Unveiling Landscape-Level Drivers of Freshwater Biodiversity Dynamics

Abstract

Human activities severely impact biodiversity, particularly in freshwater lakes. These habitats provide critical ecosystem services and, at the same time, suffer from river inflow, agricultural runoff, and urban discharge. DNA-based techniques are preferred for monitoring biodiversity due to their effectiveness. However, pinpointing the causes of biodiversity decline across landscapes poses challenges due to the complex interactions between biodiversity and environmental drivers. In this study, we used an explainable multimodal machine learning approach that can integrate different types of data, such as biological, chemical, and physical data, to discover potential causes of biodiversity dynamics. This is done by identifying relationships between environmental drivers—plant protection products, physico-chemical parameters and typology- and community biodiversity changes in 52 lake ecosystems. By analyzing benthic and pelagic lake communities, we found significant correlations between biodiversity and environmental drivers, such as plant protection products. Furthermore, our analysis allowed us to identify factors within these drivers responsible for biodiversity dynamics. Specifically, insecticides and fungicides were identified as the most important factors, followed by 43 physico-chemical factors, including many heavy metals. Our holistic, data-driven approach provides insights into large-scale biodiversity changes and could inform conservation efforts and regulatory interventions to protect biodiversity from pollution.