基于TRMM 3B43数据的川西高原月降水量空间降尺度模拟

利用2001-2013年TRMM 3B43、MODIS-NDVI、DEM、气象观测等数据，在分析植被对降水响应滞后性的基础上，构建了TRMM 3B43数据中月降水量与经纬度、海拔、坡向和NDVI因子间的多元线性回归方程式，作为川西高原月降水量资料的降尺度计算模型，采用“回归方程+残差”的插值方法获取研究区2001-2013年1km空间分辨率的月降水量空间数据，并利用区内16个气象站点的观测数据与模拟结果进行了相关分析和误差检验。结果表明：（1）各气象观测站点基于TRMM 3B43资料的降尺度模拟降水量的数据均具有很高的精度，其中，精度最高的稻城站模拟结果与站点观测值的相关系数高达0.9839，精度最低的小金站相关系数亦高达0.8781；（2）在月、年尺度上，降尺度模拟降水量的数据亦具有很高的精度，其中，5-10月的精度明显高于其它月份，湿润年份精度总体高于干旱年份；（3）降尺度模拟降水量与站点实测降水量整体上相关系数为0.9499，偏差为0.0866，两者吻合度较高，但降尺度模拟降水量值略偏高；（4）降尺度在月尺度上能基本保证TRMM 3B43原始数据的精度，而在年尺度上能有效提高原始数据的精度，加之对空间分辨率的提高，可为获得更加全面、精细的降水分布数据提供有效方法。

Spatial Downscaling Simulation of Monthly Precipitation Based on TRMM 3B43 Data in the Western Sichuan Plateau

Using TRMM 3B43 data, MODIS-NDVI, DEM, meteorological data of observation stations during 2001-2013, a multiple linear regression model was built among TRMM 3B43 monthly precipitation and longitude, latitude, altitude, aspect, NDVI factor as downscaling model of monthly precipitation data in the Western Sichuan Plateau based on the analysis of the lag of vegetation response to precipitation. Then, combined with regression equation and residuals, interpolation method was adopted to obtain monthly precipitation data with 1km spatial resolution. Finally, the accuracy of simulated data obtained by downscaling model was tested by the correlation analysis and error detection between the simulated results and the observation data of 16 meteorological stations in the study area. The results showed as follows: (1) Precipitation simulated by downscaling model based on TRMM 3B43 data had a high precision in all meteorological observation stations. Daocheng site displayed the highest accuracy, the correlation coefficient between simulation results and observed values attained 0.9839, while Xiaojin site displayed the lowest accuracy, the correlation coefficient was 0.8781. (2) The precipitation simulated by downscaling model displayed high accuracy in the whole study area at both monthly and yearly time scale. The accuracy of simulated results from May to October was significantly higher than other months, as well as typical wet year (2012) was higher than dry year (2006). (3) The precipitation simulated by downscaling model displayed high accuracy to the observation data on the whole(R=0.9499, Bias=0.0866), while the value of simulated precipitation was slightly higher. (4) Compared to the original data TRMM 3B43, the simulated data by downscaling model guaranteed the accuracy and improved the spatial resolution. So, this method could provide an effective way to produce a more sophisticated precipitation data with higher spatial resolution.