Long term modelling of the dynamical atmospheric flows over SIRTA site

The atmospheric flow knowledge is important for its role in pollutant dispersion and wind energy. In this work, the hourly atmospheric flow output (8760 states) from Weather Research and Forcasting (WRF) model for the year 2011 over SIRTA (Site Instrumental de Recherche par Teledetection Atmospherique) are analyzed and clustered into a finite number of representative atmospheric states using two clustering methods: non-controlled clustering and controlled clustering. The resulting representative situations of those clusters are used to specify boundary conditions for flow downscaling over the heterogeneous SIRTA. For flow downscaling, the CFD code Code_Saturne is used to simulate each representative atmospheric state. To assess the efficiency of WRF clustering and Code_Saturne downscaling, the measurements in SIRTA over the same year are used as reference. The Mean Absolute Error (MAE) and the Kullback-Leibler divergence (KL) metrics were computed for the distributions of the atmospheric flow features in order to: (i) compare the difference between the performance of the two clustering procedures, and (ii) compare the distribution of flow properties between WRF mesoscale model and Code_Satume. It is clearly demonstrated that the two clustering methods are comparable in benefit, and that Code_Satume improves considerably the flow features modeling in comparison to measurements.