Retrieval and validation of METOP/IASI methane

A new global IASI methane product developed at the Royal Belgian Institute for Space Aeronomy (BIRA-IASB) is presented. The retrievals are performed with the ASIMUT-ALVL software based on the Optimal Estimation Method (OEM). This paper gives an overview of the forward model and retrieval concept. The usefullness of reconstructed Principal Component Compressed (PCC) radiances is highlighted. The retrieval uncertainty of the CH4 profiles is less than 4% below 100 hPa (∼16 km). The information content study carried out in this paper shows that most IASI pixels contain between 0.9 and 5 1.6 independent pieces of information about the vertical distribution of CH4, with a good sensitivity in the mid to upper troposphere. An extended validation with ground-based CH4 observations at 10 locations was carried out. IASI CH4 partial columns are found to correlate well with the ground-based data for 7 out of the 10 Fourier Transform Infrared (FTIR) stations with correlation coefficients between 0.71 and 0.96. Mean differences between IASI and FTIR CH4 range between -1.93 and 4.40% and are within the systematic uncertainty. For 7 out of the 10 stations absolute differences are less than 1%. The standard 10 deviation of the difference lies between 1.40 and 3.99% for all the stations.

the use of the Principal Component Compressed IASI spectra is addressed. The information content of the IASI CH 4 product is presented in Sect. 4. In addition, global distributions are shown and retrieval processing details are briefly discussed. In Sect. 5 the IASI CH 4 product is compared to ground-based measurements, providing an quality assessment of the retrieved BIRA-IASB CH 4 columns. The final section summarizes the main results of this work.

5
The Infrared Atmospheric Sounding Interferometer (IASI) onboard MetOp, is a thermal cross-nadir scanning infrared sounder. IASI is a Fourier Transform Infrared (FTIR) spectrometer which measures the TIR radiation emitted by the Earth and the 10 atmosphere. With a wide swath width of 2 x 1100 km it provides near-global coverage twice a day, with a local overpass time at ∼9:30 AM and PM. It has an instantaneous field of view (FOV) at nadir with a spatial resolution of 50 km x 50 km, composed of 2 x 2 circular pixels, each corresponding to a 12 km diameter footprint on the ground at nadir ). IASI has four spectral bands in the spectral range from 645 to 2760 cm −1 (3.62 to 15.5 µm), with an apodized spectral resolution of 0.5 cm −1 and spectral sampling of 0.25 cm −1 .

15
Designed to provide highly accurate temperature and humidity profiles for numerical weather predictions, the IASI mission allows simultaneous global observations of the air composition with an excellent spatial resolution. From the atmospheric spectra recorded by the instrument, concentrations of several trace gases can be monitored, enhanced levels of pollution can be detected, and particle types can be determined to some extent. On the longer term the continuity of the program is ensured with the IASI-NG mission that will extend the IASI observation for 15-20 more years (Clerbaux et al., 2016). 20 3 The IASI CH 4 retrieval method The IASI CH 4 profiles are retrieved with the ASIMUT-ALVL software developed at BIRA-IASB (Vandaele et al., 2006).
ASIMUT-ALVL is a modular software for radiative transfer (RT) calculations and inversions in planetary atmospheres. The code has been developed with the objective to be as general as possible, accepting different instrument types and different geometries. ASIMUT-ALVL has been coupled to the SPHER/TMATRIX (Mishchenko and Travis , 1998) and LIDORT (Spurr 25 , 2006) codes to include the complete treatment of the scattering effects into the RT calculations. It has a specific interface dealing with the IASI instrument characteristics and IASI input information and is also used for the IASI aerosol dust retrievals (Vandenbussche et al., 2013). The RT simulations are performed with the ASIMUT-ALVL RT code for the IASI CH 4 data product while the LIDORT RT code is used for the IASI aerosol dust retrievals in order to include all scattering effects due to aerosols. Both IASI retrieval products use the same retrieval module, based on the formalism of the Optimal Estimation Initially developed for Earth observation missions, ASIMUT-ALVL has also been adapted for planetary atmospheres, in particular those of Venus (Vandaele et al., 2008) and Mars (Drummond et al., 2011) and is now the reference code for the NOMAD instrument on-board ExoMars TGO (Robert et al., 2016).

