High-resolution spatio-temporal monitoring of coastal hypoxia using machine learning models and GOCI-MODIS satellite...

While hypoxia is conventionally defined as a dissolved oxygen (DO) concentration below 2 mg/L, this study adopts a more conservative 3 mg/L threshold as an early-warning indicator. If hypoxia sustained for an extended period, dead zones can form, posing serious threats to marine ecosystems. The frequency of hypoxia events is increasing globally, making early detection and response critical. However, high spatiotemporal resolution hypoxia detection technologies remain limited worldwide. This study developed a machine learning-based regression model to estimate coastal bottom DO concentration and a classification model to detect hypoxia using satellite remote sensing data across Korean coastal waters (2012-2020). Through optimal model and predictor selection, as well as accuracy improvement techniques, the final regression models achieved a determination coefficient (R²) of 0.79, while the classification model demonstrated an 85.7 % accuracy in detecting hypoxia. In addition, we confirmed that model performance could be further improved by incorporating additional predictors such as bottom temperature and stratification indicators. With further improvements, the satellite- and machine learning-based hypoxia detection system developed in this study could serve as an effective tool for monitoring and responding to coastal hypoxia events.