Research on Provincial-Level Soil Moisture Prediction Based on Extreme Gradient Boosting Model

As one of the physical quantities concerned in agricultural production, soil moisture can effectively guide field irrigation and evaluate distribution water resources for crop growth various regions. However, spatial variability is dramatic, its time series data are highly noisy, nonlinear, nonstationary, thus hard to predict accurately. In this study, taking Jiangsu Province China as an example, 70 meteorological automatic observation stations from 2014 2022 were used establish prediction models 0–10 cm relative humidity (RHs10cm) via extreme gradient boosting (XGBoost) algorithm. Before constructing model, according measured characteristics, divided into three categories: sandy soil, loam clay soil. Based on impacts factors budget balance, 14 predictors chosen among which atmospheric accounted 10 4, respectively. Considering differences characteristics lagged effects environmental impacts, best influence times different types determined through correlation analysis improve rationality model construction. To better importance factors, two sets (Model_soil&atmo Model_atmo) designed by optional put XGBoost model. Meanwhile, contributions results analyzed with Shapley additive explanation (SHAP). Six effect indicators, well a typical drought process that happened 2022, accuracy. The show highest correlations between RHs10cm varied but was similar types. Among these predictors, contribution rates maximum air temperature (Tamax), cumulative precipitation (Psum), (RHa) functioned critical factor affecting variation moisture, relatively high both models. addition, adding could accuracy prediction. certain extent, performed when compared artificial neural networks (ANNs), random forests (RFs), support vector machines (SVMs). values coefficient (R), root mean square error (RMSE), absolute (MAE), (MARE), Nash–Sutcliffe efficiency (NSE), (ACC) Model_soil&atmo 0.69, 11.11, 4.87, 0.12, 0.50, 88%, This study verified applicable at provincial level, it reasonably development processes event.