Uncertainty analysis of five satellite based precipitation products and evaluation of three optimally merged multi-algorithm products over the Tibetan Plateau
Uncertainty analysis of five satellite based precipitation products and evaluation of three optimally merged multi-algorithm products over the Tibetan Plateau
复制标题
青藏高原5个星基降水产品不确定性分析及3个优化融合多算法产品评价
DOI：
10.1080/01431161.2014.960612
复制
发表时间：
2014
影响因子：
3.4
通讯作者：
Manabendra Saharia
中科院分区：
文献类型：
--
作者：
Yan Shen;Anyuan Xiong;Yang Hong;Jingjing Yu;Yang Pan;Zhuoqi Chen;Manabendra Saharia
这项研究是对2005-2007年青藏高原五种主流卫星降水产品的区域、季节、降雨率、地形和积雪的不确定性的首次综合检验。它进一步研究了三种合并方法，以便为气候和水文学研究提供可能的最佳产品。不确定性的空间分布不同，东部和南部的不确定性较高，西部和北部的不确定性相对较小。这种不确定性具有很强的季节性，在时间上是变化的，从1-4月呈下降趋势，然后保持在相对较低的水平，并在10月份之后增加，冬季峰值和夏季谷值明显。总体而言，不确定性也随着降雨率的增加而呈指数下降的趋势。当海拔超过4000米时，地形对不确定性的影响趋于迅速增大，而在低于该地形的区域，地形对不确定性的影响缓慢减小。除夏季外，海拔高度对不确定性的影响在所有季节都是显著的。进一步的交叉调查发现，不确定性趋势与基于MODIS的TP积雪覆盖率时间序列高度相关(例如相关系数≥为0.75)。最后，为了减少TP上仍然存在的相对较大和复杂的不确定性，研究了三种数据合并方法，通过对五种产品的最佳组合来提供可能的最佳卫星降水数据。三种合并方法-算术平均值、误差平方倒数和去除一个异常值的算术平均值-显示出不显著但细微的差异。三种合并方法的偏差和均方根误差随季节的不同而不同，但剔除异常值的方法更稳健，其结果在除冬季以外的所有季节都优于五种单项产品。三种合并方法的相关系数始终高于五种单个卫星估计值中的任何一种，表明了该方法的优越性。这种最佳融合多种算法的方法是一种成本效益高的方法，可以提供质量更好的卫星降水数据，而不确定度比全球降水测量任务之前的TP更少。
This study is the first comprehensive examination of uncertainty with respect to region, season, rain rate, topography, and snow cover of five mainstream satellite-based precipitation products over the Tibetan Plateau (TP) for the period 2005–2007. It further investigates three merging approaches in order to provide the best possible products for climate and hydrology research studies. Spatial distribution of uncertainty varies from higher uncertainty in the eastern and southern TP and relatively smaller uncertainty in the western and northern TP. The uncertainty is highly seasonal, temporally varying with a decreasing trend from January to April and then remaining relatively low and increasing after October, with an obvious winter peak and summer valley. Overall, the uncertainty also shows an exponentially decreasing trend with higher rainfall rates. The effect of topography on the uncertainty tends to rapidly increase when elevation exceeds 4000 m, while the impact slowly decreases in areas lower than that topography. The influence of the elevation on the uncertainty is significant for all seasons except for the summer. Further cross-investigation found that the uncertainty trend is highly correlated with the MODIS-derived snow cover fraction (SCF) time series over the TP (e.g. correlation coefficient ≥0.75). Finally, to reduce the still relatively large and complex uncertainty over the TP, three data merging methods are examined to provide the best possible satellite precipitation data by optimally combining the five products. The three merging methods – arithmetic mean, inverse-error-square weight, and one-outlier-removed arithmetic mean – show insignificant yet subtle differences. The Bias and RMSE of the three merging methods is dependent on the seasons, but the one-outlier-removed method is more robust and its result outperforms the five individual products in all the seasons except for the winter. The correlation coefficient of the three merging methods is consistently higher than any of five individual satellite estimates, indicating the superiority of the method. This optimally merging multi-algorithm method is a cost-effective way to provide satellite precipitation data of better quality with less uncertainty over the TP in the present era prior to the Global Precipitaton Measurement Mission.