Landscape connectivity has emerged as a major focus of conservation biology as landscapes have become increasingly fragmented by human land use. Several approaches to modeling locations of dispersal and migration corridors in complex landscapes have been developed and are now considered integral tools for conservation planning. However, the predictive performance of these models has never been rigorously assessed, limiting confidence in their application. To address this issue, I used GPS data from migrating elk in Madison Valley, MT to evaluate the accuracy of migration routes predicted by two modeling methods: widely used permeability (least cost path) models and more novel circuit theory models. Resource selection functions (RSF) quantifying elk habitat suitability on multiple spatial scales and in both spring and fall were first developed from a subset of the data using matched case-control logistic regression, which allows inclusion of time-varying covariates such as NDVI and snow water equivalent (SWE). Resulting habitat suitability maps served as inputs to corridor models. Corridor model outputs were evaluated against the remaining GPS data both directly in terms of spatial overlap of GPS points with predicted high-quality corridors and indirectly in terms of habitat characteristics of used and predicted paths.
Results/Conclusions
RSF model selection indicates topography, land cover, land ownership, road density, and snow water equivalent contribute to migration habitat suitability at multiple spatial scales and in both spring and fall. NDVI was also marginally important. Predicted corridor locations were generally similar between the two methods; predictive accuracy, therefore, was also similar. Preliminary analyses suggest that both methods perform very well, with over 90% of GPS test data falling within predicted corridors with quality values in the 95th percentile. Both methods also fared well when evaluated in terms of similarity of habitat characteristics of observed and predicted paths. Circuit theory models appear to slightly outperform least cost path in this high topographic relief landscape due to their greater emphasis on identifying bottlenecks such as those formed by narrow canyons linking Madison Valley summer and winter ranges. While more comprehensive analyses will be presented, these initial results indicate that although some uncertainty exists, commonly used methods of predicting corridors to be targeted in conservation planning are indeed informative. Moreover, the approach demonstrated here, which enables incorporation of phenological variables predicted to be affected by global warming, sets the stage for critical analyses of shifts in corridor locations under projected climate change scenarios.
