Retrieval of methane source strengths in Europe using a simple modeling approach to assess the potential of spaceborne lidar observations

. We investigate the sensitivity of future spaceborne lidar measurements to changes in surface methane emissions. We use surface methane observations from nine European ground stations and a Lagrangian transport model to infer surface methane emissions for 2010. Our inversion shows the strongest emissions from the Netherlands, the coal mines in Upper Silesia, Poland, and wetlands in southern Finland. The simulated methane surface concentrations capture at least half of the daily variability in the observations, suggesting that the transport model is correctly simulating the regional transport pathways over Europe. With this tool we can test whether proposed methane lidar instruments will be sensitive to changes in surface emissions. We show that future lidar instruments should be able to detect a 50 % reduction in methane emissions from the Netherlands and Germany, at least during summer.


Introduction
Although methane (CH 4 ) is the second most important anthropogenic greenhouse gas, it is arguably just as important as carbon dioxide (CO 2 ) from a policy perspective.Several studies conclude that in order to reduce net anthropogenic radiative forcing, it costs less to cut CH 4 emissions compared with CO 2 emissions (Shindell et al., 2012;Delhotel et al., 2006).To monitor future CH 4 emissions, policy makers and government officials will desire estimates of surface fluxes at a fine temporal and spatial resolution.Scientists also need this information to understand global and regional CH 4 budgets and the physical processes that control them.
Because observed emission rates of CH 4 are highly variable over small temporal and spatial scales, scientists have often resorted to a top-down approach to determine regional emissions.Total column CH 4 observations from space have been used in an inversion algorithm to estimate surface emission fluxes (Bergamaschi et al., 2009;Meirink et al., 2008).Currently, the spaceborne sources for near-surface CH 4 information are the Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY) instrument on board the late ENVISAT satellite platform (Bergamaschi et al., 2007;Frankenberg et al., 2011), and more recently the Greenhouse gases Observing SATellite (GOSAT) operated by the Japan Aerospace Exploration Agency (Schepers et al., 2012).These instruments measure changes in CH 4 spectral absorption from reflected near-infrared solar radiation, so they are susceptible to contamination from undetected clouds and aerosols.Too often global maps of CH 4 retrievals from today's passive satellite instruments have data voids over persistent cloudy regions, even if the clouds are optically thin.They also cannot make measurements in darkness or low sunlight conditions.
There are several planned space-and aircraft-based instruments that will use laser technology to measure the total CH 4 column (Riris et al., 2013;Ehret et al., 2008) and total CO 2 column (Abshire et al., 2010).This approach should remedy some of the current issues concerning highlatitude coverage and scattering from clouds and aerosols.Indeed, a recent aircraft demonstration campaign to test the feasibility of lasers to measure CO 2 has successfully retrieved CO 2 concentrations in thin cloud conditions (A. Ramanathan, personal communication, 2013;Abshire et al., 2013).In the frame of a German-French climate monitoring initiative, DLR (Deutsches Zentrum für Luft-und Raumfahrt) and CNES (Centre National d'Etudes Spatiales) proposed a MEthane Remote LIdar MissioN (MERLIN) on a small polar orbiting satellite.The DLR Institute of Atmospheric Physics is also developing an airborne lidar system for demonstration and satellite validation purposes.Performance simulations have shown the basic ability of such active remote sensing systems in improving the accuracy of global methane observations (Kiemle et al., 2011;Stephan et al., 2011).
The question remains whether these new instruments will be able to detect changes in surface CH 4 emissions that occur at the state, nation or continent spatial scale.These changes may arise from policies that hopefully reduce anthropogenic emissions or from natural processes, which include climate feedback effects on permafrost soils, ocean hydrate sediments, and wetlands, in regions that are often difficult to access.This study's approach is to first build a retrieval algorithm that estimates CH 4 emissions using the FLEXPART Lagrangian transport model constrained by hourly CH 4 surface observations.Then we evaluate our emission estimates with existing emission inventories to ensure that our emissions are realistic and the method is sound.The second part of this study applies this tool to remote sensing.We build an algorithm that inverts surface emissions from total column measurements.With this tool we test whether the proposed satellite instruments will be able to detect perturbations in column CH 4 based on the precision of future spaceborne instruments.

