天山山区TRMM降水数据的空间降尺度研究

山区降水是干旱区水资源的重要补给源,但由于山区地形复杂、监测困难造成资料缺乏,水文预报的误差较大。近年来,TRMM3B43降水数据得到了大量应用,但受其较低空间分辨率的影响,使得应用精度受到限制。论文以2001—2010年TRMM3B43数据为基础,结合提取的7个数字地形因子（经度、纬度、坡度、坡向、海拔、地形开阔度、地形起伏度）,构建了天山山区年、季的降水主成分-逐步回归降尺度模型。分析结果表明：主成分-逐步回归降尺度模型有效地将TRMM3B43数据的空间分辨率由0.25°×0.25°提高到1 km×1 km。通过站点实测降水数据对比验证,决定系数均在0.85以上,降尺度后的数据精度显著优于原始TRMM3B43数据。该方法对研究干旱区空间降水精细化具有一定参考意义。

Downscaling Method of TRMM Satellite Precipitation Data over the Tianshan Mountains

The precipitation in mountains is an important supply of water resources in arid areas. The error of hydrology forecast is still high in mountain areas because of complex terrain and lack of data. In recent years, TRMM3B43 data have been widely used in precipitation estimation. However, there is limit of TRMM data in resolution (0.25°×0.25°). Based on the TRMM3B43 data during 2001-2010, a principal component-stepwise regression model for downscaling TRMM3B43 precipitation was built with seven topographic factors (longitude, latitude, altitude, slope, aspect, terrain unobstructed factor, relief degree of land surface) in this paper. The downscaling model can transform the pixel resolution from 0.25°×0.25° to 1 km×1 km effectively. The downscaled precipitation estimations were subsequently validated by observed data at 21 raingauge stations in Tianshan Mountains during 10 years. The results showed that: 1) The downscaled precipitation results have smaller errors than the original TRMM precipitation data, and the maximal improved value of annual average precipitation is 63.16 mm, suggesting the good performance of principal component-stepwise regression downscaling model. 2) The spatial distribution characteristics of precipitation in the Tianshan Mountain can be depicted by downscaled results in detail that the high precipitation values mainly appear in the Yili Valley of the western Tianshan Mountains and the low values appear in Turpan and Hami of the eastern Tianshan Mountains. 3) The downscaled results conform well with the observed data in each elevation zone in the Tianshan Mountains, but the downscaled results overestimated the actual precipitation to some extent. So the downscaling method in this article is feasible, which could provide spatial distribution of precipitation in fine resolution in arid region.