基于GWR模型降尺度模拟蒙古高原地区TRMM降水数据

以多年月平均NDVI值高于0.1为阈值，对蒙古高原进行“植被区”与“非植被区”的子区域划分。在分析“植被区”植被对降水响应时滞性和“非植被区”陆地表面温度不同数据值与降水量相关性的基础上，子区域分别构建了TRMM3B43降水数据与海拔、坡度、坡向数据、归一化植被指数（NDVI）/陆地表面温度（LST）数据的地理加权回归（GWR）模型，得到区内2006-2015年每年5-10月1km空间分辨率的月降水量降尺度模拟数据，并利用区内141个气象站点数据对降尺度模拟数据进行精度验证。结果表明：（1）蒙古高原“植被区”植被对降水响应存在时滞性，约为一个月；“非植被区”多数月份白天与夜晚陆地表面温度差（LST＿D＿N）与降水量的相关性最显著。（2）降尺度模拟数据与气象站点数据具有较好的一致性，月尺度相关系数为0.83，各站点相关系数介于0.42～0.98。（3）在生长季、月平均尺度上，降尺度模拟数据具有较高精度，其中9月和10月数据精度优于TRMM 3B43数据。降尺度模拟数据整体精度较高，加之对原始数据在50°N以上未覆盖地区的填补以及空间分辨率的提高，可为区内水循环变化、农牧业生产、干旱监测等...

Downscaling Simulation of TRMM Precipitation Data in the Mongolian Plateau Based on GWR

The Mongolian plateau was divided into "vegetation area" and "non-vegetation area" using the threshold value of the multi-year monthly mean NDVI value above 0.1. Based on an examination of the time lagged response of vegetation to precipitation in the "vegetation area" and the association between various data values of land surface temperature and precipitation in the "non-vegetation area", a geographically weighted regression (GWR) model of TRMM 3B43 data with the elevation, slope, slope direction data, and normalized vegetation index (NDVI)/land surface temperature (LST) data was constructed to obtain monthly precipitation downscaling simulations at 1km spatial resolution from May to October 2006−2015 in the different areas. Utilizing information from 141 local meteorological stations, the accuracy of downscaled simulation data was validated. The results showed that, (1) there is a time lag in the response of vegetation to precipitation in the "vegetated area" of the Mongolian plateau, which is about one month. The most significant correlation between daytime and nighttime land surface temperature difference (LST_D_N) values and precipitation is found in the "non-vegetated area" in most months. (2) The descending-scale simulations agree with the meteorological station data, with a monthly-scale correlation coefficient of 0.83 and correlation coefficients ranging from 0.42 to 0.98 for each station. (3) On the growing season and monthly mean scales, the downscaled simulation data are highly accurate, with September and October data being more accurate than TRMM 3B43 data. The overall accuracy of the downscaled simulation data is higher, and together with the filling of areas not covered by the original data above 50°N and the increase in spatial resolution. It can provide essential data support for research on water cycle changes, agricultural and livestock production, and drought monitoring in the region.