Forward models
The two variations of the forward model used in this study start with gridded fields of CH 4 surface emissions and output either (1) concentrations at the ground, simulating observations from monitoring stations or (2) total column amounts, simulating observations from satellite.The driver for both is FLEXPART -a 3-dimensional particle dispersion model developed at the Norwegian Institute for Air Research (Stohl et al., 2005).Our approach is to divide northern Europe into 262 tiles (Fig. 1) and assume a uniform emission flux from these tile.The FLEXPART model is run independently for each tile using the NOAA Global Forecast System (GFS) meteorology fields at 3 hr time and 0.5 • resolution.The GFS uses 64 vertical sigma-pressure hybrid layers.The FLEXPART model actually transports the CH 4 as particles that have a lifetime of 20 days.We understand that CH 4 's actual lifetime is on the order of twelve years, but uncertainties of the FLEXPART model grow so that after 20 days the results are unreliable.We are implicitly assuming that after 20 days the CH 4 particles become part of a background concentration term.
For each tile we release the equivalent of 1 kg of CH 4 at 150 m above ground level, assume a climatological OH field, and an OH reaction rate that varies with temperature (3.5 × 10 −15 cm 3 s −1 at 25 • C).We set the FLEXPART model vertical domain to simulate concentrations from the earth surface to 400 hPa.We do not attempt to simulate stratospheric intrusions.
One forward model implementation (Fig. 2a) calculates perturbations in the surface CH 4 concentration at any geographical location by adding up the contribution (S i • C i ) from each of the 262 tiles.A background concentration (B) is also added to the perturbations.S i is the source strength for the i th tile and C i is the surface concentration simulated by FLEXPART using the 1 kg per day source.The background value is a retrieved quantity from the retrieval algorithm.
A second forward model implementation (Fig. 2b) calculates perturbations in the total column CH 4 using a similar approach.For all of our FLEXPART model runs, the simulated CH 4 is able to move vertically in the atmosphere according to diabatic and convective processes captured by the GFS meteorology.While the model does not calculate an absolute column amount, it provides perturbations from an unknown total column value, which is adequate for our purposes.We output the CH 4 at 20 vertical levels in the atmosphere (surface to ∼400 mb).Our configuration of the FLEXPART model does not loft many particles (molecules of methane) above the 400 hPa level so its contribution to the column perturbation is insignificant.The partial column methane above 400 hPa is spatially homogeneous and is included in the background term.
3 Source inversions from ground-based observations

Retrieval algorithm
The source retrieval algorithm estimates CH 4 emissions from data sampled hourly at nine European ground stations in 2010.As shown below, the surface CH 4 concentrations simulated by the transport model reproduce much of the short timescale (hourly) perturbations in the CH 4 concentrations, suggesting that these fast fluctuations, rather than the slowly varying background CH 4 value, provide information about the source strength over the domain.
Once the individual trajectory calculations are run for each tile, we can run the retrieval algorithm.As shown in Fig. 3, we retrieve a source strength (S i ) for each tile and a background value (B) for each station.The retrieval algorithm adjusts the source strengths of each individual tile, until the simulated (analyzed) observations best match those observed.The background value at each station is also adjusted.A standard linear inverse method (chapter 3 of Rodgers, 2000) is used and iterated until the source strengths and background values have converged.Note that the retrieved source strengths are constant over a 45 d period and the retrieved background values vary linearly with time over each 45 d period.
The observations for this retrieval algorithm are sampled at nine European stations that continually measure surface CH 4 , preferably every hour.These are listed in Table 1 and further described in the Appendix.There are additional European stations that only sample weekly or monthly (labeled by "event"), but we could see no significant change in the retrieved emissions when these data sets were included.The high temporal resolution hourly observations capture information on CH 4 filaments passing over a ground station, which are then deconvolved by our retrieval algorithm to yield the surface emission strengths.This interesting finding is due to the complexity of transport in association with a strong spatial heterogeneity of emissions.
