Runoff Confluence Simulation Supported By GPU

The frequency of sudden heavy rainfall is increasing.Rainstorm and flood disasters and landslide and debris flow disasters caused by Rainstorm and flood disasters have become the principal natural disaster in China.Rainstorm and flood disasters and landslide/ debris flow disasters will cause serious economic losses and casualties every year.Using runoff confluence model to simulate the process of runoff confluence before heavy rainfall weather is of great significance to effectively reduce the losses caused by rainstorm and flood disasters.The simulation result of distributed hydrological model can reflect the runoff confluence result of the whole basin,so it has been widely used to runoff confluence model.There are two main factors affecting the simulation accuracy of the distributed hydrological model,one is the accuracy of the basic data of the model,and the other is the scale of meshing.The former is closely related to the accuracy of measuring instruments,the accuracy of space-time observation and other hardware facilities.It is difficult to improve the accuracy of model by providing high-precision basic data.It is found that the computational complexity of the model will multiple increases with the reduction of mesh size.The contradiction between the low performance of CPU(Central Processing Unit)and the increasing computational complexity of the model has become an important factor restricting the improvement of the simulation accuracy of the distributed hydrological model.So,a runoff confluence model based on general computing is proposed to improve the efficiency of runoff confluence simulation in the article.The main contents are as follows:Firstly,according to the characteristics of GPU and the principle of runoff confluence model,the feasibility of applying GPU to runoff confluence is analyzed.Secondly,the principle and algorithm implementation of GPU runoff confluence model are introduced.The model is applied to several watersheds of different watershed scales to verify the performance of GPU in runoff confluence simulation.Thirdly,the model is used to simulate the precipitation process of the target watershed(Zhenjiangguan-Jiaochangba section in the upper reaches of Minjiang River),and then compare the simulation result and the measured hydrological data of the hydrological station(school yard dam)in the lower reaches of the basin.Based on the comparison result,the shortcomings of the model are analyzed.The factors which affect the simulation accuracy of the model are elaborated and the optimization schemes are given.The optimized runoff confluence model is used to simulate the precipitation process of the target basin again.then using the measured hydrological data of hydrological station to verify the feasibility of the optimization scheme.The main innovations of this paper are as follows:Firstly,most of the traditional distributed hydrological models simulate runoff confluence base on CPU computing platform.The lower CPU computing power results in slowly simulation speed and long time simulation process.In order to enhance the simulation performance of runoff confluence model,a runoff confluence model based on GPU general calculation is designed to improve the operation efficiency of runoff confluence simulation.Secondly,in order to reduce the terrain error in the simulation of runoff confluence in River Basin,a method of terrain correction using sampling points to correct River elevation is proposed.Through River elevation correction,the river flows smoothly from upstream to downstream,which lays a foundation for runoff confluence simulation.Thirdly,a runoff confluence simulation method based on steady flow field is proposed in this paper.In order to get more accurate water regime in the early stage of the basin,the runoff confluence model is improved by adding the flow mixing link.Experiments show that when the upstream water supply is stable,the algorithm can generate stable flow field.On this basis,a certain number of hourly upstream flows and surface rainfall before the simulation period are substituted into the model iteration to simulate the early stage water regime of the basin.Experiments show that the model in this paper not only has a high simulation efficiency,but also has a good fitting result with the measured hydrological data of hydrological stations.It shows that the simulation result can reflect the actual runoff confluence process and change rule and have practical significance for rapid simulation and assessment of storm flood disaster risk.