Assessing functional connectivity: a landscape approach for handling multiple ecological requirements


  • Anne Mimet,

    Corresponding author
    1. Conservation des espèces, Restauration et Suivi des Populations, UMR MNHN-CNRS-UPCM, UMR 7204, Paris, France
    • Laboratoire Dynamiques Sociales et Recomposition des Espaces, UMR CNRS-Paris 1- Paris 7- Paris 8- Paris 10, Paris, France
    Search for more papers by this author
  • Thomas Houet,

    1. Laboratoire GEODE, Géographie de l'Environnement, Toulouse Cedex 1, France
    Search for more papers by this author
  • Romain Julliard,

    1. Conservation des espèces, Restauration et Suivi des Populations, UMR MNHN-CNRS-UPCM, UMR 7204, Paris, France
    Search for more papers by this author
  • Laurent Simon

    1. Laboratoire Dynamiques Sociales et Recomposition des Espaces, UMR CNRS-Paris 1- Paris 7- Paris 8- Paris 10, Paris, France
    Search for more papers by this author

Corresponding author. E-mail:


  1. The improvement of tools for protecting biodiversity requires integrating habitat connectivity to build efficient ecological networks that facilitate the movement of species under pressure from global change. Several methodological and scientific challenges are faced in constructing such networks. First, ecological networks need to incorporate habitat connectivity for species with different ecological requirements. Secondly, the networks should be based on functional connectivity rather than on structural connectivity alone. Thirdly, connectivity needs to be treated as a continuous variable.
  2. We propose a non-oriented approach of landscape description to identify favourable areas and measure functional connectivity for multi-specific applications, using three groups of common bird species (farmland specialists, forest specialists and generalists) as indicators of biodiversity.
  3. In the highly anthropized region of Seine-et-Marne, we defined 20 landscape types based on composition and configuration. We used statistical modelling to obtain a value of favourability for each landscape type for each bird group. We then mapped landscape favourability, for the three groups in 1982 and 2003 to identify favourable entities (adjacent favourable landscape units) and determine connectivity. We then examined temporal changes in the favourable areas and their connectivity and determined the sensitivity of the favourable landscape types to land cover change.
  4. Composition and configuration both influenced landscape favourability. Some landscape types were favourable for several groups of species and could potentially serve as junction landscapes in ecological networks that accommodate a variety of ecological requirements. Increasing urbanization and fragmentation between 1982 and 2003 resulted in a decrease in favourable landscape units, as well as consequent decreases in favourable areas and connectivity, for the three species groups. Connectivity loss was greatest for farmland and generalist species, as it was already high for forest species in 1982.
  5. Such a non-oriented landscape description could be used to delineate multi-specific ecological networks at regional and national scales and could be further developed to study the connectivity of communities. The maps of favourability produced here could also be used in combination with other methods, such as graphs or circuits, to detect ecological corridors and stepping stones to habitat connectivity.


Anthropogenic land use and land cover change (LUCC) have led to a global biodiversity crisis (Pimm et al. 1995) through the degradation of ecosystem functionality (Hector & Bagchi 2007) and the depletion of ecosystem services (Chapin et al. 2000). This loss of biodiversity has provoked the implementation of various conservation policies. Since the introduction of the Habitat Directive in Europe in 1992, policies have been developed with the aim of building an ecological network composed of protected areas based on valuable species, habitats or ecosystems (Natura 2000). Recent studies show that the ecological network developed as a results of these policies also protects common species (Devictor et al. 2007) that are recognized as important for ecosystem services and human well-being (Gaston & Fuller 2007). Indeed, since the 1980s, the importance of connectivity between habitat patches for biodiversity conservation has been demonstrated numerous times (Taylor et al. 1993; Devictor, Julliard & Jiguet 2008; Grashof-Bokdam et al. 2009). However, the connectivity between protected areas is not easy to take into account because of the variation in ecological requirements for different species and because corridors that enhance connectivity for some species may serve as barriers for others. The diversity in ecological requirements implies that developing an ecological network for multiple species with different ecological requirements or community types is a challenge. Despite these challenges, we argue that an efficient ecological network should enhance connectivity for all non-invasive species.