Figure 4 shows the observed hourly CH 4 concentrations at the nine European ground stations used for this study (black trace).The FLEXPART simulated (analyzed posterior) concentrations that best match the observations are also shown (red trace).The FLEXPART model is often able to reproduce the weekly variability and sometimes captures hourly spikes in the observations.The model performs best at the Mace Head Ireland site and the Pallas-Sammaltunturi site in northern Finland.These are remote sites situated far away from any pollution sources; moreover, Pallas is usually above the convective boundary layer.Therefore, the variability at these sites is largely influenced by regional transport from European sources.Differences between FLEXPART simulated concentrations and the observations can be attributed to parameterization schemes employed by FLEXPART, errors in the GFS winds, the boundary conditions, the varying background concentration and/or errors in the observations themselves.
The model is unable to capture the minimum values at mountain stations: the Jungfraujoch site (elevation of 3580 m), the Plateau Rosa site (3480 m) and Monte Cimone Also note that while the model is able to reproduce the timing of the spikes at the Kollumerwaald site, it only simulates half the amplitude.Issues with the GFS fields and FLEX-PART's simulation of boundary layer height and mixing may explain why the model captures the timing of spikes but not the amplitude.If a local source is present, a shallower nighttime boundary layer capped by an inversion will foster higher CH 4 concentrations.However, a time series plot of hourly CH 4 concentrations at Kollumerwaald sorted by nighttime and daytime measurements (not shown) shows that spikes are not limited to nighttime conditions.These spikes require further study.
The squared correlation coefficients between the observed and simulated concentrations range from .48 to .77.This suggests that at least half of the variability in the observed surface concentrations can be explained by the regional transport simulated by the FLEXPART model.The remaining unexplained variability in the surface concentrations is from short temporal scale variability in the emissions strengths (we assume a constant emission over a 45 d time period), or weaknesses in the trajectory model, or weaknesses in the GFS meteorological fields.

Retrieved source strengths
Figure 5 shows our retrieved source strengths for each tile.Note that our forward model assumes that these emissions are unchanged over a 45 d period.The strength values and uncertainties for tiles with significant emission are shown in black and white, respectively.The uncertainties are derived from the averaging kernels (see Rodgers 2000, chapter 4) which are a diagnostic quantity from the retrieval algorithm.
Ideally, each of the nine ground stations (Table 1) would have back trajectories transporting CH 4 from each one of the 262 source tiles during a 45 d period.Instead, there are times when no trajectories from a source tile pass over one of the nine observing ground stations and the strength uncertainty (as determined by the averaging kernel) will be high.This was too often the case when we attempted to resolve the strengths at temporal resolutions less than 45 days; a significant number of tiles had strength uncertainties that were larger than the actual values.
Several other studies have inverted methane surface emissions using observations from ground-based monitoring stations.One study reports UK emissions using observations from the Mace Head site (Manning et al., 2011).Bergamaschi et al (2010) use data from many monitoring sites along with a nested Eulerian transport model to estimate emissions over NW Europe.
We compare our retrieved emission strengths with reanalysis fluxes from the European Monitoring Atmospheric Composition and Climate (MACC) collaborative project for 2010 (Fig. 6a).These surface fluxes are inverted from total column CH 4 amounts from SCIAMACHY (Bergamaschi et al., 2009) and should include both natural and anthropogenic sources.Our retrieved surface fluxes are qualitatively consistent with the much smoother MACC reanalysis over the UK and central Europe.(Their inversion grid has a much coarser spatial resolution than ours.)However, the MACC reanalysis does not show the strong Fenno-Scandinavia emissions that we retrieve in July.This may be due to the lack of SCIA-MACHY data at high latitudes.
