Ecological Informatics (Dec 2025)
Combining environmental DNA and remote sensing variables to model fish biodiversity in tropical river ecosystems
Abstract
Tropical river ecosystems face substantial threats, leading to a sharp decline in their biodiversity. High-resolution data on the spatial distribution of biodiversity is essential for devising effective conservation strategies. However, biodiversity information is limited because traditional assessment methods often face challenges in these vast, inaccessible environments. Here, we aim to assess whether combining large-scale environmental DNA (eDNA) data with environmental variables generated from remote sensing images in machine learning models can overcome this limitation. We used a fish eDNA dataset of 264 samples collected from major tropical rivers—the Casamance, Cuando, Cunene, Okavango, and Zambezi (Africa); the Magdalena, Maroni, and Oyapock (South America); and the Kinabatangan (Southeast Asia)—together with aquatic and terrestrial variables derived from remote sensing imagery. Based on this data, we constructed both river-specific and multi-river Random Forest models to predict fish species richness and the Shannon diversity index. The models exhibited a good fit to the data, indicating the suitability of variables in capturing the determinants of fish biodiversity in these rivers. Moreover, the models effectively predicted the metrics during cross-validation, underscoring their utility in generating biodiversity maps along large tropical rivers. Although predictions for unencountered rivers remain challenging, the models are able to capture large-scale patterns. With further refinement and expansion through additional data, this integrated approach holds promise for generating biodiversity insights without extensive on-site sampling requirements. Our study highlights the potential of combining eDNA with remote sensing variables to model biodiversity patterns in tropical river ecosystems.
Keywords
