Data assimilation at local scale to improve cfd simulations of dispersion around industrial sites and urban neighbourhoods

Wind fields around industrial sites and in urban neighbourhoods have very complex structures, which are sensitive to geometrical features such as topography and buildings. These wind field structures are difficult to simulate with CFD models. Yet, these simulations are important to address various issues related to micrometeorology and dispersion of pollutants. To perform small scale simulations, CFD models use inputs (initial and boundary conditions) that usually are meteorological data obtained from measurements or larger-scale model outputs. These data often lack precision, may not contain all necessary information, and are not adapted to the detailed features of local scale, especially the topography and the presence of buildings. A few measurements inside the domain, although very local, have the potential to greatly enhance the precision of the simulations and thus the prediction of pollutants concentrations. If some concentration measurements are available, they can also be used to improve the simulations. Using measurements to improve the estimation of the system state is the goal of data assimilation. Data assimilation techniques developed so far in meteorology are generally applied to larger scale simulations that are mainly driven by initial conditions and deal with simple geometries without obstacles. The present work aims at developing local-scale data assimilation techniques that focus on boundary conditions rather than initial conditions and may deal with very complex geometries. Two data assimilation techniques (back and forth nudging algorithm and iterative ensemble Kalman smoother) are evaluated at small scale, at first, with a simple flow, using 1D/2D solutions of the shallow water equations. This simple test-case allows in particular to verify that small-scale simulations are more sensitive to errors on boundary conditions than errors on initial conditions. The data assimilation techniques are adapted to take boundary conditions into account and their performance and limitations are compared using the simple case described above.