assessing the capacity of different urban forms to preserve the conncetivity of ecological
TRANSCRIPT
-
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
1/12
Landscape and Urban Planning 105 (2012) 128139
Contents lists available at SciVerse ScienceDirect
Landscape and Urban Planning
journal homepage: www.elsevier .com/ locate / landurbplan
Assessing the capacity ofdifferent urban forms to preserve the connectivity ofecological habitats
Ccile Tannier,Jean-Christophe Foltte, Xavier Girardet
ThMA,CNRS University of Franche-Comt, 32 rueMgevand, F-25 030Besancon Cedex, France
a r t i c l e i n f o
Article history:
Received 24 March 2011
Receivedin revised form 1 December 2011Accepted 15 December 2011
Available online 10 January 2012
Keywords:
Residential development
Spatial simulation
Compact city
Fractal city
Landscape connectivitySpatial indexes
a b s t r a c t
This paper addresses the relationship between anthropogenic forest habitat fragmentation and the formof urbanpatterns. Using a two-step methodology we first generate 40 theoretical residential developmentscenarios following a repeatable procedure; the simulated urban forms are either moderately compact or
fractal. Then, we compare the scenarios according to the functional connectivity ofthe remaining foresthabitat using a graph-based approach. The methodology is applied to the urban region of Besancon
(France), where forest surfaces are considered as a generic habitat for several animal species. Resultsobtained show that fractal scenarios of residential development are almost equivalent to moderately
compact scenarios regarding the connectivity offorest habitat when the residential development is weak.Inthe case ofamore intense residential development,fractal scenarios are superior to nonfractalscenarios
when low dispersal distances ofanimals are concerned. 2011 Elsevier B.V. All rights reserved.
1. Introduction
Managing urban sprawl is a major concern in urban planning
since it has marked negative environmental (air pollution, noise,and destruction of natural resources) and socio-economic (higherhousing and commuting costs leading to social segregation andsocial inequity) effects. Yet urban development is a real necessity
in many countries owing to the growing numbers of inhabitantsand households. Consequently, land consumption is a major plan-ning issue. Given that new residential buildings often requireonly moderate amounts of land compared to the associated road
infrastructures (Camagni, Gibelli, & Rigamonti, 2002), the recur-ring question is where might urban expansion be located withoutworsening the effects of urban sprawl?
One major impact of urban sprawl on natural ecosystems is
the fragmentation of wildlife habitats (Forman, 1995). The spread
of artificial surfaces reduces available habitats through the loss offavorableareasand thebreak-upof theremaining habitat areas intoseparate patches. The viability of a species in a fragmented habi-
tat depends on the ability of individuals to reach one patch fromanother by crossing unsuitable habitat. Consequently, landscapeconnectivity, combined with the size and the quality of habitat
Corresponding author. Tel.: +33 381 66 54 81; fax: +33381 66 53 55.
E-mail addresses: [email protected](C. Tannier),
[email protected] (J.-C. Foltte),
[email protected](X. Girardet).
patches, proves to be a key notion for the conservation of animalspecies. Alongside structural connectivity, functional connectiv-ity is recognized as being ecologically relevant (Taylor, Fahrig, &
With, 2006). Functional connectivity may be defined as the inter-action between a given species and the elements of a landscape.Methods for evaluating functional connectivityoften use landscapemetrics (Magle, Theobald, & Crooks, 2009) or spatial simulation
models (Tischendorf & Fahrig, 2000). In order to set up tools thatare easy for planners and landscape managers to use, we turn tograph theory, which provides a happy compromise between theneed for intensive measurement as part of a biological approach
and the constraints associated with data acquisition (Calabrese &Fagan, 2004; Fall, Fortin, Manseau, & OBrien, 2007). Moreover,graph theory is a preferable alternative to spatially explicit popula-tion models for species conservation in heterogeneous landscapes
(Minor & Urban, 2007).
