Species distribution modeling is widely used in ecology to estimate the relationship between target species and environmental variables associated with the species' habitat. To estimate the habitat map, Maxent is recognized as one of the most powerful tools, in which the model's parameters are estimated by sequentially maximizing the likelihood. However, computing the likelihood becomes extremely expensive when the number of locations in the study area is large. Therefore, we propose a new species distribution modeling approach based on gamma-divergence, in which the normalization term in the gamma-loss function is approximated using cumulant coefficients for the study area. This results in a computationally efficient estimation of the model's parameters, as well as high estimation accuracy. Moreover, by using the cumulant-based approximation, we show that the estimation of Maxent is equivalent to Fisher linear analysis when the normality assumption holds. We demonstrate the proposed method using simulation studies and a dataset of Japanese vascular plants.

Please click here for participating the talk online.