Hypoxia is an escalating environmental concern in low-latitude river-dominated shelf regions, where strong vertical stratification suppresses ventilation, thereby limiting the replenishment of dissolved oxygen from the surface to subsurface waters. The northern Indian Ocean exemplifies a region where persistent low-oxygen conditions prevail. In recent years, intensified stratification has been accompanied by declining primary productivity and a continued decline in subsurface dissolved oxygen, which has now approached the hypoxic threshold.
To investigate the mechanisms underlying this transition, this study develops a Biogeochemistry-Informed Neural Network (BINN) modeling framework for northern Indian Ocean. BINN integrates a vectorized process-based physical-biogeochemical model into a neural network structure, capturing the coupled dynamics of nutrient cycling and oxygen consumption, and enables robust uncertainty quantification in the estimation of key biogeochemical parameters. BGC Argo observational datasets are integrated with machine learning reconstructions of oxygen to inform the BINN framework, allowing us to build a high-resolution model that better resolves the mechanistic coupling between riverine inputs, nutrient cycling, and oxygen dynamics. This study will advance our understanding of how riverine inputs and upper-ocean stratification jointly regulate the onset and intensification of hypoxia, and provide a scientific basis for ecological risk assessment and hypoxia mitigation in river-influenced marginal seas.