Research On Parallel Computing Method Of Flow Routing Process Based On WEP-l Hydrological Model

The WEP-L hydrological model is a distributed physical model that simulates the natural water cycle and the artificial side water cycle.It can be applied to the evaluation of water resources in river basins and the analysis of the evolution law of water resources under the influence of human activities.With the rapid development of remote sensing(RS),digital elevation model(DEM),geographic information system(GIS)and other technologies.The simulation range of WEP-L model is larger and larger,the resolution is higher and higher,and the process is more and more complex.As a result,the serial computation time of the model is long and the computational efficiency is low.Therefore,it is necessary to introduce highperformance parallel computing technology.This study focuses on the special temporal and spatial dependence between the sub-watershed in the WEP-L hydrological model.To solve the problem that the flow routing process can not directly realize parallel computing.The parallel computing method is designed and applied to WEP-L model flow routing process.The main achievements include the following:1.The parallel computation of layered simulation unit based on sub-watershed is designed.This paper studies sub-watershed division by Pfafstetter code in WEP-L model,and establishes the spatial topological relation table of sub-watershed according to the coding rules.By spatial stratification of sub-watershed,the sub-watershed in the same layer have no interdependent relationship,and the sub-watershed in each layer are assigned to each thread for calculation.The optimal scheduling,running time and parallel performance and the influence of threshold of sub-watershed number on parallel computation are analyzed.The experiment shows that the parallel computation of layered simulation unit based on sub-watershed can effectively reduce the simulation time.2.The parallel computation of domain decomposition based on greedy scheduling is proposed.A domain decomposition algorithm based on main and branch is designed.The original flow routing process was decomposed into tributary flow routing process composed of several independent hydrological domains and main flow routing process.On this basis,in order to realize the load balance on the thread,the optimal scheduling based on greedy algorithm is integrated.Finally,the tributary tasks were assigned to different threads for calculation,and the main stream sub-watershed used serial computation.The experiment shows that both of the two methods of domain decomposition before and after greedy scheduling can shorten the running time of the confluence,and the parallel algorithm based on greedy scheduling is better.3.The temporal-spatial discretization parallel computation based on Pfafstetter coding is proposed.According to Pfafstetter coding does not have the property of spatial layering and cannot directly use temporal-spatial discretization methods.Therefore,a temporal-spatial discretization algorithm based on Pfafstetter coding is proposed.The differences between multitree coding and Pfafstetter coding were analyzed.The theoretical time consumption of serial computation,layered simulation unit parallel computation and temporal-spatial discretization parallel computation are studied respectively.The experiment shows that the temporal-spatial discretization parallel computation based on Pfafstetter coding further explores the parallelism degree of flow routing process,and its parallel performance is better than that of spatial decomposition parallel computation.4.Developed the WEP-L distributed hydrological model parallel computing system.The system simplifies the process of model parameter calibration.Implant the above three parallel computing methods,and provide the function of generating watershed topology,which can more easily and intuitively design algorithms,execute parallel programs,and understand the load state of the runtime,etc.