同化AMSR2数据提高HYDRUS-1D模型土壤湿度模拟精度

同化遥感监测数据提高土壤剖面湿度模拟精度,对区域农业发展等实践与理论领域具有重要意义。该文结合了集合卡尔曼滤波(ensemble Kalman filter,EnKF)方法与HYDRUS-1D模型,同化降尺度后的AMSR2(advanced microwave scanningradiometer2)微波土壤湿度数据,开展榆社、荫城2个实验站点的土壤剖面湿度模拟。结果表明:在2个实验站点,与直接使用HYDRUS-1D模型相比,同化具有一定误差的AMSR2土壤湿度数据对不同深度土壤湿度的模拟精度提高都发挥了作用,尤其是对于同化前模拟方案S1(4月1日站点实测含水量)与S4(4月1日遥感含水量),由于HYDRUS-1D模拟时输入了较少数量的土壤湿度数据,数据同化效果与土壤湿度模拟精度提高更为显著;同化前后不同深度的土壤湿度精度对比结果表明,同化效果随深度增加而逐渐减弱。

Assimilation of AMSR2 soil moisture by ensemble Kalman filter and HYDRUS-1D model

Abstract: The spatiotemporal distribution of soil moisture plays an important role in many fields including hydrological processes, agricultural management and climate change. Due to the limitation in in-situ measurement of soil moisture and its dynamic process, accurately estimating it at different soil depth by assimilating remote sensing data into soil hydraulic model has received increased attention. Combining the ensemble Kalman filter (EnKF) method with the hydrological simulation model HYDRUS-1D, this paper simulated soil moisture dynamics at soil profile scale and evaluated its precision by assimilating soil moisture retrieved from spatial resolution of 1 km and soil moisture downscaled from microwave sensor of the advanced microwave scanning radiometer 2 (AMSR2). The downscaled soil moisture was calculated using a scale-independent multi-parameter linear model by combining the multi-kinds of the optical MODIS image data including land surface temperature, albedo, and the normalized difference between vegetation index products. Using field measurement data and the downscaled soil moisture at resolution of 1km as the initial condition to the HYDRUS-1D model, we designed six assimilation schemes and compared them with the associated results simulated from the HYDRUS-1D model. The soil moisture at different depth from April 1 to August 31 in 2013 at Yushe and Yincheng in Shanxi province was simulated using the six designed schemes, and the simulated results were compared with the data measured on 1, 11 and 21 of each month at soil depth of 10, 20 and 40 cm, respectively. The results indicated that the precision of the estimated soil moisture at the two location was comparable, and the assimilated downscaled AMSR2 data can effectively improve soil moisture estimation, especially in the surface soil. When there were not enough field measurement data or remotely sensed soil moisture to drive the HYDRUS-1D model as initial condition, the HYDRUS-1D simulation could give rise to significant errors and assimilation results were more precise. Compared with the simulation schemes S1, S2, the root mean square error (RMSE) of the assimilation schemes A1 and A2 was low, and the effectiveness coefficients of A1 and A2 at different soil depth are higher than 19% and 13% respectively. Compared to S3, the effectiveness coefficients of A3 are negative due to some uncertain errors associated with the assimilated AMSR2 soil moisture. For schemes S4, S5 and S6 simulated directly from the HYDRUS-1D model using the AMSR2 monitored soil moisture, their effectiveness coefficients at different depth are all positive and greater than that of schemes A1, A2 and A3. For temporal change in soil moisture, the correlation between different schemes after assimilation are higher than that simulated directly from the HYDRUS-1D model, with the correlation decreasing with soil depth because the AMSR2 only captured the soil moisture in top soil. Sensitivity analysis reveals that the precision is impacted mostly by observation frequency and its errors, and it was insensitive to the background errors and the model simulation errors.