Abstract
Optical remote sensing is an important tool in the study of animal behaviorproviding ecologists with the means to understand species–environment interactions in combination with animal movement data. However, differences in
spatial and temporal resolution between movement and remote sensing data limit their direct assimilation. In this context, we built a data-driven frameworkto map resource suitability that addresses these differences as well as the limitations of satellite imagery. It combines seasonal composites of multiyear surfacereflectances and optimized presence and absence samples acquired with animalmovement data within a cross-validation modeling scheme. Moreover, itresponds to dynamic, site-specific environmental conditions making it applicable to contrasting landscapes. We tested this framework using five populations of White Storks (Ciconia ciconia) to model resource suitability related to foraging achieving accuracies from 0.40 to 0.94 for presences and 0.66 to 0.93 for absences. These results were influenced by the temporal composition of the seasonal reflectances indicated by the lower accuracies associated with higher day differences in relation to the target dates. Additionally, population differences
in resource selection influenced our results marked by the negative relationship between the model accuracies and the variability of the surface reflectances associated with the presence samples. Our modeling approach spatially splits
presences between training and validation. As a result, when these represent different and unique resources, we face a negative bias during validation. Despitethese inaccuracies, our framework offers an important basis to analyze species environment interactions. As it standardizes site-dependent behavioral and environmental characteristics, it can be used in the comparison of intra- andinterspecies environmental requirements and improves the analysis of resource selection along migratory paths. Moreover, due to its sensitivity to differences in resource selection, our approach can contribute toward a better understanding of species requirements.