Work

DarcyWarp: A graphics processing unit (GPU)-accelerated multigrid solver for repeated 2D fixed-transmissivity groundwater solves validated against reference solvers

Year: 2026

Type: article

Abstract: Repeated steady-state groundwater solves for calibration and uncertainty analysis are often limited by runtime. We present DarcyWarp, a graphics processing unit (GPU)-accelerated geometric multigrid solver for two-dimensional, steady-state, single-layer groundwater flow using a fixed-transmissivity formulation on structured grids. DarcyWarp implements an explicit cell-integrated five-point finite-difference operator and multigrid cycle as Python kernels just in time compiled for GPU execution with NVIDIA Warp, and supports areal source/sink terms, prescribed head, no-flow, and general-head boundaries. Validation against a sparse finite-difference direct solve and MODFLOW 6 on matched grids shows close head agreement, consistent residuals, and mass-balance closure. Benchmarks span matched-grid single-instance cases from 1 0 4 to 9 × 1 0 6 active cells and two 48-case repeated-solve ensembles. DarcyWarp reaches about 70 times end-to-end speedup over MODFLOW 6 in the single-instance suite and order-of-magnitude throughput gains in ensembles, enabling reproducible repeated-solve workflows on commodity GPUs.

Source: Environmental Modelling & Software

DOI: 10.1016/j.envsoft.2026.107081

Authors: Patrick Durney, Jaco Fourie

Institutions: Lincoln University, Lincoln Agritech (New Zealand)

Language: English

Topic: Groundwater flow and contamination studies

Subfield: Environmental Engineering

Field: Environmental Science

Domain: Physical Sciences

SDG: Clean water and sanitation

Open Access status: hybrid

Estimated APC paid: $3,500