Ecological Niche Modeling for Rare Species

Abstract

Ecological niche modelling (ENM) -- using species occurrence records and environmental predictor layers to characterise the climatic and abiotic conditions within which a species can maintain viable populations -- is the primary tool for predicting species distributions across unsampled areas and future climate scenarios, with particular value for rare and data-deficient species where conventional distribution mapping from exhaustive surveys is impractical. Rare species with few occurrence records present a fundamental ENM challenge: standard MaxEnt and ensemble SDM approaches require minimum 20-50 records for stable model performance, yet the most conservation-critical species are often those with fewer records available. This study evaluated and compared seven ENM approaches for rare-species applications across 284 rare European plant, invertebrate, and vertebrate species (1-49 occurrence records per species after BDHF cleaning; from GBIF + national expert databases) with independent validation data held out from model fitting, identifying the algorithms, predictor selection strategies, and regularisation settings that maximise ENM performance at low sample sizes. Ensemble modelling (weighted mean of MaxEnt + BRT + Random Forest; weights from 5-fold CV AUC) consistently outperformed single-algorithm approaches for all sample size categories (AUC 0.884 at n = 20-49 records vs. AUC 0.724 for MaxEnt alone). A Rare Species Niche Model Reliability Index (RSNMRI) integrating model AUC, niche breadth stability, predictor collinearity, and sample size predicted independent validation AUC with r = +0.84, providing a standardised quality indicator for rare species ENM outputs used in conservation planning.