喀斯特山区TRMM降水数据降尺度研究

Spatial Downscaling of TRMM Precipitation Data in Karst Mountainous Area

获取高精度的卫星降水数据,为喀斯特区域旱涝灾害评估、水文预报等各研究领域提供数据基础。以热带降雨卫星(tropical rainfall measuring mission, TRMM)为数据源,采用多元线性回归法(ordinary least square, OLS)和地理加权回归法(geographically weighted regression, GWR),综合考虑高程、坡度、坡向、经纬度和归一化植被指数(normalized difference vegetation index, NDVI)等6个因子构建OLS和GWR降尺度模型进行年降尺度研究,并比较OLS和GWR两种降尺度模型在喀斯特山区的适用性。结果表明:1)TRMM数据与站点观测数据之间精度较好;2)降尺度后数据空间分辨率提升到1 km, GWR降尺度年降水量在多数年份比原始TRMM数据更接近实测值,高估现象得到改善;与OLS降尺度数据相比,其三项指标表现更好;3)单站点中,OLS降尺度数据在高程和NDVI突变区域易出现假性更优相关性。综合多指标评价,GWR降尺度数据在喀斯特山区总体精度更高。后续可通过划分植被区、岩溶区、增加环境因子、校正等使降水更贴合实测值。

This study obtains high-resolution satellite precipitation data to provide a data base for various research fields in karst regions such as drought and flood disaster assessment and hydrological forecasting.Taking tropical rainfall measuring mission as the data source,using the ordinary least square and geographic weighted regression,this study constructed downscaling model of precipitation with elevation,slope,aspect,longitude,latitude and normalized difference vegetation index for annual downscaling research.And the applicability of OLS and GWR downscaling methods in the karstic mountainous areas of Guizhou province was compared.The results were as follows.1)The accuracy between TRMM data and station observations was good.2)The spatial resolution of the downscaling data was greatly improved to 1 km.The GWR model downscaling annual precipitation was closer to the measured value of meteorological stations than the original TRMM data in most years,and the overestimation phenomenon was ameliorated.Compared with the OLS downscaling annual data,the three indicators of GWR downscaling data performed better.3)At single station,OLS downscaling data tended to have spurious better correlations in cases where elevation and NDVI value changed suddenly.Based on comprehensive evaluation of multiple indicators,the accuracy of GWR downscaling data was generally higher in karst mountainous areas.The subsequent research can get the precipitation data closer with the measured data by dividing vegetation areas and karst areas,adding environmental factor,calibrating downscaling precipitation,and so on.