Connectivity is ‘the degree to which the landscape facilitates or impedes movement along resource patches’ (Taylor et al. 1993) and is both species-specific and landscape-specific (Tischendorf & Fahrig 2000). Connectivity depends on the characteristics of the habitat patches and the distance between patches (Ewers & Didham 2006) but also on the suitability and permeability of the matrix (Powney et al. 2011; Vergara 2011). Landscape connectivity is also dependent on some landscape characteristics, which modify interspecific relationships (Ewers & Didham 2006; Wakano et al. 2011) and mortality risks (Tischendorf & Fahrig 2000). Thus, species success or failure depends on features of landscape patches and landscape characteristics that need to be taken into account when estimating connectivity (Kadoya 2009; Sawyer, Epps & Brashares 2011). Using movement data to estimate connectivity within a species' territory requires very important logistical and economical resources (Zeller, McGarigal & Whiteley 2012), which become even more important when multi-species connectivity is considered. For these reasons, connectivity is often inferred from resistance values computed by regression of species distribution on landscape characteristics (With 1997; Chetkiewicz & Boyce 2009). The underlying hypothesis of this method is that the species distribution in a given landscape is a proxy for connectivity. This hypothesis states that the absence of a species in a patch is explained by aspects of landscape composition and configuration which make that patch inaccessible or unsuitable for the species of concern (Tischendorf & Fahrig 2000). In this case, the landscape is resistant to the species' movement. While this hypothesis has not yet been widely validated (Zeller, McGarigal & Whiteley 2012), a few studies indicate that it is not completely unfounded. Newby (2011) showed that the habitat preference of dispersers of a cougar species was similar to the preference of resident adults. Goodwin & Fahrig (2002) showed that landscape structure was strongly correlated to connectivity, especially habitat area and inter-patch distance.

This study aims to better incorporate the effects of landscape characteristics on multi-specific connectivity to improve the design of ecological networks. In this study, we propose to work at the landscape level to infer landscape favourability (Powney et al. 2011). This landscape-based approach assumes that (i) a given patch of habitat is not necessarily favourable for a species and (ii) favourability depends on landscape composition and configuration.

Instead of validating a built network based on structural connectivity using biological data, our approach uses observed relationships between changes in landscape in terms of composition and configuration and changes in species distribution (Beier & Majka 2008; Chetkiewicz & Boyce 2009) to compute a landscape favourability value (continuous value). In other words, the model uses landscape and biodiversity data to assess the landscape context of patch connectivity. If the connectivity assessment is initially based on the landscape structure, then the calculated connectivity is functional because the favourability level for the landscape depends on species responses to landscape characteristics (Tischendorf & Fahrig 2000; Goodwin 2003; Kadoya 2009). With this approach, the transition from the patch scale to the landscape scale allows for direct interpretation of the favourability map as a connectivity map because the characteristics of the landscape that explain patch occupancy are incorporated into the description of the landscape.

The overall methodological approach was based on a non-oriented typology of landscapes (i.e., a typology that is independent of specific species requirements) and relies on landscape composition and configuration. The non-oriented approach makes the method suitable for use with species with various ecological requirements. First, relationships between the distribution of 36 species (belonging to three groups of common bird species: forest specialists, farmland specialists and generalists) and landscape types were used to assign a favourability level to each landscape type for each species group (a favourable landscape type significantly increases the probability of a species group being present on that landscape). Secondly, we used maps of favourability levels in 1982 and 2003 to delineate favourable entities and their connectivity, using two levels of favourability for each group of species and for the two dates. Thirdly, we compared the results between the 2 years to characterize changes in the favourable entities over time and corresponding changes in their connectivity. Finally, we used this temporal analysis to determine the sensitivity of the favourable landscape types to recent LUCC.

Following the methods described above, the aim of this paper was to (i) identify favourable landscape configuration and composition for the three bird groups, both independently and overall; (ii) map favourable entities and their connectivity in space and time; and (iii) identify negative and positive landscape changes for each of the groups and identify areas and landscape types sensitive to LUCC to help protect connectivity.

Materials and methods

Study area: the seine-et-marne region

The Seine-et-Marne region is located east of the metropolitan area of Paris. The region has a 5,915 km² area and contained 1,289,524 inhabitants in 2007. Historically, this region was dominated by agriculture, and 60% of the land is still devoted to this activity. The region also includes a sizeable area of forest (137,000 ha in 2003), including protected forests. Due to its location, the region is under strong anthropogenic pressure, resulting in dynamic urban sprawl and the development of transportation networks (Berger 2004). Agriculture and urbanization directly impact biodiversity, making this area of particular interest (Filippi-Codaccioni et al. 2010a).

Landscape description

The process of landscape description was the basis of the method and aimed to provide a non-oriented description which could be used to map the distributions of species with different ecological requirements.

Land cover and land use data

Land use and land cover change data sets from 1982 and 2003 were provided by the Urbanism Institute of the Ile-de-France region (IAU) (Filippi-Codaccioni, Clobert & Julliard 2009). Vector maps were derived from the interpretation of aerial photographs and additional data on building type. Because technical performance and resolution increased between 1982 and 2003, the classification was actualized so that the same classes were used for both years. The resolution of the land cover data was estimated by IAU to be 1/5000 for a minimum mapping unit of 625 m². The thematic precision of the data set was high, with the data set encompassing 83 land cover types. We searched for a simplification of land cover classes (smaller number of classes) informing about the type and level of human appropriation, but also about the type of natural habitats. Understanding landscape changes requires a land cover classification that includes human land use and can detect land use changes such as urbanization or agricultural development. For this study, these data were collapsed into six land use/land cover classes:

  • Agricultural areas: areas devoted to agricultural activities such as farming or pasture.
  • Urbanized areas: built areas, urban parks, gardens, building grounds or swimming pools.
  • Forest: natural woodlands, forests and poplar groves.
  • Water and wetlands: rivers, other bodies of water and wetlands.
  • Transportation areas: roads, railroad, parking.
  • Natural open areas: wasteland and cut forest.

