卫星降雨数据在高山峡谷地区的代表性与可靠性

Representativeness and reliability of satellite rainfall dataset in alpine and gorge region

以长江上游金沙江流域典型高山峡谷地区为研究对象,用该区域地面观测降雨量数据对TRMM PR 3B42 V6产品进行了3 h、日、月3个时间尺度的有效性评估,旨在为开展区域卫星与地面降水数据融合的流域水文模拟及预报奠定数据基础。分别采用了线性回归方法分析降雨量相关性、经验正交函数-奇异值分解方法(EOF-SVD)分析降雨量主要模态空间分布特征、相对偏差Bias、错报率RFA和探测率PD指标对该卫星产品进行了精度评定。研究结果表明:研究区该卫星产品与地面观测数据在3个时间尺度存在显著的线性时间和空间相关性,但相关程度随时间尺度的减小而减弱;二者在空间分布上总体具有一致性特征,但在高海拔、大坡度区域表现出较为显著的差异;相对偏差指标显示2008—2010年降雨量均值相对偏差在±10%的概率密度百分数为36.08%;随高程的增加,卫星数据RFA呈逐渐增加趋势变化,PD呈逐渐减小趋势变化;总体上小雨对误差的贡献最大,大雨峰值误差贡献次之,时段降雨量偏差随时间尺度的增加逐渐减小,而随高程的增加卫星数据的探测精度下降。因此,对于类似的高山峡谷流域,要应用该卫星产品进行日、3 h尺度水文模拟及预报,有必要对流域卫星数据和地面观测数据进行融合,充分发挥两种数据的优势。

In order to apply the satellite and gauges rainfall datasets in hydrological simulation and forecasting, the TRMM PR 3B42 V6 product is evaluated by ground-based observation rainfall in alpine and gorge region of Jin-sha River Basin, which involves 3-hourly, daily and monthly temporal scales. Concretely, the linear regression analysis is used to analyze correlation of the two datasets. The empirical orthogonal function and singular value decomposition methods （EOF-SVD） are adapted to examining the spatial patterns of precipitation. The indices of relative deviation （B^as） , false alarm ratio （RFA） and the probability of detection （PD） are applied to evaluating the accuracy of the satellite dataset. The results indicate that a significant linear correlation is derived between the two datasets at different spatiotemporal scales, but the correlation coefficient will be decreased with the reduction on temporal scales. Good consistency is achieved in the overall spatial patterns of precipitation, even though local discrepancies are found in high altitude and high- gradient areas. The TRMM 3B42 data are accurate about 36.08% within 10% of the station locations. The altitude shows certain influence on the indices of RFA and PD, which demonstrate increasing on R^A but decreasing on PD with elevation rising. Overall, the accuracy of TRMM 3B42 data declined with elevation rising; the periods rainfall error decreased with temporal scale increasing and the light rain have the greatest contribution to satel- lite deviation but the heavy rainfall is the second. These results suggest that merge and correction would be necessary when the satellite data are used for hydrological simulation and forecasting on finer temporal scales in similar investigation areas.