Urbanization is characterized by high levels of sealed land-cover, and small, geometrically complex, fragmented land-use patches. The extent and density of urbanized land-use is increasing, with implications for habitat quality, connectivity and city ecology. Little is known about densification thresholds for urban ecosystem function, and the response of mammals, nocturnal and cryptic taxa are poorly studied in this respect. Bats (Chiroptera) are sensitive to changing urban form at a species, guild and community level, so are ideal model organisms for analyses of this nature.
We surveyed bats around urban ponds in the West Midlands conurbation, United Kingdom (UK). Sites were stratified between five urban land classes, representing a gradient of built land-cover at the 1 km2 scale. Models for bat presence and activity were developed using land-cover and land-use data from multiple radii around each pond. Structural connectivity of tree networks was used as an indicator of the functional connectivity between habitats. All species were sensitive to measures of urban density. Some were also sensitive to landscape composition and structural connectivity at different spatial scales. These results represent new findings for an urban area. The activity of Pipistrellus pipistrellus (Schreber 1774) exhibited a non-linear relationship with the area of built land-cover, being much reduced beyond the threshold of ∼60% built surface. The presence of tree networks appears to mitigate the negative effects of urbanization for this species.
Our results suggest that increasing urban density negatively impacts the study species. This has implications for infill development policy, built density targets and the compact city debate. Bats were also sensitive to the composition and structure of the urban form at a range of spatial scales, with implications for land-use planning and management. Protecting and establishing tree networks may improve the resilience of some bat populations to urban densification.
Citation: Hale JD, Fairbrass AJ, Matthews TJ, Sadler JP (2012) Habitat Composition and Connectivity Predicts Bat Presence and Activity at Foraging Sites in a Large UK Conurbation. PLoS ONE 7(3): e33300. https://doi.org/10.1371/journal.pone.0033300
Editor: Brock Fenton, University of Western Ontario, Canada
Received: November 25, 2011; Accepted: February 10, 2012; Published: March 12, 2012
Copyright: © 2012 Hale et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This research was undertaken as part of the Urban Futures (SUE2) program (EP/E0216030) (grant holder JPS) (http://www.epsrc.ac.uk/) and the OPAL (Open Air Laboratories) project (http://www.opalexplorenature.org/), which is funded by The National Lottery (UK) through the Big Lottery Fund (http://www.biglotteryfund.org.uk/) (for additional volunteer field assistants). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Fifty years of agricultural intensification, fragmentation and urbanization have radically altered the landscape composition of the UK . Urban areas have grown substantially over the last 20 years  and now support the majority of the global population . Urbanization is characterized by an increase in sealed land-cover density , geometric complexity, fragmentation of land-use patches ,  and a reduction in patch size , . Combined with varying disturbance levels , this results in a spatio-temporally complex land-use mosaic ,  with far-reaching consequences for species dispersal , ecological function  and ecological service provision . In some countries, urbanization has resulted in urban sprawl into agricultural land , while in others policies favour compact city forms . Where greenbelts constrain urban sprawl , there is evidence that urban landscapes have ‘densified’ and lost greenspace, especially over the last decade . Little data exist that indicate how much densification the urban ecosystem can withstand before ecosystem function is substantially impaired. In terrestrial habitats increased urbanization generally has a negative effect on species richness, although this pattern is not universal . Organism responses to increasing urban land-cover are species and trait-specific, but generally differentiate between generalist species that thrive or show humped abundance patterns, and specialist species that exhibit declines , , , , , .
Urban density thresholds for species presence and abundance are likely to be contingent upon sampling methodology and the spatial scale at which built density and landscape composition are measured. There is currently a multiplicity of approaches evident in the literature , , clarification is needed in order to improve comparability between studies and to aid the translation of results into conservation practice. Clarity may be gained by studying taxa whose species are sensitive to different measures of urbanization at a range of spatial scales, as well as to the surrounding landscape. There is already a considerable literature on birds and urbanization , , , but their life histories and responses to urbanization do not always reflect those of other groups . Mammals, nocturnal and cryptic taxa are poorly studied in this respect and bat (Chiroptera) communities are ideal candidates for research. They typically include species that exploit built structures  are sensitive to landscape scale, patch effects ,  and to changes in structural connectivity , .
The few studies focusing on the effect of urbanization on bat species indicate variability in response to changing urban form at a species, guild and community level. In studies of cities in the Czech Republic , Mexico  and Australia  bat activity was lower in high density residential areas, than in low density areas (e.g. suburban, urban fringe) and semi-natural areas. In addition, lower species richness was reported in the urban centre and densely developed areas. This contrasts with other studies in the USA ,  where positive relationships have been reported between both overall bat activity and species richness of natural habitat fragments and the urban density of the surrounding landscape. Several studies have identified positive relationships between urbanization and the activity of certain species , , , , but other species clearly favoured semi-natural areas or exhibited a broad tolerance of urbanization , . It has been suggested that these responses reflect differences in wing and call morphology, with species specialising in cluttered habitats avoiding brightly lit and poorly vegetated urban areas .
Given the intensity of compositional change in urban areas ,  and the associated high levels of fragmentation , one might expect that connectivity and linkage would be a central theme in urban ecology, as it is in other landscapes , , , . Indeed, connective features such as green networks and corridors have been influential in guiding city planning in many areas of the world , , and the creation and preservation of wooded corridors does seem to present an ideal opportunity for restoration aimed at enhancing spatial population resilience in cities . However, there are very few studies that focus on this element. Studies on plants and invertebrates ,  in UK greenways identified multiple structural and functional roles that were species specific in terms of habitat provision, but did not indicate a strong functional conduit role that enhanced movement and dispersal. Although evidence that wooded linear features such as streets and riparian corridors facilitate connectivity for birds in urban areas ,  there are few studies pertaining to urban bats, although several have identified relationships between linear features and bat activity in agricultural areas , , . Such features appear to have roles in both feeding and movement and thresholds for loss of functional connectivity are still unclear , .
Here we explore the influence of urban landscape composition and structural connectivity on the presence and activity of bats at a range of spatial scales. We stratified sampling sites evenly across classes of urban form whose composition and extent were clearly defined a priori using a wide range of environmental data captured in a Geographical Information System (GIS). Foraging sites with similar local land-cover were selected to reduce the effect of confounding local variation in habitat type, and their landscape context measured consistently at multiple spatial scales. A proxy measure of functional connectivity was developed for each scale based on the traits of the species encountered. Both walking surveys and fixed position detectors were used to record bat activity. Using the assemblage and environmental data we addressed the following research objectives: (1) To characterize bat activity and presence in relation to urban density and landscape composition; (2) To assess the spatial scale at which species respond to the urban landscape; (3) To establish the significance of connectivity for bat activity in a heavily urbanized landscape.
