【摘要】 为监测青藏高原区域大范围、长时序土壤水分变化,基于极轨卫星被动微波传感器AMSR2(Advanced Microwave Scanning Radiometer 2)和倾斜轨道卫星被动微波传感器GMI(Global Precipitation Measurement Microwave Imager)的微波亮温,以多通道协同反演算法(Multi-channel Collaborative Algorithm,MCCA)获取的极轨卫星SMAP(Soil Moisture Active Passive)高精度土壤水分为目标,建立精细化的逐网格机器学习模型。将MCCA SMAP数据优势由AMSR2转移至GMI,以反映青藏高原日内土壤水分变化。训练期平均皮尔逊相关系数R达到了0.82,均方根误差RMSE为0.050 m~3/m~3。测试期的平均R为0.81,RMSE为0.055 m~3/m~3。重塑的GMI土壤水分显著增加了有效反演个数,与地面观测数据高度一致,平均R为0.81,无偏均方根误差ubRMSE为0.039 m~3/m~3。同时,GMI与SMAP的联合可为青藏高原短期极端气候变化及长期变化趋势的监测提供了一种选择。更多还原

【Abstract】 In order to monitor the large-scale and long time-series soil moisture changes in the Tibetan Plateau region, a refined grid-by-grid machine learning model is established based on the microwave bright temperatures of the polar-orbiting satellite passive microwave sensor AMSR2(Advanced Microwave Scanning Radiometer 2) and the inclined-orbiting satellite passive microwave sensor GMI(Global Precipitation Measurement Microwave Imager), aiming at the polar-orbiting satellite SMAP(Soil Moisture Active Passive) high-precision soil moisture obtained by the Multi-Channel Collaborative inversion Algorithm(MCCA). The advantage of MCCA SMAP data was transferred from AMSR2 to GMI to reflect the intra-day soil moisture changes on the Tibetan Plateau. The average Pearson correlation coefficient R for the training period reached 0.82, and the root mean square error RMSE was 0.050 m~3/m~3. The average R for the test period was 0.81 and the RMSE was 0.055 m~3/m~3. The remodeled GMI soil moisture significantly increased the number of valid inversions and was highly consistent with the ground observations, with an average R of 0.81 and an unbiased root mean square error ubRMSE of 0.039 m~3/m~3. Meanwhile, the combination of GMI and SMAP may provide an option for monitoring short-term extreme climate change and long-term trends on the Tibetan Plateau.更多还原