Improvement of Performance of MODFLOW toward Better Management of Hydrologic Systems

Numerical modeling is an essential tool for better management of hydrologic systems. In that regard, MODFLOW has been widely used for many years to investigate groundwater flow systems and is a widely accepted standard for groundwater modeling. In most of numerical models including MODFLOW, more than 80% of memory and execution time is spent in the matrix solver; thus, improving matrix solver performance is a key to enhancing simulation performance. The preconditioned conjugate gradient type method has shown its robustness for solving sparse matrix systems. The method’s preconditioning component is paired with an acceleration part. Even if the acceleration methods guarantee that the solution will be obtained within a finite number of iterations, a higher quality of preconditioning is necessary to reduce the number of iterations and the computational cost. A χMD solver package was developed for higher robustness, faster execution speed, and better memory efficiency. The preconditioning module of χMD consists of level-based incomplete LU (ILU) factorization with a drop tolerance scheme that can reduce memory usage and lead to faster execution speed. The acceleration module of χMD includes conjugate gradient, ORTHOMIN, and Bi-CGSTAB accelerations. The χMD solver package is adapted for both an Un-Structured Grid Version of MODFLOW and MODFLOW-NWT. Preliminary results show that level-based ILU factorization with a drop tolerance scheme greatly reduces memory usage compared to ILU-only factorization by a factor of two or more. In addition, execution speed increases by 40% or more.