IMPROVEMENT OF PERFORMANCE AND APPLICABILITY OF MODFLOW-2005: NEW NWT SOLVER AND xMD MATRIX SOLVER PACKAGE

MODFLOW has been widely used for many years to investigate groundwater flow systems. When a simulation domain involves unconfined and non-linear boundary conditions, Picard's method is used to linearize the governing flow equation. Although Picard's method demonstrates good performance in most cases, its efficacy degrades when a model system is strongly non-linear. This is caused by the method’s weak linearization capability. In order to improve the performance of the MODFLOW model for non-linear problems, the Newton linearization method, which is widely used for solving a system of non-linear equations and more robust than Picard's method, is employed in a newly-developed NWT solver package for MODFLOW-2005. The NWT solver package could greatly extend the applicability of the MODFLOW model, particularly to problems involving unconfined aquifers and surface-water/groundwater interaction. In addition to poor performance for strong non-linear problems, the MODFLOW model shows difficulties in obtaining solutions when a simulation problem involves larger hydraulic conductivity contrasts among its geological units. These difficulties are caused by a “hard-to-solve” matrix that is assembled through numerical discretization of the governing flow equation. A new χMD solver package, in addition to the existing matrix solvers of MODFLOW-2005, is used in order to improve the model’s performance. The χMD package is based on a preconditioned conjugate gradient type matrix solver that includes CGSTAB. The preconditioning scheme permits various levels of incomplete LU factorization and reordering of unknowns. In addition to these features, the solver package allows the use of a drop tolerance scheme which dramatically reduces computational cost and storage. This talk will present the preliminary results of the new package, including the model’s ability to provide a solution for a difficult unconfined groundwater flow problem as well as improvement of matrix solver performance.