Characterizing landscape types

To be suitable for species with different ecological requirements, the landscape characterization was built using a non-oriented (i.e., independent to specific species) approach that included landscape structural variables known to be linked to ecological processes. The landscape characterization relied on two independent descriptions of landscape: composition and configuration. The landscapes were mapped using homogeneous gridding (Tischendorf & Fahrig 2000), which was also applied to all species data. Because landscape scaling is species-specific (Tischendorf & Fahrig 2000), we selected a scale that would allow for unique yet comparable landscape descriptions. On the basis of the literature, we retained a grid of hexagonal units that were 87 ha (500-m radius) in size. This scale was found to be suitable for multi-species studies of bird–landscape relationships for forest species (Caprio, Ellena & Rolando 2009) as well as farmland species (Filippi-Codaccioni, Clobert & Julliard 2009; Smith, Fahrig & Francis 2011). The entire study area comprised of 6524 cells (Appendix S1).

The first step of the landscape characterization consisted of two separate reductions in the composition and configuration data using multivariate analyses. These reductions allowed us to reduce the number of variables and limit collinearity. Landscape composition was calculated using CA (Correspondence Analysis) with the proportions of the six land cover types from 1982 to 2003. Landscape configuration was computed by PCA (Principal Component Analysis) using the 10 landscape metrics for the 2 years (Table 1).

Table 1. Landscape metrics used for the configuration description of landscape
ScaleLandscape metric
LandscapeShannon equitability
Number of land cover classes
Dominant land cover classesAveraged perimeter of patches of each land cover class
Averaged number of patches of each land cover class
Averaged distance between the patches of the same land cover class
Averaged shape (ratio perimeter/area) of each patches of each land cover class
Minority land cover classesAveraged perimeter of patches of each land cover class
Averaged number of patches of each land cover class
Averaged distance between the patches of the same land cover class
Averaged shape (ratio perimeter/area) of each patches of each land cover class

Dominant land cover was defined as land cover representing more than 20% of the landscape unit. The computation of landscape metrics usually based on a land cover class (distance, patches number, shape and perimeter of patches) was done in this study for dominant and minority land covers. This shift from the land cover class to the dominant/minority classes permit to obtain a landscape configuration description independent from the landscape composition. This independence assured an equal level of precision of landscape configuration in all composition contexts.

The composition classification was determined using the first three axes of the CA, which accounted for 86% of the total variation (Appendix S2). The configuration classification was determined using the first seven axes of the PCA, which accounted for 91% of the total variation (Appendix S3). The classifications were performed using the CLARA (Clustering LARge Applications) method (Maechler et al. 2005) which is well-suited to large data sets. Separability of the composition and configuration classes was assessed using the cluster silhouette, which is an indicator of the number of top-ranking individual clusters (Maechler et al. 2005). To facilitate discrimination among landscape types, we considered a maximum number of 10 landscape classes. Among the classifications composed of less than 10 classes, we selected the classification with the best silhouette values.

The selected composition classification was composed of nine classes (Fig. 1), and the selected configuration classification was composed of seven classes that represented a gradient of increasing landscape fragmentation, based on our interpretation of the two-first axes of the PCA (Table 2).

Table 2. Characteristics of the landscape metrics of the seven configuration classes
 Class numberClass description
Highly fragmented landscapesClass 1High number of land covers, high equitability, small patches
Class 2Like class 1, but with more moderated values
Moderately fragmented landscapesClass 3Moderate values of all variables and complex patches shape
Class 4Moderate values of all variables, but few minority land covers
Homogeneous landscapesClass 5Low equitability, large and scarce patches of dominant land covers, high number of land covers
Class 6Low equitability, few land covers, few patches, patches of minority land covers distant from each others
Class 7Low equitability, few land covers, few patches, low distance between patches of minority land covers
Figure 1.

Triangular representation of the composition of the nine landscape composition classes in a triangle delineated by homogeneous landscapes (the three main land covers).

Finally, each landscape unit was characterized by a combination of the composition and configuration classes to define landscape type (Fig. 2). For a given composition class, landscape configuration varied extensively, typically between two and four configuration classes (Appendix S4). In the following text, a landscape type is referred to by the dominant land cover(s) (name of the composition class) and a number designating its configuration class. Higher configuration class numbers represent more homogeneous landscapes.

Figure 2.

