Abstract

This study proposes a high-precision method for retrieving sea surface salinity (SSS) using GF-1 satellite imagery, focusing on Qinzhou Bay along the Guangxi coast. The analysis identified the spectral index B3×B4/(B1×B2) as having the strongest correlation with SSS (R = 0.929). To enhance the performance of the Back Propagation Neural Network (BPNN) model, optimization algorithms including Improved Grey Wolf Optimization (IGWO), Particle Swarm Optimization (PSO), and White Shark Optimization (WSO) were applied. Comparative results show that IGWO significantly optimized network weights and thresholds, yielding superior test performance metrics (MAE = 0.906 psu, MAPE = 4.124%, RMSE = 1.067 psu, and R2 = 0.953), demonstrating strong generalization ability. Validation using third-party data indicated accuracy reductions of 10.9% and 8.6% in Qinzhou Bay and Tieshan Port, respectively, highlighting the model’s robustness and broad applicability. SSS retrieval results for Qinzhou Bay in 2023 revealed significant spatial and seasonal variations: the Inner Bay exhibited lower salinity (average 14 psu) from April to September due to freshwater inflows, while salinity increased (average 22 psu) from November to February. The Outer Bay, influenced by its connection to the South China Sea, maintained consistently high salinity levels (25–30 psu) year-round. Additionally, different models showed varying levels of effectiveness in Qinzhou Bay’s complex salinity environment; the IGWO–BPNN model, with its dynamic weight adjustment mechanism, demonstrated superior adaptability in areas with high salinity variability, outperforming other models. These findings suggest that the IGWO–BPNN model provides high accuracy and stability, supporting real-time, precise monitoring in Qinzhou Bay and similar coastal waters, thereby offering robust support for water quality management and marine conservation.