Urban-scale variational flux inversion for CO Using TROPOMI total-column retrievals: A case study of Tehran

Emission inventories of trace gases and aerosols are a key input for air pollution modelling, yet are subject to significant uncertainties. The aim of this study is to refine urban-scale, gridded estimates of carbon monoxide (CO) emissions for Tehran, Iran; this was done using an inverse-modelling approach. We assimilated total column mixing ratios of CO retrieved from the TROPOspheric Monitoring Instrument (TROPOMI) using a four-dimensional variational assimilation system based on the Community Multi-Scale Air Quality Model. The inversion system allows for two representations of the temporal profile of emissions at each grid-point: scaling a fixed diurnal profile, or with separate, constant estimates spanning four diurnal time-categories. We evaluate the inversion capability by applying independent surface measurements. We find that the model forced with prior emissions substantially underestimates day-to-day variation in satellite data. However, the posterior emissions show significant improvements in simulating the assimilated data, reducing the bias of modelled mixing ratios by 72% to 91% when using categorised temporal variation in the system and by 35% to 83% with a fixed temporal variation. Comparisons with surface measurements indicate that the posterior emissions also show some improvements in simulating surface observations, although not in all cases. For example, in February 2019, the magnitude of the bias is reduced from 1.1 ppm to 0.9 ppm and 0.2 ppm using the categorised and fixed temporal variation schemes, respectively. The posterior results are sensitive to the specification of temporal emission patterns, and the spatiotemporal pattern of the posterior varies depending on whether we assume the fixed or categorised temporal variation. It is worth noting that TROPOMI samples only once per day, so it cannot define the full diurnal cycle of emissions. Finally, the inability to match simultaneously the surface and column-integrated observations may be due to issues in modelling the vertical tracer transport, however, this was not evaluated in the present work. Notwithstanding these weaknesses, the approach shows considerable promise for regions lacking local inventories.