Variational regional inverse modeling of reactive species emissions with PYVAR-CHIMERE-v2019

Abstract

pUp-to-date and accurate emission inventories for air pollutants are essential for understanding their role in the formation of tropospheric ozone and particulate matter at various temporal scales, for anticipating pollution peaks and for identifying the key drivers that could help mitigate their concentrations. This paper describes the Bayesian variational inverse system PYVAR-CHIMERE, which is now adapted to the inversion of reactive species. Complementarily with bottom-up inventories, this system aims at updating and improving the knowledge on the high spatiotemporal variability of emissions of air pollutants and their precursors. The system is designed to use any type of observations, such as satellite observations or surface station measurements. The potential of PYVAR-CHIMERE is illustrated with inversions of both carbon monoxide (CO) and nitrogen oxides (NOspan classCombining double low line"inline-formula"ix/i/span) emissions in Europe, using the MOPITT and OMI satellite observations, respectively. In these cases, local increments on CO emissions can reach more than span classCombining double low line"inline-formula"+/span50 %, with increases located mainly over central and eastern Europe, except in the south of Poland, and decreases located over Spain and Portugal. The illustrative cases for NOspan classCombining double low line"inline-formula"ix/i/span emissions also lead to large local increments (span classCombining double low line"inline-formula">/span 50 %), for example over industrial areas (e.g., over the Po Valley) and over the Netherlands. The good behavior of the inversion is shown through statistics on the concentrations: the mean bias, RMSE, standard deviation, and correlation between the simulated and observed concentrations. For CO, the mean bias is reduced by about 27 % when using the posterior emissions, the RMSE and the standard deviation are reduced by about 50 %, and the correlation is strongly improved (0.74 when using the posterior emissions against 0.02); for NOspan classCombining double low line"inline-formula"ix/i/span, the mean bias is reduced by about 24 % and the RMSE and the standard deviation are reduced by about 7 %, but the correlation is not improved. We reported strong non-linear relationships between NOspan classCombining double low line"inline-formula"ix/i/span emissions and satellite NOspan classCombining double low line"inline-formula"2/span columns, now requiring a fully comprehensive scientific study.