Study Of Precipitation Data Merging Using Hierarchical Bayesian Network Algorithm In The Source Region Of Yangtze River

The source region of Yangtze River is one of the most vulnerable places impacted by global climate change and is one of the most fragile eco-environment areas around the world.The variation of water resources in the region aroused widely attentions from government and academic community.Scientific assessing the spatial-temporal variation of precipitation and predicting the future trend will be of great importance in protecting the environment,in preventing disasters and reducing damages and keeping sustainable development.However,the limitation of sparse gauged data causes high uncertainty in studying the climate and water resources variation in the region.In recent years,many global/large regional scale gridded hydrometeorological datasets are developed and widely used to promote the study on the water resources evaluation and the impact of climate change on the source region.However,many studies focused on the assessment of the gridded datasets indicated that obvious errors existed in the datasets and the spatial distribution of the errors performs regional characteristics.Therefore,to solve the sparse gauged data problem that existed in hydrological researches in the source region of Yangtze River,this study takes use of the Hierarchical Bayesian Network method to merge the gauged precipitation and gridded APHRODITE precipitation data to produce the merged daily precipitation data with high spatial resolution of 0.25°Lat×0.25°Lon.And based on the merged data,this doctoral thesis investigate the spatial and temporal variation of the precipitation in the catchment controlled by Gangtuo hydrometric station in the source region of Yangtze River.The main achievements of this study are:(1)This study build a Hierarchical Bayesian Network model which contains nine parameters to merge the gauged and gridded precipitation data.And take use of Mean Absolute Error(MAE)as evaluation criteria to train the Hierarchical Bayesian Network to accomplish the study,verification and prediction processes in the model.The WinBUGS software platform is applied to merge the gauged and gridded precipitation data,and use the Markov Chain Monte Carlo sampling method to infer the values of merged precipitation and the corresponding posterior probabilities,and finally obtains 555 grids merged daily precipitation dataset from 1980 to 2007 that covers the whole study catchment with spatial resolution of 0.25°Lat×0.25°Lon.(2)The significance test results from Student t test,F test and Kolmogorov-Smirnov test show that the merged precipitation data has the same mean values,variation and distribution in compare with the gauged precipitation at most of the verification points under the 95%confidence level in the study catchment.The relative errors between the merged and gauged precipitation data range from-6.8%to 5.9%,and the coefficients of determination derived from linear regression between the merged and gauged precipitation are all above 0.95.The absolute errors computed from each grid at each day during the computation period between the merged and gauged precipitation are mostly range from-0.5 mm to 0.5 mm.Meanwhile,the number of precipitation days are generally similar between the merged and gauged precipitation at every verification point.(3)After forcing the gauged,gridded and merged precipitation into the GBHM hydrological model,the runoff simulation results indicate that the model performance improves obviously using merged precipitation than using gridded data,while the merged precipitation based model gives the similar performance as that of the gauged precipitation based model.The runoff simulation results show that the gauged and merged precipitation based models give the Nash-Sutcliffe model efficiency coefficient values are higher than 0.85 and the relative errors are smaller than 5%.Furthermore,the probability of detection and the false alarm rate computed from the runoff simulated by using gauged and merged precipitation are 85.9%and 3.1%,respectively,under the window phase of one day.At last,the modeled discharges analyzed by using Partial Duration Series indicate that the model using gauged and merged precipitation give similar cumulative frequencies in the Partial Duration Series,while the cumulative frequencies are lower than that of the observed runoff when runoff is larger than 2500m3/s.(4)Based on the merged precipitation dataset,it is found that the precipitation in the study area is concentrated in summer,archives 238mm and takes account 59%of the annual precipitation.In winter,precipitation is poor,only 12mm.The spatial distribution of precipitation is mainly concentrated in the alpine valleys zone in southeast of the catchment,and the annual precipitation achieves 550mm or higher.However,the middle and north part of the catchment is dry,the annual precipitation is only 350mm or less,meanwhile,the inhomogeneity and concentration ratios of precipitation in this area are higher than those in other part of the catchment,achieve 1 and 0.65,respectively.In the south part of the catchment,the inhomogeneity and concentration ratios of precipitation are low as climate in this region is humid.(5)The variation trend of precipitation in the study area analyzed by Mann-Kendal method during 1980-2007 shows that the annual precipitation increases significantly,especially in summer.Meanwhile,the precipitation variation trend performs obvious differences spatially.Precipitation increases in every season in the alpine valleys zone in southeast part of the catchment,especially increases significantly in spring and summer.Precipitation decreases in most part of the catchment,and only shows significant decrease trend in western part of the catchment.(6)The analysis of vertical distribution of summer precipitation indicates that summer precipitation shows the trend of decrease first and increase later with the increase of the altitude in general.Altitude at 5000m is the boundary of the variation of summer precipitation at vertical direction.When the altitude is lower than 5000m,the summer precipitation reduces with the elevation rising,and the variation rate is-8.3mm/100m;when elevation is higher than 5000m,summer precipitation increases obviously,and the variation rate is 25.2mm/100m.However,due to the comprehensive influence of many factors such as terrain,meteorological and climate,etc.,the vertical distribution of summer precipitation shows obvious personality characteristics at different areas.