Rainfall Spatial Estimation Using Multi-source Information And Its Hydrological Application

Accurately mapping precipitation spatial distribution has been an important task inhydrometeorology. It also plays a fundamental role in hydrology analysis, naturaldisasters control, and so on. Recently, combining ground observed and satellite sensedprecipitation to obtain regional estimates has inspired a flurry of research. Thisdissertation therefore aims to estimate precipitation spatial distribution usingmulti-source information including surface rainfall observation and satellite sensing.Particularly, selecting the Gangjiang River Basin as the study area, we carried outresearch from4major aspects. First, we studied traditional precipitation interpolationmodels. Second, we evaluated the accuracy of representative satellite precipitationproducts. Third, rainfall data merging methods were proposed and evaluated. Finally,the merged data was fed to drive hydrology models to understand the merits ofprecipitation merging in improving runoff simulation accuracy.Based on the geographically weighted regression (GWR), we proposed aprecipitation spatial interpolation scheme. This scheme improves the traditionalrepresentation of precipitation locally spatial autocorrelation and its cross-correlationwith geographic factors. Using it, we explored the relationship between the accuracy ofprecipitation spatial estimation and the ground gauges density. When the ground gaugesdensity is approximately lower than1300km~2per gauge, the estimation accuracy variesdramatically with the gauges number; whereas if the density is above380km~2per gaugeapproximately, the accuracy is insensitive to the gauge numbers variation.We then comprehensively evaluated the spatial and temporal accuracy of4representative satellite precipitation products，and analyzed their seasonal andspatial variability. The selected satellite products include TRMM3B42/3B43V7,3B42RTV7, CMORPH and PERSIANN. We focused on2different temporalscales, i.e., daily and monthly scales, and3spatial scales, i.e.,0.25°×0.25°grid,sub-basin, and the whole study area scales. Analyses indicate that thequantitative error of satellite precipitation is significant. However, they coulddynamically provide useful information for rainfall spatio-temporal distributionand thus can complement relatively sparse ground observations. In order to combine the advantages of both ground observed and satelliteprecipitation, we then developed GWR-based data merging methods. With thesemethods, we conducted precipitation merging tests. We combined daily andmonthly ground observations from gauge networks of different densities, withTRMM3B43/3B42V7and CMOPRH satellite data, respectively. When thegauge network density is below2500km~2per gauge approximately, the accuracyof spatial precipitation estimation obtained by data merging is gradually greaterthan that obtained by traditional interpolation methods only using groundobservations. Pertaining to daily precipitation, when the ground gauges densityis about7500km~2per gauge, compared with the traditional spatial interpolationmodel, spatial estimation obtained by merging ground observations andCMORPH increases approximately by33%in spatial correlation coefficient anddecreases by16%in average absolute error.Through extensive hydrologic simulations in19catchments within thestudy area, we further investigated the effect of merged precipitation data onrunoff simulation. When the gauge network is relatively sparse, combinedprecipitation can significantly improve runoff simulation accuracy. Moreprecisely, when the ground gauge number throughout the study area is11, thedetermination coefficient of simulated daily runoff can be improved by morethan0.15with a maximum value over0.4; the relative error of total runoffvolume can be reduced by greater than5%with a maximum values over20%.