基于GWR模型的青藏高原地区TRMM数据降尺度研究

Research on downscaling of TRMM data in the Tibetan Plateau based on GWR model

青藏高原地区复杂的气候环境与地形构造给水文气象数据观测造成了很大的困扰,缺少有效的高时空分辨率观测数据成为区域气象预报与预测的重要阻碍。以2001—2015年间热带降雨测量卫星(tropical rainfall measurement mission,TRMM)降水产品为基础,使用1 km空间分辨率增强型植被指数(enhanced vegetation index,EVI)空间数据,基于地理加权回归(geographically weighted regression,GWR)模型实现TRMM产品的降尺度计算,并以地面站点实测数据对降尺度结果在年、月尺度上进行结果检验与对比分析。研究结果表明:TRMM产品在降尺度处理前后的空间分布特征整体上趋于一致,但降尺度后的结果精度明显高于原始TRMM产品;2001—2015年间各年TRMM产品降尺度后的降水量与实测降水量的相关系数R 2均高于原始TRMM降水量,均方根误差和平均绝对偏差分别平均降低了21.652 mm和16.379 mm;2001—2015年间TRMM产品原始降水量准确率相对较低,在使用时需进一步进行修正,经过降尺度处理的TRMM降水量与实测降水量的拟合程度有显著性提高,除6月、8月和11月外,其他月份R 2均达到了0.65以上,表现出较好的一致性和适用性。

The complex climatic environment and topographical structure of the Tibetan Plateau region have caused great troubles for the observation of hydrometeorological data.The lack of effective high-temporal resolution observation data has become an important obstacle to regional meteorological forecasting and prediction.Based on the 2001—2015 tropical rainfall measurement mission(TRMM)precipitation product,the authors used the 1 km resolution enhanced vegetation index(EVI)spatial data to calculate the downscaling based on the geographically weighted regression(GWR)model.The downscaling results at the annual and monthly scales were tested and compared with the measured data from the ground stations.The results show that the spatial distribution characteristics of TRMM products before and after downscaling are generally consistent,but the accuracy of the results after downscaling is significantly higher than that of the original TRMM products.From 2001 to 2015,the correlation coefficient R 2 of the precipitation of TRMM products after downscaling and the actual ground precipitation was higher than that of original TRMM,and the RMSE and MAE decreased by 21.652 mm and 16.379 mm,respectively.During these years,the accuracy of the original precipitation of TRMM products was relatively low,and hence further correction is required in utilization.The degree of fitting of the TRMM precipitation with the measured precipitation was significantly improved,except for June,August and November,R 2 in other months was 0.65 or even higher,showing good consistency and applicability.