TRMM卫星降水产品降尺度及其在湘江流域水文模拟中的应用

高时空分辨率降水数据对准确刻画区域降水时空变化特征、精准模拟区域生态和水文过程具有重要的现实意义。以湘江流域为例，在考虑地理、地形和植被等多重要素的基础上，建立了基于地理加权回归法（Geographic Weighted Regression，GWR）的热带降雨测量卫星（Tropical Rainfall Measuring Mission，TRMM）降水降尺度模型，并采用比例指数法反演得到星地融合日降水Ⅰ、Ⅱ、Ⅲ三种产品，用来驱动土壤和水评估模型（Soil and Water Assessment Tool，SWAT），分析评估其在水文模拟中的应用潜力。结果表明：1）GWR降尺度后，在TRMM降水空间分辨率由0.25°提升至0.05°的同时，同气象站点观测月降水之间的决定系数（R2）平均提升了0.33，均方根误差（RMSE）平均降低了43.30 mm，平均相对偏差（Average Relative Error，ARE）平均降低了38.71个百分点，表明该降尺度模型在湘江流域TRMM月降水降尺度研究中具有良好的适用性；2）与TRMM日降水量相比，星地融合日降水Ⅲ产品同气象站点观测日降水量的R2提高了0.81，RMSE降低了10.27 mm，ARE降低了0.11个百分点，表明以气象站点观测日降水量作比例指数展布星地融合月降水是可行有效的；3）星地融合日降水Ⅲ产品在SWAT模型日、月径流模拟中的纳什效率系数最大，分别为0.79、0.93，相对误差最小，分别为0.12%、1.10%，水文模拟效果最优，可替代气象站点和TRMM卫星降水进行水文模拟。研究结果可为气象站点稀缺区域的高精度降水资料获取和高效水文模拟提供数据支撑和方法借鉴。

Downscaling of TRMM satellite precipitation products and its application in hydrological simulation of Xiangjiang River Basin

Precipitation data with a high temporal and spatial resolution is of great practical significance to accurately characterize the spatiotemporal changes of regional precipitation in ecological and hydrological processes. However, traditional meteorological station observations cannot meet the high requirements of data acquisition. The purpose of this research was to deal with the time-scale extension in Tropical Rainfall Measuring Mission (TRMM) downscaling, particularly for a higher spatial resolution of TRMM satellite precipitation products under continuous observation and wide coverage. A variable parameter spatial regression model, Geographic Weighted Regression (GWR), was selected for the spatial downscaling of annual and monthly TRMM. Specifically, the parameters were estimated between the dependent and the independent variables at each location via the local weighted least squares method. The study area was taken as the Xiangjiang River in the Dongting Lake water system of the Yangtze River Basin in Hunan Province of China. The specific procedure was as follows. The precipitation data of meteorological stations was first embedded into the TRMM satellite precipitation grid. Then the longitude, latitude, digital elevation models were selected, with the slope, aspect, and normalized difference vegetation index as auxiliary variables. Finally, a TRMM satellite precipitation downscaling model was established using GWR and multiple factors, such as geography, topography, and vegetation. In addition, a variety of scale indexes were used to invert for three products of satellite-ground fusion daily precipitation I, II, and III. The precipitation input data were selected to drive the SWAT distributed hydrological model, further to evaluate the application potential in hydrological simulation. The coefficient of determination, root mean squared error, and average relative error were used to quantitatively evaluate the accuracy of TRMM data before and after downscaling. Moreover, the relative error and Nash-Sutcliffe coefficient of efficiency were also used to quantitatively evaluate SWAT simulation. The results showed that the spatial resolution of TRMM precipitation increased from 0.25° to 0.05°, while the coefficient of determination between the monthly precipitation observed by the meteorological station increased by 0.33 on average, and the root mean square error decreased by 43.30 mm on average, and the average relative deviation decreased by 38.71 percentage points on average after the GWR downscaling, indicating excellent applicability in the TRMM downscaling. Compared with the TRMM daily precipitation, the coefficient of determination between the satellite-ground fusion daily precipitation III product and the meteorological station observation daily precipitation increased by 0.81, the root mean square error decreased by 10.27 mm, and the average relative deviation decreased by 0.11 percentage points, indicating that it was feasible and effective for the meteorological station observation daily precipitation as a proportional index to spread the satellite-ground fusion monthly precipitation. The satellite-ground fusion daily precipitation III product presented the largest Nash efficiency coefficient, the smallest relative error, and the best hydrological simulation effect in the soil and water assessment tool’s daily and monthly runoff. It infers to replace meteorological stations with the TRMM satellite precipitation for hydrological simulation. The finding can provide potential support to high-precision precipitation data acquisition and efficient hydrological simulation in scarce areas of meteorological stations.