Occurrence-based models reveal greater spatial details in tropical species richness

Abstract

Species richness maps are essential for biodiversity assessment and conservation planning, but their accuracy is limited by uneven sampling effort and large data gaps, especially in tropical regions. Here, we combined the Uniform Sampling from Sampling Effort (USSE) framework with deep neural networks to generate occurrence-based richness maps for tropical vertebrates, invertebrates and plants. The USSE approach incorporates sampling intensity during model calibration and projects richness under a standardized high-sampling scenario, reducing the influence of spatial sampling bias. Model predictors included climatic, topographic and remotely sensed habitat-conditions, allowing richness predictions to reflect both environmental gradients and habitat alteration. Predictive performance was high across taxonomic groups, with spatially blocked cross validated R² values ranging from 0.60 to 0.95, and a value of 0.90 for the combined multi taxon model. Independent validation against 107 published field inventories yielded R² values between 0.55 and 0.87. For terrestrial vertebrates, predicted richness patterns were broadly consistent with International Union for Conservation of Nature polygon-based maps in areas of high ecological integrity, while revealing greater fine-scale spatial heterogeneity associated with habitat variation. Across groups, richness was generally highest in tropical forest regions. Similarity analyses indicated that invertebrate groups, particularly Lepidoptera, showed strong agreement with overall multi-taxon richness patterns. Our results show that USSE combined with deep learning provides a scalable framework for producing spatially detailed biodiversity maps that account for sampling bias and current habitat conditions, offering a complementary tool for conservation prioritization in tropical regions.