Understanding and predicting the effect of landscape composition on biodiversity and ecosystem service delivery is a key issue for environmental science. Most existing approaches arbitrarily choose one (sometimes a few) scales at which to analyse these effects. However, it is well known that different species or taxonomic groups use landscapes differently, thus must respond to landscape effects at different scales. In addition, environmental variables also vary in the distance at which they can impact a species or some ecosystem process. Here we present a method to directly estimate the distance at which a set of variables should be extracted to study landscape effects. The method relies on Bayesian estimation and incorporates landscape effects as smoothly decaying with distance. We test the effect of different scale kernels and different prior distribution for the scale parameter on simulation data with known “true” values. We then apply the method to model the response of farmland bird abundance to landscape composition, using monitoring data from Sweden.
The method captures the variability of species’ responses to their surrounding landscapes. When tested on simulated data, the models estimate the known values with high accuracy in most cases, but fails to do so with extremely high zero inflation. The estimates of landscape scale of effect for farmland birds reflect well what we know of the species. In addition to getting a clear measure of the landscapes’ scale of effect, using Bayesian Inference allows us to quantify uncertainty around the estimates and to incorporate prior knowledge about the system. Here we discuss our results emphasizing both ecological and methodological aspects. Joining in the call for more functional approaches in landscape ecology, we provide a method that is conceptually simple, and yet powerful. Our approach brings new insights into landscape ecology, and offers a promising path to model biodiversity and ecosystem services in complex landscapes.