Integrating Multi-Source Environmental Variables with Sentinel-3 OLCI Imagery for Interpretable Retrieval of Eutrophication Parameters in Bohai

Abstract

Satellite remote sensing technology, characterized by high spatiotemporal resolution and long-term continuous observations, has become a critical tool for water quality monitoring. Excessive inputs of nutrients, particularly nitrogen and phosphorus, into marine environments can induce eutrophication and a range of associated ecological problems. Dissolved Inorganic Nitrogen (DIN) and Soluble Reactive Phosphorus (SRP) are key parameters governing water quality. However, their inherently weak spectral response poses a persistent challenge for the construction of accurate satellite-based inversion models. Incorporating environmental variables into model input features alongside spectral reflectance represents a promising approach to improving inversion performance. Taking the Bohai Sea as a case study, multi-source environmental variables were integrated with remote sensing spectral features, and the optimal feature combination and corresponding model were identified using Bayesian optimization and an iterative feature-importance screening strategy. Bayes-CatBoost demonstrated superior performance in estimating DIN and SRP, achieving R2 values of 0.82 and 0.68, respectively. After incorporating multi-source environmental variables, inversion errors decreased by 17.41% for DIN and 16.79% for SRP relative to models based solely on multispectral remote sensing data. Application of the proposed model to Sentinel-3 OLCI imagery enabled the generation of daily DIN and SRP distributions for the Bohai Sea. The results indicate that from 2018 to 2023, DIN and SRP concentrations in the Bohai Sea exhibited an overall stable but slightly declining trend with pronounced seasonal variability, accompanied by a gradual improvement in water quality.