Stacking Gaussian Process Regression for Soil Moisture Estimation Over the Continental U.S.

To overcome the limitations posed by sparse and insufficient sample sizes in soil moisture estimation, ensemble learning—specifically the integration of multiple individual estimation models—has emerged as a promising approach. In this paper, we introduce a novel soil moisture estimation method based on ensemble learning, termed Stacking Gaussian Process Regression (SGPR). This method incorporates Gaussian Process Regression (GPR) within a stacking strategy that utilizes gradient boosting and employs K-fold cross-validation, leveraging 11 multi-source remote sensing features. Experiments conducted across the Continental U.S. from April 2015 to March 2016 demonstrate that the proposed SGPR method significantly outperforms existing state-of-the-art models, achieving a correlation coefficient $ R = 0.9097 $ and a root mean square error $ RMSE = 0.0474 \, cm^{3}/cm^{3} $ . By harnessing the strengths of various regression models and fully utilizing the prior information embedded within the sample data, the SGPR model effectively enhances both the accuracy and stability of soil moisture estimation.