We achieved our objectives and explore in the Discussion section the issues surrounding quantifying land-cover, land-use, functional connectivity and species specific responses to landscape change. Despite some progress with mapping key variables at a high spatial resolution and large spatial extent, our proxies for roost potential (large trees and residential buildings) and lighting (road area) did not add any explanatory power to the models and could be improved on. Future studies may benefit from data on tree species, building age and densities of lighting columns.
We recorded bat calls within a total of 14,176 survey minutes using the fixed-point automatic detector. Of these, 11,545 minutes contained calls identifiable to species or guild level. These included 9,950 active minutes (86% of the identifiable bat calls) of P. pipistrellus calls, 1,330 (11%) of P. pygmaeus and 345 (3%) belonging to the NSL (Nyctalus noctula, Eptesicus serotinus, Nyctalus leisleri) guild, with some minutes including calls from more than one species/guild. Walking surveys added an additional 1178 active minutes (75%) for P. pipistrellus, 190 (12%) for P. pygmaeus, 49 (3%) for the NSL guild and 163 (10%) Myotis calls, all of which where from bats observed feeding over the pond surface and were therefore confirmed as Myotis daubentonii. Bats were recorded at all of the thirty survey sites. Only P. pipistrellus was recorded at all sites, although P. pygmaeus was recorded at 93.3% of sites. The NSL guild was recorded at 73.3% of sites but was virtually absent from those within the Dense Urban land class (Fig. 1). M. daubentonii was present at 33.3% of sites and was negatively associated with increased urbanization (Fig. 1). Full-night activity for both P. pipistrellus (p = 0.043) and the NSL guild (p = 0.035) was significantly higher in the Rural compared to the Dense Urban land class (Fig. 1). Evening activity for P. pipistrellus and the NSL guild followed similar patterns, although the differences between classes were not significant.
Land classes follow an urbanization gradient from Rural (R) to Dense Urban (DU). Box plots represent total active minutes for (A) P. pipistrellus, (B) P. pygmaeus, (C) a group comprising N. noctula, E. serotinus, N. leisleri and (D) M. daubentonii. Boxes that do not share a letter showed significant differences (P<0.05) between land classes.
Multiple models were created for all species and guilds and the best of these are presented in Table 1. These models included data extracted using concentric buffers applied both to the landscape around each pond (concentric landscape) and restricted to the landscape intersected by a connectivity mask (connected landscape) (Fig. 2). In general, P. pipistrellus activity was highest at sites surrounded by low or moderate levels of built land-cover or at well-connected sites in highly urban areas. The best models all included the area of built land-cover within 350 m of survey ponds, with activity peaking at intermediate levels of built land-cover and being lower but more variable at high levels of urbanization (Fig. 3). These models included a positive association with connected tree cover (>6 m high) within a radius of 150 m for sites in Dense Urban and Dense Suburban land classes (Fig. 4). Evening activity was also positively associated with connected garden area within both 50 and 500 m. These garden parameters were not present in the full-night model, but were replaced by a measure of connected vegetation cover within 200 m of the site (Table 1).
(A) An unrestricted extraction of landscape data using concentric buffers. (B) Connected (available) landscape mask created using tree networks buffered by 50 m. This polygon was used as a mask to restrict the landscape analysis to the area within this network. In both examples, landscape data were extracted at distances of 50, 100, 150, 200, 350 and 500 m from the pond centre.
This was included in the final model for evening activity of P. pipistrellus (see Table 1 for full model).
(A) sites within Dense Urban and Dense Suburban (B) Dense Suburban and Suburban (C) Suburban and Light Suburban, (D) Suburban and Rural land classes.
Pipistrellus pygmaeus activity was highest at ponds located within a highly vegetated landscape but poorly connected at a local level. All models included negative parameter coefficients for connected tree cover (>6 m high) within a radius of 100–200 m. Evening models included a positive relationship between activity and connected vegetation within 1 km, and the full-night models included a negative association with the area of built land-cover within 1 km.
No valid GLM or GAM activity models were identified for M. daubentonii, or the NSL guild, but several logistic regression models were selected for evening presence data. The evening presence model for M. daubentonii included a positive relationship with natural land-cover (concentric) within 150 to 350 m of the survey sites (Table 1). For the NSL guild evening activity was negatively associated with built land-cover (concentric) within 350–750 m (Table 1). For these species, restricting the landscape analysis to the areas adjacent to tree networks (connected landscape) did not result in any valid models. Candidate all-night NLS presence-absence models all exhibited residual spatial patterning (e.g. Figure S1). We used a spatial correlation structure to compensate for this, but it did not improve either the residual spread or the AIC of the models, so all the models were rejected.
We investigated the response of a bat community to urbanization, landscape composition and structural connectivity at a variety of spatial scales using standardized samples across five urban landscape classes (Figs. 5 & 6), targeting small ponds with consistent levels of adjacent riparian woodlands. All species were found to be sensitive to at least one measure of urbanization and some were additionally influenced by landscape composition and structural connectivity at different spatial scales. For two species, habitat associations differed between evening and full-night models.
It includes the metropolitan borough centres of A Wolverhampton, B Dudley, C Sandwell, D Walsall, E Birmingham and F Solihull. Bat survey ponds are indicated by a black circle and were stratified by urban land classes, which are represented by a grid of 902×1 km2pixels covering the study area. These range from Dense Urban (white) to Rural (dark green). Canals and railways are indicated by fine black lines.
(A) Mean area and SD for 3 of the 13 land-cover and land-use types derived from the Ordinance Survey Mastermap and used to assign ponds to land classes (see Table S1). (B) Mean area and standard deviation for vegetation cover and trees>4 m high.
