Exposure of population to degraded air quality generates adverse effects on public health and the environment. Therefore, regulation activities at national and international levels are enforced to ensure countries take the appropriate measures to control emissions and/or to develop and establish mitigation strategies. One of the basis to establish such measures is the correct knowledge of the actual emissions of the pollutants under consideration. The emission inventories are developed through bottom-up, top-down and downscaling approaches (eg. Guevara et al.,2103; EEA 2013, Kuenen et al., 2014), being the bottom-up one of the most common. This approach relies on the emission factors and activity rates (periodically updated) issued by different sources and often obsolete and associated to uncertainties (Hanna et al., 2001; Menut and Bessagnet, 2010) which are  difficult to address and that will ultimately affect the assessments and conclusions derived from them. These uncertainties may include unknowns in the disaggregation of budgets, over/under estimation of sources, wrong timing distribution of the emissions and even the non-consideration of undeclared emissions (Vautard et al., 2003; Viana et al., 2005; Stohl et al., 2013).

Optimized top-down approaches with inverse modelling, combining model output and observational data (ground-based, aircraft-based or satellite), are developed and used to overcome this limitation and to complement, evaluate and improve existing emission models (e.g. Bergamaschi et al., 2000; Queló et al., 2005; Brioude et al., 2011; Saide et al., 2011). This approach is conceptually based on the minimization of the differences between the modelled concentrations and the observations. For this method and for linear or linearised systems, in addition to the measurements, the self-adjoint Lagrangian particle dispersion models (LPDM) provide an effective way to generate the model sensitivities needed to estimate the modelled concentrations. LPDM, nonetheless, suffer as well from limitations, especially in the treatment of background and initial conditions. Approaches to overcome these limitations are developed based on observational data (e.g., using background station measurements) or using previous/simultaneous Eulerian runs (e.g. Thompson et al.2014).

Currently, inverse modelling with ground-based measurements has been successfully used to estimate anthropogenic fluxes (e.g. Thompson et al. 2014) mostly at global, continental and country-wise/regional scales with monthly to yearly basis. Methane has been one of the target pollutants for such studies since it is an important climate forcing gas with a large global warming potential, relatively short atmospheric lifetime and contributor to many atmospheric chemistry processes. In order to complement the weaknesses, compensate the estimates or check consistency from direct flux measurements or other process-based modelled bottom-up approaches, inverse modelling is also applied to estimate methane fluxes.


The IMPLiCIt (IMProving inversion modeL Capability in Ireland) project is a Research and Development Service funded by the Irish Environmental Protection Agency (EPA). The project aims at developing a combined measurement and modelling system to verify CH4 sources over Ireland and regions affecting the Irish domain and to  improve the Irish national  capacities  to estimate  and verify national methane and other greenhouse gases (GHG)  emission inventories. A  relatively new inverse modelling system based on the atmospheric transport model FLEXPART (Stohl, 2005) together with the methodology and open-source system developed by Thompson et al. (2014, 2015), FLEXINVERT ( (Figure 1) has been implemented and tested for the Irish domain. Briefly, FLEXINVERT (and its newest and improved version FLEXINVERT+) is an inverse modelling system based on a Bayesian approach to optimise fluxes of atmospheric gases and or pollutants by combining the model sensitivities obtained by an atmospheric transport model and ground-based measurements. One crucial aspect of the system is to obtain the so-called Source Receptor Sensitivities or Relationships (SRS) with an atmospheric transport model. Conceptually, the SRS describe the relationship between the measurements at the receptor point and receptor time and the emission sources (i.e. fluxes) at the emission location and emission time. In FLEXINVERT, FLEXPART, in a receptor-oriented approach, is used to obtain the SRS fields for a set of selected European measurement stations, including  the Irish Mace-Head, Malin Head and Carnsore Point, and for one year of study (2012).


Figure 1: schematic diagram of the  FLEXINVERT system that combines observations,
a priori and background information and model sensitivities to provide CH4 surface flux estimates.


