Construction of long-term (2003–2024) DIN dataset in the Chinese coastal ocean using interpretable ML

Monitoring dissolved inorganic nitrogen (DIN) concentration in coastal ocean is essential for better understanding eutrophication processes, biogeochemical cycling, and ecosystem responses to anthropogenic and climate changes. Traditional field-based monitoring is limited by sparse coverage and high costs, while satellite retrieval of DIN remains challenging due to its complex correlation with optical signals. We developed an interpretable machine learning (ML) approach to generate the first 1-km resolution, two-decade (2003–2024) monthly surface DIN dataset for the entire Chinese coastal ocean, filling a critical gap in long-term large-scale coastal nutrient monitoring. Our model integrated spectral remote sensing reflectance ( R rs (λ)) with physical and biogeochemical parameters, leveraging CatBoost algorithm and SHapley Additive exPlanations (SHAP) to achieve accurate and interpretable DIN retrieval. The satellite-derived DIN product obtained strong performance (RMSE = 0.12 mg L −1 ) on independent validation dataset. SHAP analysis revealed distinct nearshore-offshore driving mechanisms, with nearshore DIN dominated by turbidity ( R rs (667)) and freshwater input (SSS), and offshore DIN controlled by stratification (MLD), clarifying the biogeochemical drivers of DIN dynamics. Along the Chinese coastal ocean, DIN generally peaks in winter due to vertical mixing and weakened stratification, exhibiting a distinct inshore-to-offshore gradient from May to September near major estuaries driven by monsoonal rainfall and expanded river plumes. During 2003–2024, DIN increased at rates of 1.5 to 1.8 μg L −1 yr −1 in the Yangtze River and Pearl River estuaries, but decreased at rates of −1.9 to −0.7 μg L −1 yr −1 in the Yellow River Estuary and the outer shelf waters of the East China Sea (ECS). This interpretable ML model facilitates large-scale and continuous DIN monitoring, delivering new insights to support the sustainable management of coastal marine ecosystems.