Predictions on arrival times of water of the St. Francis dam break flood using ANUGA

ANUGA software is used to simulate the 1928 St. Francis dam break flood. ANUGA (see https://anuga.anu.edu.au/) is a free and open source software, which is designed to simulate shallow water flows. Our simulation assessment in this paper is based on the arrival times of water at several stations. We use the BreZo hydrodynamic algorithm results of Begnudelli and Sanders ("Simulation of the St. Francis Dam-Break Flood," Journal of Engineering Mechanics, Vol. 133, pp. 1200-1212, 2007) to compare with, in addition to measured arrival times of the real event. Note that in simulations, arrival time can be measured with respect to either the water front or water discharge peak. Our simulations are of two types, catchment and detailed. The catchment type simulation considers the area extending from the dam site to the Pacific Ocean. It has the dimension 81.3 km by 43 km containing the entire 87 km river reach (about domain diagonal) with around 300,000 triangular computational cells. Our results of the catchment type simulation agree with those of measured arrival times of the real event. These measured arrival times of the real event were investigated by a number of researchers, such as, Outland in 1963 (" Man-made disaster: The story of St. Francis Dam", published by The Arthur H. Clark Company). Note that Begnudelli and Sanders computed arrival times of water at specified stations by checking the arrival of water front and the water momentum (discharge) peak. They obtained that the arrival times of water discharge peaks agree more with measured arrival times of the real event than the arrival times of the water front. However, our ANUGA simulation results in the opposite, that is, arrival times in terms of the water front match better with measured arrival times of the real event. The detailed type simulation focuses on the upstream reach area. It has the dimension 8.1 km by 6.9 km containing 6 km portion immediately downstream of the dam wall and the 4 km reservoir, making it in total a 10 km (about domain diagonal) model. The aforementioned paper by Begnudelli and Sanders reports on shock waves and sloshing behaviour as the dam break flood wave progressed down the initial portions of the valley. In order to fully investigate these phenomena the detailed ANUGA mesh is set up and includes triangular cells down to the size of 100 m(2). Therefore, we have around 161,000 triangular computational cells for this detailed type simulation. Once again, our computational results indicate that arrival times in terms of the water front match better with measured arrival times of the real event rather than water discharge peaks. These results confirm that arrival time predictions should be viewed with some scepticism unless modellers have accurate values of computational parameters, such as topography roughness, as Begnudelli and Sanders suggested.