While numerous studies have analyzed responses of animals toanthropogenic habitat fragmentation, little research has addressedthe relationship between habitat fragmentation and form of urban
patterns (Bierwagen, 2005, 2007). The limited transfer of knowl-edge between the eco-physical and spatial planning domains,underlined by Termorshuizen, Opdam, and van den Brink (2007),may partly explain the lack of knowledge about the relationship
between urban forms and ecological systems. As Alberti (2005)points out, we do not know how clustered versus dispersed andmonocentric versus polycentric urban structures differently affectenvironmental conditions, nor how urban development patterns
influence ecological systems along the gradient of decreasing
0169-2046/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.landurbplan.2011.12.008
http://localhost/var/www/apps/conversion/tmp/scratch_3/dx.doi.org/10.1016/j.landurbplan.2011.12.008http://localhost/var/www/apps/conversion/tmp/scratch_3/dx.doi.org/10.1016/j.landurbplan.2011.12.008http://www.sciencedirect.com/science/journal/01692046http://www.elsevier.com/locate/landurbplanmailto:[email protected]:[email protected]:[email protected]://localhost/var/www/apps/conversion/tmp/scratch_3/dx.doi.org/10.1016/j.landurbplan.2011.12.008http://localhost/var/www/apps/conversion/tmp/scratch_3/dx.doi.org/10.1016/j.landurbplan.2011.12.008mailto:[email protected]:[email protected]:[email protected]://www.elsevier.com/locate/landurbplanhttp://www.sciencedirect.com/science/journal/01692046http://localhost/var/www/apps/conversion/tmp/scratch_3/dx.doi.org/10.1016/j.landurbplan.2011.12.008 -
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
2/12
C. Tannieret al./ Landscape andUrban Planning105 (2012) 128139 129
density from urban center to its periphery. Alberti (2005) also
remarks that ecological studies dealing with urbanization simplifythe consideration of urban structures to such an extent that theresults are no longer useful to urban planners and managers. Forexample, Tratalos, Fuller, Warren, Davies, and Gaston (2007) com-
pare and contrast several urban density measures with a series ofmeasures of environmental quality and biodiversitypotential. Theirstudy yields no conclusive results and shows that similar urbanforms may induce a varying environmental quality. Conversely,
Bierwagen (2005) shows that urban areas that differ visually maynonetheless have similar ecological connectivity scores. Her studyalso shows that ecological connectivity declines with the increas-ing size of the urban area. However, it cannot be inferred from such
a statistical relationship that there is any functional relationshipbetween certain characteristics of urban forms and the ecologicalsystem.Urban forms arehighly complex,whichmay account forthedifficulty in identifying key variables for use as a lever for wildlife
conservation. To overcome this difficulty, it may be helpful to workon simulated rather than real-world urban forms, since their mor-phologicalcharacteristics can be controlled.For instance, GeursandvanWee (2006) use a system calledEnvironment Explorer in which
land-use and transport modules are dynamically interconnected tosimulate scenarios of urban development. The 500m resolution of
the land-use cells, however, was too coarse for precise measure-ments of environmental impacts at local level.
In this paper, we aim to better understand how differentpatterns of residential development may impact the shape of animal habitats, and therefore affect their connectivity. Two cat-egories of built patterns are considered: compact built patterns
characterized by high built densities, uniformity, and sharp (i.e.non-sprawling) boundaries (Geurs & van Wee, 2006); fractal builtpatterns that are intrinsically nonuniform across scales, and exhibitlonger and more sinuous boundaries (Frankhauser, 2004). In urban
planning, the compact city model is the common answer tothe problem of urban sprawl. But the models limitations havebeen pointed out, especially the congested roads, reduced accessto green and natural areas, higher housing prices, and reduced
living space (Breheny, 1992; Burton, 2000). Accordingly an alter-native urban model has come to the fore combining reasonabledensification, as in the wisely compact city (Camagni et al. ,2002), and a polycentric urban organization (Davoudi, 2003). The
fractal city model, in keeping with this tendency, appears promis-ing since several commentators have suggested that the fractalcity could satisfy people who consume various urban and ruralamenities by improving access to both built and nonbuilt spaces
(Cavailhs, Frankhauser, Peeters, & Thomas, 2004; Frankhauser,2004).
In this paper,we adopt a two-step method.First,we generate 40theoretical scenarios of residential development using a repeatable
procedure that explicitly takes into account fractal or nonfractalurban development models. Then we compare the scenarios in
terms of the functional connectivity of the remaining habitat usinga graph-based approach. Our aim is to identify the urban form thatbest preserves habitat connectivity.
2. Studyareaanddata
The studyarea includes the cityof Besancon and its metropolitanarea in eastern France(Fig. 1). With a surface areaof 116 827ha,the
studyareanumbers about234 000inhabitants.Except forthe urbancore, the study area is not densely urbanized but urbanization istending to grow. This provides considerable scope for simulatingnumerous scenarios of residential development.
Forested zones, threatened by urban sprawl, dominate the
landscape. They provide a suitable ecological habitat for several
Fig. 1. Metropolitan area of Besancon,France (4714N, 601E).
mammal species whose home range lies within the study area.Examples of target species relevant for ourstudy arethe Red Squir-rel (Sciurus vulgaris): maximal dispersal distance of 1.5 km (Haleet al., 2001), the Western Barbastelle (Barbastella barbastellus):
maximal dispersal distance of 10km (Sierro & Arlettaz, 1997), theLynx (Lynx lynx): maximal dispersal distance of 40km (Kramer-Schadt, Revilla, Wiegand, & Breitenmoser, 2004), and the CommonGenet (Genetta genetta): maximal dispersal distance of 90km
(Zuberogoitia, Zabala, Garin, & Aihartza, 2002). Those species sharethe following characteristics: (1) they preferentially use forestedlandscape elements; (2) they have difficulties moving in nonforest
landscapes; and (3) they are threatened by the extension of builtareas.
The land-cover data used are derived from the BD Topo vec-tor database provided by the Institut Gographique National (IGN2009, BD TOPO http://professionnels.ign.fr/ficheProduitCMS.
do?idDoc=5667214). The BD Topo includes forest and buildingsurfaces as well as the road network. Maps of forest habitat wereobtained by rasterizing the forest layer of the BD Topo. To pre-serve a high level of spatial detail, rasterizing was performed at a
resolution of 20m (0.04ha per grid cell).
