Abstract:

In this paper, based on the non-stationary relationship between precipitation and terrain, and combined with the limited observed precipitation data, the Geographically Weighted Regression (GWR) model was used for downscaling and calibrating the TRMM 3B43 precipitation data in the karst mountainous area of Guizhou Province. Finally, the precipitation data with a spatial resolution of 1 km×1 km was produced and verified. The results were shown as follows: (1) The spatial resolution and accuracy of TRMM 3B43 precipitation data in Guizhou karst mountainous area could be improved by the GWR model. (2) The validation results for different time scales showed that the error of downscaling and calibrating TRMM precipitation data was much smaller than TRMM 3B43 precipitation data, which was closer to ground observation precipitation. The downscaling algorithm was closer to the real precipitation on the time scale with less precipitation in Guizhou. (3) If the TRMM 3B43 could be accurately predicted by the resampled relief degree of land surface, the deviation of the TRMM 3B43 data was the main source of error in the GWR downscaling algorithm. If precipitation was weakly related or irrelevant to relief degree of land surface, other spatial variables should be considered to modify this spatial non-stationarity relationship.

Key words: GWR, Guizhou, karst mountainous area, TRMM, downscaling, spatial non-stationarity

摘要：

基于降水与地形起伏之间的非平稳关系，结合有限的观测降水数据，利用GWR回归模型，对贵州喀斯特山区的TRMM 3B43降水资料进行降尺度和校准，最终得到空间分辨率为1 km×1 km的降水量分布数据并进行了验证。结果显示：（1）考虑地形起伏和降水空间非平稳性的GWR模型，提高了贵州喀斯特山区TRMM 3B43遥感降水资料的空间分辨率和准确度。（2）不同时间尺度的验证结果表明，在与观测降水的相关统计中，TRMM降尺度降水具有较TRMM 3B43降水更高的统计精度和更小的误差，更接近于地面观测降水；该降尺度算法在贵州降水较少的时间尺度更加接近真实值。（3）当TRMM 3B43可以被重采样的地形起伏度（RDLS）进行准确预测时，TRMM 3B43的精度是GWR降尺度算法中的主要误差源；当区域的降水与地形起伏弱相关或无关时，应考虑引入其他影响降水的空间变量来修正这一空间非平稳性关系。

关键词: GWR, 贵州, 喀斯特山区, TRMM, 降尺度, 空间非平稳性