Four examples of landscape types (composition plus configuration) from the composition class agricultural (around 80% of agricultural land cover in the landscape unit).

Characterizing landscape dynamics

Landscape dynamics refer to LUCC that occurs over time (Houet, Verburg & Loveland 2010a). Characterizing landscape dynamics in terms of composition and configuration was an important step in understanding how LUCC affected landscape characteristics and connectivity. Landscape dynamics were determined by comparing the landscape types of the landscape units for 1982 and 2003 (Hietel, Waldardt & Otte 2004). Landscape units that showed no change in composition or configuration showed ‘stable dynamics’. Conversely, those illustrating change showed ‘conversion dynamics’. All stable and major conversion dynamics were retained as needed to reach 95% of the total of landscape units (Appendix S4).

Favourability of landscapes for groups of habitat specialists and generalists

Bird data

The bird data base was provided by the ‘Dynamic Biodiversity Atlas in Seine-et-Marne’ project, which aims to understand the spatial and temporal relationships between society and biodiversity at the regional scale. The target species of this project were common bird species whose populations are strongly decreasing in Europe and in France (Jiguet 2008) and provide important ecosystem services (Gaston & Fuller 2007). Common bird species are recognized as good indicators of disturbance and community change, especially when intermediate to high commonness and guild representation are taken into account (Jiguet & Julliard 2006; Koch, Drever & Martin 2011). The distribution of these birds can be used to identify widely favourable areas and connectivity for rarer species.

Bird data were collected using a standardized scheme based on a grid of 4 km² cells across the study area. To adequately sample the variation in environmental and agricultural patterns and obtain a good sampling distribution, cells were sampled randomly in 16 areas that were characterized as homogenous in terms of environment and agriculture (French Agricultural District; Appendix S5). The data were collected by skilled volunteer ornithologists who inventoried common breeding birds by identifying song or visual characteristics (Julliard, Jiguet & Couvet 2003; Gregory et al. 2007). A total of 168 cells (2 × 2 km) were selected from the study area. In each cell, the observers designated between five and ten evenly distributed (at least 200 m apart) observation sites (200-m radius) from which the abundance and species of all individual birds were recorded over two successive 5-min periods. Observations were made between 5 and 11 AM during the breeding season. Each site was visited twice between 2006 and 2009. Species abundance was computed for each species as the maximum observed abundance across the different observation sessions.

We divided the common bird species into three groups of habitat specialization, following the European Bird Indicator (Gregory et al. 2005) adapted for France ( According to this indicator, a species is considered to be habitat specialist for one of the three main habitats in France (forest, farmlands and urban areas) when that species is at least twice as abundant in one habitat than the other two habitats.

Model selection framework: identifying favourable and unfavourable landscape types for groups of common bird species

We matched each observation site with the landscape type for its surrounding location. At the outset, 915 observation sites were distributed among 28 landscape types. Only those landscape types that were sufficiently represented in the data set to allow fitting the model were used in the analysis; therefore, the final data set consisted of 870 observation sites distributed within 20 landscape types and species abundance for 36 bird species distributed among the three preferred-habitat groups. The total number of landscape units for all landscape types in 1982 and 2003 was 5602 units (Fig. 3).

Figure 3.

Schematic representation of the methodology used to map favourable area and their connectivity. The grey boxes represent input data. The white boxes represent results. The numbers assess the three steps of the analysis and the associates results: (1) Step 1: Detecting favourability of landscape types; (2) Step 2: Measuring favourability area and connectivity; (3) Step 3: Identifying sensitive landscape types based on observed landscape dynamics.

The identification of favourable landscape types for the 36 species was performed in two steps. The first step involved construction of a GLM (General Linear Model) to predict species abundance for each landscape type. We used the negative binomial family distribution (to accommodate overdispersion) with a logarithmic link (R.10·2, R Development Core Team 2010) and corrected for spatial autocorrelation by including an autocovariate term (Augustin, Mugglestone & Buckland 1996; Betts et al. 2009) that was computed using the autocov_dist function of the spdep package (Bivand 2011):

display math

The second step identified the favourable landscape types for each species. The abundance of each species was predicted for each landscape type using the modelling results, then converted into percentages representing overall predicted abundance.

Favourable landscapes were defined using the following approach. If landscape type did not have an effect on species abundance, we would expect an equal distribution of the total abundance of all species across the 20 landscape types, with 5% of the total abundance in each landscape type. Thus, a landscape with more than 5% of the total abundance of the species could be considered to have a positive influence on species abundance (favourable landscape type), and inversely, a landscape with less than 5% of the total abundance could be considered to have a negative influence on species abundance (unfavourable landscape type).

