A Method for Sea Surface Temperature Retrieval Based on XGBoost Optimized by the Improved Sparrow Search Algorithm

Sea surface temperature (SST) serves as a critical indicator of energy exchange between the Earth’s surface and the atmosphere. Accurate retrieval of SST from satellite remote sensing data plays a pivotal role in investigating the physical and chemical properties of the ocean and monitoring sea surface thermal variations at both regional and global scales. The Chinese Atmospheric Environment Monitoring Satellite (DQ-1), launched on April 16, 2022, is equipped with a wide-swath imager (WSI), enabling daily monitoring of sea/land surface properties and atmospheric conditions at fixed time. In this study, a novel SST retrieval framework based on eXtreme Gradient Boosting (XGBoost) optimized by the Improved Sparrow Search Algorithm (ISSA) was proposed for DQ-1/WSI data. Given the decisive impact of XGBoost’s hyperparameters on predictive accuracy, ISSA is improved by integrating Fuch chaos mapping, a sine wave dynamic disturbance mechanism, and firefly search optimization strategy into the original SSA, which significantly enhances the average fitness. Leveraging the adaptive hyperparameter optimization capability of ISSA, global dynamic optimization is performed on key hyperparameters of XGBoost model. Based on the DQ-1/WSI simulated dataset, XGBoost regression model is constructed to quantify the relationship between sea surface characteristic parameters and SST. The results indicate that the XGBoost model optimized by the ISSA achieves an RMSE of 0.46 K., realizing high-accuracy SST retrieval from the DQ-1 satellite’s thermal infrared data. Validation using in-situ data from the Array for Real-Time Geostrophic Oceanography (Argo) global ocean observation network shows that the proposed model yields an RMSE of 0.51 K, outperforming the split-window algorithm (0.68 K) and the angle-corrected split-window algorithm (0.63 K). Cross-comparison with MODIS SST products reveals that the proposed retrieval algorithm reaches its performance peak in summer with an RMSE of 0.19 K, while exhibiting a slight degradation to 0.77 K in autumn, nevertheless maintaining satisfactory accuracy.