TRMM和GPM卫星降水数据在中国三大流域的降尺度对比研究

Comparative Study on Downscaling of TRMM and GPM Satellite Precipitation Data in Three Major River Basins in China

以中国三大流域为研究区域,分别针对TRMM 3B43和GPM IMERG降水数据,构建基于归一化植被指数(NDVI)和数字高程模型(DEM)数据的地理加权回归(GWR)模型,以2014年夏季(6-8月)数据为例,得到了三大流域1 km分辨率的降尺度降水数据,并进一步通过气象站点数据对降尺度结果进行验证。研究表明:(1)GPM原始数据整体精度优于TRMM数据,两种数据在黄河流域的精度评价指标优于长江流域与珠江流域;(2)经降尺度计算,两种数据空间分辨率得到了显著提高,且表现出更多降水变化的细节与趋势,但精度指标对原始数据依赖性大,导致提升并不显著;(3)当联合使用NDVI及DEM作为GWR降尺度模型的辅助变量时,两种降尺度结果的差异值在长江流域与珠江流域的高海拔地区中更为显著;(4)黄河流域中,使用不同辅助变量下的降尺度结果在空间分布上总体相似。TRMM数据使用地形作为单独辅助变量的降尺度结果最佳,而GPM数据不同辅助变量下降尺度结果差异并不明显。

Taking three major river basins in China as the study areas, this study tries to reveal the difference between downscaled results of TRMM 3 B43 and GPM IMERG precipitation based on different Geographically Weighted Regression(GWR) models. Normalized Difference Vegetation Index(NDVI) and Digital Elevation Model(DEM) were selected as the basic ancillary variables of GWR models. Different combinations of NDVI and DEM were employed to downscale TRMM and GPM precipitation of 2014 summer, which were then evaluated based on rain gauge observations. The results demonstrate that,(1) Overall accuracy of the original GPM precipitation data is better than TRMM. Accuracy indices imply that both data have higher accuracy in the Yellow River Basin than in the Yangtze River Basin and the Pearl River Basin.(2) After downscaling, the spatial resolutions of both precipitation data have been significantly improved. The downscaled results demonstrate more details and trends of precipitation variation, but their accuracy is more dependent on the original data, which makes the accuracy improvement not significant.(3) Using GWR model with both NDVI and DEM as ancillary variables, difference between downscaled results of TRMM and GPM is significant in high altitude areas of the Yangtze River Basin and Pearl River Basin.(4) The downscaled results with different auxiliary variables in the Yellow River Basin are generally similar in spatial distribution. Downscaling TRMM using terrain as the only auxiliary variable generates the best results, while for GPM, GWR models with different auxiliary variables show no obvious difference.