For each of the bird species groups, we distinguished between two groups of favourable landscape types. One group included landscape types that were significantly favourable across all species in the bird group, and the other group included landscapes that were favourable for the species in the group on average, but not significantly favourable overall. We first calculated the average of the predicted abundances (percent) for each bird group by landscape type. We then tested whether the distribution of the predicted abundances of the group of species (in percent) was significantly greater than 5% using a Wilcoxon test (5% represents the null hypothesis). The test distinguished between the favourable landscape types, which were significant at the 5% level, and the neutral landscape types, which were mostly favourable for the species of the group, but not significantly different from the 5% level.

Deducing connectivity from favourable landscape types

Identifying the favourable landscape types for farmland, forest and generalist birds was the first step in mapping the favourable connected areas for the three groups for 1982 and 2003 (Fig. 3). Maps were based on (i) favourable landscape types only and (ii) favourable landscape types and mostly favourable neutral landscape types (hereafter called favourable and neutral landscape types). Connected areas were defined at two levels of landscape favourability. The first level included only the favourable adjacent landscape types adjacent to each other. The second level included favourable landscape types and neutral adjacent landscape units, with the goal of increasing connectivity between areas that were strictly favourable for an entire bird species group.

The total area for favourable entities and the total area for favourable and neutral entities, as well as their dynamics, were calculated for 1982 and 2003. Changes in total favourable area from 1982 to 2003 were then determined for the three bird species groups.

Two connectivity indices at the regional level were also computed. The number of isolated favourable entities and the number of favourable and neutral entities in the region were used as indicators of fragmentation and disconnection between entities. The average area of the two types of entities and the variation in average area were used as indicators of the spatial homogeneity of favourable areas and favourable and neutral areas. Moderate values and a small variance for the average area of favourable entities indicated a homogeneous area size and a high level of global disconnection. Moderate values and a large variance indicated heterogeneity in size among the connected entities, suggesting high connectivity within large areas and high disconnectivity among smaller entities.

Finally, the maps of favourability for the three groups of species were overlaid to highlight sensitive areas that were beneficial for at least two groups of species. When located in strategic locations between areas suitable for species with different ecological requirements (e.g., farmland species and forest species), these areas could act as habitat junctions, allowing a double connection between favourable landscapes.

Effects of landscape dynamics on favourable areas and connectivity

Changes in the total favourable area in the region between 1982 and 2003, and comparisons of connectivity indicators among the groups of species for these 2 years characterized the evolution of favourability and connectivity in the context of important land cover change. Temporal evolution in favourability and connectivity was analysed using landscape-type dynamics. This step in the study identified landscape types that were favourable for one or more groups of species, but sensitive to recent changes in land cover; it is these landscape types where efforts must be focused to protect vulnerable species groups.


Landscape dynamics

Changes in land cover composition during the 1982–2003 period mainly involved the artificialization of agricultural landscapes, but also a loss of forest area in forest-dominated landscapes (Appendix S6). With respect to landscape configuration and composition, the dominant landscape conversions led to an increase in the fragmentation of landscapes dominated by forests and farmlands, while landscapes dominated by artificialized areas showed decreasing fragmentation.

Favourability of landscapes for groups of habitat specialists and habitat generalists

The predicted abundance of all bird species for the different landscape types varied from 0% to 22% of the total species abundance (Appendix S7).

Farmland species

Four favourable landscape types were identified for farmland species (Table 3) and 10 landscape types were identified as favourable and neutral for farmland species.

