One of the main benefits of MODFLOW-USG is that it is capable of achieving a high level of accuracy, with much fewer grid cells than what would be needed with MODFLOW-2005. In order to quantify the performance gains, we ran a benchmark test that compared the run time of MODFLOW-USG to the run time of MODFLOW-2005 for a real world regional scale groundwater flow model, with the same grid refinement around the flow boundaries. Our benchmark tests demonstrate that MODFLOW-USG is faster than MODFLOW-2005, reducing run times by 90% while delivering the same level of accuracy.
A regional watershed model spanning 225 km2 was generated using Visual MODFLOW Flex. This model includes several meandering rivers, recharge, and a few water supply wells that provide potable water to a nearby community.
Watershed boundary and boundary conditions in the conceptual model
A conceptual model was generated in Visual MODFLOW Flex which included the boundary conditions (rivers, pumping wells, and recharge) and the subsurface characterization. This conceptual model consisted of six layers with layers 3 and 4 pinching out in the southern area of the model. The conductivity of the model ranges from 3.0E-7 m/s to 1.0E-5 m/s.
Left: property zones in the conceptual model.
Right: cross-sectional slice through the unstructured grid, colored by property zone ID.
Two numerical models were created from the conceptual model. Both numerical models had the same level of grid refinement around the wells and rivers. The first was a uniform finite difference grid for MODFLOW-2005, consisiting of 50 meter cell spacing and 6 deformed vertical layers. The result was a grid with 290 rows and 306 columns, approximately 346,000 active cells. Since the model structure included two layers that pinchout, a minimum thickness of 0.01 m was enforced. The second numerical model was an unstructured grid for MODFLOW-USG. The Voronoi Cells honored a 50 meter cell refinement around the rivers and pumping wells. The cells in the discontinuous regions have zero thickness and are assigned as inactive. The result was a grid with approximately 31,000 active cells.
Boundary condition cells for the MODFLOW-2005 grid (well colored in red, river colored in blue)
Boundary condition cells for the MODFLOW-USG grid
MODFLOW-2005 and MODFLOW-USG numerical models were created and run using the specified numerical engines. The MODFLOW-2005 model displayed dry cells in layer 1 which likely resulted in an unstable solution and longer run times. MODFLOW-USG used layertype=4, with Upstream Weighting and as a result, does not have the dry cells problem. The BCF property package was used for both model runs. The MODFLOW-2005 run used the WHS solver with hclose and rclose equal to 0.0001. MODFLOW-USG used the xMD solver with the default settings.
Plan view of calculated heads from the MODFLOW-2005 run (left) and MODFLOW-USG run (right)
The end result of this model comparison showed a nearly identical mass balance result with the discrepancy of less than 0.5%. The model results were very comparable with small discrepancies. The significant differences come from the efficiency and reliability of the solution. MODFLOW-USG was able to solve this model ten times faster than MODFLOW-2005 and with only one tenth the number of cells. The detailed results are presented in the chart below*:
Volumetric Budget summary from the steady-state run
and comparisons of runtimes and grid size between MODFLOW-2005 and MODFLOW-USG
The release of MODFLOW-USG is clearly a game-changer for groundwater modeling. MODFLOW-USG can substantially improve run times and model accuracy, making groundwater models much more efficient. The integration with Visual MODFLOW Flex adds another level of efficiency to groundwater modeling, allowing users to generate multiple numerical models from the same conceptual model data set. This means that comparisons, such as the one discussed in this article are easily and quickly completed with the creation of a conceptual model, leading to an improved level of model acceptance and reliability.
*Model runs were executed using the x64 version of Visual MODFLOW Flex v.2014.1 (v.188.8.131.52), on a Windows 7 x64 computer, with a dual-core Intel(R) Xeon(R) CPU 5110 @ 1.60GHz, and 10 GB RAM.