Extrapolation of random forest models shows scale adaptation in egret colony site selection against landscape complexity

• We create random forest models for colony site selection of herons and egrets.
• We apply those models to two landscapes with different complexity.
• We extrapolate models by swapping landscapes and scales.
• Accurate model for one landscape is accurate to the other with different scales.
• The landscape with higher complexity favors smaller scales for site selection.

Availability of certain habitats or landscape configurations can cause differential habitat selection in animal species. Landscape complexity can affect foraging scales, home ranges and movement, but its effect on habitat selection is not well documented. We aimed to examine differences in colony site selection of herons and egrets in different regions. We studied whether landscape complexities could affect their scale of selection and habitat preferences. We used colony distribution data and land-use maps for two neighboring regions, Ibaraki and Chiba prefectures in Japan, to create random forest models for analyzing habitat preferences and important scales of selection. We did cross-validation of the models, adjusted for its respective region's land-use maps with changing scales. The scales that best explained colony distribution were 1-, 4-, 10- and 15-km in the Ibaraki region, and 1- and 10-km in the Chiba region. Evergreen forest was the most important variable for Ibaraki at 4-km and for the Chiba at 1-km. The importance of other variables differed for other models. Cross-validation showed that herons and egrets had the same habitat preferences at a 4-km scale in Ibaraki and at a 1-km scale in Chiba. The scale of selection was reduced in Chiba, where the main foraging resources for herons and egrets was more complex. Differences in landscape complexities did not affect habitat preferences but resulted in differences in the scale of selection. Habitat selection models created at the landscape level can be useful to study behavioral aspects difficult to describe with direct observation in detail.