Spatial Downscaling Of TRMM Precipitation Data In The Red River Basin,China

Precipitation is an important part of hydrological cycle,and accurate precipitation data is of great significance for hydrological forecasting,water resource management and drought and flood disaster assessment.Traditionally,the spatial distribution of precipitation is usually acquired based on the spatial interpolation of point precipitation measurements,which are obtained from rain gauge stations.However,due to the sparse distribution of site and limited observation range,it is difficult to accurately reflect the spatial distribution of precipitation in the whole region.Satellite-based precipitation data with high spatial resolution and wide coverage,has obvious advantages in obtaining the temporal and spatial distribution of precipitation data.But its spatial resolution is still relatively low when applied to the basin scale,thus the spatial downscaling and calibration method need to be developed to solve this problem.In this study,we firstly evaluate the applicability of TRMM(Tropical Rainfall Measuring Mission)3B43V7 satellite precipitation data with observed data from meteorological stations over the Red River Basin with monsoon climate and complex topography.Then,based on the relationships between TRMM precipitation and the digital elevation model(DEM),Normalized Difference Vegetation Index(NDVI),a regression model with a residual correction method is applied to downscale the TRMM product from coarse resolution(25 km)to fine resolution(1 km).Two methods,geographical difference analysis(GDA)and geographical ratio analysis(GRA),are used to calibrate the downscaled TRMM precipitation data.Next,monthly 1 km precipitation data are obtained by disaggregating 1 km annual downscaled and calibrated precipitation data using monthly fractions derived from original TRMM data.Finally,the downscaled annual and monthly TRMM data are used to analyze the spatial and temporal precipitation distribution regularity of the Red River Basin.Main conclusions are as follows:(1)In terms of time,TRMM precipitation data has a high consistency with the observed data on the annual,seasonal,and monthly scales(R2 > 0.64).Furthermore,there is a 7.7% overestimation by the TRMM data in the Red River Basin.Spatially,the trajectory of the gravity center of precipitation indicates that TRMM precipitation data can basically reflect the spatial distribution and evolution of precipitation.On the station scale,the correlation coefficients are all greater than 0.84,and stations with relatively large bias are mainly distributed in valleys and basins.In general,TRMM precipitation data has a good applicability in the Red River basin.(2)Based on the relationships between TRMM precipitation data and DEM,NDVI,we compare three different regression techniques(multiple linear regression model,artificial neural network model and geographically weighted regression model)of estimating precipitation at 0.25° resolution and five different interpolation methods(the inverse distance weighted interpolation,the simple spline regularized interpolation,the simple spline tension interpolation,the ordinary kriging method and the simple kriging method)of regression residual.The results indicate that the geographically weighted regression model plus the simple kriging residual correction method can obtain ideal results.Therefore,the geographically weighted regression kriging(GWRK)method is introduced to conduct the spatial downscaling of TRMM data.Compared with the original TRMM precipitation,the downscaled TRMM data of 1 km resolution describes the spatial patterns of precipitation reasonably well with more detailed information.(3)Validation results with rain gauge data show that both GDA and GRA calibration methods can provide more accurate annual precipitation distributions.For the comparison of the two methods,the GRA method provides results with larger R2,smaller RMSE,MAE,and Bias than that of the GDA method,indicating that the GRA calibration method is better than the GDA in the Red River basin.In addition,the density of rain gauge stations has an influence on the calibration of GDA and GRA.(4)Monthly 1 km precipitation data are produced by disaggregating the TRMM annual downscaled precipitation using a simple fraction disaggregation method,and then compared with the original TRMM data based on the monthly precipitation from rain gauge stations.The results show that the downscaled monthly precipitation not only has significant improvement in spatial resolution,but also agrees well with data from the validation rain gauge stations(R2 = 0.91,RMSE = 22.2 mm,MAE = 13.5 mm,and Bias = 0.048).(5)Based on the downscaled TRMM data,the spatial-temporal distribution characteristics of precipitation in the Red River Basin are as follows: on the annual scale,the precipitation generally shows a trend of decreasing from south to north and west to east.In the wet and dry season,the spatial distribution of precipitation in the wet season is similar to that on the annual scale.While the precipitation in the dry season is mainly characterized by spatial variability in the south-north direction.On the monthly scale,precipitation is mainly concentrated in the period from May to October.The spatial distribution changes of precipitation clearly reflect the influence of monsoon activities(both onset and decline)on precipitation.