Forward model
The ASIMUT-ALVL RT module simulates atmospheric transmittances and radiances for cases under local thermodynamical 5 equilibrium and where scattering can be neglected. A detailed description of the radiative transfer model is given in Vandaele et al. (2006). The spectral range considered for the CH 4 retrieval is the 1210-1290 cm −1 range covering part of the ν4 spectral band. EUMETSAT IASI L2 temperature and water vapour profiles are used as input for the radiative transfer calculations. The spectroscopic parameters for CH 4 , N 2 O and other species are taken from the HITRAN 2012 database (Rothman et al., 2013).
The IASI Instrument Line Shape (ILS) is characterized by a Gaussian function with a 0.5 cm −1 FWHM. Frequency dependent 10 emissivity maps are provided by Zhou et al. (2011). Fig. 1 shows an example of measured and simulated radiances in the 1210-1290 cm −1 spectral range and its residual (difference between measured and simulated radiances). We have a negligible bias and a 1-σ standard deviation comparable to the radiometric noise of 2x10 −8 W/(cm 2 sr cm −1 ) (see Sect. 3.2). Certain spectral ranges in the considered spectral band are not well simulated by the radiative transfer model, leading to outliers in the residuals with absolute differences larger than 5x10 −8 W/(cm 2 sr cm −1 ), for example at 1246 cm −1 and 1252 cm −1 . These 15 spectral ranges are masked in the retrieval set-up, i.e. the radiometric noise is set to zero at these spectral points, so that no information is lost.
Only IASI L1C spectra with a cloud fraction < 10% based on the EUMETSAT IASI L2 fractional cloud cover product are processed.

Retrieval and error characterization 20
The ASIMUT retrieval module is based on the OEM (Rodgers, 2000) where the Jacobians are calculated analytically. The characteristics of the IASI retrieval are summarized in Table 6. The state vector includes the skin temperature (T skin ), 23-level CH 4 , N 2 O and H 2 O profiles and a CO 2 total column. The T skin a priori is taken from the EUMETSAT IASI L2 T skin product.
The a priori profiles x a and covariance matrices S a for CH 4 and N 2 O are based on a climatology from the WACCM model. The H 2 O a priori uncertainty covariance matrix is characterised by an uncertainty covariance matrix with a 10% standard deviation 25 on the diagonal and an exponential decaying correlation width of 6 km. The EUMETSAT IASI L2 water vapour profile is used as the H 2 O a priori profile x a . The interfering species HNO 3 and O 3 are included in the RT calculations, its a priori values are provided by the WACCM model.
A diagonal measurement uncertainty covariance S e is taken, with the radiometric noise set to 2x10 −8 W/(cm 2 sr cm −1 ). This value is conservative, about a factor 5 higher than the estimated radiometric noise in this spectral region of 4x10 −9 W/(cm 2 sr 30 cm −1 ) ). It includes not only the measurement uncertainty, but also the uncertainties in the temperature and water vapor profile, the spectroscopic parameters and surface emissivity (De Wachter et al., 2012). The horizontal bars represent the retrieval uncertainty. The averaging kernel (AK) is given in the middle figure. It shows that the sensitivity of the IASI CH 4 product lies in the 800-100 hPa (∼2-16 km) range.
The right plot of Fig. 2 diplays the vertical profiles of the retrieval uncertainties together with the CH 4 a priori variability (black line). The square root of the diagonal elements of the uncertainty is plotted. The CH 4 a priori variability is calculated from the square root of the diagonal of the a priori uncertainty covariance matrix. The retrieval is quite constrained with of an a priori 5 variability of a few percent at the surface going up to 7-8% at 20 km. Following Rodgers (2000) the error sources contributing to the total retrieval uncertainty are 1) the smoothing error, which accounts for the vertical resolution of the retrieved CH 4 , 2) the error due to uncertainties in forward model parameters such as spectroscopy, the temperature profile, surface emissivity and 3) the IASI measurement uncertainty. For IASI the forward model uncertainties are included in the measurement uncertainty.
As we can see from Fig. 2, the dominant source of uncertainty is the smoothing uncertainty. The total retrieval uncertainty 10 declines from 3% at the su rface to ∼2% between 800 and 200 hPa, the altitude range of maximum sensitivity. Above 200 hPa the total retrieval uncertainty increases rapidly up to ∼4% at 100 hPa and ∼6% at 60 hPa.