In brief, the IMPLiCIt  objectives are:

  • Implementation, development and optimisation of an inverse modelling system (FLEXINVERT / FLEXINVERT+) for CH4 emissions.
  • Independently verify emissions and sinks of CH4 in Ireland based on data from key boundary sites to produce estimates for 1 or 2 years.
  • Provisional assessment of the relative contribution from individual sources using modelling and observational data analysis techniques.
  • Increase the expertise in Ireland on inverse modelling for emission estimates.
  • Engagement with the community to ensure the best practices are implemented and provide the starting point for future project collaborations on modelling and assessment of greenhouse gas emissions in Europe.


IMPliCIt has taken the first steps to increase the capabilities of Ireland in the field of inverse modelling of greenhouse gases, in particular for methane, have been taken. IMPLiCIT has implemented and tested the FLEXINVERT Bayesian inversion framework using different sets of observational data and prior information combined with atmospheric transport modelling calculations made with the widely used FLEXPART model. FLEXINVERT has been adapted and prepared to be used for the Irish domain (Figure 2) with minor adaptations of the code and with the preparation of the programming environment to arrange and perform the necessary runs and the corresponding outcome evaluation.


Figure 2: innermost domains used in the SRS calculations, nested into a global 1x1 degree domain. Green corresponds to 0.5x0.5, red 0.2x0.2 and blue 0.125x0.125 degree respectively. The blue domain is only to be used in the potential FLEXINVERT+ inversions. The red dots are the observation stations used in the inversions.


Two a priori datasets have been used, that of Thompson et al. 2014 paper (Figure 3) and the high resolution EDGAR v 4.3.1 (Figure 4) .


Figure 3: plot of the emission inventory provided by Thompson, which includes both anthropogenic and natural CH4 sources, over a wide domain (top) and zoomed into the European domain (bottom).


Figure 4: plot of the EDGARv4.3.2 emission inventory, which only includes anthropogenic sources, over a wide domain (top) and zoomed into the European domain (bottom).

Two different inversion modelling systems have been used for comparison, FLEXINVERT and LOGINVERT (Brioude et al. 2011), compared among themselves, with independent data through quantitative evaluation (RMSE, R, NSD) and with the results of the MapEIre project (

With the combinations of inversion modelling schemes, observational data and prior information, a set of sensitivity runs have been performed. Those include the following:

  • Influence of the horizontal resolution in the FLEXPART runs
  • Influence of the prior information
  • Influence of the station data used in the inversion
  • Influence of the observation time window
  • Influence of the correlation length
  • Influence of the errors definition (observational error, global error and error in prior)

The inversion results show a significant increase of the total emissions for Ireland as compared to the MapEIre 2015 high resolution emission inventory, almost doubling the total emissions but without clearly improving the statistical scores when compared with independent site observations. However, the estimates fit within the range of uncertainties of other inversion systems. It is concluded that additional work is needed to improve the results and evaluate the different contributors to the uncertainty and identify procedures to either correct them or include them properly in the inversions.


Example results   “All sites” & “All sites with half of the Irish data”

The inversion results using all the data available as well as keeping half of the Irish sites data for evaluation yield both quite large total emissions (Table 1) when compared to the declared amount in 2012, 502.84Gg (including LULUCF and memo items). The differences are almost double.  Although the geographical patterns are realistic (Figure 5) compared to the MapEIre results, the magnitudes are much large. However, when adding the total emissions of Ireland and the Uk, the results are 3.4 Tg, which is similar to what is found in the literature (Bergamaschi et al. 2015) although quite at the high end. These results lead to the conclusion that additional resources should be invested in investigating further the differences and associated uncertainties. It seems probable that the atmospheric transport modelling uncertainties should be larger and better addressed and as well better understood the planetary boundary layer height definition that may affect the poor representation of the concentration maxima. Also, given the fact that only three sites are in the region of interest, it would be advisable to use a better emission inventory as a priori, for example one as that provided in MapEIre, to have already a better starting point for the inversion system to correct it.

Table 1: total emissions in kt for each of the inversions with two inventories and “all_sites” runs and “all_sites -1/2 Irish sites”

Total 2012 emissions (kt)











Figure 5: total emission maps for the different runs with the two emission inventories as a priori and using all the sites (left panels) or all the sites with half of the Irish data (right panels) compared to the MapEIre emission inventory for 2015 (bottom panel).

