Interpretable machine learning reveals the importance of geography and landscape arrangement for surface water quality across China

Abstract

Elucidating the influence of land use patterns on surface water quality is crucial for effective watershed management. Despite numerous studies in individual watersheds, factors influencing water quality in diverse geographical environments are less understood due to data and methodological constraints in large-scale studies. This study employs Interpretable Machine Learning (IML) to explore the drivers of water quality variations across 234 watersheds in China. Results reveal that urban land is the primary source of nitrogen pollution, while rural residences contribute substantially to phosphorus pollution. Water bodies are key sinks for both nutrients. Climate and land use compositions show substantial variations across watersheds with distinct geographical locations. These geography-related factors together contributed 82%-89% relative importance to water quality variations across China, implicating the dominant role of geography in shaping water quality. Additionally, the spatial arrangements of source-sink landscapes exhibit greater variations within the same geographic zone, whose impact on water quality is also inevitable. This highlights the potential to enhance water quality via optimizing landscape spatial arrangements given current land use composition and production routines that have been adapted to geographical conditions. Our study demonstrates the utility of IML in discerning key factors affecting water quality in large-scale assessments, offering valuable insights for targeted watershed management strategies.