology 89 (10),
27122724.
Minor, E.S., Urban, D.L., 2007. Graph theory as a proxy for spatially explicit popula-
tion models in conservation planning. Ecol. Appl. 17 (6), 17711782.
National Landsurvey of Sweden, 2006. GSD-Landcover and GSD-Property Map
[GSD Data]. National Landsurvey of Sweden (retrieved June 18, 2006 [grant I
2008/1818] https://butiken.metria se/digibib/).
OBrien, D., Manseau, M., Fall, A., Fortin, M.J., 2006. Testing the importance of spatial
conguration of winter habitat for woodland caribou: an application of graph
theory. Biol. Conserv. 130 (1), 7083.
Opdam, P., Steingrover, E., van Rooij, S., 2006. Ecological networks: a spatial concept
for multi-actor planning of sustainable landscapes. Landscape Urban Plan. 75
(34), 322332.
Pascual-Hortal, L., Saura, S., 2006. Comparisonanddevelopment of newgraph-based
landscape connectivity indices: towards the priorization of habitat patches and
corridors for conservation. Landscape Ecol. 21 (7), 959967.
Pinto, N., Keitt, T.H., 2009. Beyond the least-cost path: evaluating corridor redun-
dancy using a graph-theoretic approach. Landscape Ecol. 24 (2), 253266.
Rae, C., Rothley, K., Dragicevic, S., 2007. Implications of error and uncertainty for an
environmental planning scenario: a sensitivity analysis of GIS-based variables
in a reserve design exercise. Landscape Urban Plan. 79 (34), 210217.
Ray, N., Lehmann, A., Joly, P., 2002. Modeling spatial distributionof amphibianpopu-
lations: aGISapproachbasedonhabitat matrixpermeability. Biodivers. Conserv.
11 (12), 21432165.
Rothley, K.D., Rae, C., 2005. Working backwards to move forwards: graph-based
connectivity metrics for reserve network selection. Environ. Model. Assess. 10
(2), 107113.
Saura, S., Pascual-Hortal, L., 2007. A new habitat availability index to integrate con-
nectivity in landscape conservation planning: comparison with existing indices
and application to a case study. Landscape Urban Plan. 83 (23), 91103.
Sjgren-Gulve, P., 1998. Spatial movement patterns in frogs: target-oriented disper-
sal in the pool frog, Rana lessonae. Ecoscience 5 (1), 3138.
Sollentuna kommun, 1993. Naturvrdsplan [GSD data].
Stockholm Municipality, 1999. Database for the Biotopes of Stockholm [GSD Data].
Department of Physical Geography and Quaternary Geology, StockholmUniver-
sity, and Stockholm Municipality.
Theobald, D.M., 2006. Exploring the functional connectivity of landscapes using
landscape networks. In: Crooks, K.R., Sanjayan, M.A. (Eds.), Connectivity Con-
servation. Cambridge University Press, Cambridge, pp. 416443.
Theobald, D.M., Hobbs, N.T., Bearly, T., Zack, J.A., Shenk, T., Riebsame, W.E., 2000.
Incorporating biological information in local land-use decision making: design-
ing a system for conservation planning. Landscape Ecol. 15 (1), 3545.
Theobald, D.M., Norman, J.B., Sherburne, M.R., 2006. FunnConn v1 Users Manual:
ArcGIS Tools for Functional Connectivity Modeling [Users Manual]. Natural
Resource Ecology Lab, Colorado State University, Fort Collins, CO.
Urban, D., Keitt, T., 2001. Landscape connectivity: a graph-theoretic perspective.
Ecology 82 (5), 12051218.
Urban, D.L., Minor, E.S., Treml, E.A., Schick, R.S., 2009. Graph models of habitat
mosaics. Ecol. Lett. 12 (3), 260273.
Uy, P.D., Nakagoshi, N., 2007. Analyzing urban green space pattern and eco-network
in Hanoi, Vietnam. Landscape Ecol. Eng. 3 (2), 143157.
Walston, L.J., Mullin, S.J., 2008. Variation in amount of surrounding forest habi-
tat inuences the initial orientation of juvenile amphibians emigrating from
breeding ponds. Can. J. Zool. 86 (2), 141146.
van Langevelde, F., 2000. Scale of habitat connectivity and colonization in frag-
mented nuthatch populations. Ecography 23 (5), 614622.
Vitousek, P.M., Mooney, H.A., Lubchenco, J., Melillo, J.M., 1997. Human domination
of Earths ecosystems. Science 277 (5325), 494499.
Zetterberg, A., 2009. Network based tools and indicators for landscape ecological
assessments, planning, and design. Licentiate Thesis. Royal Institute of Technol-
ogy, Stockholm.
Zhang, L.Q., Wang, H.Z., 2006. Planning an ecological network of Xiamen Island
(China) using landscape metrics and network analysis. Landscape Urban Plan.
78 (4), 449456.