TRMM偏最小二乘降尺度降水模型在新疆不同地貌的适应性

Adaptability of precipitation estimation method based on TRMM data combined with partial least squares downscaling in different landforms of Xinjiang,China

高分辨率降水数据有利于客观描述区域降水时空分异特征,也可为区域生态和水文等过程高精度模拟提供重要参数。该研究旨在研究采用偏最小二乘法降尺度方法提高热带降雨测量卫星(Tropical Rainfall Measuring Mission,TRMM)数据空间分辨率用于新疆不同地貌区降水量估算的可行性。基于多源遥感数据和气象站点数据,引入相对湿度、NDVI、地形和经纬度等地理环境因子,构建偏最小二乘法降尺度模型,使遥感数据的空间分辨率由0.25°×0.25°(27.5 km×27.5 km)提高至250 m×250 m,模拟对比降尺度前后降水量估算精度差异,并探讨新疆降水时空分布特征及其对地形地貌的响应规律。结果表明:1)相比TRMM原始数值,由偏最小二乘法降尺度模型修正后的降水量与气象站点数值的拟合度显著提高,其决定系数(R^2)由0.74提高到0.85,均方根误差降低了0.26 mm,较好解决了原始数据低值高估和高值低估等问题;用乌鲁木齐河流域不同海拔高度和地形地貌区站点实测值验证该模型,发现修正后的所有站点其精度均有提高(R^2从0.06~0.91提高到0.39~0.95,均方根误差从0.20~0.44 mm降低到0.18~0.40 mm),表明了该模型在不同海拔和地形地貌区降水分异刻画的可靠性及大尺度推广的可行性。2)研究期内各地形地貌区的年降水量排序为中高山、极高山、高山、低山、中山、平原、盆地;中高山及以上山区降水集中发生在6-9月,低海拔山区及平原区的降水较多发生在5-8月,但年内降水量分配较均匀。新疆降水呈“北多南少”的空间分布格局特征,各山体(山群)均存在2个较明显的降水峰值区,但各山体绝对高度和规模不同,以致峰值区的海拔也不尽相同。研究结果可为气象站点稀缺区域的降水数据提供有益补充。

High-resolution precipitation data is beneficial to describe the spatial-temporal differentiation characteristics of regional precipitation.The purpose of this study was to explore the feasibility of using partial least squares downscaling method to improve the spatial resolution of Tropical Rainfall Measuring Mission’s(TRMM)data for estimating precipitation in different landform regions of Xinjiang,China(73°40′-96°23′E,34°22′-49°10′N).The spatial resolution of remote sensing data was increased from 0.25°×0.25°to 250 m×250 m based on multi-source remote sensing data and meteorological station data,by introducing environmental factors such as relative humidity,topography,and latitude and longitude,and constructing a partial least squares downscaling model.The accuracy of precipitation estimation before and after downscaling was compared,and the temporal and spatial distribution characteristics of precipitation in Xinjiang and its responses to topography and geomorphology were discussed.The results showed that the original TRMM could estimate precipitation well for the whole Xinjiang and mountain and plain areas of Xinjiang.However,the estimation accuracy in basin was low.Compared with that of the original TRMM value,the accuracy of precipitation estimation was improved by the partial least squares downscaling model with the coefficient of determination(R^2)increased from 0.74 to 0.85 and the root mean square error(RMSE)decreased by 0.26 mm.Moreover,the estimation after downscaling solved the problems of overestimation in the areas with low precipitation values and underestimation in the areas with high precipitation values.By downscaling method,the accuracy of precipitation estimation in stations at different altitudes and topography in the Urumqi River Basin was improved greatly.Compared to that before downscaling,the R^2 increased from 0.06-0.91 to 0.39-0.95 and the RMSE decreased from 0.20-0.44 mm to 0.18-0.40 mm.It indicated that the downscaling method used for TRMM data was reliable in estimating precipitation in different altitudes and topography.The annual precipitation in different landform areas of Xinjiang was the highest in the areas with medium high mountain,followed by those with extremely high mountain,high mountain,low mountain,medium mountain,plain and basin.The ratio of area of basin,plain and mountainous was about 1:1.6:2.5.The precipitation in the areas with medium high mountain mainly occurred from June to September.The rich precipitation in low-altitude mountain areas and plain areas was from May to August and during this period the precipitation distribution was relatively uniform.For example,the months with the least precipitation in low mountain areas were only 10%less than the months with the most precipitation.The spatial distribution pattern of precipitation in Xinjiang was characterized by more in the north and less in the South.The multi-year average monthly precipitation in the northern of the Altai Mountains was higher than 20 mm,while the precipitation in the southeast of Tarim Basin and Tu-Ha Basin was little.In the study area,two obvious precipitation peak areas in each mountain(mountain group)were observed,however the height and scale of each mountain were different.Thus,the altitude of the peak area was different.The results can provide a valuable method to estimate regional precipitation for areas with scarce meteorological stations.