3. Generation of scenarios of residential development
Forty scenarios of residential development were created using
MUP-City 0.5.3 software (Tannier, Vuidel, Houot, & Frankhauser, inpress). MUP-City can be used to generate residential developmentscenarios starting from an existing built pattern. The creation ofnew residential locations is simulated, but not the creation of newroadsthat often accompanies them. MUP-City requires two types of
dataas its input: detailed road network(lines) and buildings (poly-gons). As its output, MUP-City provides a raster map with threetypes of cells: initially built, newly built (simulated), and nonbuilt(natural or artificial land uses).
The simulated scenarios may take into account two planningrules.The first rule is that each newresidential cell hasto be locatedclose to both a nonbuilt cell and a built cell; moreover, buildingthe cell must not impede access to open spaces for neighboring
built cells. This planning rule is designed to satisfy house-
hold preferences by offering a rural environment while ensuring
http://professionnels.ign.fr/ficheProduitCMS.do?idDoc=5667214http://professionnels.ign.fr/ficheProduitCMS.do?idDoc=5667214http://professionnels.ign.fr/ficheProduitCMS.do?idDoc=5667214http://professionnels.ign.fr/ficheProduitCMS.do?idDoc=5667214 -
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
3/12
130 C. Tannier et al. / LandscapeandUrban Planning105 (2012) 128139
Fig. 2. Multi-scale spatial modeling.
possible interactions with neighboring households (Caruso,Peeters, Cavailhs, & Rounsevell, 2007). It also answers two plan-
ning objectives, which areto reduce automobile trips (for accessingopen spaces), and to limit landscape fragmentation.
Thesecond planning rule says that new residential cells have tobe as close as possible to existing roads in order to limit the use ofspace. Land consumption for road infrastructure is extremely high,representing 25% of the total urban area in Europe and 30% in the
United States (Camagni et al., 2002).Formalization of the two planning rules is based on assessment
criteria taking the form of fuzzy variables ranging from 0 (bad) to 1(good). Forrule #1 Proximity to built andopen spaces, the assess-
ment criterionis the numberof nonbuilt cells aroundeach built celldirectly contiguous to the cell under assessment (in a 33 Mooreneighborhood). A fuzzy variable (x) describes the assessment cri-terion through a membership value to the good evaluation fuzzy
set:(x) =x/34with(x) [0;1],andxbeingthenumberofnonbuiltcells contiguous to at least one nonbuilt cell in the neighborhoodof the cell being assessed.
This formulation refers to the fact that a cell counts no more
than 34 nonbuilt cells in its 33 neighborhood (Tannier et al., inpress).
The assessment criterion for rule #2 Proximity to existingroads is the distance to the closest road. When the cell being
assessed iseithercrossedby a roador close toa roadthe value oftheassessment criterion is 1, and it declines as the distance (measuredin cells) increases.
Either one or both of the planning rules were applied under
the different scenarios. The results obtained are either one or twoassessment values between 0 and 1. The arithmetic mean of these
values yields a synthetic evaluation of the usefulness of each cellfor residential use. Ultimately, the system dynamically selects the
cells it would be most beneficial to urbanize. When there is a tiebetween cells, one is chosen at random.
3.1. Fractal scenarios of residential development
For the creation of fractal development scenarios, planning rules
have been applied considering a series of nested scales. Initially,the study area is covered by a regular coarse-grained grid. Eachgrid square contains a fixed number of cells determined by areduction factor r. We then apply multi-scale modeling that con-
sists in reducing the size of the grid square from one level of
analysis to the next (Fig. 2). Initially, the grid square size is l1. At
the next level of analysis, each grid square of size l1 is subdividedinto grid squares of size l2, corresponding to the cells of level l1.
l2 =
1
r
l1
This procedure of decomposing the grid squares into cells isreiterated until the cell size is close to that of the buildings (here20m).
From a fractal point of view, twoparameters determine the self-
similarity dimension D of a pattern: the reduction factor rand thenumber of elements N(r) (Mandelbrot, 1982).
D =log(N(r))
log(r)
Different values of fractal dimension can be used to differen-tiate built patterns with different topological properties: related,
unrelated, or partially related across scales (De Keersmaecker,Frankhauser, & Thomas, 2003). In MUP-City, the choice of a fractaldimension of the future built pattern is determined by the maxi-mum numberNmaxof cells that can be built per grid square and the
reduction factor r. For the current application, the reduction factor
ris always 3. By fractal logic, if a grid square of size l1 is not built,building is prohibited in the cells of size l2 belonging to that gridsquare.