Table 3. Cumulative values of the connectivity indices computed integrating the next less favourable landscape type at each line for the three groups of species. Connectivity indices are computed in 1982 and 2003 and the changes between 1982 and 2003 are also represented
FavourabilityComposition classConfig. classAveraged predicted abundance (%)Number of connected areas in 1982Total area in 1982 (cumulated km²)Entities averaged area
19822003Change (%)19822003Change (%)Entities averaged area2003Change (%)
Farmlands species
Favourable landscape typesStrictly agricultural59·80280247−11·8341·2291·8−14·51·2 ± 0·61·2 ± 0·6−3·0
Agriculture48·29397387−2·5574·2560·3−2·41·4 ± 1·11·4 ± 1·40·1
Strictly agricultural77·78199187−6·01719·01647·1−4·28·6 ± 35·98·8 ± 41·82·0
Agriculture/Forest47·58221205−7·21927·71804·7−6·48·7 ± 36·98·8 ± 42·40·9
Neutral landscape typesAgriculture/Artificialized16·81247236−4·52074·11932·9−6·88·4 ± 44·58·2 ± 42·2−2·5
Agriculture/Forest25·931811831·12488·92337·3−6·113·8 ± 83·812·8 ± 81·9−7·1
Strictly agricultural45·43117114−2·62839·62673·3−5·924·3 ± 176·623·5 ± 161·1−3·4
Agriculture25·3989934·53048·32877·7−5·634·3 ± 222·430·9 ± 206·9−9·7
Agriculture55·35577022·83322·83160·9−4·958·3 ± 321·245·2 ± 278−22·5
Agriculture15·18506224·03444·13251·0−5·668·9 ± 363·152·4 ± 305·2−23·9
Forest species
Favourable landscape typesForest/Agriculture411·44146106−27·4160·2114·3−28·61·1 ± 0·61·1 ± 0·6−1·7
Forest48·79170150−11·8249·4262·45·21·5 ± 1·51·7 ± 2·319·2
Forest78·49151125−17·2458·1396·6−13·43 ± 103·2 ± 10·94·6
Forest/Agriculture28·21210192−8·6696·3645·2−7·33·3 ± 9·93·4 ± 101·3
Agriculture/Forest26·88290275−5·21111·11049·6−5·53·8 ± 11·83·8 ± 11·5−0·4
Agriculture/Forest46·44266263−1·11319·81207·2−8·54 ± 13·74·6 ± 12·9−7·5
Neutral landscape typesForest/Agriculture17·89266262−1·51402·91286·0−8·35·3 ± 14·34·9 ± 13·2−6·9
Mixed15·072532613·21628·91544·1−5·26·4 ± 17·85·9 ± 15−8·1
Generalist species
Favourable landscape typesForest/Agriculture47·07146106−27·4160·2114·3−28·61·1 ± 1·61·1 ± 1·1−1·7
Agriculture/Forest26·34352331−6·0575·0518·7−9·81·6 ± 5·71·6 ± 5·2−4·1
Forest/Agriculture25·95335324−3·3813·2767·3−5·62·4 ± 8·12·4 ± 7·7−2·4
Agriculture/Artificialized15·393723730·3959·5895·4−6·72·6 ± 9·62·4 ± 9−6·9
Neutral landscape typesForest/Agriculture16·18368367−0·31042·7974·3−6·62·8 ± 10·42·7 ± 9·7−6·3
Agriculture16·003453573·51163·91064·3−8·63·4 ± 11·63 ± 10·6−11·6
Forest75·843163459·21372·61198·5−12·74·3 ± 13·73·5 ± 12−20·0
Mixed15·7526130316·11598·61456·6−8·96·17 ± 164·83 ± 14·6−21·5
Agriculture/Forest45·5622227724·81807·31614·2−10·78·1 ± 185·8 ± 16·1−28·4
Wetlands15·4320524620·01871·41706·9−8·89·17 ± 18·76·9 ± 17−24·0
Agriculture45·3122327523·32104·41975·3−6·19·4 ± 217·2 ± 19·7−23·9

As expected, the favourable landscape types for farmland species were dominated by agricultural land use. However, not all agricultural-dominated landscapes were favourable for farmland species, as Agricultural/Artificialized landscape types were neither favourable nor neutral. Farmland species were strictly favoured only in some landscape configuration classes. In Strictly Agricultural landscapes, farmland species abundance responded positively to the less fragmented configuration classes (5 and 7). In Agricultural and Agricultural/Forest composition classes, only configuration class 4 had a positive effect on farmland species (this configuration class showed moderate fragmentation with close-to-average values for all metrics, but with a few patches of the minority land cover types). Therefore, in these less agricultural landscapes, landscape types that were favourable for farmland species were characterized by an intermediate level of fragmentation.

Forest species

Six landscape types were identified as strictly favourable for forest species, and two additional types were identified as neutral. The favourable landscape types for forest species were mostly dominated by forest, but also included non-forest dominated composition classes that were greater than 20% forest cover, including Mixed and Agriculture/Forest classes. As in the case of the farmland species, the Agricultural/Forest composition class of configuration class 4 was favourable for forest species.

Generalist species

Four landscape types were identified as favourable for generalist species and seven others were identified as neutral. Generalist species were mostly found in landscape types dominated by two or more types of land cover, especially landscape types dominated by both forest and agriculture. The neutral landscape types showed medium to high levels of fragmentation.

Favourable area and connectivity

As expected, the group of farmland species had the largest strictly favourable area (1800 km²; Table 1), while forest species had only 1200 km² of favourable areas. Surprisingly, generalist species only had about 900 km² of strictly favourable areas.

On the favourability maps for 2003, the favourable areas for farmland species were broadly connected by mostly favourable neutral landscape types (Fig. 4). The favourable areas for forest species were more disconnected, primarily because only few landscape types were favourably neutral for forest species. In terms of favourable areas, the results for the generalist species were similar to those obtained for the forest species. The main difference was that the favourable areas for the generalist species appeared to be largely connected by favourable neutral landscape types.

Figure 4.

Maps of landscape favourability in 2003 for the three groups of species. The oval forms point out examples of disconnected favourable entities. The rectangular forms point out examples of favourable entities connected by neutral to favourable path of landscape units.

Effects of landscape dynamics on favourable areas and connectivity

Land cover changes induced a global decrease in neutral and favourable areas for the three groups of species and an increase in landscape types that did not favour any species groups (+156% between 1982 and 2003; Table 3). The decrease was more pronounced for farmland and forest species than for generalist.