Three broad measures of urbanization were derived for this study (i) the area of built (sealed) land cover parcels derived from OS digital data, (ii) the area of natural (vegetated) land-cover parcels and (iii) the area of vegetated (remotely sensed) land-cover (Table 2). It is notable that all of the species/guilds in our study were found to be sensitive to at least one of these measures of urban density, given the limited evidence on the response of bat communities to urbanization , . Each measure provided significant explanatory power to different models, supporting calls for the use of multiple measures of urbanization in gradient studies . It is likely that these measures differ in their representation of key resources or ecological disruptors as they are broad and indirect measures of a complex anthropogenic gradient . Although most species demonstrated a negative association with urbanization, we found a non-linear relationship between P. pipistrellus activity and built surface cover. Activity peaked at ∼40% built cover, yet at levels above ∼60% activity rapidly reduced, implying the existence of a threshold or tipping point . This is the first report of such a relationship for a bat species, although non-linear relationships with urbanization have been identified for other taxa . Pipistrellus pipistrellus could be described as an “urban adapter” , whilst the remaining species would be “urban avoiders” of varying sensitivity. These results broadly agree with those of other studies, for example, that Myotis species tend to avoid villages  and urban centres . European work  suggests that small bats with low wing loadings are tolerant of even dense urbanisation, but large bats with high wing loadings generally avoid urban centres. This is, however, at odds with work in Australia which suggests the reverse  and also in disagreement with studies of urban bird traits, which suggest large size/wings are a trait of urban adapters . These differences may be an artefact of differences in urban composition and morphology between European and Australian cities, the scale that urbanisation is measured at, how urban density is defined and the degree to which different species are willing to accept human subsidised resources (e.g. building roosts).
The often contradicting findings from studies of urban bat communities illustrate some of the broader challenges associated with attempting to identify ecological patterns along urbanization gradients . The descriptions of the urban form provided by Gaisler et al. , Avila-Flores and Fenton  and Gehrt and Chelsvig  varied considerably in their detail. We addressed this issue by accessing high-resolution parcel based and remotely sensed data for land-cover and land-use for the entire study area.
Although all species demonstrated a negative response to broad measures of urbanization, considerable differences in activity were evident between sites at similar points along these gradients. This suggests that more subtle variations in landscape composition may also be important. Pipistrellus pipistrellus evening activity was found to be positively associated with gardens, which might be expected given their propensity for roosting in buildings  and for early evening emergence. An additional explanation is that gardens might typically provide tree cover that facilitates early emergence and feeding. These findings pose further questions about the mechanisms behind associations between bat activity and residential land-covers reported elsewhere , , , . As with other studies, differences were found between the landscape composition preferred by P. pipistrellus and P. pygmeaus. A positive association between P. pipistrellus and local tree cover was expected given their known use of edges as commuting and feeding areas ,  and their role in increasing the attractiveness of adjacent roost sites . The association of P. pygmeaus with aquatic habitats described elsewhere , ,  was not observed in this study. However, this may reflect variations in the quality of riparian vegetation , which we were unable to measure at a landscape scale.
We expected M. daubentonii to demonstrate a strong positive association with the area of water in the vicinity of the survey sites , yet we did not find support for this in our results. It is possible that by surveying ponds we removed water as a limiting variable or that the social dynamics of this species served to mask important habitat associations . Members of the NSL guild are reported to seek out pastures, parks and other open green spaces . Their relatively high wing loadings, medium-high aspect ratios and low call frequencies permitting them to hawk in the open, typically feeding on large insects . Our presence models broadly support this, although we were unable to differentiate between ground vegetation types and management for the extent of our study area.
In general, we expected bat presence and activity to reflect the local availability of roosts  and foraging sites . Several of our landscape variables were intended to be proxy measures for these key resources, yet the value of our roost metrics was limited. There is an inevitable trade-off between the detail and availability of urban habitat data and the spatial scale of analysis  and it is likely that greater effort is required to map roost potential effectively. In addition, both roosting bats , commuting bats  and their insect prey are sensitive to variations in microclimate, which in urban areas will be heavily influenced by human activity. Additional data that improves the integration of human processes such as land management and intensity of use into urban ecological models may therefore clarify how roosting or feeding potential varies within each land-cover type.
Our approach employed proxy measures of functional connectivity to estimate the areas of the landscape theoretically available to bat species that commute along tree networks and is an extension of the accessible habitat model , . Functional connectivity is concerned with the ability of individuals to move between resource patches within the landscape rather than explicitly measuring the structure of landscape elements, although structure is frequently used as a proxy for function , . The measures of structural landscape connectivity used to extract landscape variables from the GIS appeared to be a good approximation of functional connectivity for the two Pipistrellus species. Euclidian distance may be a more appropriate for measuring accessible habitat for M. daubentonii and the NSL group. This supports previous studies highlighting the importance of linear landscape features , , , but this study is unique in demonstrating the functional importance of structural connectivity of tree cover for bat species in urban landscapes. As M. daubentonii has a similar wing aspect ratio and loading to the two Pipistrellus species we had expected models for M. daubentonii to include measures related to connectivity. It is possible that structural connectivity is relevant to this species, but that our landscape measures were insufficient to detect this relationship. However, given that our site data for this species is limited to activity within the first 1.5 hours after dusk and that this species is a late emerger, such interpretations should be treated with caution.
Whilst the concept is relatively straightforward, measuring functional connectivity in urban landscapes is challenging, particularly as the patch/matrix distinction is often unclear and actual movement paths are not easily observed. Previous studies have successfully employed expert judgement to estimate landscape resistance values for different urban matrix types . Our approach is easily replicable and scalable, but relies on the accurate mapping of individual trees in three dimensions and on a consistent response by bats within a population to gaps in tree networks. For studies of highly mobile bird species, estimating the path of movement within the urban matrix has delivered improved models compared to more general landscape measures  and it is worth noting that for both Pipistrellus species studied, the only valid models we identified were those that included variables measured using a connectivity mask. At sites where the built land-cover of the surrounding landscape was over 40%, structural connectivity was critical for maintaining high levels of P. pipistrellus activity. Urban density dependent relationships with connectivity such as this have not been demonstrated before.
Gehrt and Chelsvig  and Lookingbill et al.  located bat survey sites within natural reserves along an urbanization gradient. However, the direct comparison of these studies is difficult as the spatial extent used to define urbanization and characterize the landscape differed between studies. We attempted to avoid such issues by measuring urban density at a wide range of spatial scales. This identified broad patterns, with edge specialists (Pipistrellus spp.) being sensitive to landscape composition even at small spatial scales (50–100 m). Their relatively fast and agile flight appears to allow them to utilise relatively small foraging areas, supported by a high wing aspect ratio, low wing loading and small size . This may well explain their presence in densely built areas, as patch sizes tend to decrease with urbanization . That P. pygmaeus responds to the landscape at radii of up to 1 km may reflect a need to travel further to access preferred feeding habitats  such as highly structured riparian vegetation . The NSL aerial hawkers guild seeking un-built land-cover at larger scales (≥500 m) would be expected, given that their large size and high wing loading is suited to efficient flight over large open areas .
