Abstract: As a critical parameter for energy exchange in the ocean-atmosphere system, Sea Surface Temperature (SST) is vital for disaster early warning and climate change research. The Yellow Sea, as a typical marginal shelf sea in the Northwest Pacific, exhibits significant spatiotemporal heterogeneity. To address the limitations of single models in predicting SST in the Yellow Sea, such as insufficient representation of complex dynamic mechanisms and limited ability to capture spatiotemporal scale features, this study proposes a Stacking-based multi-model ensemble prediction method.The workflow involves: 1) denoising data through exponential smoothing, constructing and standardizing multidimensional features; 2) employing eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Random Forest Regression (RFR), Multilayer Perceptron (MLP), and Support Vector Regression (SVR) as base models, with RFR designated as the meta-model for Stacking ensemble training; 3) predicting SST using the trained model.Results show that the optimized model achieves a mean root mean square error (RMSE) of 0.173 5 ℃, mean absolute error (MAE) of 0.130 1 ℃, and coefficient of determination (R²) of 0.999 1, outperforming single models and traditional regression methods. Comparison with Optimum Interpolation Sea Surface Temperature (OISST) products confirms high spatiotemporal consistency between predicted and observed SST distributions. The study demonstrates that this model can effectively integrate the advantages of multiple models, significantly improve the short-term prediction accuracy of SST, and provide reliable technical support for marine environmental monitoring and climate system modeling in the Yellow Sea.