A novel ensemble framework for background selection to improve species distribution models

Species distribution modeling (SDM) is widely used to predict the spatial distribution of species. However, it faces challenges in terms of model performance and reliability because of the uncertainties associated with presence records and the representation of species absence through background points. Here, we propose a novel ensemble-based background selection method that extracts background points from areas with low habitat suitability, as identified using CLIMEX model predictions. This method was developed as an effective strategy for developing high-performance ensemble-based SDMs. We applied the ensemble method three species with different sample sizes and regional distribution patterns: the taro caterpillar ( Spodoptera litura Fabricius, 1775), the spotted lanternfly ( Lycorma delicatula White, 1845), and the red imported fire ant ( Solenopsis invicta Buren, 1972). The method was compared with traditional methods (random and bias-based approaches) using MaxEnt and Random Forest under various conditions (spatial filtering and evaluation scenarios). The ensemble-based approach minimized the uncertainty in background point selection by drawing points from areas with low predicted habitat suitability, thereby increasing the likelihood of representing true absences. The novel approach consistently outperformed the traditional methods across all evaluation scenarios. The average true skill statistic value across all conditions were 0.63 for the random selection method, 0.34 for the biased selection method, and 0.81 for the ensemble-based method, demonstrating the superior predictive performance of the ensemble approach. The model-based ensemble background selection method can be used as a practical framework for developing high-performance SDMs, ultimately enhancing the applicability of SDMs for predicting the spatial distribution of species. • The novel ensemble-based method was developed to select background points. • The ensemble method is compared with traditional methods (e.g. random and biased). • The novel approach outperformed the traditional methods across evaluation scenarios. • This study provides a framework for improving the accuracy and utility of SDMs.