Soil Moisture Retrieval Using Stacked Generalization: An Ensemble Machine Learning Method

Ensemble learning has been used to solve classification and regression problems in many fields, but its application in remote sensing is rarely explored. In this study, an ensemble machine learning algorithm named stacked generalization (or stacking) was developed to forecast surface soil moisture over the Tibetan plateau. The algorithm combined multiple learning techniques including Random Forests (RF), Extreme Gradient Boosting (XGBoost) and linear regression (LR). Normalized difference vegetation index (NDVI), modified soil-adjusted vegetation index (MSAVI), shortwave angle normalized index (SANI) and Shortwave Infrared Transformed Reflectance (STR) derived from Landsat-8 were used as the input variables for the soil moisture retrieving model. The experimental results show that the stacking algorithm is able to achieve an acceptable accuracy, the average root mean squared error (RMSE), mean absolute error (MAE) and coefficient of correlation (r) were 0.052 cm3 cm−3, 0.039 cm3cm−3and 0.82 respectively, which is much better than its base models.