Conventional habitat selection models using log-linear or more sophisticated regressions often failed to predict species distribution out of the region where the models were fitted. We created colony site selection models using a random forest algorithm for herons and egrets in two nearby regions (Ibaraki and Chiba prefecture) where landscape patterns are slightly different. We found that the best models for explaining the colony distributions share the common land use variables between the two regions, but on the different scales. Herons and egrets showed some flexibility to adapt their scale for habitat selection depending on complexity of landscape structure in human influenced landscapes.