Using regression trees to predict alpha diversity based upon geographical and habitat characteristics

Different environmental factors act as driving forces of diversity at different scales of analysis; and also the effect of one environmental factor changes as the scale of analysis changes. Most studies rely on multiple regression models, and such models tend to mix-up the effect of all factors and assume that factors effects are additive. We believe that the effect of environment on diversity should be characterized by a hierarchical structure with coarse scale factors, like geographical tropics to poles gradients, defining the envelope of possible diversity conditions, and other more local factors, like habitat structure, being responsible for the fine tuning of diversity. This structure is most efficiently modeled with regression trees. We show that for six habitat types in Greek protected areas regression tree models were able to describe plant species richness based upon environmental factors considerably more efficiently than multiple regression models. More importantly when the models were extrapolated to other sites in Greece, outside their domain, the differences between the predictive ability of the two approaches was magnified. The tree models picked up important ecological characteristics, and a hierarchical structure that used coarse scale factors, like latitude and longitude, for the coarse scale estimate of alpha diversity, and finer scale factors like fragmentation, for the fine-tuning of the estimation. Therefore, we advocate that the regression tree methodology is most appropriate for modeling the relationship between diversity and environmental factors, and the use of the classical regression approaches might be misleading.