高分辨率青藏高原历史月降水数据重建  

Reconstruction of high resolution monthly precipitation data of the Tibetan Plateau

作　　者：徐明[1] 石玉立[1] 王彬 XU Ming;SHI Yuli;WANG Bin(School of Geography and Remote Sensing,Nanjing University of Information Science and Technology,Nanjing 210044,China)

机构地区：[1]南京信息工程大学地理与遥感学院,南京210044

出　　处：《地理科学进展》2018年第7期923-932,共10页Progress in Geography

基　　金：国家自然科学基金项目(41471312)~~

摘　　要：青藏高原对全球气候研究具有重要意义,而降水数据对水文、气象和生态等领域的研究也至关重要,且随着研究内容和尺度的变化,对高时空分辨率的历史降水数据的需求越发迫切。本文基于TRMM 3B43降水数据,采用随机森林算法,引入归一化植被指数(AVHRR NDVI)、高程(SRTM DEM)、坡度、坡向、经度、纬度6个地理因子,建立历史降水重建模型,获得1982-1997年分辨率为0.0833°的青藏高原年降水数据,然后根据比例系数法计算出月降水数据。为提高精度,利用站点数据对月降水数据进行校正。结果表明,该方法能简单有效地获得高时空分辨率的历史降水数据,决定系数R2大部分在0.4~0.9之间,平均值为0.6767,其中夏季效果最好,冬季效果最差;均方根误差RMSE和平均绝对误差MAE均在50 mm以下,RMSE均值为22.66 mm,MAE均值为15.97 mm;偏差Bias较小,基本在0.0~0.1之间。The Tibetan Plateau is of great significance to the study of global climate. Precipitation data are also important to hydrological, meteorological, and ecological research. With the change of research content and scale, the demand for historical precipitation data with high spatial and temporal resolutions is increasingly more urgent. This study selected six factors including the normalized vegetation index（NDVI）, elevation, slope,longitude, latitude, and precipitation to obtain the historical precipitation data with high spatial and temporal resolutions of the Tibetan Plateau from 1982 to 1997. This research is based on the precipitation data of TRMM3 B43, AVHRR NDVI, and SRTM DEM. The factors were entered into the random forest model. A historical precipitation model was constructed to obtain the spatial resolution of 0.0833° for annual precipitation. Monthly precipitation was estimated according to the proportional coefficient. In order to improve estimation accuracy,the simulated monthly precipitation was corrected using station data. The results show that the method can simulate historical precipitation with high spatial and temporal resolutions. The coefficient of determination R2 is between 0.4 and 0.9, with an average value of 0.6767. The model performed better in summer and worse in winter. The root mean square error and average absolute error is below 50 mm. RMSE mean is 22.66 mm. MAE mean is 15.97 mm. Deviation bias is between 0.0 and 0.1.

关 键 词：随机森林 历史降水 重建 TRMM 3B43 青藏高原 

分 类 号：P426.6[天文地球—大气科学及气象学]