The statistical scores for the independent set of data from the Irish sites (Table 2) shows that the increase in the correlation is very small when using the source term estimate by FLEXINVERT. The RMSE is also not much modified but with a larger decrease when the EDGAR_0.1x0.1_ANT emission inventory is used. The more flexible LOGINVERT system has an improved correlation, decreased RMSE and increased NSD. It is clear that in the current set-up and with the data available, FLEXINVERT is largely influenced by the prior information. The statistics for all the remaining sites and data show a similar behaviour with a larger increase in the correlation when using the high resolution prior EDGAR_0.1x0.1_NAT (Table 2).

Table 2: statistical scores and their averages for the Irish sites (using only the data not used in the inversion)  and the EDGAR_0.1x0.1_NAT inventory. The scores are provided for the prior, FLEXINVERT and LOGINVERT.















Malin Head




















Mace Head




















A visual representation of the averaged statistical scores is provided in Figure 6. It is also clear there that, although there is an improvement in the scores, the improvement is not as large as desired.


Figure 6: summary of the averaged statistics for the runs with all the sites and those without half of the Irish data for the dependent sites
(i.e. sites and data used in the actual inversions).

The time series (Figure 7) of the runs with all the sites support the scores obtained. The time series of the modelled concentrations using the FLEXINVERT estimates do show some differences  for some of the dates. However, they tend to follow the prior. Whereas the LOGINVERT system seem to sometimes capture a better representation of the maxima concentration of methane in the observational sites at the expense, however, of increasing even more the total emissions for the Irish domain.


Figure 7: time series at Mace head, Carnsore Point and Malin Head with the observed values (black), those modelled using the prior as emission inventory (gray), the FLEXINVERT estimates (red) and the LOGINVERT ones (blue).



It is clear from the work performed in the IMPLiCIt project that additional actions are required to understand and bridge the gap between the fluxes estimated through a bottom-up approach, and used in the nationally provided total emissions, and those estimated through a top-down approach. It is also to be considered that Ireland has only three measurement sites located on the coast and in some areas that could have quite large influence from the local emission sources and this makes it difficult to be constrained through the inversion. Therefore a road-map of actions and potential approach is designed.


Figure 8: diagram of additional activities to be performed to have a better estimate of the emission inventories in a top-down approach and to bridge the gap with the bottom up assessments.


To improve the surface emission estimates and provide constraints on complex bottom‐up calculations of emission inventories, different strategies are proposed that are summarised below and in diagram in Figure 36:

  1. Constrain/estimate the model error

More work needs to be done to further assess and understand the uncertainties involved. The observation and prior covariance matrices are needed to provide uncertainty estimates in the inversion framework. The model error is particularly difficult to estimate, and using a single meteorological model can bias the posterior estimates. The use of an ensemble of transport models would help to better characterize the transport model error, and the error in the posterior estimates allowing for a more realistic assessment of the estimate of the uncertainty in the posterior. In our case, a transport model is defined as a combination of an Eulerian meteorological model (e.g. ECMWF) and a Lagrangian model (e.g. FLEXPART).

  1. Constrain on surface emissions per emission sectors through measurement campaigns

While ground sites can help monitor monthly variations in emissions from large regions, they can be difficult to use in an inversion framework for several reasons: It is difficult to separate variations due to local emissions from the background, the density of ground sites need to be significant to constrain surface emissions at fine scale (an Ireland, being a small country, requires to work in a high resolution) Therefore, it is difficult to identify the source of a potential bias in a bottom-up inventory.

In the USA, such problems have been addressed, for example, in aircraft campaigns led by NOAA to better constrain sources from urban areas. Theses campaigns can become complex and require that the measurement devices are portable and easy to handle in a mobile platform. A portable light-weight system such as a Picarro , which measures both CO2 and CH4, could be carried on a light plane with an inlet sampling the air outside the cockpit. Preparing a careful flight plan, under well mixed conditions, different source sectors can be isolated and then compared with those in the bottom-up emission inventories.

Profiles with the aircraft may as well help obtain information on the vertical extent of the PBL, which in turn can affect the estimates and can provide input to the atmospheric transport. Transects upwind of the sources will provide the background CH4 concentration. If the flight track is done with a steady wind (same wind direction and wind speed), a mass balance approach could be used to estimate a total flux (see Peischl et al., 2013). Otherwise, those observations can be used in an inversion framework, such as FLEXINVERT or LOGINVERT, to optimize the emission inventory at the mesoscale.

