Impact of model complexity on cross-temporal transferability in Maxent species distribution models: An assessment using paleobotanical data

摘要: Abstract Maximum entropy modeling (Maxent) is a widely used algorithm for predicting species distributions across space and time. Properly assessing the uncertainty in such predictions non-trivial requires validation with independent datasets. Notably, model complexity (number of parameters) remains major concern relation to overfitting and, hence, transferability Maxent models. An emerging approach validate cross-temporal using paleoecological data. In this study, we assess effect on performance projections time two European plant ( Alnus glutinosa (L.) Gaertn. Corylus avellana L.) an extensive late Quaternary fossil record Spain as study case. We fit 110 models different levels under present tested AUC (area receiver operating characteristic curve) AICc (corrected Akaike Information Criterion) through standard procedure randomly partitioning current occurrence then compared these results by projecting mid-Holocene (6000 years before present) climatic conditions their ability predict pollen presence–absence abundance. find that calibrating default settings result generation overly complex While increased when distributions, it was higher intermediate distributions. Hence, resulted best trade-off Reliable temporal especially relevant forecasting future climate change. Consequently, species-specific tuning should be control complexity, notably data independently projections. For which not available, selected.