Spatiotemporal Analysis of River Health Using Remote Sensing and Machine Learning Approaches

Abstract

Disclosing the spatiotemporal (ST) fluctuations of nutrients in coastal waters is essential for comprehending and assessing the coastal environment, consequently offering practical recommendations for aquatic restoration. This study introduced a Machine Learning (ML) that integrates ST data, facilitating the establishment of quantitative connections between determined external variables and extensive satellite imagery, while reducing estimation errors exceeding 40% compared to traditional ML models lacking ST integration. The ST patterns of Dissolved Inorganic Nitrogen (DIN) and Dissolved Inorganic Phosphorus (DIP) throughout 45000 km² of the Sea were acquired on an 8-day interval. The ST variations illustrated the water quality trends, revealing fluctuations of two critical nutrients in harbors affected by complex anthropogenic impacts, typical waterways with multiple river components, and open oceans with significant fisheries. Despite a 25% and 20% reduction in DIN and DIP concentrations over nine years, the inshore ocean's water condition has not improved, particularly during fall and winter. The research conducted a quantitative analysis of the primary causes of water degradation and offered scientific recommendations for focused surveillance and regional cooperation in governance.