Data Assimilation (4D-VAR) for Shallow-Water Flow: The Case of the Chicoutimi River

This is a web presentation of the work presented at the 10th Annual Conference of the CFD Society of Canada, “CFD 2002”, at the University of Windsor on June 9-11, 2002. This discussion paper presents the four-dimensional variational data assimilation (4D-VAR) technique as a tool to forecast floods. This discussion will be limited to hydrological forecast. We assume that the weather, here a large rainstorm, had already been forecasted by the meteorological services. In the 4D-VAR technique, we need to minimize, in the sense of Lagrange, a cost function which measures the difference between the forecast and the observations. The physical equations acts as a set of constraints. Here, the model is the shallow-water equations modified to include sediment transport. The minimum was found by using the steepest descent algorithm. This is made possible because the gradient of the cost function can be calculated analytically by using the adjoint equations of the model. To illustrate the 4D-VAR technique, the bypass of a simple theoretical dam as well as the more complex overflowing of the Chicoutimi River at the Chute-Garneau dam (during the 1996 flood) are investigated.