Hazardous Atmospheric Dispersion in Urban Areas: a Deep Learning Approach for Emergency Pollution Forecast

International audience Today, Computational Fluid Dynamics approaches have a high level of spatial/temporalaccuracy in modelling atmospheric transport and dispersion in very complex environments.Several numerical models require, however, heavy computational resourcesand prolonged simulation time up to several days. This time constraint is specificallycrucial for intervention planning in case of accidental or malevolent toxic releasesin a city. In this paper, we propose to use synthetic data generated by a realistic 3-D transport/dispersion simulator, to train a learning framework called MCxM. Thelatter relies on a sequence of masking and correction operations to progressivelyapply the spatial constraints and underlying physics of transport and dispersion. Thelearning phase uses the urban geometry of the French city Grenoble.We then test theeffectiveness of the trained MCxM in a different French city: Paris. The results showthat the MCxMs forecasts are virtually instantaneous and generalize successfully tounseen conditions.