祁连山TRMM降水数据降尺度不同方法比较研究

Comparative Study on Different Downscaling Methods of TRMM Satellite Precipitation Data over the Qilian Mountains,China

降水数据对研究山区陆面过程、水文模型、生态模型至关重要,而山区地形复杂降水数据获取困难,且TRMM(Tropical Rainfall Measuring Mission)降水数据空间分辨率难以进行小尺度区域降水研究。本文以祁连山为研究区,应用2005-2016年的TRMM降水数据,从不同时间尺度对TRMM降水数据在祁连山的适用性进行分析,利用多元逐步回归模型、地理加权回归模型、随机森林模型三种降尺度方法,分别建立了降水与高程、植被指数、纬度、经度、坡度、坡向的关系,获得1 km高分辨率的TRMM降尺度数据,并通过对比三种模型的降尺度结果来选取适用于祁连山的降尺度方法。结果表明:(1)TRMM降水数据适用于祁连山。TRMM降水数据和实测数据在不同时间尺度上具有显著的相关性,年、季、月相关系数分别为0.88、0.92、0.88。(2)三种降尺度模型都能有效地获得1 km高分辨率的TRMM降尺度数据,其中随机森林模型更适用于获取祁连山TRMM降尺度降水数据。(3)相比于原始TRMM降水数据,随机森林模型降尺度结果整体偏小,同时具有更高的空间分辨率和更小的偏差。本研究可以为进一步获取西部山地高空间分辨降水数据和开展水文研究提供参考与借鉴。

The high-spatial-resolution precipitation data plays a crucial role in the studies of hydrological models,ecological models and land surface processes.However,it is difficult to obtain high-spatial-resolution precipitation data in mountainous areas due to some factors,such as varied topography,sparse rainfall gauges and complicated precipitation environment.Meanwhile,the coarse spatial resolution of the Tropical Rainfall Measuring Mission(TRMM)precipitation data is difficult to apply in small-scale regions.In this paper,TRMM precipitation data at different time scales were used to analyze its applicability in the Qilian Mountains.Based on the TRMM precipitation data from 2005 to 2016,Multiple Stepwise Regression Model,Geographical Weighted Regression Model and Random Forest Model were used to obtain downscaled precipitation data precipitation data by establishing relationships between precipitation and environmental variables(elevation,vegetation index,latitude,longitude,slope and aspect).TRMM precipitation data with 1 km high spatial resolution in the Qilian Mountains was obtained by comparing the three downscaling models and further to select the most suitable one.Results showed as follows:(1)TRMM precipitation data was well matched with rain gauge data in the Qilian Mountains.TRMM precipitation data and measured data had significant correlation on different time scales.The annual,seasonal and monthly correlation coefficients were 0.88,0.92 and 0.88,respectively.(2)Three downscaling models can effectively obtain downscaled precipitation data with 1 km high spatial resolution in yearly time scale,the Random Forest Model was the most suitable model for downscaling of TRMM precipitation data in Qilian Mountains.(3)Compared with the original TRMM precipitation data,the downscaled results of the Random Forest Model were relatively small,with higher spatial resolution and smaller deviation.Thus,this study has certain reference significance for obtaining high spatial resolution precipitation data and cucting hydrological research in the arid mountainous areas of Northwest China.