Retrievals with PCC L1C data
The CH 4 profiles are retrieved from IASI radiances recomposed from the EUMETSAT Principal Component Compressed (PCC) L1C dataset (Hultberg , 2009). The use of PCC data allows both noise filtering and a large reduction in data volume 15 compared to the use of raw radiances. Our main motivation is the large reduction in data storage. One year of the original IASI L1C (BUFR format) data amounts to 10 Tb which is reduced to 1 Tb/year for the PCC data. Fig. 3 shows the raw and PCC radiances for a random pixel in the CH 4 ν4 spectral band. Differences between raw radiances and PCC radiances lie in the IASI radiometric noise level (4x10 −9 W/(m 2 sr m −1 )) as given by the IASI radiometric noise figure from Clerbaux et al. (2009). This is a factor 5 lower than the conservative radiometric noise level of 2x10 −8 W/(m 2 sr m −1 ) used in the CH 4 retrieval (see 20 Sect. 3.1). Fig. 4 compares the CH 4 concentrations retrieved with the PCC L1C data with those retrieved with the raw radiances for March 2011 and September 2013 for daytime retrievals between 60 • S and 70 • N. We find an excellent correlation (R=1) between the retrieved concentrations and negligibles biases of 0.0026% and 0.025% with a 1-σ standard deviation of 0.12%.
With these results we are confident to use the PCC-reconstructed radiances.

Information content
For correct interpretation of the data one needs to consider the vertical sensitivity of the retrieved CH 4 profile. This information is contained in the averaging kernel (AK), which is provided with each retrieved CH 4 profile. The peak of each AK gives the altitude of maximum sensitivity. Its full width at half maximum can be interpreted as the vertical resolution of the retrieval.
For the 3 geographical locations, the sensitivity is reduced in the boundary layer, which is typical for thermal infrared sounders.
In each figure the Degree of Freedom for Signal (DOFS) is given, which is an estimate of the number of independent pieces of information contained in the measurement. It is the trace of the AK. One independent piece of information (1.01 < DOFS the DOFS) is dependent on the thermal contrast, which exhibits significant geographical, seasonal, and diurnal variability. The 10 retrieval sensitivity is favourable, and hence the DOFS is high, when thermal contrast is high. In general the thermal contrast or the DOFS is higher during the day, over land, and over dry, sparsely vegetated regions ). This pattern is visible in Fig. 6

Global distributions
Monthly mean global daytime distributions for the year 2013 are presented in Fig. 8. IASI CH 4 partial columns between 4 and 17 km between 60 • S and 70 • N are shown for the IASI morning overpass. CH 4 concentrations are averaged over the four, 2 x 2 circular IASI pixels, which are measured simultaneously (see Sect. 2) and binned on a 1 • x 1 • grid. Areas with missing data correspond to areas which were identified as cloudy by the EUMETSAT IASI L2 fractional cloud cover product, or correspond 25 to areas where not all of the 4 simulateneously measured pixels converged in the retrieval.
We see a latitudinal gradient with higher concentrations in the Northern Hemisphere (NH) than in the Southern Hemisphere (SH), which is consistent with the fact that most of the methane sources are located in the Northern Hemisphere. In the NH, higher CH 4 concentrations are found during boreal summer than during boreal winter. This summer increase of mid to upper tropospheric CH 4 has also been observed by AIRS (Xiong et al., 2010).  Methane observed in the boundary layer by surface stations from the NOAA network displays a reversed seasonal cycle in the NH (Dlugokencky et al., 2009). These results demonstrate the added value of thermal infrared CH 4 measurements which have a sensitivity at higher altitudes.

Retrieval output and processing
The BIRA-IASB IASI CH 4 product is delivered in HDF5 format. Daily daytime and nighttime observations are provided in 5 separate files. The HDF files contain CH 4 profiles, the retrieval uncertainty, the CH 4 a priori profiles and averaging kernels.

