A0291
Title: Cumulant-based approximation for fast and efficient prediction of species distribution
Authors: Osamu Komori - Seikei University (Japan) [presenting]
Shinto Eguchi - The Institute of Statistical Mathematics (Japan)
Yusuke Saigusa - Yokohama City University (Japan)
Abstract: Species distribution modeling plays a crucial role in estimating the habitat suitability of species based on environmental variables. Popular and powerful methods, such as Maxent and the Poisson point process, are extensively employed across ecological and biological sciences. However, these methods face significant computational challenges, particularly when dealing with large background datasets, as is often the case with fine-resolution data or global-scale estimations. To address this issue, a computationally efficient species distribution model is proposed using a cumulant-based approximation (CBA) applied to the loss function of -divergence. Additionally, a sequential estimation algorithm is introduced, incorporating an L1 penalty to identify key environmental variables strongly associated with species distribution. The regularized geometric-mean method, derived from the CBA, demonstrates both high computational efficiency and robust estimation accuracy. Furthermore, by applying CBA to Maxent, an equivalence is established between Maxent and Fisher's linear discriminant analysis under a normality assumption. This equivalence enables the development of a highly efficient computational approach for estimating species distributions. The effectiveness of the proposed methods is demonstrated through simulation studies and analyses of datasets, including 226 species from the National Centre for Ecological Analysis and Synthesis and 709 Japanese vascular plant species.