There are few studies that attempt to characterize the response of bats to urban landscapes at multiple radii that extend over a large spatial extent, although the response of groups such as birds has been explored . Our data suggest that individual species may be sensitive to changes in landscape composition at multiple spatial scales. For example, evening models for P. pipistrellus include connected garden area within a radius of 500 m and connected tree cover within 150 m. We speculate that the 500 m radius may indicate the “roost catchment” of the pond i.e. that bats using the pond, tend to roost in houses within 500 m. The area within 150 m of a pond may be relevant to the quality and accessibility of the local feeding area, with ponds surrounded by a high density of tree networks being particularly desirable. These results corroborate other studies that conclude multiple spatial scales may be relevant to bats  and urban mammals .
We have demonstrated that the density of landscape urbanization, its composition and configuration are important to the urban bat community. These relationships are scale dependent and species-specific. The broadly negative associations with urbanization for all species imply that a transition to more compact urban forms that reduce greenspace and habitat would inevitably impact the species richness of the urban bat community. This work informs the continuing debate about the sustainability of this approach to development . The presence of thresholds for ecological function raises the possibility that development densities could be specified with ecological thresholds in mind, and that tipping points should be explored in more detail for other taxa. The importance of connectivity for the Pipistrellus species suggests that some ecological function could be retained even within high-density developments and that protected tree networks may deliver some spatial resilience  to the impacts of increased urban densities. It remains to be seen whether tree networks play a similar role for other organisms (but cf. ). Our data on ecologically important land-covers, land-uses and spatial scales should support urban planners and managers in making spatially explicit decisions about urban conservation . We recommend that a multi-scale approach to planning and management be adopted, whilst recognising that this may be challenging given the typical spatial scales of urban land ownership ,  and decision-making . In particular, we suggest that when creating new urban bat habitats, consideration is given to ensuring that they remain functionally connected and therefore available to at least part of the urban bat community into the future.
Materials and Methods
The landowners gave permission for access to the sites. All Bat species are protected in the UK and licenses are needed if they are handled, mist-netted, or disturbed in their roosts. As our sampling involved only monitoring at foraging sites there were no licensing issues.
Study Area and site selection
The West Midlands metropolitan county (population ∼2.3 million) is a highly urbanized region of the United Kingdom (UK), covering 902 km2. As a centre of the industrial revolution it has undergone multiple cycles of development and distinct zones can be identified representing pre, wartime and post-war regeneration. The study area includes several urban centres (Fig. 5) with high levels of sealed land-cover, canals, railways, residential areas of varying housing density, industrial zones, parks, nature reserves and agricultural land on the urban fringe. Existing survey records for the study area indicate that several species of bat were present, including: Pipistrellus pipistrellus (Schreber 1774), Pipistrellus pygmaeus (Leach 1825), Myotis daubentonii (Kuhl, 1817), Eptesicus serotinus (Schreber, 1774), Nyctalus leisleri (Kuhl, 1817) and Nyctalus noctula (Schreber, 1774). These species vary considerably in their roosting, commuting and feeding behaviour (Table S1).
In order to stratify the survey sites along an urbanization gradient we first classified the landscape using land-use and land-cover data from OS Mastermap (OSM) , which is a high-resolution parcel based GIS dataset (Table S1). OSM polygon data were converted into a 2 m pixel resolution raster and displayed in a GIS (ArcGIS 9.2, ESRI Redlands, USA). A grid of 1 km2 cells was used to extract raster summaries using Hawth's Analysis Tools , as this is close to the average minimum foraging areas of both P. pipistrellus and P. pygmaeus , which are the smallest foraging areas for the species we expected to encounter (Table S2). Five land classes were identified using a cluster analysis of landscape variable percentages (Table S1) in SPSS 18.0 [c.f. 76] and excluding squares with greater than 30% water cover or 80% tree cover. These represented a gradient from Rural (R), Light Suburban (LS), Suburban (S), Dense Suburban (DS) and Dense Urban (DU) land classes (Figs. 5&6). In order to reduce the potential for variations in local habitat composition to obscure the effect of landscape context , , survey sites were restricted to small (515–2146 m2) unlit ponds with at least 30% riparian edge tree cover. This choice was a reflection of the attractiveness of aquatic, riparian and woodland edge habitats for foraging to all the species we expected to record  and the need to identify a foraging habitat patch that would be present in all land classes. Candidate ponds were assigned to one of the five land classes based on the land-cover and land-use percentages for a 1 km2 circle surrounding each pond (Fig. 6) and six survey sites were then selected from each land class. All ponds were separated by at least one kilometre (pond centre to pond centre).
Bat sampling methods
Ponds were surveyed for bat activity fortnightly between May and August 2009. We avoided nights where strong rainfall or wind were predicted and surveyed several sample points within each site . A variety of techniques have previously been applied to compare bat species presence and activity between sites, with detectors generally regarded as superior . We used a combination of walking transects ,  and fixed point detector surveys , which allowed multiple microhabitats habitats to be surveyed and activity to be recorded from dusk to dawn.
Evening walking surveys were undertaken for a period of 1.5 hours following sunset using a Pettersson D240x ultrasound bat detector (Pettersson Electronic, Sweden), in heterodyne mode, alternating between 20 and 50 kHz. Sample calls of 3.4 seconds were recorded in time expansion mode and transferred to a Sony MZ_NH6000 Minidisk recorder (Sony, Japan). Walking routes circled each pond at varying distances (0- 50 m from edge), with the purpose of detecting and observing bats that were active in the close vicinity, as well as directly over the pond. Fixed point surveys were initiated at dusk and terminated at dawn, using an AnaBat SD1 frequency division bat detector (Titley Scientific, Australia) installed at the edge of the pond at a height of 1 m, using an acoustic reflector . Tests confirmed that bats active within at least 15 m (horizontal distance) and up to ∼10 m above ground level were detectable, with bats calling at low frequencies (20–30 kHz) recorded at an unknown but greater distance.
