In general, the higher the resolution of the finite difference grid (smaller grid cells), the more accurate the solution will be. While theoretically the number of grid cells that can be used for GSSHA modeling is unlimited, there are some practical limitations. In order to define all model parameters, approximately 3,500 bytes of memory are required for each grid cell. This means that one megabyte of RAM is required for each 300 grid cells of a GSSHA simulation. In addition to the amount of memory, the time to display and work with the model inside of WMS and the time required for the GSSHA simulation to run will increase.
The issue of appropriate grid sizes has not been adequately answered to date. Several research papers address the range of applicability of the diffusive wave form of the equations of motion that are used for overland flow routing in GSSHA. However, the picture is complicated by the difficulty in including the effects of microtopography on runoff routing. Given this fact, the use of any physically based routing technique is merely an approximation of reality. GSSHA has been successfully used with grid sizes ranging from 30 to 1,000 m (~100 ft to 3,280 ft). However, experience by the GSSHA developers has shown that grid sizes smaller than 200 m (660 ft) produce more robust calibrations. Downer et al. (2002b) discuss how grid size should be sufficiently small to capture the essential features in the study watershed. In general, the philosophy of distributed hydrology is that “smaller is better” provided that the problem remains computationally feasible and the quality of the data warrants the use of small grid sizes.