Species Distribution Models Overfitting Effect on Hidden Animal Diversity Estimation: Implications for Conservation Decision-Making

作     者：Matus-Olivares, Camilo Carrasco, Jaime Yela, José Luis Meli, Paula Weintraub, Andrés Gil, Fulgencio Lisón 

作者机构：Wildlife Ecology and Conservation Lab Departamento de Zoología Facultad de Ciencias Naturales y Oceanográficas Universidad de Concepción Casilla 160-C Concepción Chile Doctorado en Ciencias Agroalimentarias y Medioambiente Facultad de Ciencias Agropecuarias y Forestales Universidad de La Frontera Temuco Chile Laboratorio de Ecología del Paisaje y Conservación Departamento de Ciencias Forestales Facultad de Ciencias Agropecuarias y Medioambiente Universidad de La Frontera Box 54-D Temuco Chile Department of Industrial Engineering Universidad de Chile Santiago Chile Complex Engineering System Institute – ISCI Santiago Chile Departamento de Industria Facultad de Ingeniería Universidad Tecnológica Metropolitana Santiago Chile Grupo DITEG Facultad de Ciencias Ambientales y Bioquímicas Universidad de Castilla-La Mancha Avda. Carlos III s.n. Campus Real Fábrica de Armas Toledo45005 Spain Laboratorio de Estudios del Antropoceno Facultad de Ciencias Forestales Universidad de Concepción Concepción Chile 

出 版 物：《SSRN》 

年 卷 期：2024年

核心收录：

主　　题：Conservation 

摘      要：The geographical distribution patterns of animal species communities are essential to design and implement ecological applications and conservation measures successfully. However, to know accurately the species distribution is complex mainly due to three factors: 1) undersampled areas, 2) sampling methodologies not always effective, and 3) uncertainty in the true absence. Therefore, conceptually, there exists a hidden animal diversity (HAD) that is possible to measure and quantify, especially through species distribution models (SDMs). Here, we analyzed the overfitting effect of different SDM algorithms over the HAD estimates. Specifically, we a) compare and assess the overfitting levels and predictive performance of different SDM algorithms;b) examine the overfitting effect of different SDM algorithms over HAD estimates and;c) create and assess an evaluation method that allows choosing the most suitable SDM algorithm according to a jointly evaluation of metrics of predictive performance and overfitting. Our results showed that those algorithms with high overfitting levels were worse at predicting hidden animal diversity. We found a significative negative relationship between the expected value of HAD and the level of overfitting of an SDM algorithm. This result is reflected in the distribution maps and it becomes an issue in the decision-making process for the conservation policies. Our findings help improve HAD estimation and have applications in conservation and restoration, providing critical issues on how allocating resources in space, taxonomic groups, or functional guilds. Moreover, we provide a flexible evaluation method that considers the overfitting when evaluating an SDM algorithm. © 2024, The Authors. All rights reserved.