Bat calls were identified to species level where possible, using parameters given in Russ . Where species identification was not possible in the field, bat calls detected on walking surveys were recorded and analyzed using BatSound 3.31 (Pettersson Electronic, Sweden). Calls recorded using fixed point Anabat detectors were processed automatically using filters within AnalookW . Although Myotis sp. calls were identified at several sites the call quality was highly variable. Subsequent tests confirmed that using a reflector on the Anabat dramatically reduced the detectable range for this group, so Myotis calls from the fixed detector were excluded from analysis. Species or guild specific call filters were developed and their results compared to a 10% sample of the call dataset to estimate the percentage of bat calls incorrectly rejected by filters, calls allocated to the incorrect species/group, and files incorrectly identified as a bat call. Considerable caution was applied, preferring filters that discarded a greater percentage of calls. As considerable overlap in call parameters has been reported for Nyctalus noctula, Eptesicus serotinus, Nyctalus leisleri, and these species were rarely observed in flight (which would aid identification), we processed these (NSL) calls as a single functional group of large, early emerging bats with similar foraging behaviours (Table S2).
Landscape and connectivity environmental variables
Using the GIS we selected a range of variables that related to roosting, commuting and feeding resources or that could be used as broad measures of urbanization (Table 2). Summaries of the area of built (sealed manmade) or natural (vegetated) surface cover (derived from the OSM landscape parcels) and the (remotely sensed) vegetation layer provided broad indications of urbanization density. Whilst the parcel based mapping was useful for estimating dominant land-cover, parcel types such as gardens were excluded as they contained varying levels of built and semi-natural land-cover. The remotely sensed vegetation layer was therefore used to gain a better reflection of vegetation cover. For the species we encountered, roost sites are likely to be located either in buildings or trees. Buildings provide a variety of roost opportunities due to their varied age, materials, architectural style and degree of maintenance. In addition to buildings, we included gardens from the OSM as an indirect measure of residential building availability, which we hypothesized might offer an enhanced roosting resource compared to commercial or industrial structures. Tree cover with a minimum height of either 15 m or 20 m was also included as a variable, as this was expected to indicate roost potential in mature woodland.
Several bat species are reported to fly along tree-lines when commuting and feeding , . We estimated suitable commuting habitats by identifying areas of the landscape where vegetation was greater than 4, 6, 10 or 15 m high, which correspond to the range of typical flight heights for these species (Table S2). In addition, the raster representing vegetation greater than 4 m high was converted to a polygon feature class and buffered by a distance of 20 m. This resulted in a layer representing tree networks separated by gaps of no more than 40 m, which was used as a measure of structural connectivity and a proxy for functional connectivity. Although gaps of this size are not likely to be problematic for most species when commuting within rural landscapes , we hypothesized that with increasing urbanization, an increase in artificial lighting could be sufficient to deter bats from crossing such gaps . Whilst we were unable to access spatial lighting datasets, we used a road dataset derived from the OSM as an indirect indicator of lighting and traffic disturbance. Insect feeding potential was represented by OSM derived polygons depicting water bodies (canals, streams and still waters), natural land-cover (predominantly vegetated) and supplemented with a high-resolution remotely sensed vegetation layer. Finally, the perimeter length of tree patches within tree networks connected to each pond was estimated, as many species are known to feed along woodland edges , .
Two approaches were taken to extract landscape variable summary data for the landscape surrounding each survey site. Firstly, using the GIS we created multiple concentric circular buffers around the ponds and extracted complete summaries of the underlying landscape data (Fig. 2A). Such a multi-scale approach is increasingly common , , although we used a particularly large number of radial extents (14, between 50 m and 4 km) in an attempt to accurately identify the spatial scales of relevance for each species. This approach assumes that all of the landscape is potentially available to the species concerned. Our second approach was to restrict the landscape analysis to the parts of the landscape adjacent to tree-lines. As both P. pipistrellus and M. daubentonii activity has been reported to occur predominantly within ∼50 m of water and woodland edges , , we buffered the tree networks connected to each pond by 50 m, creating a connectivity mask. Landscape variable summaries were again extracted at multiple radii around each pond centre, but this time the available landscape data were limited to the areas intersected by the connectivity mask (Fig. 2B).
The measure of bat activity used for each site was the total number of minutes in which a call was recorded, for each species or functional guild (hereafter termed active minutes). We make no assumptions that this is a measure of individual bat abundance. Variation in bat activity between land classes was analyzed in SPSS using either a one-way ANOVA or a Kruskal-Wallace test if variances were heterogeneous. Tukey or Nemenyi post hoc tests were used to identify which classes differed in activity .
The relationships between the environmental measures and bat activity were modelled using a combination of Brodgar v2.6.4 (Highland Statistics, Newburgh, UK) and R (version 2.11.1)  using the mgcv and nlme libraries. We used up to three response variables per species (or guild): (i) the total minutes of bat activity recorded by the fixed detector for the first hour and a half following dusk (evening) (ii) total bat minutes recorded by the fixed detector from dusk to dawn (full-night) and (iii) presence using either the fixed detector or walking transect surveys (when bat activity data were either unavailable or did not produce valid models).
Data exploration was undertaken prior to statistical analyses, and as a result, we included an additional nominal explanatory variable differentiating between sites in highly urban (Dense Urban and Dense Suburban) and less urban land classes (Suburban, Light Suburban and Rural). This was used as a conditional variable in the fixed element of the models for some species (e.g. Table 1).
Initially we developed species or guild specific models independently for each spatial scale (50, 100, 150, 200, 350, 500, 750, 1000, 1250, 1500, 1750, 2000, 3000 and 4000 m). First, we used co-plots to inspect co-variation between explanatory variables at each scale and where variables had high correlations (>0.5) one of the pair was removed. We then assessed these against the response variables, removing explanatory variables with correlation scores of <0.3. This reduced the potential pool of explanatory variables considerably. None of the explanatory variables at higher spatial scales (>1000 m) showed significant relationships with any of our response variables. This left a pool of seven spatial variables at eight spatial scales (56 in total). We entered these into all the models and used Akaike Information Criterion (AIC) to identify the most parsimonious model for each species or guild, ensuring that model variables had variation inflation scores (VIFs) of <3 . Where initial co-plots suggested linear relationships between the response and explanatory variables we used generalised linear modelling (GLM) . Non-linear relationships were analyzed using generalised additive modelling (GAM) . The deviance/degrees of freedom ratio was used to assess possible over-dispersion in the models . We used negative binomial distributions to account for over-dispersion (Ne)  and logistic regression with a binomial (Bi) distribution for presence data . Finally, this process was repeated with the variables from the best models at each scale combined to derive multi-scale models, pooling site based, concentric and connected variables for each species. All Models were validated using graphical visualisation tools in Brodgar and R. We plotted the residuals against fixed values to assess model homogeneity, QQ-plots for normality and plotted residuals against environmental co-variables to test for independence . Lastly, we used bubble plots in the gstat R library to examine each individual model for spatial autocorrelation . Where patterns indicated no spatial patterning the models were accepted (e.g. Figure S2). After validation we were left with a pool of 51 models: Pipistrellus pipistrellus evening (9), all-night (8); Pipistrellus pygmaeus evening (9), all-night (12), M. daubentonii evening (5) and NSL guild evening (8). Of these, candidate models with AIC≤2 of the optimum model were retained  (Table 1).
