【摘要】 研究建立了一个基于集合卡尔曼滤波方法和HYDRUS-1D模型相结合的数据同化方案，利用2021年4月26日—10月5日内蒙古巴彦淖尔五原站0～80 cm的土壤水分观测数据进行模拟验证，以提高土壤水分的模拟精度。结果表明：（1）集合数大小和观测误差的选取对同化系统性能有较大影响。集合数在100以上，土壤含水量的同化精度不再有明显提高。观测误差越小土壤水分模拟精度越高，因此观测误差为0.025时的同化精度最高。（2）数据同化后，各层土壤水分模拟精度较同化前均有明显提高，各层土壤水分同化值与观测值间的相对误差、均方根误差、平均绝对误差分别减少至0.025～0.063、0.01～0.017 cm³·cm^-3、0.008～0.016 cm³·cm^-3。证明数据同化方法能够有效改善土壤水分模拟效果。（3）0～20 cm土壤水分同化效果最好，20～40 cm次之，40～80 cm土壤水分同化效果较差。模拟精度分析值优于同化预报值，同化预报值优于HYDRUS-1D预报值。更多还原

【Abstract】 In this study, a data assimilation scheme based on the combination of the ensemble Kalman filter method and HYDRUS-1D model was established. The soil moisture observation data of 0～80 cm from April 26 to October 5 in 2021 in Wuyuan Station of Bayannur were used for simulation verification, in order to improve the simulation accuracy of soil moisture. The results showed that:（1） The size of set number and the selection of observation error had great influence on the performance of assimilation system. The accuracy of soil moisture simulation on longer significantly improved when the set number was more than 100. The smaller the observation error was, the higher the simulation accuracy of soil moisture was, so the assimilation accuracy was the highest when the observation error was 0.025.（2） After data assimilation, the simulation accuracy of soil water in each layer was significantly improved compared with that before assimilation. The relative error, root mean square error and mean absolute error between the assimilation value and the observation value of soil water in each layer were reduced to 0.025～0.063,0.01～0.017 cm³·cm^-3,0.008～0.016 cm³·cm^-3, respectively. It is proved that the data assimilation can effectively improve simulation effect of soil moisture.（3） The soil water assimilation effect of 0～20cm was the best, followed by 20～40 cm, and the soil water assimilation effect of 40～80 cm was poor. In terms of simulation accuracy, the analysis value is better than the assimilation forecast value, and the assimilation forecast value was better than HYDRUS-1D forecast value.更多还原