Testing of the Subtiming and Subgridding Algorithms Incorporated into HydroSphere to Handle Intensive Computational Demand
Recent advances in distributed modeling of integrated surface/subsurface flow and transport processes have made it possible to assess Reclamation water supply and water quality concerns in an accurate and rigorous, physically-based, spatially distributed manner. However, computational demands limit routine application to large basins for long simulation periods.
* Can the incorporation of nested-gridding and sub-timing techniques improve computational efficiency sufficiently (by an order of magnitude or more), resulting in significant reductions in simulation time--thereby make accurate computations at large scales feasible?
Need and Benefit
Hydrologic models are used by Reclamation and other agencies for a wide variety of water resource planning and management purposes, including streamflow prediction, flood and drought forecasting and reservoir operations. Many commonly used models were first formulated in the 1960s and 1970s when computer speed and memory imposed severe restrictions on the level of sophistication that was feasible. These models, such as HSPF and the Sacramento soil moisture accounting model, represent watersheds as simplified, lumped parameter systems; further, they ignore interactions between the surface and subsurface water regimes altogether, or treat them as separate with a source-sink linkage. The limitations of such simplifications are increasingly recognized and documented (e.g., Labolle et al, . Review of the Integrated Groundwater and Surface-Water Model, Groundwater, 41, No. 2, 238-246, 2003). They include mass-balance errors and, in some cases, predicted outcomes that are physically unreasonable.
The shortcomings described above have been overcome by a model called HydroSphere, developed at the University of Waterloo and acquired by the Reclamation in fiscal year (FY) 2003. HydroSphere is a comprehensive and fully-integrated mechanistic model that accounts for three-dimensional (3D) variably-saturated subsurface flow and two-dimensional (2D) overland/stream flow. It has been applied to lab, field, and sub-basin scale problems; however, to make large basin scale simulations practicable, issues of computational efficiency for simulating large scales need to be overcome. To further enhance the numerical efficiency of HydroSphere, for water supply forecasting and assessment of large basins over long time periods, it is proposed that advanced algorithms for nested-gridding and sub-timing be incorporated into the software. The expected benefit is a 10 to 100-fold increase in computational speed. Nested-gridding in three dimensions allows a relatively coarse numerical finite element or finite difference grid to be used for the entire basin, with finer resolution, only where needed.
Hence, a model can be developed with the most optimal spatial resolution in all portions of the domain for best simulation efficiency. For instance, the overland flow domain can use a finer resolution in regions of steep slopes or around topographical features of importance, while the subsurface domain can use finer resolution only in regions of differing material properties, faults, or around wells. Sub-timing allows different time-step sizes to be used for different parts of the modeling domain, in a fully implicit fashion. For instance, smaller time steps may be used for surface water flow, with larger time steps for ground water flow. This technique results in a large increase in computational speed for long term simulations, while still maintaining full coupling between surface and subsurface flow and transport processes.
Contact the Principal Investigator for information about these documents.