Residual bubble plot for NSL all-night Anabat data from logit binomial presence-absence data. The plot shows clumping of similar size positive residuals in the middle of the plot, indicative of spatial structuring in the data. Negative residuals in black and positive residuals are grey. The size of the circles indicates the size of the residuals.
Bubble plot for P. pipistrellus all-night Anabat residuals from a GAM of bat activity minutes. The plot indicates no spatial structuring in the data. Negative residuals in black and positive residuals are grey. The size of the circles indicates the size of the residuals.
Mean area (m2) and standard deviation of Ordinance Survey (OS) land-cover type for each urban land class.
We would like to thank the following people and organisations that have helped support this research. Volunteers from the Birmingham and Black Country Bat Group and the Open Air Laboratories (OPAL) (www.opalexplorenature.org) project assisted the bat surveys. We acknowledge the numerous landowners who have permitted access to their land for the sampling and EcoRecord (BBCWT) for access to information concerning bat records in the sample region. Geospatial data were provided by the Ordnance Survey (GB) and comprised: OS Mastermap Topography Layer [GML geospatial data], coverage: Birmingham, Black Country and Solihull, Updated: November 2008, Ordnance Survey (GB), using: EDINA Digimap Ordnance Survey Service, http://edina.ac.uk/digimap, downloaded: December 2008; OS Mastermap Topography Layer [GML geospatial data], coverage: West Midlands, updated: May 2010, Ordnance Survey (GB), using: Ordnance Survey Research, http://www.ordnancesurvey.co.uk, ordered: May 2010. The tree/vegetation data were derived from aerial near-infrared photography, colour photography and photogrammetrically derived height measurements provided by Bluesky International Limited (2007). We thank Adam Bates, Mark Ledger and Chris Sherlock for their valuable comments on early drafts of this manuscript.
Conceived and designed the experiments: JDH JPS. Performed the experiments: JDH AJF TJM. Analyzed the data: JDH JPS. Wrote the paper: JDH JPS. Contributed to writing the manuscript: AJF TJM.
- 1. Robinson R, Sutherland W (2002) Post-war changes in arable farming and biodiversity in Great Britain. Journal of Applied Ecology 39: 157–176.
- 2. Grimm NB, Faeth SH, Golubiewski NE, Redman CL, Wu JG, et al. (2008) Global change and the ecology of cities. Science 319: 756–760.
- 3. UN (2010) World Urbanization Prospects: The 2009 Revision. Highlights: Population Division of the Department of Economic and Social Affairs of the United Nations Secretariat.
- 4. McKinney ML (2002) Urbanization, biodiversity, and conservation. Bioscience 52: 883–890.
- 5. Luck M, Wu J (2002) A gradient analysis of urban landscape pattern: a case study from the Phoenix metropolitan region, Arizona, USA. Landscape Ecology 17: 327–339.
- 6. Zhang LQ, Wu JP, Zhen Y, Shu H (2004) A GIS-based gradient analysis of urban landscape pattern of Shanghai metropolitan area, China. Landscape and Urban Planning 69: 1–16.
- 7. Sadler JP, Small EC, Fiszpan H, Telfer MG, Niemela J (2006) Investigating environmental variation and landscape characteristics of an urban-rural gradient using woodland carabid assemblages. Journal Of Biogeography 33: 1126–1138.
- 8. Williams NSG, Schwartz MW, Vesk PA, McCarthy MA, Hahs AK, et al. (2009) A conceptual framework for predicting the effects of urban environments on floras. Journal of Ecology 97: 4–9.
- 9. Cadenasso ML, Pickett STA, Schwarz K (2007) Spatial heterogeneity in urban ecosystems: reconceptualizing land cover and a framework for classification. Frontiers in Ecology and the Environment 5: 80–88.
- 10. McDonnell MJ, Pickett STA (1990) Ecosystem structure and function along urban-rural gradients: an unexploited opportunity for ecology. Ecology 71: 1232–1237.
- 11. Tremblay MA, St Clair CC (2009) Factors affecting the permeability of transportation and riparian corridors to the movements of songbirds in an urban landscape. Journal of Applied Ecology 46: 1314–1322.
- 12. McDonnell MJ, Pickett STA, Groffman P, Bohlen P, Pouyat RV, et al. (1997) Ecosystem processes along an urban-to-rural gradient. Urban Ecosystems 1: 21–36.
- 13. Eigenbrod F, Bell VA, Davies HN, Heinemeyer A, Armsworth PR, et al. (2011) The impact of projected increases in urbanization on ecosystem services. Proceedings of the Royal Society B- Biological sciences 278: 3201–3208.
- 14. Irwin EG, Bockstael NE (2007) The evolution of urban sprawl: Evidence of spatial heterogeneity and increasing land fragmentation. Proceedings of the National Academy of Sciences of the United States of America 104: 20672–20677.
- 15. Burton E (2000) The Compact City: Just or Just Compact? A Preliminary Analysis. Urban Studies 37: 1969.
- 16. Sadler JP, Bates AJ, Hale J, James P (2010) Bringing cities alive: the importance of urban greenspaces for people and biodiversity. In: Gaston KJ, editor. Urban Ecology. Cambridge: CUP. pp. 230–260.
- 17. Dallimer M, Tang Z, Bibby PR, Brindley P, Gaston KJ, et al. (2011) Temporal changes in greenspace in a highly urbanized region. Biology Letters 7: 763–766.
- 18. McKinney ML (2008) Effects of urbanization on species richness: a review of plants and animals. Urban Ecosystems 11: 161–176.
- 19. Bates AJ, Sadler JP, Fairbrass AJ, Falk SJ, Hale JD, et al. (2011) Changing Bee and Hoverfly Pollinator Assemblages along an Urban-Rural Gradient. PLoS ONE 6: e23459.
- 20. Tratalos J, Fuller RA, Evans KL, Davies RG, Newson SE, et al. (2007) Bird densities are associated with household densities. Global Change Biology 13: 1685–1695.