For farmland and generalist species, the loss of favourable areas resulted from a loss of small landscape entities; therefore, there was not a large change in the average area of favourable entities for these groups. In contrast, the loss of favourable areas for forest species was unrelated to the number of entities, and the average entity area was diminished through trimming. The loss of favourable area for farmland and generalist species was accompanied by an increase in the number of favourable landscape entities and a strong decrease in their average area. These patterns indicate extensive loss of connectivity between large favourable entities resulting from the conversion of important landscape units which were connecting entities in 1982. For forest species, the decrease in favourable area resulted in a slight decrease in the average entity area, with no loss or gain in the number of entities. This pattern suggests a slow rate of decrease in surrounding favourable landscape units rather than a widespread loss of connecting landscape units.

Multi-favourable landscape types

Some landscape types appeared to be favourable for at least two groups of species (Table 3 and Table A4). Agriculture/Forest configuration class 2 and Agriculture configuration class 2 were neutral or favourable for all three groups, and the Agriculture/Forest configuration class 4 was favourable for both farmland and forest species. The number of landscape units of the latter landscape type, however, decreased strongly between 1982 and 2003. The loss of 24% of the landscape units made this type especially sensitive to recent landscape dynamics, in which the favourable area lost by the each groups was approximately 8%. Generalist and forest species also shared six landscape types.


Favourable areas and connectivity

The approach used in this study was based on independent descriptions of landscape composition and configuration. Therefore, the landscape favourability levels can be interpreted as a measure of landscape resistance (Sawyer, Epps & Brashares 2011; Zeller, McGarigal & Whiteley 2012). The values of landscape favourability can be directly translated into connectivity levels because the landscape scale accounts for spatial characteristics that explain the distribution of species. This approach to connectivity analysis provides some advantages: its simplicity allows one to directly generate and compare functional connectivity maps for species with different ecological requirements as well as groups of species, and it allows for the transition from the species-specific level to the multi-specific level.

However, one drawback to the simplicity of this connectivity approach is that it could underestimate connectivity. For instance, if two favourable entities are disconnected by a unique unfavourable landscape unit, one could expect that birds would still be able to cross this unfavourable area and therefore the two areas would not be completely disconnected. We here faced limitations due to the choice of the grid size (Zeller, McGarigal & Whiteley 2012). Nonetheless, a disconnection caused by an unfavourable landscape would have important effects on bird movements, and our approach to connectivity would take this change in movement into account by treating connectivity as a continuous variable.

The overall method and the resulting favourability maps could be combined with other connectivity methods, such as graph theory or circuit models (Kadoya 2009), to overcome the above mentioned problems and build connectivity paths at the landscape scale, thereby developing a more complete model of connectivity. It could also be improved using concomitant bird and LUCC data.

Our method also provides a better way to identify the location of potential habitat with regards to landscape characteristics, which is a crucial issue in connectivity studies (Chetkiewicz & Boyce 2009). However, the favourable landscape types identified in this study should not be considered as habitat by the usual definition, given that our study was not conducted at the habitat patch scale. Therefore, we propose the concept of ‘landscape-habitat’ to refer to a group of landscape characteristics that are favourable for the movement and survival of species in habitat patches.

Landscape types and common bird species

Landscape configuration appeared to be important for assessing favourable landscape types, favourable areas and connectivity. Contrary to a study by Goodwin (2003), who found that matrix composition was of little importance to connectivity, our results suggest that matrix characteristics (composition and configuration) could make the difference between a favourable and an unfavourable landscape.

The observed responses of the three groups of species to the landscape types indicated that the groups had different sensitivities to the proportion of preferred habitat and provide insight into group sensitivity to landscape composition and configuration. The favourable landscape types for farmland species are strongly defined in terms of both composition (the amount of agricultural land in the landscape) and configuration. Configuration appears to become more important as the amount of farmland decreases. This finding suggests that the farmland specialist species are responding to landscape characteristics in a nonlinear way. However, inconsistent with the nonlinear fragmentation hypothesis (Andrén 1994; Betts et al. 2006), farmland species appear to be positively affected by an intermediate level of fragmentation when the amount of habitat decreases. In the Agricultural/Forest and Agricultural composition classes, farmland species are not only positively affected by less fragmented landscapes, but also by configuration classes with less diversified land cover. These results confirm the proposed difference between species that are observed in farmlands and favoured by landscape heterogeneity (Devictor & Jiguet 2007; Fahrig et al. 2010) and farmland specialists that are negatively affected by landscape heterogeneity and are more directly associated with the proportion of farmland in the landscape (Filippi-Codaccioni et al. 2010b; Fischer et al. 2011). The dependency of forest species on composition and configuration is less complex than for farmland species and is consistent with the landscape composition hypothesis (Betts et al. 2006; Tscharntke et al. 2012). Forest species are more abundant in forested and less fragmented landscapes, and their abundance decreases globally with loss of forest area and increased fragmentation. When landscape composition positively affects forest species, the influence of configuration is not sufficient to overwhelm the effects of forest land cover proportion (Mortelliti et al. 2010). In contrast with the two groups of specialist species, the generalist species appeared to have a stronger positive response to fragmentation and heterogeneity. Therefore, if these species are generalists with respect to habitat, they are also specialists of highly fragmented landscapes, and landscape fragmentation may be the main driver of their distribution.

