拉萨河流域TRMM降水数据统计降尺度及其有效性研究

Statistical downscaling of TRMM data and its applicability in the Lhasa River Basin

利用2001—2014年TRMM 3B43、MODIS等数据,采用“统计降尺度+GDA校正”的方法对拉萨河流域TRMM 3B43降水数据进行降尺度校正,构建高分辨率TRMM降水数据。在对TRMM 3B43降水数据进行适用性检验的基础上,构建了TRMM月降水量与气温、NDVI因子间的多元线性回归关系,获得更为精细的0.05°×0.05°TRMM 3B43降水数据。结合实测数据,通过GDA法对0.05°×0.05°降水数据进行误差校正,提升TRMM降水数据精度。并与实测数据进行相关分析和误差检验,验证“统计降尺度+GDA校正”方法对于获得高分辨率TRMM降水数据的可行性。最后应用降尺度校正数据驱动VIC模型作为降水数据未来使用潜力的的初步探究。结果表明:实测数据与TRMM 3B43降水数据具有较高的相关关系,但整体上有高估现象存在;降尺度校正模型模拟的降水在提高空间分辨率的基础上具有较高的精度,与实测站点的相关系数均在0.91以上,均值达到0.95;在月尺度上,降尺度校正模型模拟的降水月相关系数较TRMM原始数据均有所提升,其中6—10月明显好于其他月份;降尺度校正后的降水数据在径流模拟中有较好的表现,NSE率定期0.75,验证期0.72,在无资料地区径流模拟中具有良好前景。

By using TRMM 3 B43 and MODIS data from 2001 to 2014, the method of ′statistical downscaling plus GDA correction′ is applied to correct the error of TRMM 3 B43 in the Lhasa River basin to obtain high resolution precipitation data. The multiple linear regression relationship between precipitation, temperature and NDVI are constructed on the basis of the applicability test of the TRMM 3 B43 preciprcation data to obtain more refined 0.05°×0.05° precipitation data over the basin. Combined with the observed data, the error correction of 0.05°×0.05° TRMM 3 B43 is carried out by the GDA method to improve data accuracy. The correlation analysis and error test are carried out in order to illustrate the feasibility of method. Finally, the downscaling corrected precipitation is utilized to drive the VIC model as a preliminary exploration for the future application. The result shows that the simulated precipitation has a high correlation with the observed data, but there is an overall overestimation phenomenon. The simulated precipitation by the downscaling correction model has high accuracy compared with the observed data, and the correlation coefficients are all above 0.91 with an average value of 0.95. At the monthly scale, the simulated data has higher correlation compared with original data. Especially, it is better from June to August than other months. The downscaling corrected precipitation also has a good performance in runoff simulation, with NSE of 0.75 in the calibration, 0.72 in the verification. It would be a good prospects of runoff simulation in ungauged basins.