Parallel Groundwater Modeling using MODFLOW 6
Abstract
When hyper-resolution accuracy is required to support water management decisions at regional or national scales, computation times and memory requirements prevent effective use of numerical models. An efficient technique for obtaining reasonable run times and memory requirements is parallel computing, where the problem is solved using multiple processor cores simultaneously.
MODFLOW 6 is the latest version of the U.S. Geological Survey's modular hydrologic model. The object-oriented framework and underlying numerical solver make it possible to tightly couple any number of models at the matrix level. The inherent multi-model framework of MODFLOW 6 naturally lends itself to distributed memory parallel computing, where each model can be uniquely assigned to a processor core that is connected to other cores through the standardized and portable Message Passing Interface (MPI). The MODFLOW 6 code is expanded to support distributed memory parallel computing using MPI. MPI communication between the models is implemented by revising the existing MODFLOW 6 exchange object and introducing the concept of a halo (or ghost) model along each shared boundary between two adjacent models. With this approach of halo models, each exchange object has access, in parallel, to all of the required data in adjacent models. This also allows the existing serial code for simulating groundwater flow with the traditional conductance or Newton-Raphson formulation to be used without modification. The Iinear accelerators of the MODFLOW 6 Iterative Model Solution have been expanded to support the parallel additive Schwarz preconditioner (block Gauss-Jacobi). Numerical experiments for structured grids were carried out on the Cartesius Dutch supercomputer using up to 480 cores. A synthetic test corresponding to Poisson's equation on a circular model domain with up to 1.25 billion active cells was simulated. The PCRaster GLOBal Water Balance groundwater model with 4.5 million 10 km x 10 km active cells was also simulated. Initial tests show that a) significant speedups (> 40) can be obtained, b) the parallel simulations converge to the specified tolerances, and c) parallel computing makes MODFLOW 6 suitable for running extremely large groundwater models.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMDI31B0014V
- Keywords:
-
- 0545 Modeling;
- COMPUTATIONAL GEOPHYSICSDE: 0560 Numerical solutions;
- COMPUTATIONAL GEOPHYSICSDE: 1932 High-performance computing;
- INFORMATICSDE: 3260 Inverse theory;
- MATHEMATICAL GEOPHYSICS