Validation
Ground-based data was collected from 10 FTIR stations from the Network for the Detection of Atmospheric Composition Change (NDACC). The stations chosen are operated on a quasi-continuous basis and deliver CH 4 vertical profiles. Certain 15 stations provide limited observations since they only recently entered the NDACC network or since they only make campaign measurements. We therefore excluded stations with less than 200 collocations due to insufficient collocation points for a statistically significant comparison. NDACC FTIR CH 4 profiles have good sensitivity in the troposphere and stratosphere with 2 to 3 independent pieces of information. Note, the NDACC CH 4 retrieval is not fully harmonized yet for all the NDACC stations. This work is ongoing as part of the Horizon 2020 Gap Analysis for Integrated Atmospheric ECV CLImate Monitoring 20 (GAIA-CLIM) project (http://www.gaia-clim.eu/).
We performed a detailed comparison between IASI and NDACC CH 4 partial columns between 4 and 17 km at these 10 NDACC stations for the period 2011 to 2014. Since the two retrievals have been computed with a different a priori, the NDACC retrieved profiles are adjusted for the comparison. Following Rodgers and Connor (2003) (equation 10), the term (A NDACC − I) · (x a,NDACC − x a,IASI ) is added to each NDACC retrieval to adjust for the different a priori profile used in the IASI 25 retrieval. Here A NDACC is the NDACC averaging kernel, I the unity matrix, x a,NDACC the NDACC CH 4 a priori profile and x a,IASI the IASI CH 4 a priori profile.
In addition, to account for the different resolution between the IASI and the higher resolved NDACC FTIR profiles, a smoothing is applied to the (a priori adjusted) NDACC profile x NDACC by the IASI averaging kernel:  It is important to compare these results with the uncertainty budget of the IASI and the NDACC CH 4 partial columns. As given by Rodgers and Connor (2003) (equation 30), S ∆ , the covariance of the difference IASI-NDACC, can be calculated as: The first term is the smoothing uncertainty of the comparison ensemble (the smoothed and a priori-corrected NDACC and IASI product) with S a,IASI the IASI a priori uncertainty covariance matrix. S IASI is the IASI retrieval uncertainty covariance exluding the smoothing uncertainty and S NDACC is the NDACC retrieval uncertainty covariance without the smoothing uncertainty. We 20 compare the systematic and random uncertainty on the difference directly to the mean difference and standard deviation of the difference between IASI and NDACC. NDACC provides systematic and random uncertainty covariances for the different stations, with exception of Jungfraujoch. For IASI we set the random uncertainty S rand a,IASI equal to the IASI apriori uncertainty covariance matrix used in the IASI retrieval. We calculated a systematic component S syst a,IASI with a 2% standard deviation of the a priori profile values. Separating the systematic and random component of S IASI is less straightforward. Here we only consider 25 the IASI measurement uncertainty as the random uncertainty and we do not consider the systematic component. Table 2 lists the results of this analysis with the relative mean differences, the standard deviation of the differences and the mean values of the systematic and random uncertainties. We have a good agreement where the IASI-NDACC mean differences lie within the systematic uncertainty. Also for Thule the mean difference of 4.40% is within the systematic uncertainty of 5.14%. The standard deviation of the difference is close to the random uncertainty, but for certain stations slighlty larger. This 30 discrepancy could be due to an additional error associated with the grid conversions or a geolocation error which are not taken into account (Calisesi et al., 2005;Vigouroux , 2009). Another reason could be the current underestimation of the random uncertainty of the NDACC CH 4 retrievals. They were found to be less than 1% for 6 out of the 10 NDACC stations. The ongoing work in the GAIA-CLIM project will harmonize the error characterization for all NDACC stations in the coming These results demonstrate the ability of the IASI data to well capture the temporal variation of CH 4 .

