Abstract: Accurate measurement of soil moisture is critical for hydrology, agriculture, and environmental science. Among various indirect measurement techniques, frequency-domain reflectometry (FDR) soil moisture sensors are low-cost and widely used. To address the problems of difficult field calibration and susceptibility to environmental interference that lead to systematic biases in FDR soil moisture sensors, this study proposes an automatic calibration method based on the ensemble Kalman filter and meteorological data. This method couples the ensemble Kalman filter algorithm with a bias calibration model, uses meteorological data to drive a soil hydrodynamic model, and inverts sensor biases and soil hydraulic parameters as joint variables to be estimated. Outdoor experiments were conducted to systematically evaluate the effects of different bias models and assimilation settings on calibration performance. Uncalibrated sensors exhibited significant systematic biases in soils of different textures. The average statistical results for soil at depths of 10, 20, and 50 cm showed that the average relative error (RE), root mean square error (RMSE), and mean absolute error (MAE) of uncalibrated sensors were 32.7%, 0.045 cm³/cm³, and 0.037 cm³/cm³, respectively. After calibration with the linear model, the average RE, RMSE, and MAE were 18.6%, 0.027 cm³/cm³, and 0.023 cm³/cm³, respectively, while those calibrated with the power function model were 19.2%, 0.029 cm³/cm³, and 0.024 cm³/cm³, respectively. This indicates that both models can improve sensor accuracy, with the linear model achieving superior performance. The optimal assimilation duration was found to be 20–30 days, and the suitable ensemble size ranged from 100 to 150. The proposed method can effectively correct systematic biases without destructive sampling, providing an efficient and low-cost technical solution for soil moisture monitoring.