Long-term urban aerosol simulation versus routine particulate matter observations
Résumé
The ability of chemistry-transport models (CTMs)to accurately simulate particulate matter in urban areas is still to be demonstrated. This study presents a statistical evaluation of the performances of a mesoscale aerosol CTM over the Paris area, calculated over a long time period. Model simulations are compared to measured particulate matter PM(10) and PM(2.5) levels at monitoring ground stations. In summer, the PM10 daily mean levels are fairly well predicted by the model at all stations with correlation coefficients exceeding 0.67, relatively low biases (<2.5 mu g m(-3)) and normalized errors (<27%). The relatively uniform negative biases suggest that the background PM(10) levels are underestimated. In winter, discrepancies between the model and observations are more important, in particular at urban sites where several erroneous peaks are simulated. Consequently, the correlation coefficient drops down to 0.59 at urban sites and PM(10) values are overestimated by about 10 mu g m(-3) with normalized errors exceeding 55%. We assume that discrepancies between simulated and observed PM levels are due to (i) TEOM (tapered element oscillating microbalance) measurement underestimation (35% in winter) caused by the evaporation of ammonium-nitrate, (ii) the under-prediction of the model vertical mixing over the urban heat island and (iii) possible overestimation of local PM emissions. We use corrections for the urban boundary layer height and we subtract ammonium-nitrate from model PM(10) concentrations. These modifications significantly improve the comparison statistics at urban sites in winter: the mean bias (< 2 mu g m(-3)) and normalized error (<30%) are reduced, while the correlation coefficient increased to 0.64. However, the overestimation at urban sites is inconsistent with the underestimation of PM10 background concentrations. The analysis of the total model biases at urban sites reveals that the underprediction of PM(10) background levels is largely compensated by their local overprediction due to the overestimation of anthropogenic emissions.