兼顾多要素空间非平稳性特征的卫星降水数据精度提升方法

Improvement of the Accuracy of Satellite-Derived Precipitation Data by Considering the Spatial Non-stationarity of Multifactor:A Case Study of Sichuan Province

卫星降水数据存在的较大时空误差制约了其更深入应用。该文引入多个降水的主要影响因子作为辅助信息,通过半参数地理加权回归(SGWR)+局部薄盘光滑样条插值(PTPS)构建模型,从月、年尺度上实现了两套卫星降水数据(GPM-IMERG和CMORPH-CPA)在四川地区的融合,并通过气象站点的实测数据对不同融合方法的结果进行了验证。研究结果表明:1)融合后降水数据的精度较高且比融合前有较大改善:年、月尺度融合结果的平均绝对误差率分别达8.12%和12.94%;融合结果的数据精度比单一卫星降水产品提升25%以上。2)该融合模型对降水数据精度的提升效果优于没有加入影响因子的同类模型,说明加入影响因子等辅助信息有助于进一步提高融合结果的精度。3)该模型的融合效果在时空上存在差异:干季月份的效果明显优于湿季月份;融合后90%地区的数据精度得到提升,少数区域融合效果不佳,可能是部分时段其中一种卫星降水数据精度较低所致。

As a novel data source,satellites-derived precipitation data is playing more and more important role on many environmental research applications since its own advantages of spatial continuity,high-precision measurement and timely information delivery.However,further application on this precipitation data is restricted by the problems of great calculating error in some time period and some regions.Considering multiple influencing factors including location,terrain and vegetation coverage combined with their characteristics of spatial non-stationary,a case study on the fusion of two satellites-derived precipitation produces(GPM-IMERG and CMORPH-CPA)in two different time-scale(year and month)was undertaken at Sichuan Province of Southwest China by the method combination of semiparametric geographically weighted regression(SGWR)and partial thin plate smoothing splines interpolation(PTPS).The study results demonstrated that:1)There was an obvious improvement on the data accuracy of fusion result compared to the pre-merged version.It was found that the annual and the monthly bias between the fusion results and the measured data of meteorological stations were 8.12%and 12.94%,respectively.And the data accuracy markedly raised on average by 25%compared with the single source of satellites-derived precipitation produce.2)The model built in this paper had the specialties of great openness on the selection of factors and lower resulting error,which proved a higher accuracy compared with similar models without the introduction of influencing factors.In this study,it showed the effective serviceability of auxiliary information on the improvement of data accuracy.3)There were temporal and spatial differences on fusion performance:the degree of improvement on data accuracy at the months of dry season was obviously superior to that at the months of wet season.There was a raising of data accuracy at almost 90%of the study area after the fusion of the two satellites-derived precipitation produces,however,a falling-off of data accuracy in a small amount of stations occurred simultaneously,which could be attributable to the poor accuracy on one of the satellites-derived precipitation data in these areas.