Section: New Results

Multiscale modeling of air quality

Participants : Vivien Mallet, Irène Korsakissok.

Classical large-scale models in air quality are based on Eulerian approaches. In particular, it is usually assumed that emissions from the point sources mix immediately within the grid cell, whereas a typical point source plume (e.g., from a power plant) does not expand to the size of the grid cell for a substantial time period. Hence, there is a need for a subgrid-scale modeling of the key phenomena (emissions, transport and chemistry). The plume-in-grid modeling technique, that consists in coupling a local-scale model with an Eulerian model, has been developed to allow a more accurate representation of sub-grid processes. A sensitivity study was carried out for passive tracers with ETEX experimental data in order to investigate the influence of the parameterizations for standard deviations in the puff model, as well as the feedback methods (i.e., the way the puff is injected in the Eulerian model). Results for chemically reactive plumes have been obtained in the Paris area.

Statistical approaches were also studied. Based on a single large-scale model or on an ensemble of large-scale models, statistical downscaling allows to accurately forecast air quality (that is, pollutant concentrations) at observed locations. The methods rely on regressions. The use of an ensemble leads to problems due to the colinearities between the regressors (the models). This issue is addressed with reduction based on principal component analysis or, preferably, with principal fitted components.