Accurate and efficient prediction of soil moisture content (SMC) is vital for field water management. In this study, two types of ensemble learning models (RF and GBM) were used to compare their applicability in SMC prediction based on the automatic hourly SMC data at 10–40 cm during 2018—2021 from three self-built sites in the western Liaoning area, the prediction results were also compared and verified at annual and seasonal scales. The SHAP (Shapley Additive Explanations) method was introduced to quantitatively characterize the effects of five input variables (precipitation, sunshine hour, average relative humidity, wind speed and average temperature) on SMC prediction. Interval division rules were developed to identify the interval of maximum contribution threshold of variables. The results show that R² of GBM and RF models are 0.982 and 0.888 respectively on annual scale, temperature is the most important factor with the maximum contribution range of 21–23℃, while R² of the two models are 0.935 and 0.863 respectively on seasonal scale, sunshine hour is the most important factor with the maximum contribution range of 2–4 hours. This study innovatively applied SHAP method to analyze the contribution rates of input variables of machine learning, and verified the results of RF and GBM methods in SMC prediction, which can provide reference for related study on SMC.