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

Precise wind fields simulated by CFD models are used for many environmental and safety micro-meteorological applications, such as dispersion modelling or wind potential assessment. Atmospheric simulations at local scale are largely determined by boundary conditions, which are provided, for instance, by meso-scale models (e.g., WRF). In order to improve the accuracy of the boundary conditions (BC), especially in the lowest levels more perturbed by high resolution topography, data assimilation methods might be used to take available observations into account. Data assimilation methods have been generally developed for larger scale meteorology and initial conditions. Among the existing methods, the iterative ensemble Kalman smoother (IEnKS) has been chosen as it is independent of the atmospheric model and it is able to handle non-linear operators. The IEnKS has been adapted to local scale atmospheric simulations by taking BCs into account. This adapted version has previously been tested on a simple shallow-water model in 1D. In the present study, we analyse the performances of the IEnKS in 3D with the CFD model Code Saturne using both twin experiments and field observations over a realistic, very complex topography. We propose a method to determine the first estimate of the control vector, which corresponds to the BCs, and to construct the associated background error covariance matrix, from the statistical analysis of three years of WRF simulations. The IEnKS is proved to greatly reduce the error and the uncertainty on the BCs and thus on the simulated wind field over the small-scale domain. The IEnKS is also tested in urban conditions with observations provided by the Mock Urban Setting Test field campaign. This study case allows to evaluate the possibility to assimilate either wind observations (speed and direction) or pollutant concentration values. We present here the first results obtained in this urban configuration.