Comparison of new finite-volume SWMM5+ Saint-Venant solver to SPRNT and SWMM5

A novel parallelized, explicit, finite-volume solver alternative to the established serial, implicit, link-node SWMM5 Saint-Venant has been prepared for alpha release. The performance of the new computational engine (SWMM5+) was tested on a large scale river network, the entire HUC-2 Texas River Basin, and compared with SWMM5. Some discrepancies were expected from the differences in computational approach. The established Simulation Program for River Networks (SPRNT), was included in the comparison to provide a range of tolerable disagreement between accepted models. SPRNT uses a link-node finite-difference form of the full Saint-Venant equations solved with the implicit Preissmann stencil and Newton solution of the resulting nonlinear matrix equation. SWMM5 represents the Saint-Venant equations as coupled node head and link flow equations using an implicit Euler stencil and Picard iteration with a relaxation factor of 0.5. SWMM5+ uses a mass-conservative finite-volume approach for the Saint-Venant equations with explicit Runge-Kutta 2nd order time-marching.

In an ensemble model comparison analysis, the SPRNT 5-month continuous simulation results were used as a baseline. SWMM5+ and SWMM5 (both Kinematic and Dynamic wave) model outputs were quantitatively compared to the SPRNT output on the basis of Normalized Nash-Sutcliffe Efficiency Index (NNSE), Normalize Peak Time Difference, and Normalized Peak Magnitude Difference. The comparative metrics were modified to account for model behavior spatially distributed across the Texas River Basin, ensuring that the models were in agreement at a variety of hydraulically independent sites, as well as temporally for the duration of the 5-month continuous simulation. The conclusions of this analysis will be available at the time of presentation.

Remaining work toward the full release of the SWMM5+ engine includes completing the implementation of an Artificial Compressibility solver for surcharged pipes, as well as expanding the parallelization potential of the solver through the inclusion of a graph partitioning method, BIPquick. Parallelized simulation results are not yet available, but preliminary diagnostic tests suggest a speed-up of 20x or more.

 Click here to download a static PDF version of the presentation.

 Click here to watch recorded presentation on YouTube.