This multi-scale fractal modeling differs from the iterated func-tion systems ofBarnsley (1988) used for example by Milne (1991)to create fractal landscapes. It also differs from the midpoint dis-placement algorithm ofSaupe (1988) adopted by Gardner (1999)
for creating multi-scale maps with RULE software.Following the application of the fractal rule of urbanization, we
know the number of cells in each grid square that can potentiallybe urbanized. The two planning rules are then used to select whichcells to urbanize among the eligible cells. Multi-scale modelingimplies that each planning rule is applied considering each levelof analysis, i.e. each cell size successively. In the case of planning
rule #2, the calculation of the assessment criterion varies with thesize of cell considered (Table 1).
Fifteen fractal scenarios have been simulated. Nmax varies from3 to 7, corresponding to a fractal dimension varying from 1 to 1.8.
The form of the simulated patterns is typically fractal, being nei-ther dense nordispersed (Fig.3). This is consistentwith a numberofpublicationsshowing that urban growth engenders a fractal spatialorganization (Batty & Xie, 1996; Benguigui, Czamanski, Marinov,
& Portugali, 2000). Scenarios that take into account planning rule
#1 generate more compact built patterns than the other fractal
-
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
4/12
C. Tannieret al./ Landscape andUrban Planning105 (2012) 128139 131
Table 1
Assessment values for planning rule #2 proximity to existing roads.
Distance to nearest
road (number of
cells)
Size of cells (m)
>500 500200 20050 5020 200
0 1 1 1 1 1
1 0 0.5 0.67 0.75 0.8
2 0 0 0.33 0.5 0.6
3 0 0 0 0.25 0.4
4 0 0 0 0 0.2
5 0 0 0 0 0
scenarios.Scenariosthat take into account planning rule #2 exhibitmore elongated built forms.
3.2. Nonfractal scenarios of residential development
In creating nonfractal development scenarios, planning ruleshave been applied at a single scale corresponding to a grid of cellswith sides 20m long. The number of cells worth urbanizing was
set a priori as the number of cells worth urbanizing identified by
MUP-City for each of the 15 corresponding fractal scenarios. Non-fractal simulated built patterns combine moderately compact andlinear developments. Moderately compact developments are char-
acterized by the presence of nonbuilt cells inside each built cluster.Fewer and smaller open spaces are preserved within the built pat-tern than is the case in fractal scenarios. Linear extensions are ina straight line, in the case of planning rule #1, or along existing
roads, in the case of planning rule #2 (Fig. 3). Real-world patternsof the sortare found for example in Belgium (Thomas, Frankhauser,& Biernacki, 2008) and inItaly(Camagni et al., 2002).
3.3. Neutral scenarios of residential development
Neutral landscape patterns are usually created starting from a
blank or randomized initial situation (Gardner, Milne, Turner, &ONeill, 1987). Since they do not account for the effect of the initial
situation, we created neutral landscape scenarios starting from theinitial land-use pattern (Hagen-Zanker & Lajoie, 2008). Five neu-tral nonfractal scenarios were created by locating newly built cellsrandomly. Neither the multi-scale fractal modeling nor the plan-
ning rules were applied. Five neutral fractal scenarios were createdby applying the multi-scale fractal modeling but not the planningrules. Instead, a simple random function was used to position thenewly built cells in each grid square. Neutral nonfractal scenarios
generate purely dispersed built patterns. Neutral fractal scenar-ios generate the most dispersed instances of all the fractal builtpatterns (Fig. 3).
4. Assessing change in habitat connectivity
4.1. Defining the landscape map
Starting from the raster map of forest habitat, forest patches
were identified using the Morphological Spatial Pattern Anal-ysis (MSPA) available in the free software package GUIDOS(http://forest.jrc.ec.europa.eu/biodiversity/GUIDOS/). MSPA usesmathematical morphology to classify structural patterns on a
binary map of land cover (Vogt et al., 2007). The input map is com-posed of a foreground, which is the focal habitat (here the forestland cover), anda complementarybackground.The method appliesa sequence of morphological operators (erosion, dilation and skele-
tonization) using a square-like structuring element. The size of this
structuring element (edge width in MSPA) was defined at two
cells (40m). Only the forest cells surrounded with at least 40m of
other forest cells were considered to be patch cores.Seven classes of landscape elements are identified by MSPA
(Vogt et al., 2007): core, islet, bridge, loop, branch, edge, and perfo-ration. MSPA classes have to be interpreted as classes of functional
connectivity to create a structural map of potential connectivity(Vogt et al., 2009). In this perspective, forest surfaces were splitinto two classes: forest patches, resulting from the merging ofthe MSPA core, perforation, and edge classes, and other for-
est surfaces favorable to animal movements defined as the MSPAbranch, bridge, loop, and islet classes. Patches of less than1 ha (containing only one core cell surrounded by edge cells)were reclassified as surfaces favorable to animal movements. This
minimum patch size is somewhat smaller than in other studies:5hain Minor and Urban (2007), 10hain Vogt et al. (2009); 25hainVogt et al. (2007); however, it seems to be appropriate in view ofboththe fine-grained data used andthe generic function associated
with the habitat patches.When overlaying the forest habitat map with a scenario map
of residential development, a synthetic landscape map is obtainedthat includes the simulated built pattern and the remaining forest
pattern. To compare the landscape maps, we followed the threeprinciples set out by Gardner (1999): map grain and extent must
be the same; maps must be of sufficient size to minimize boundaryeffects; comparisons must account for differences in the propor-
tion of the landscape covered by habitat, designated by p below(Gardner et al., 1987).
