Fig 1.
Map of the species’ roost records used from the Lake District National Park, NW England.
Sample size: P. pipistrellus—129; P. pygmaeus—80; N. noctula—10; M. brandtii / mystacinus—24; M. daubentonii—51; M. nattereri—23; P. auritus—102. NB. Drawn at this scale, overlap between roosts masks many species’ roost locations. See Fig 5 for separate species’ roost maps. Crown database right 2010. An Ordnance Survey/EDINA supplied service.
Table 1.
The fifteen habitat variables used for analysis.
All layers were produced at 8 different spatial scales except the two distance variables.
Fig 2.
Distance to water and woodland edge response curves.
These graphs show probability of a species’ roost (p) at a location based on these distances and are based on the results of univariate models to prevent any other interacting or collinear variables affecting the relationships found. Variables which were found to have poor predictive power for a species (AUC ≤ 0.5 or test gain < 0.01) are not shown.
Fig 3.
These graphs show the strength of association (as test AUC) between each species’ presence and individual environmental variables at different spatial scales. The average predictive power of the distance variables is shown as a dashed line: these were independent of scale. Environmental variables with a predictive power ≤ 0.5 are no better than random. Only variables retained in pruned models are shown. NB. The scale range is not linear for improved clarity at small scale.
Table 2.
Measured using random 5-fold cross validation (P. pipistrellus, P. pygmaeus, M. daubentonii and P. auritus), or jackknife validation (N. noctula, M. brandtii/mystacinus and M. nattereri) tests.
Table 3.
Mean model performance from spatially constrained 5-fold cross validation.
rSAC lag = largest distance within which data pairs retained significant, positive spatial autocorrelation.
Fig 4.
Representative MaxEnt response curves.
These graphs show the probability of a species’ presence at a location for a range of parameters. These graphs are based on univariate models to prevent interacting or collinear variables from affecting the relationships modelled. Variables found to have poor predictive power for a species (AUC ≤ 0.5 or test gain < 0.01) are not shown.
Fig 5.
Habitat suitability maps made using each species’ pruned set of variables.
HSI = Habitat Suitability Index. The Lake District National Park boundary is marked in white. Species roost locations are coloured by test subsets where appropriate. Crown database right 2010. An Ordnance Survey/EDINA supplied service.