Performing Multiple Simulations for Multiple Watersheds in Flood Forecasting Using the GSSHA Distributed Hydrologic Models in Large Basins

Thursday, 18 December 2014
Fidel Perez1, Norm Jones1, E. James Nelson1 and Scott D. Christensen2, (1)Brigham Young University, Provo, UT, United States, (2)Brigham Young University, Civil and Environmental Engineering, Provo, UT, United States
When massive storm events are imminently expected over a given area, flood forecast centers often need to execute hydrologic model simulations for multiple watersheds some of which may be large in size for distributed models to be able to complete in a relatively short time. And since it is advantageous to subdivide the watersheds for a distributed model like GSSHA to be able to have short run times, the operational solution needs to consider simultaneous simulations runs for the models of the individual sub-basins. The number of simulations in each watershed would increase if there are different scenarios that consider several weather forecasts with storm tracks and direction that produce many rainfall patterns that would differ in the magnitude and spatial distribution of rainfall. The problem is further increased when it is necessary to analyze several possible watershed conditions and several emergency operation alternatives for flood control. Some computer hardware solution is needed to perform the many that would result from the combination of the forecasted scenarios, the operation alternatives, watershed conditions in an already large number of watersheds which might also be subdivided into many sub-basins. The first problem is finding a suitable watershed sub-division that breaks down the size of the areas for the models in a way that saves run-time. This is accomplished in two test-case watersheds using a workflow of basin-model orchestration where arrangements of parallel and cascade simulations are done for the sub-basins of tributary rivers and the main river. The second problem of running all the possible simulations is dealt with using distributed computing options such as HT Condor and Microsoft Azure. The results are compared using operational performance indicators.