If we assume that common bird species are good indicators of community dynamics and rare species distribution (Jiguet & Julliard 2006; Koch, Drever & Martin 2011), the connectivity maps obtained for farmland and forest specialists could be applied to specialist communities and rarer species. The proposed multi-specific approach could be a valuable tool to delineate connectivity networks that go beyond common species' habitats.

Effects of landscape dynamics on connectivity

The increased artificialization of landscapes and the global increase in the fragmentation of forested and agricultural landscapes caused a decline in total favourable areas and connectivity for all three groups (Hepinstall, Alberti & Marzluff 2008). Recent landscape dynamics played an especially important role in altering the more favourable landscape types of the two specialist groups. The loss of neutral landscapes types also had important consequences. The loss of neutral landscape units increased disconnectivity through changes in key landscape units in the territory (namely, those landscape units that connected two large landscape entities in 1982). Area losses changed the degree of connectivity for forest species less dramatically than for the other species groups because forest entities were already highly disconnected in 1982. Subtle land cover changes appeared to have potentially important consequences for connectivity (Houet et al. 2010b).

A new idea: junction landscapes, or how to combine species habitat sensitivities

One common discussion in the literature that remains largely unresolved both at the multi-specific level and the community level is that a landscape with high connectivity for one species may be disconnected for another. Moreover, a network built to connect habitat areas for certain species could act as a barrier for others. Yet, using the approach developed in this study, we show that some landscape types actually support two or three groups of species. Indeed, these landscape types could permit multi-connectivity in a network built for multi-habitat sensitive species (Fig. 5).

Figure 5.

Three examples of double benefit in 2003 or loss of double benefits for farmlands and forest species between 1982 and 2003 (E.G. loss of the Agriculture/Forest of configuration class 4).

The results for Agricultural/Forest configuration class 4 are interesting because this landscape type is favourable for both farmland specialist and forest specialist species, two groups for which conservation pressure is strong. Finding a way to combine the connectivity of these two groups is an important conservation issue. The loss of landscapes favourable for two or three groups (e.g., farmland and forest birds, or forest and generalist birds) is an important consideration, as their loss has greater consequences for global connectivity. Multi-group landscape types are characterized by mixed composition and configuration (moderate to high fragmentation levels and intermediate heterogeneity). For landscapes favourable for forest and farmland species, the importance of the configuration, and especially heterogeneity, was apparent in this study.

Our results suggest that for land use management, both composition and configuration need to be accounted for to maintain suitable conditions for biodiversity (Tischendorf & Fahrig 2000; Pellissier et al. 2012). Heterogeneous landscapes appear to be useful in allowing, in some places, connectivity between entities of two or three area favourable for different groups of species. This can be viewed as a junction or a roundabout for species movement. Fragmented landscapes may be important for allowing multi-connectivity and providing an adapted spatial configuration. These junction landscapes could be especially useful when constructing regional or national networks. Indeed, one risk of constructing such a network is the failure to consider species that inhabit more anthropogenic landscapes, such as farmland species. There are few data on connectivity for farmland species, and studies of connectivity usually exclude farmland species (Grashof-Bokdam & Langevelde 2004). In France, farmland bird species are in decline, with an average decrease of approximately 28% between 1989 and 2008 (Jiguet 2008), making these populations more sensitive to recent land cover dynamics than forest birds.


The approach presented in this study holds promise as a tool for describing landscapes for land managers and for improving biodiversity conservation. This method can be used to visualize connectivity paths and identify sensitive areas to improve habitat connectivity for species that live in different habitats or have very different ecological requirements. The resulting landscape descriptions can inform ecological networks because they account for the sensitivity of different habitat specialist species. We have also shown that recent landscape dynamics in Seine-et-Marne potentially led to a decrease in suitable landscapes for the three study groups, including specialist and generalist species. We have not yet attempted to compare this approach with graph theory approach or other methods, but we think that the combination of the present method with one of these methods would lead to important and interesting developments in this method.


The authors wish to thank the IAU (Institut d'Aménagement et d'Urbanisme) for the land use and cover-data. We also wish to thank Guy Pe'er and an anonymous reviewer for their careful review of this manuscript and permit a significant improvement of its quality. This study was funded by the program ‘Atlas dynamique de la biodiversité en Seine-et-Marne’ of the Seine-et-Marne region.