Inverse modeling of emissions

The emission inventories of pollutants used in atmospheric models are best estimates based on the current knowledge of the processes causing the emissions, like economic activities, vegetation fires, etc. These estimates have large uncertainties, especially in the case of natural emissions. An alternative way of estimating these emissions relies on the use of atmospheric observations in conjunction with an atmospheric model.

The principle of inverse modeling is to search for the emissions which minimize the overall discrepancy between the model calculations and the observations, while taking the uncertainties on the observations into account. In the so-called Bayesian approach, an additional constraint is provided by the requirement that the optimized emissions are not too different from the a priori guess given by the “bottom-up” emission inventory. Of course there are a lot of difficulties with this method, e.g. related to the non-linear behavior of reactive gases (like CO and NO2) or with the choice of the emission parameters to be optimized and of the a priori errors on these parameters. Possible model or measurement biases need also to be taken into account.

In the last few years, we have developed a powerful tool for optimizing the emissions of tropospheric pollutants, based on the adjoint of the global tropospheric model IMAGES. The adjoint provides the derivative of the overall model/data bias with respect to a set of control parameters. This method is computationally expensive, but it is perfectly adapted to the optimization of emissions of reactive gases. We have demonstrated the feasibility and interest of the approach.

  • In Muller and Stavrakou, Atmos. Chem. Phys., 2005, the annual emissions of CO, NOx and the biogenic hydrocarbons over large regions (Europe, Africa, etc.) are simultaneously optimized, while taking their chemical interactions into account. These feedbacks are significant due to the large impact of these compounds on the abundance of the hydroxyl radical (OH) which is the main cleansing agent of the atmosphere. The observations used in this case were ground-based observations of CO and satellite-derived tropospheric NO2 columns.
  • In Stavrakou and Muller, J. Geophys. Res., 2006, we use CO columns from the MOPITT satellite instrument in order to provide improved estimates of CO and biogenic VOC emissions. In this case, we optimize the emissions from every model pixel and every month and for each of three main emission categories (anthropogenic, biomass burning, biogenic). The number of parameters optimized is about 30,000. Correlations between the errors on the emissions from different locations and times are also introduced. They are found to be necessary in order to produce meaningful solutions to the optimization problem.

More recently (Stavrakou et al., Geophys. Res. Lett., 2008), the same technique has been applied to the inversion of long time series of measurements provided from the combination of GOME and SCIAMACHY measurements. The NO2 vertical columns of the TEMIS project have been used to infer the distribution, interannual variability and trends in the emissions of NOx.

A 10-year data series of HCHO columns retrieved by I. De Smedt and M. Van Roozendael at IASB-BIRA (De Smedt et al., Atmos. Chem. Phys. Discuss.,2008) is now being used to constrain the biomass burning emissions of non-methane volatile organic compounds as well as the biogenic emissions from isoprene (Stavrakou et al., Proceedings of the 2nd ACCENT Symposium, Urbino, July 2007).