- 21. Croci S, Butet A, Clergeau P (2008) Does urbanization filter birds on the basis of their biological traits? Condor 110: 223–240.
- 22. Evans KL, Chamberlain DE, Hatchwell BJ, Gregory RD, Gaston KJ (2011) What makes an urban bird? Global Change Biology 17: 32–44.
- 23. Niemela J, Kotze DJ (2009) Carabid beetle assemblages along urban to rural gradients: A review. Landscape and Urban Planning 92: 65–71.
- 24. Hahs AK, McDonnell MJ (2006) Selecting independent measures to quantify Melbourne's urban-rural gradient. Landscape and Urban Planning 78: 435–448.
- 25. McDonnell MJ, Hahs AK (2008) The use of gradient analysis studies in advancing our understanding of the ecology of urbanizing landscapes: current status and future directions. Landscape Ecology 23: 1143–1155.
- 26. Gagne SA, Fahrig L (2007) Effect of landscape context on anuran communities in breeding ponds in the National Capital Region, Canada. Landscape Ecology 22: 205–215.
- 27. Altringham JD (2003) Corbet SA, Walters SM, West R, Streeter D, Ratcliffe DA, editors. London: Harper Collins. 218 p. British Bats;.
- 28. Gehrt SD, Chelsvig JE (2003) Bat activity in an urban landscape: Patterns at the landscape and microhabitat scale. Ecological Applications 13: 939–950.
- 29. Gehrt SD, Chelsvig JE (2004) Species-specific patterns of bat activity in an urban landscape. Ecological Applications 14: 625–635.
- 30. Verboom B, Huitema H (1997) The importance of linear landscape elements for the pipistrelle Pipistrellus pipistrellus and the serotine bat Eptesicus serotinus. Landscape Ecology 12: 117–125.
- 31. Verboom B, Spoelstra K (1999) Effects of food abundance and wind on the use of tree-lines by an insectivorous bat, Pipistrellus pipistrellus. Canadian Journal of Zoology 77: 1393–1401.
- 32. Gaisler J, Zukal J, Rehak Z, Homolka M (1998) Habitat preference and flight activity of bats in a city. Journal of Zoology 244: 439–445.
- 33. Avila-Flores R, Fenton MB (2005) Use of spatial features by foraging insectivorous bats in a large urban landscape. Journal of Mammalogy 86: 1193–1204.
- 34. Threfall CG, Law B, Banks PP (2011) Sensitivity of insectivorous bats to urbanization: Implications for suburban conservation planning Biological Conservation (on-line early view) doi:10.1016/j.biocon.2011.11.026.
- 35. Lookingbill TR, Elmore AJ, Engelhardt KAM, Churchill JB, Gates JE, et al. (2010) Influence of wetland networks on bat activity in mixed-use landscapes. Biological Conservation 143: 974–983.
- 36. Boughey K, Lake I, Haysom K (in press) Biological Conservation : Improving the biodiversity benefits of hedgerows: How physical characteristics and the proximity of foraging habitat affect the use of linear features by bats. Biological Conservation.
- 37. Cornulier T, Robinson RA, Elston D, Lambin X, Sutherland WJ, et al. (2011) Bayesian reconstitution of environmental change from disparate historical records: hedgerow loss and farmland bird declines. Methods in Ecology and Evolution 2: 86–94.
- 38. Lawton JH, Brotherton PNM, Brown VK, Elphick C, Fitter AH, et al. (2010) Making Space for Nature: a review of England's wildlife sites and ecological network. London: DEFRA.
- 39. Turner T (2006) Greenway planning in Britain: recent work and future plans. Landscape And Urban Planning 76: 240–251.
- 40. Fleury A, Brown R (1997) A framework for the design of wildlife conservation corridors with specific application to southwestern Ontario. Landscape and Urban Planning 37: 163–186.
- 41. Marzluff JM, Ewing K (2001) Restoration of fragmented landscapes for the conservation of birds: A general framework and specific recommendations for urbanizing landscapes. Restoration Ecology 9: 280–292.
- 42. Angold PG, Sadler JP, Hill MO, Pullin A, Rushton S, et al. (2006) Biodiversity in urban habitat patches. Science of the Total Environment 360: 196–204.
- 43. Small E, Sadler J, Telfer M (2006) Do landscape factors affect brownfield carabid assemblages? Science Of The Total Environment 360: 205–222.
- 44. Fernandez-Juricic E (2000) Avifaunal use of wooded streets in an urban landscape. Conservation Biology 14: 513–521.
- 45. Shanahan DF, Miller C, Possingham HP, Fuller RA (2011) The influence of patch area and connectivity on avian communities in urban revegetation. Biological Conservation 144: 722–729.
- 46. Limpens HJGA, Kapteyn K (1991) Bats, their behaviour and linear landscape elements. Myotis 29: 63–71.
- 47. Downs NC, Racey PA (2006) The use by bats of habitat features in mixed farmland in Scotland. Acta Chiropterologica 8: 169–185.
- 48. Oprea M, Mendes P, Vieira TB, Ditchfield AD (2009) Do wooded streets provide connectivity for bats in an urban landscape? Biodiversity and Conservation 18: 2361–2371.
- 49. Scheffer M (2010) COMPLEX SYSTEMS Foreseeing tipping points. Nature 467: 411–412.
- 50. Blair R (1996) Land use and avian species diversity along an urban gradient. Ecological Applications 6: 506–519.
- 51. Vaughan N, Jones G, Harris S (1997) Habitat use by bats (Chiroptera) assessed by means of a broad-band acoustic method. Journal of Applied Ecology 34: 716–730.
- 52. Threfall CG, Law B, Penman T, Banks PB (2011) Ecological processes in urban landscapes: mechanisms influencing the distribution and activity of insectivorous bats. Ecography 34: 814–826.
- 53. Simon M, Hüttenbügel S, Smit-Viergutz J (2004) Ecology and Conservation of Bats in Villages and Towns. Bonn: Bundesamt für Naturschutz.
- 54. Hourigan CL, Catterall CP, Jones D, Rhodes M (2008) A comparison of the effectiveness of bat detectors and harp traps for surveying bats in an urban landscape. Wildlife Research 35: 768–774.
- 55. Jenkins EV, Laine T, Morgan SE, Cole KR, Speakman JR (1998) Roost selection in the pipistrelle bat, Pipistrellus pipistrellus (Chiroptera : Vespertilionidae), in northeast Scotland. Animal Behaviour 56: 909–917.
