The first 1-year-long estimate of the Paris region fossil fuel CO2 emissions based on atmospheric inversion
- 1Laboratoire des Sciences du Climat et de l'Environnement, LSCE/IPSL, CEA-CNRS-UVSQ, Université Paris-Saclay, Gif-sur-Yvette, France
- 2AIRPARIF Surveillance de la Qualité de l'Air en Île-de-France, Paris, France
- 3CNRM-GAME (CNRS-Météo-France), UMR3589, Toulouse, France
- anow at: Laboratoire de Physico-Chimie de l'Atmosphère, Université du Littoral, Côte d'Opale, Dunkerque, France
- bnow at: Environment and Climate Change Canada, Toronto, Canada
- cnow at: Institut für Umweltphysik, Ruprecht-Karls-Universität Heidelberg, Heidelberg, Germany
Abstract. The ability of a Bayesian atmospheric inversion to quantify the Paris region's fossil fuel CO2 emissions on a monthly basis, based on a network of three surface stations operated for 1 year as part of the CO2-MEGAPARIS experiment (August 2010–July 2011), is analysed. Differences in hourly CO2 atmospheric mole fractions between the near-ground monitoring sites (CO2 gradients), located at the north-eastern and south-western edges of the urban area, are used to estimate the 6 h mean fossil fuel CO2 emission. The inversion relies on the CHIMERE transport model run at 2 km × 2 km horizontal resolution, on the spatial distribution of fossil fuel CO2 emissions in 2008 from a local inventory established at 1 km × 1 km horizontal resolution by the AIRPARIF air quality agency, and on the spatial distribution of the biogenic CO2 fluxes from the C-TESSEL land surface model. It corrects a prior estimate of the 6 h mean budgets of the fossil fuel CO2 emissions given by the AIRPARIF 2008 inventory. We found that a stringent selection of CO2 gradients is necessary for reliable inversion results, due to large modelling uncertainties. In particular, the most robust data selection analysed in this study uses only mid-afternoon gradients if wind speeds are larger than 3 m s−1 and if the modelled wind at the upwind site is within ±15° of the transect between downwind and upwind sites. This stringent data selection removes 92 % of the hourly observations. Even though this leaves few remaining data to constrain the emissions, the inversion system diagnoses that their assimilation significantly reduces the uncertainty in monthly emissions: by 9 % in November 2010 to 50 % in October 2010. The inverted monthly mean emissions correlate well with independent monthly mean air temperature. Furthermore, the inverted annual mean emission is consistent with the independent revision of the AIRPARIF inventory for the year 2010, which better corresponds to the measurement period than the 2008 inventory. Several tests of the inversion's sensitivity to prior emission estimates, to the assumed spatial distribution of the emissions, and to the atmospheric transport modelling demonstrate the robustness of the measurement constraint on inverted fossil fuel CO2 emissions. The results, however, show significant sensitivity to the description of the emissions' spatial distribution in the inversion system, demonstrating the need to rely on high-resolution local inventories such as that from AIRPARIF. Although the inversion constrains emissions through the assimilation of CO2 gradients, the results are hampered by the improperly modelled influence of remote CO2 fluxes when air masses originate from urbanised and industrialised areas north-east of Paris. The drastic data selection used in this study limits the ability to continuously monitor Paris fossil fuel CO2 emissions: the inversion results for specific months such as September or November 2010 are poorly constrained by too few CO2 measurements. The high sensitivity of the inverted emissions to the prior emissions' diurnal variations highlights the limitations induced by assimilating data only during the afternoon. Furthermore, even though the inversion improves the seasonal variation and the annual budget of the city's emissions, the assimilation of data during a limited number of suitable days does not necessarily yield robust estimates for individual months. These limitations could be overcome through a refinement of the data processing for a wider data selection, and through the expansion of the observation network.