A Study For Applying Parallel Computing In The MODFLOW-A Groundwater Modeling Software For Numerically Simulating Groundwater Flow

The numerical simulation of groundwater flow is an important method for evaluating the groundwater resources. As researches go further, urgent requirements have been posed on modeling groundwater with much-refined spatial grids and long temporal series. However, the use of the serial computing would slow down calculating rate of the model, and even lead to incorrect running of the model because of the limitation from the memory and the data channel bandwidth of the software.This research focused on the speeding up of the modeling process, in which the calculation and storage problem of the large data during the modeling was investigated. Based on the superiority of the parallel computing to the large date processing and calculation and also the parallel programming particularity of the different computer architectures, an efficient parallel algorithm for the large-scale groundwater flow numerical simulation with different parallel implementation methods was designed. The following results were obtained: (1) The program interface between the groundwater numerical simulation software MODFLOW and the JASMIN framework was established. By decomposing the research area into different small patches and using lots of processors (at least hundreds, even thousands) at one time, the calculation efficiency was improved. The researchers can selectively use all kinds of highly effective parallel solvers under the JASMIN framework according to the practical situation of the established modeling, by which the flexibility and convenience of the model process was largely enhanced. (2) The design principles of the parallel P-PCG subroutine package were demonstrated in detail, and method for solving the problem of ineffective accelerating of the model with less layer was proposed. The improved parallel P-PCG subroutine package can decrease the inhibition effect of the model layer to the parallel function of the program and can utilize each processor of the computer efficiently. The realization of the parallel function of the improved subroutine package would obviously increase the simulating ability of the MODFLOW to the groundwater numerical simulation with large-scale area, long time series and complicated hydrogeology conditions.(3) Based on the character of the distributed memory parallel programming, We analyzed the parallel feature of the GMRES(m) algorithm of Krylov subspace under MPI environment and proposed an parallel implementation method through the domain decomposition, and compared the effects of different combinations of the method with different preconditioners. The outcome showed that the parallel GMRES(m) algorithm can highly increase calculation speed of the groundwater flow numerical simulation, and even better speed-up and running efficiency were obtained when the algorithm integrated with the Jacobi preconditioner. In summary, parallel GMRES(m) algorithm has a preferable generality and extensibility.