4.2. Patch-based graphs of landscape connectivity
Each landscape map was used as an input data layer in graph-based modeling of connectivity. Graphs were built by using forestpatches as nodes having a two-dimensional geometry (Galpern,
Manseau, & Fall, 2011), and by computing edge-to-edge least-costdistances betweenall pairs of patches to define the links. Least-costdistances were calculated using resistance values assigned to eachland cover class:
- habitat patch: 1
- forest surface favorable to movement: 1- built surface: 10- background: 5
Sinceour approach is generic, theresistance valuesweredefinedby assuming that built surfaces involve the higher resistance toindividual movement, and that background (surfaces other than
forest) plays an intermediary role.The scenarios were compared by assessing their global connec-
tivity but nottheir local connectivity.For further explanation aboutthechoiceof an analysis level see(Rayfield, Fortin,& Fall, 2011). The
analysis was based on the complete graph, containing all the infor-mationabout thepotentialpathsbetween thenodes (Galpern et al.,2011). Many metrics can be applied to estimate connectivity at theentire graph level. Direct metrics are more relevant when dealing
with changes in habitat connectivity since they take into accountthe graph structure. Among all existing direct metrics, we chosethe probability of connectivity (PC) index proposed by Saura andPascual-Hortal (2007). The PC index measures the probability that
an animal will remain in the same set of connected patches whenmoving:
PC=
ni=1
naiajp
ij
A2
where n is the total number of patches, ai and aj are the areas of
the patches i and j, p
ij is the maximal probability of the potential
http://forest.jrc.ec.europa.eu/biodiversity/GUIDOS/http://forest.jrc.ec.europa.eu/biodiversity/GUIDOS/ -
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
5/12
132 C. Tannier et al. / LandscapeandUrban Planning105 (2012) 128139
Fig. 3. Types of residential development patterns generated by simulation focus on a 2052 ha zone north-west of Besancon. Nmaxis 5 (14% of thelandscape is built-up).
-
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
6/12
C. Tannieret al./ Landscape andUrban Planning105 (2012) 128139 133
Table 2
Percentage of land cover classes calculated for each series of scenarios.
Forest
(p)
Built
initially
Built
simulated
Total
built
Background
Initial landscape 0.43 0.05 0.00 0.05 0.52
Simulated landscapes
Nmax=3 0.42 0.05 0.01 0.07 0.51Nmax=4 0.41 0.05 0.04 0.09 0.50
Nmax=5 0.40 0.05 0.09 0.14 0.47
Nmax=6 0.36 0.05 0.17 0.22 0.42Nmax=7 0.30 0.05 0.31 0.37 0.33
paths between i andj, andA is the total area under study.pijcan becomputed with an exponential function such that:
pij = ekdij
where dijis the least cost distance between i andj, and k (0
-
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
7/12
134 C. Tannier et al. / LandscapeandUrban Planning105 (2012) 128139
Fig. 4. Initial landscape. The built area represents 5% of thetotalarea.
forest pattern in nonfractal scenarios than in fractal scenarios.
Considering the neutral scenarios, fewer differences are observedbetween neutral fractal scenarios and other fractal scenarios thanbetween neutral nonfractal scenarios and other nonfractal scenar-ios. Neutral nonfractal scenarios exhibit a residential development
full of holes and the remaining forest patches are very small. Theyare numerous when residential development is only slight; theyare very sparse, even absent, when the residential development is
intense.Only three fractal scenarios are characterized by better values
of basic indexes than the nonfractal scenarios of the same series(Table 3). Here, we consider that a scenario is better when it isnearer to the initial situation. When the built area represents 14%
of the total study area, the fractal scenario with planning rule #2has the highest maximum patch size; the fractal scenario withboth planning rules has the fewest patches. When the built arearepresents 0.09% of the total study area, the fractal scenario with
planning rule #2 has the highest maximum patch size.Figs. 46 display three examples of the maps obtained. The ini-
tial landscape exhibits local variations in the form of the foresthabitat. In the North, forest zones favorable to animal movements
are mainly located in the background. In the South-West, habitat
patches and zones favorable to animal movements are inter-penetrating. In the East of the study area, patches of backgroundoccur withinthe large habitat patches. Simulated residential devel-
opment essentially concentrates on background cells of the initiallandscape. A fractal residential development creates new back-ground cells surrounded by forest zones favorable to movementswithin habitat patches (Fig. 5). A nonfractal residential develop-
ment creates ribbon-likebackground or builtcells corresponding totheroads along which the residential development is concentrated.The ribbons are not surrounded by forest zones favorable to animalmovements (Fig. 6). Figs. 5 and 6 nicely illustrate the crucial dif-
ference between fractal and nonfractal urban patterns: a clear-cutlimit between built and nonbuilt patterns characterizes nonfractalurban forms; a fuzzy limit marked by the development of zones
favorable to animal movements characterizes fractal urban forms.
