This thesis has been carried out at the Inria branch in Montpellier within the LEMON team, an interdisciplinary team working on the design and implementation of accurate and computationally inexpensive models for natural processes occurring in the littoral area. Urban-scale flood models are increas ingly being studied in the context of risk prevention, to this end, for several years, the LEMON team has been developing so-called porosity models for the shallow wa ter equations to provide rapid, large-scale simulations of these phenomena. Porous shallow water equations extend the classical shallow water model by introducing porosity to represent subgrid-scale effects caused by obstacles such as buildings and vegetation in floodplains. This work explores the use of variational data as similation to improve the knowledge on the spatial distribution of porosity, and, ultimately, to increase accuracy in the prediction of the flood flow. Variational data assimilation is implemented using a gradient descent optimization approach, which efficiently computes the gradient through the adjoint problem. We present both the forward and adjoint formulations, along with their one-dimensional numerical solvers. Numerical experiments demonstrate the effectiveness of this approach for improving porosity estimation. We emphasize the importance of addressing the well-posedness of the assimilation problem to ensure reliable results.