In addition to these two approaches it would be needed to have a much closer cooperation and collaborative discussions with the responsibles of the bottom-up inventories to jointly address discrepancies and approaches.


Relevant Links



  • Guevara, M., et al: Inter-comparison between HERMESv2.0 and TNO-MACC-II emission data using the CALIOPE air quality system (Spain) , Atmos. Environ., vol. 98, pp. 134-145, 2014, ISSN 1352-2310,, 2014.
  • EEA: EMEP/EEA air pollutant emission inventory guidebook 2013, Technical guidance to prepare national emission inventories. EEA Technical report 12/2013. Publication Office of European Union, Luxembourg, ISBN: 978-92-9213-403-7, 2013.
  • Kuenen, J. J. P. , et al.: TNO-MACC_II emission inventory; a multi-year (2003–2009) consistent high-resolution European emission inventory for air quality modelling, Atmos. Chem. Phys., vol. 14, pp. 10963-10976, 2014, doi:10.5194/acp-14-10963-2014, 2014
  • Hanna, S. R., et al.: Uncertainties in predicted ozone concentrations due to input uncertainties for the UAM-V photochemical grid model applied to the July 1995 OTAG domain, Atmos. Environ., vol. 35, issue 5, pp. 891-903, 2001, ISSN 1352-2310,, 2001.
  • Menut L. and B. Bessagnet: Atmospheric composition forecasting in Europe, Ann. Geophys., vol. 28, pp. 61-74, 2010, doi:10.5194/angeo-28-61-2010, 2010.
  • Peischl, J. et al., 2013, Quantifying sources of methane using light alkanes in the Los Angeles basin, California
  • Vautard, R., et al: Paris emission inventory diagnostics from ES- QUIF airborne measurements and a chemistry transport model, J. Geophys. Res., vol, 108(D17), pp. 8564–8585, 2003.
  • Viana, V., et al.: Spatial and temporal variability of PM levels and composition in a complex summer atmospheric scenario in Barcelona (NE Spain), Atmos. Environ., vol. 39, issue 29, pp. 5343-5361, 2005, ISSN 1352-2310,, 2005.
  • Stohl, A., et al.: Black carbon in the Arctic: the underestimated role of gas flaring and residential combustion emissions, Atmos. Chem. Phys., 13, 8833-8855, 2013, doi:10.5194/acp-13-8833-2013.
  • Bergamaschi, P., et al.: Inverse modeling of the global CO cycle: 1. Inversion of CO mixing ratios, J. Geophys. Res., vol. 105(D2), pp. 1909–1927, 2000, doi:10.1029/1999JD900818, 2000.
  • Quélo, D., et al.: Inverse modeling of NOx emissions at regional scale over northern France: Preliminary investigation of the second-order sensitivity, J. Geophys. Res., vol. 110, D24310, 2005, doi:10.1029/2005JD006151, 2005.
  • Brioude, J., et al. :Top‐down estimate of anthropogenic emission inventories and their interannual variability in Houston using a mesoscale inverse modeling technique, J. Geophys. Res., 116, D20305, doi:10.1029/2011JD016215, 2011.
  • Saide, P., et al.: Constraining surface emissions of air pollutants using inverse modelling: method intercomparison and a new two-step two-scale regularization approach. Tellus B, North America, vol. 63, 2011.
  • Thompson, R.L. and A. Stohl, FLEXINVERT: An atmospheric Bayesian inversion framework for determining surface fluxes of trace species using an optimized grid, Geosci. Model Devel., 7, 2223- 2242, doi:10.5194/gmd-7-2223-2014, 2014.
  • Thompson, R.L., et al., Methane emissions in East Asia for 2000 - 2011 estimated using an atmospheric Bayesian inversion, J. Geophys. Res., 120, 9, 4352-4369, 2015
  • Stohl, A., et al.: Technical note: The Lagrangian particle dispersion model FLEXPART version 6.2, Atmos. Chem. Phys., 5, 2461-2474, doi:10.5194/acp-5-2461-2005, 2005.

Acknowledgements: this project is funded by the Irish Environmental Protection Agency.


Print Email