Fig. 5. Simulated landscape resulting from a fractal residential development con-
strained by twoplanning rules.The built area represents 14% of thetotalarea.
Fig. 6. Simulated landscape resulting from a nonfractal residential development
constrained by two planning rules.The built area represents 14% of thetotalarea.
5.2. Assessing the loss of connectivity of animal habitat
The complete graphs that model the 40 landscape maps counta maximum of 2361 nodes (2785 980 links). Fig. 7 displays thesquare root of the PC index (sqrPC) for each series of scenarios.
It shows that forest connectivity decreases as the number of built
cells increases. The maximal value of the sqrPC is 0.33 when the
-
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
8/12
C. Tannieret al./ Landscape andUrban Planning105 (2012) 128139 135
Fig. 7. sqrPC values foreach series of scenarios.
fraction of built cells is 0.07%, and 0.26 when the fraction of built
cells reaches 0.37%. This decrease in the sqrPC varies in the differ-ent scenarios. In general, the PC index decreases more rapidly thanthe fraction of forest. This reveals a clear influence of the shape of
the residential development on habitat connectivity.
Nonfractal scenarios with planning rule #2, with or without
rule 1, obtainthe best sqrPC value,and their curvesare very similar.Fig. 7, however, shows slight differences in sqrPC values betweenfractal and nonfractal scenarios when urbanization is not intense
(from 0.07% to 0.09% of built cells in the landscape). In nonfractal
-
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
9/12
136 C. Tannier et al. / LandscapeandUrban Planning105 (2012) 128139
Fig. 8. Changerate of thePC index for short dispersaldistances.
-
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
10/12
C. Tannieret al./ Landscape andUrban Planning105 (2012) 128139 137
scenarios, the connectivity in scenarios with planning rule #2,
combined or not with the rule #1, is better than the connectivityin scenarios omitting this rule. Conversely, in fractal scenarios,the connectivity in scenarios with planning rule #1, combinedor not with rule #2, is better than the connectivity in scenarios
without this rule. The neutral random scenarios, and especiallythe nonfractal random scenarios, exhibit little connectivity.
For the same intensity of urbanization, the ranking of the sce-narios may vary with the maximum dispersal distance considered.
When the intensityof urbanizationis lowor quite lowand forshortdispersaldistances (less than 1250m when thepercentage ofnewlybuilt cells in the landscape is 0.09; less than 2000 m when the per-centage of newly built cells is 0.14), fractal scenarios have a better
PCindexthan nonfractal scenarios (Fig.8). However, thePC index ofnonfractalscenariosincreasesdramatically for longerdispersaldis-tances and finally exceeds the PC index of fractal scenarios. Whenthe number of built cells is highest (between 0.22% and 0.37% of
built cells in the landscape) fractal scenarios never have a betterPC index than nonfractalscenarios whatever the dispersal distanceconsidered.
6. Discussion andperspectives
6.1. Reliability and interest of the methodology for urban
planning and design
We proposed a simple and repeatable method requiring little inthe way of data and parameters:
Create residential development scenarios using MUP-City soft-
ware; Use the MSPA method for identifying habitat patches; Calculate basic spatial indexes of habitat fragmentation; Calculate the PC index for assessing the global habitat connectiv-
ity.
We systematically explored the effect of two urban models, twoplanning rules, and a varying intensity of urbanization on certainstructural and functional aspects of ecological habitat. The simu-lated scenarios of residential development were theoretical, more
orlesscompactand linear, andfractalor nonfractal.In orderto iden-tify general rules for landscape planning and design, we created 40scenarios of residential development allowing us to simulate andanalyze a progressive decline in the amount of ecological habitat in
the landscape, responding to the requirement of a generalized gra-dient of habitat pattern in view of an explanatory theory (Gardner,1999). Except forthe neutral scenarios,the simulated scenarios rep-resent stylized facts which attempt to mimic observed phenomena.
They are unrealistic because they omit environmental constraints(slopes, soil types, etc.), planning laws and regulations, and even
householdbehaviors. But they do allow us to test the impacts of thevariables determining a built form on the surrounding pattern ofhabitat, all other things being equal. Whereas other scholars sim-ulate landscape of habitats (Bierwagen, 2007; Gardner & Urban,2007), we simulate landscape of residential development. We con-
centrate on the urban development process (more or less intense,fractal or not) to provide new insights into the transformation ofa single ecological condition i.e. connectivity between patches.Bierwagen (2007) stated that the change in habitat connectivity
caused by urbanization is relatively large for forest habitat patternsinitially exhibiting a high level of aggregation. Our results are inaccordance with this.
The urban modeling used for creating the residential develop-
ment scenarios clearly differs from cellular automata modeling of
urban growth: it is morphologically explicit but not predictive.