Discussion, conclusion and outlook
Although CH 4 is a more effective greenhouse gas than CO 2 , it has a much shorter atmospheric lifetime than CO 2 that can remain in the atmosphere for hundreds or thousands of years. Therefore the mitigation of CH 4 emissions provides an opportunity for alleviating climate change in the short-term future (Kirschke et al., 2013). Global monitoring of CH 4 is essential to study the evolution of atmospheric CH 4 and to help increase our knowledge on how the different sources and sinks influence 15 its atmospheric abundance.
In this paper, we presented a new IASI CH 4 retrieval product developed at BIRA-IASB. Global distributions of CH 4 were derived from IASI radiances with the ASIMUT-ALVL software based on the OEM. A detailed description of the forward model, the retrieval strategy and the use of PCC L1C data was given. CH 4 concentrations retrieved from raw radiances and PCC-reconstructed radiances showed an excellent correlation and negligible mean differences of < 0.026% (< 0.46 ppb). 20 We presented the latitudinal distribution of the DOFS for different seasons. We showed that, between 60 • S and 70 • N, the DOFS values range between 1 (0.9) and 1.8 (1.6) for daytime (nighttime) retrievals for NH summer. In NH winter values can become less than 1 for latitudes > 40 • N. In tropical scenes DOFS values are typically around 1.4, with a good sensitivity in the mid to upper troposphere.
A quality assessment of the retrieved IASI CH 4 product was given by a detailed comparison with ground-based FTIR ob-25 servations recorded at 10 NDACC stations. The BIRA-IASB product was compared to smoothed NDACC FTIR CH 4 partial columns between 4 and 17 km for the years 2011 to 2014. We found a very good agreement between both products with differences within the systematic uncertainty. Mean difference values range between -1.93 and 4.40% for the 10 stations. Absolute differences are less than 1% for 7 stations out of 10. The standard deviation of the difference lies in the range 1.91 to 3.99% for all the stations. These values are close to the random uncertainty of IASI and NDACC, but for certain stations slightly 30 larger. Possible reasons for this discrepancy could be an underestimation of the NDACC CH 4 random uncertainty or additional error sources not considered in the calculation of the random uncertainty, such as a regridding or geolocation error. Very good correlations are found for 7 out of the 10 NDACC stations with correlation coefficients between 0.71 and 0.96. Particularly for the 3 high-latitude stations we find high correlations, as well as for the 2 high-quality mid-latitude stations Jungfraujoch and Zugspitze. With these results we are confident that the IASI data is capturing the mid to upper CH 4 variability well.
Future work will focus on extending the validation with additional datasets. Validation measurements for atmospheric vertical profiles for CH 4 are limited and very diverse. An innovative atmospheric sampling system called AirCore (Karion et al., 2010;Membrive et al. , 2016) has been demonstrated to be a reliable concept to make vertical profile measurements of CO 2 , CH 4 5 and CO from the surface up to ∼30 km. Although campaign-based, these high precision measurements provide a promising and novel validation tool. One of the next steps is to compare the IASI CH 4 with AirCore CH 4 profiles. Further, a global scale comparison with the neural network IASI-CH 4 product (Crevoisier et al., 2009) or with one of the new OEM IASI-CH 4 products recently published and under revision (Siddans et al., 2016;García et al., 2017) would be of particular interest.
IASI provides day-and nighttime measurements over land and sea and has a high spatial coverage. Its follow-up missions guar-10 antee a long continuity of observations and its successor, the IASI-NG next-generation instrument, will ensure a continuity of data until after 2040. IASI-NG's spectral resolution and signal-to-noise ratio will be improved by a factor of two. It will fly on the three second-generation MetOp-SG-A series, scheduled to launch in 2021, 2028 and 2035. IASI provides therefore a great opportunity for continuous monitoring of the atmospheric composition on a fine spatio-temporal scale.
CH 4 is a challenging component to retrieve in the thermal infrared. Water vapour interferes strongly in the CH 4 thermal band.

15
The same holds for the greenhouse gas N 2 O which is strongly correlated to CH 4 . In addition, spectroscopic uncertainties and mixing of spectral lines are still important issues in the CH 4 ν4 band. Continuous efforts will be made on improving the IASI CH 4 retrievals as to these issues and enhancing their precision. Furthermore, future work will focus on comparing the IASI concentrations with tagged simulations of CH 4 to see whether the model output is supported by the IASI data. With this research we want to provide a better understanding of the CH 4 budget, which can help target the pertinent sources for reducing 20 CH 4 emissions and the associated climate impact of this greenhouse gas.
The BIRA-IASB IASI CH 4 dataset is available through the European Space Agency (ESA) CCI-GHG project and can be downloaded from http://iasi.aeronomie.be/. Data is available for the years 2011-2014 between 60 • S and 70 • N and CH 4 profiles, a priori profiles, retrieval uncertainties and averaging kernels are provided.
Acknowledgements. The IASI mission is a joint mission of Eumetsat and the Centre National d'Etudes Spatiales (CNES, France). The IASI 25 L1C data are distributed in near real time by Eumetsat through the Eumetcast system distribution. This work was conducted as part of the IASI.flow (Infrared Atmospheric Sounding with IASI and Follow-on missions) project, funded by the Belgian Science Policy Office and the European Space Agency (ESA-Prodex program). Additional support was provided by the ESA GHG-CCI project through the Optional Workpackage 706 TIRS (CO2 and CH4 from Thermal Infrared Sounders: IASI and ACE-FTS). The ground-based data used in this publication were obtained as part of the Network for the Detection of Atmospheric Composition Change (NDACC) and are publicly available (see Atmos. Meas. Tech. Discuss., https://doi.org/10.5194/amt-2017-165 Manuscript under review for journal Atmos. Meas. Tech.