- 56. Davidson-Watts I, Walls S, Jones G (2006) Differential habitat selection by Pipistrellus pipistrellus and Pipistrellus pygmaeus identifies distinct conservation needs for cryptic species of echolocating bats. Biological Conservation 133: 118–127.
- 57. Sattler T, Bontadina F, Hirzel AH, Arlettaz R (2007) Ecological niche modelling of two cryptic bat species calls for a reassessment of their conservation status. Journal of Applied Ecology 44: 1188–1199.
- 58. Scott SJ, McLaren G, Jones G, Harris S (2010) The impact of riparian habitat quality on the foraging and activity of pipistrelle bats (Pipistrellus spp.). Journal of Zoology 280: 371–378.
- 59. Dietz M, Encarnacao JA, Kalko EKV (2006) Small scale distribution patterns of female and male Daubenton's bats (Myotis daubentonii). Acta Chiropterologica 8: 403–415.
- 60. Kapfer GT, Rigot T, Holsbeek L, Aron S (2008) Roost and hunting site fidelity of female and juvenile Daubenton's bat Myotis daubentonii (Kuhl, 1817) (Chiroptera : Vespertilionidae). Mammalian Biology 73: 267–275.
- 61. Eigenbrod F, Hecnar SJ, Fahrig L (2008) Accessible habitat: an improved measure of the effects of habitat loss and roads on wildlife populations. Landscape Ecology 23: 159–168.
- 62. Pascual-Hortal L, Saura S (2006) Comparison and development of new graph-based landscape connectivity indices: towards the priorization of habitat patches and corridors for conservation. Landscape Ecology 21: 959–967.
- 63. Tischendorf L (2001) Can landscape indices predict ecological processes consistently? Landscape Ecology 16: 235–254.
- 64. Tischendorf L, Fahring L (2000) On the usage and measurement of landscape connectivity. Oikos 90: 7–19.
- 65. FitzGibbon SI, Putland DA, Goldizen AW (2007) The importance of functional connectivity in the conservation of a ground-dwelling mammal in an urban Australian landscape. Landscape Ecology 22: 1513–1525.
- 66. Hostetler M, Holling CS (2000) Detecting the scales at which birds respond to structure in urban landscapes. Urban Ecosystems 4: 25–54.
- 67. Garden JG, McAlpine CA, Possingham HP (2010) Multi-scaled habitat considerations for conserving urban biodiversity: native reptiles and small mammals in Brisbane, Australia. Landscape Ecology 25: 1013–1028.
- 68. Neuman M (2005) The compact city fallacy. Journal of Planning Education and Research 25: 11–26.
- 69. Nystrom M, Folke C (2001) Spatial resilience of coral reefs. Ecosystems 4: 406–417.
- 70. Ernstson H, van der Leeuw SE, Redman CL, Meffert DJ, Davis G, et al. (2010) Urban Transitions: On Urban Resilience and Human-Dominated Ecosystems. Ambio 39: 531–545.
- 71. Goddard MA, Dougill AJ, Benton TG (2010) Scaling up from gardens: biodiversity conservation in urban environments. Trends in Ecology & Evolution 25: 90–98.
- 72. Conroy MJ, Allen CR, Peterson JT, Pritchard LJ, Moore CT (2003) Landscape change in the southern Piedmont: challenges, solutions, and uncertainty across scales. Conservation Ecology 8: 3.
- 73. Ordnance Survey (2008) OS Mastermap Topography Layer [GML geospatial data], coverage: Birmingham, Black Country and Solihull, Updated: November 2008, Ordnance Survey (GB), using: EDINA Digimap Ordnance Survey Service, http://edina.ac.uk/digimap, downloaded: December 2008.
- 74. Beyer HL (2004) Hawth's Analysis Tools for ArcGIS. Freeware from http://wwwspatialecologycom/htools.
- 75. Davidson-Watts I, Jones G (2006) Differences in foraging behaviour between Pipistrellus pipistrellus (Schreber, 1774) and Pipistrellus pygmaeus (Leach, 1825). Journal of Zoology 268:
- 76. Owen SM, MacKenzie AR, Bunce RGH, Stewart HE, Donovan RG, et al. (2006) Urban land classification and its uncertainties using principal component and cluster analyses: A case study for the UK West Midlands. Landscape and Urban Planning 78: 311–321.
- 77. Duguay S, Eigenbrod F, Fahrig L (2007) Effects of surrounding urbanization on non-native flora in small forest patches. Landscape Ecology 22: 589–599.
- 78. Russ J (1999) The Bats of Britain and Ireland - Echolocation Calls, Sound Analysis, and Species Identification. Totnes: Alana Ecology. 80 p.
- 79. Fischer J, Stott J, Law BS, Adams MD, Forrester RI (2009) Designing effective habitat studies: quantifying multiple sources of variability in bat activity. Acta Chiropterologica 11: 127–137.
- 80. Corben C (2007) Weather protection for Anabat SD1 bat detector. http://userslminet/corben/Weather Protectionhtm - Weather.
- 81. Corben C (2009) AnalookW bat call analysis software. Version 3.7u, freeware from http://www.hoarybat.com.
- 82. Stone EL, Jones G, Harris S (2009) Street Lighting Disturbs Commuting Bats. Current Biology 19: 1123–1127.
- 83. Cushman SA, McGarigal K (2004) Hierarchical analysis of forest bird species-environment relationships in the Oregon Coast Range. Ecological Applications 14: 1090–1105.
- 84. Wheater CPC (2000) Using statistics to understand the environment. London: Routledge.
- 85. R Development Core Team (2010) R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN3-900051-07-0, URL http://www.R-project.org.
- 86. Zuur AF, Ieno EN, Elphick CS (2010) A protocol for data exploration to statistical problems. Methods in Ecology and Evolution 1: 3–14.
- 87. McCullagh P, Nelder JA (1983) Generalized Linear Models. London: Chapman and Hall.
- 88. Hastie T, Tibshirani RJ (1990) Generalised Additive Models. London: Chapman and Hall.
- 89. O'Hara RB, Kotze DJ (2010) Do not log-transform count data. Methods in Ecology & Evolution 1: 118–122.
- 90. Zuur AF, Ieno EN, Walker NJ, Saveliev AA, Smith GM (2009) Mixed Effects Models and Extensions in Ecology with R. New York: Springer.
- 91. Pebesma EJ (2004) Multivariate geostatistics in S: the gstat package. Computers and Geoscience 30: 683–691.
- 92. Burnham KP, Anderson DR (2002) Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. New York: Springer.