Modeling stylized facts allowed us to create 40 scenarios of res-
idential development based on the systematic variations of just afew key variables. Conversely, cellular automata applications allowthe creation of only a few scenarios: for instance two in Mitsova,Shuster, and Wang (2011), three in Syphard, Clarke, and Franklin
(2005) and Aguilera, Valenzuela,and Botequilha-Leito(2011), andfive in Herold, Couclelis, and Clarke (2005). The comparison of eco-logical habitat patterns resulting from simulated scenarios of urbangrowth is most often based on landscape metrics potentially asso-
ciated with connectivity indexes (Mitsova et al., 2011). We chosethe same indexes forcomparing oursimulated scenarios.However,we went deeperinto the assessmentof habitat connectivity by con-sidering a wide range of dispersaldistances(Bierwagen, 2007). This
allowed us to measure a general mobility potential for a range ofanimal species by analyzing a generic ecological habitat, know-ing that the effective mobility of each of the considered speciesis different.
The study ofBierwagen (2007) suggests that certain aspects ofurban configurations can be used to increase connectivity in somelandscapes. This implies that policies to modify patterns of urbangrowth can play an important role in maintaining or restoring con-
nectivity. Our study shows that, in general, habitat connectivityseems to be better preserved in nonfractal scenarios of residen-
tial development than in fractal scenarios. However, for speciescharacterized by low dispersal distances, some fractal scenarios
may be more favorable to their mobility. In the light of theseresults, our study confirms that the relationship between urbanform and ecological processes is equivocal (Bierwagen, 2007). Itseems impossible to identify a single threshold or a unique rule for
residential development with which one can conserve all specieslivingin a landscape(With & Crist,1995). Different planning strate-gies should be used to increase the likelihood of persistence fordifferent groups of species (Minor & Lookingbill, 2010). This chal-
lenges the idea that greater dispersal distances offset some ofthe changes in connectivity caused by urbanization (Bierwagen,2007). To sum up, our study shows that urbanizing along theroads combined with a moderately compact urbanization may
be a good way to preserve habitat connectivity and should bemore carefully tested. However, focusing on animal species havingshort dispersal distances, some fractal forms of residential devel-opment best preserve habitat connectivity presumably because
they increase the number of forest zones available for animalmovements.
The fragmentation effect of the road network is taken intoaccount through the decrease in patch area and the isolation of
resource patches. The road networkis indirectly taken into accountthrough planning rule #2 which favors residential developmentclose to existing roads. No friction value is assigned to the roadsin the calculation of least-cost distances between patches. As in
Mitsova et al. (2011) and Aguilera et al. (2011), scenarios are basedon the actualroad network, but its extension in relation to residen-
tial development is not simulated. No friction value is assigned totheroadsin thecalculationof least-costdistances betweenpatches.Many variables other than those taken into account in our researchinfluence the way roads affect the environment, in particular roadwidth, equipment (i.e. fences), and volume of traffic (Coffin, 2007).
The barrier effect of roads may be stronger for some species thanothers (Minor & Lookingbill, 2010) and the ecological role of theroad network varies in recent studies in the field: Vasas, Magura,
Jordn, and Tthmrsz (2009) considerroads to be absolute barri-
ers for forest carabid species; Fall et al.(2007) basethe friction valueof roads on expert opinion concerning their effect on woodlandcaribou; Zetterberg, Mortberg, and Balfors (2010) assign roads to aclass with friction value related to energyexpenditure alone,rather
than mortalityrisk.These recent works support thehypothesis that
the effect of roads is species-specific.
-
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
11/12
138 C. Tannier et al. / LandscapeandUrban Planning105 (2012) 128139
6.2. Going deeper into the assessment of landscape connectivity
We defined the habitat patches by morphological preprocess-ing applied to the focal habitat class (forest surfaces), using MSPA.This preprocessing is justified by the assumption that a core zone
is required to definehabitat patches in a relevant manner, the edgepart of a patchbeing subjected to several disturbances such as pre-dationor anthropicactivities. This assumption is probably notvalidin the case of some species which are less sensitive to these distur-
bances, but the assumption seems to be justified in our genericapproach. Nevertheless, two critical points have to be outlinedwith the use of MSPA. The first point concerns the edge widthparameter of MSPA, which defines the size of the structural ele-
ment used in the erosion/dilation process to distinguish the coreclass from all other classes. The edge width plays a prominentrole in defining habitat patches and in making derived connec-tivity measurements. This point is discussed in Ostapowicz, Vogt,
Riiters, Kozak,and Estreguil (2008), Riitters, Vogt, Soille, Kozak, andEstreguil (2007), and Vogt et al. (2007). In comparison with theedge width of 40m used in the present study, Vogt et al. (2009)choose an edge width of 75 m , arguing that this value is typi-
cal for a wide range of species. However, it appears difficult torepresent the relationships between species and the disturbances
they perceive from the matrix by a single value of edge width.In this respect, our choice of a width of 40 m may restrict the
generic value of the results. As suggested in Vogt et al. (2009), itwould be useful to perform a multi-scale sensitivity study, withthe help of expert knowledge of species, to identify species-specificscales for which it would be possible to set an appropriate edge
width.The second critical point regarding the use of MSPA concerns
the merging of the seven morphological classes. In our study, thehabitat patches resulted from the merging of the core, edge,
and perforation classes, as in Vogt et al. (2009). All the otherMSPA classes were considered to be zones favorable to move-ment, whichdiffers somewhatfrom previous works wherethe classislet was removed and where only the loop class contributed to
the connector features. As these classifications induce differencesin the computed least-cost distances, the resistance values or costassigned to each class needs further consideration. This question ofcost values may indeed contribute to increasing the degree of error
in a graph model (Rayfield, Fortin, & Fall, 2010; Spear, Balkenhol,Fortin, McRae, & Scribner, 2010). Field observations could be usedas guideline to define the costs (Belisle, 2005) but this methodoften raises practical problems and cannot by applied in a generic
approach.One solutionmightbe totake intoaccountmultipleleast-cost pathways as suggested in Rayfield et al. (2010).
