Modeling Bottom Dissolved Oxygen on the East China Sea Shelf Using Interpretable Machine Learning

Monitoring bottom dissolved oxygen (DO) is crucial for understanding hypoxia, a threat to marine ecosystems and fisheries. However, traditional observations are limited in spatiotemporal coverage, while numerical models consume tremendous computing resources. This study develops an interpretable machine learning framework to simulate the bottom DO distribution on the East China Sea (ECS) shelf and explore its potential driving mechanisms. By integrating remote sensing, in situ observations, and numerical model outputs, the framework generates high-resolution (4 km) DO estimates from 1998 to 2024. Validation against independent datasets confirms the improved accuracy and spatial resolution, with an RMSE below 1 mg/L. The results reveal a persistent decline in DO, strongly linked to rising sea surface temperature (SST), underscoring the role of surface warming in bottom water deoxygenation. Model interpretability further identifies the SST and bathymetry as key factors. This framework provides a robust tool for assessing bottom DO trends, hypoxia, and their ecological impacts, supporting future monitoring and management of the ECS shelf.