The scenarios of residential development were compared con-sidering a global assessment of forest patch connectivity. The PC
index was used because of its capacity to include several basicaspects of ecological connectivity (distances between patches, dis-
persal distance, and patch areas) while having numerous expectedproperties favoring a high level of connectivity (Saura & Pascual-Hortal, 2007). In the present study, the choice of the PC index isconsistent with the use of a complete graph. However, one couldargue that the comparison made here remains at a global level,
with no access to more precise aspects of connectivity. To supple-ment the analysis, several courses of action might be investigated.One possibility could be to focus on the spatial impact of the lossof connectivity induced by residential development. By applying
the PC index locally as a delta parameter (Saura & Pascual-Hortal,2007; Urban & Keitt, 2001), it would be possible to determine theplaces where a given scenario of residential development couldexert a stronger effect on the global connectivity. Another possi-
bility would be to distinguish aspects of connectivity which refer
only to the loss of patch areas (PC intra) from aspects that refer to
the loss of potential dispersal flux (PC flux), or the loss of potential
crossing (PC connector) (Saura & Rubio, 2010).Fractal scenarios sometimes exhibit better connectivity than
nonfractal scenarios when the dispersal distances are shortwhereas the inverse phenomenon occurs for long dispersal dis-
tances. This observation may support the hypothesis that a fractalurban form allows local animal movements from one patch toanother, but does not facilitate crossing of the entire landscape.Conversely, the existence of continuous ribbons of buildings along
roads in nonfractal scenarios creates a partitioningof theforest pat-tern. Asa result we mayhypothesize a strong barrier effect of roadsthat is not highlighted by global connectivity indexes, measuredat the landscape level, but that could be revealed by patch-level
connectivity metrics.
7. Conclusion
In this paper, we have explored the relationship between urbanforms and ecological processes. We showed that the decrease inhabitat connectivity caused by fractal or nonfractal scenarios of
residential development is almost the same when urbanization isnot intense. In this case, a fractal residential development may be
as helpful as a wisely compact development in maintaining bio-diversity. With more intense residential development, nonfractal
scenarios are better than fractal scenarios except when consider-ing short dispersal distances. The effect of two planning rules onthe landscape connectivity varies with the form of the simulatedresidential development. Planning rule #2 Proximity to existing
roads improves the landscape connectivity of nonfractal scenarioswhereas the landscape connectivity of fractal scenarios is mainlyimproved by planning rule #1 Proximity to built and open spaces.Finally, our research suggests that the interest of each urban form
and each planning rule for biodiversity conservation varies withboth the dispersal distances of animal species and the intensityof urbanization. The assessment of the landscape connectivity foreach scenario was, however, only global. It would be worthwhile
comparing the simulated landscapes more closely by analyzingbarrier effects of roads, identifying local variations of connectiv-ity within each landscape, and distinguishing different aspects oflandscape connectivity (e.g. loss of landscape traversability versus
loss of habitat patch area).Our research provides some insight into how to combine the
well-being of animals and humans with the aim of achievingsustainableurban development.Factors of human residential satis-
faction that have been taken into account arethe proximityto greenareas and to other individuals. The results obtained suggest thatthere is no one optimal solution for landscape and urban planning.The choice of a solution (here one form of residential development
associated with one or two planning rules) will necessarily resultfrom a compromise. The difficulty in reaching a compromise will
increase if more types of ecological organisms (e.g. plants) and ahigher diversity of human needs (e.g. access to local and centralservices) are taken into account.
Acknowledgements
The software application MUP-City has been developed in theframework of the French program PREDIT (research program on
innovation in transport), funded by the French Ministry of Ecology,Energy, Sustainable Development and Sea. The graph analysis wasconducted using the software Graphab, developed by GillesVuidel(UMR 6049 ThMA), in the framework of the Graphab project of
the USR 3124 MSHE Ledoux, funded by theFrench Ministry of Ecol-
ogy, Energy, Sustainable Development and Sea. Computations have
-
7/25/2019 Assessing the Capacity of Different Urban Forms to Preserve the Conncetivity of Ecological
12/12
http://dx.doi.org/10.1068/b37132