We can also compare our surface fluxes with the Emission Database for Global Atmospheric Research (EDGAR, 2011), understanding that this data set does not include natural emissions.Figure 6b shows the total anthropogenic emissions based on government and commercial statistics for 2008.The fine 0.1 • by 0.1 • spatial scale of this data set is able to resolve the strong emissions from the major metropolitan, industrial, mining and agricultural areas.To facilitate comparison we show our average retrieved emissions for May-December www.atmos-chem-phys.net/14/1/2014/Atmos.Chem.Phys., 14, 1-13, 2014 Fig. 7. Our approach to test the sensitivity of a proposed laser CH 4 instrument to changes in surface emissions.First we produce values of simulated column CH 4 observed by a laser instrument (green boxes) from gridded CH 4 surface emissions (black box).We include the effect of a proposed satellite platform orbit, field-of-view instrument sampling, rejection of data for cloudy conditions, and instrument noise.Finally, we retrieve surface emissions (red and blue boxes) from the simulated column CH 4 observations.Gridded arrays of the starting (true) and retrieved surface emissions are shown in Fig. 9. See text for further explanation.
2010 in Fig. 6c.The fine resolution EDGAR emissions are integrated over our 262 surface tiles in Fig. 6d.Almost all of Poland's coal mining activities are concentrated in Upper Silesia (shown by the dark pentagon symbol in Fig. 6).It is one of the largest in Europe and produces almost all of Poland's coal.During the extraction process significant CH 4 is released from the coal and surrounding rock.This CH 4 must be quickly removed from underground mines through ventilation systems.Although some of the mines recover the CH 4 , a significant amount is still emitted directly into the atmosphere.The EDGAR database reports that "fugitive emissions from solid fuels" (i.e., coal production) constitute half of Poland's CH 4 emissions.So the strong emission in southern Poland reported by EDGAR (113 mg day −1 m −2 , Fig. 6d) is largely from venting coal mines.Our retrieved surface fluxes also show a strong source near Upper Silesia of 57 mg day −1 m −2 shown in Fig. 6c.
The EDGAR database also shows the Netherlands as a strong source of CH 4 (Fig. 6b, d).This is not associated with the tulip industry; instead, the EDGAR database cites "enteric fermentation and manure management" as the largest contributors to the Netherlands CH 4 emission.Our retrievals consistently show strong emissions in this location.
The strong summer surface fluxes we retrieve over Fennoscandia (Fig. 5) are from natural sources (wetlands) so they are not reported in the EDGAR database.In the very northern parts of Fennoscandia, the natural wetlands are only moderately active during the period 15 May to 30 June; snow is melting and temperatures are cold.The active tiles to the east of Pallas are probably leaks from CH 4 gas production fields located just outside the tile domain in the Russian Federation.From 1 July to 14 August, many northern tiles, especially over Finland, become active.The strongest source location (468 mg day −1 m −2 ) is in southern Finland (tile #202) and is entirely dominated by lakes and wetlands.We show negligible CH4 sources during October over Fennoscandiaconsistent with the colder temperatures.The active tiles over Finland in December are probably anthropogenic sources but they are not yet understood.

Application to remote sensing
Remote sensing laser instruments designed to measure CH 4 from air and spaceborne platforms are being developed in Europe and the United States.The spaceborne and aircraft version of the MERLIN (DLR/CNES) instrument is expected to have a precision of ∼18 ppb over 50 km spatial averaging ( Kiemle et al., 2011).The Methane Sounder, being developed at NASA Goddard Space Flight Center, was tested on the NASA DC-8 in the summer of 2011 (Riris et al., 2012).While this is a breadboard instrument, only designed to demonstrate the use of laser technology, its precision is ∼50 ppb.The expected precision from the mature instrument design is ∼14 ppb.But will these new instruments (MER-LIN and NASA Methane Sounder) be able to detect changes in the current emission rates?
To study this, we need a model that accurately simulates the spatial distribution of emissions and uses source strengths that are at least in qualitative agreement with accepted values.Our model meets these criteria so we can proceed with some sensitivity experiments.We pose the question of whether the proposed instruments will be able to detect a 50 % reduction in surface emissions from Germany and the Netherlands (G&N).
Figure 7 charts the approach.We start with the methane surface emissions retrieved from the surface observations from 1 July to 14 August (also shown in Fig. 5) and label them "truth" (black box in Fig. 7).A forward model (Fig. 2b) generates total column CH 4 on a 1 × 1 • grid.Although the proposed laser instruments will be able to measure total column cloud in between clouds, measurements under fully overcast conditions are not possible.For this study we use cloud optical depths observed by the CALIPSO instrument (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations; Winker et al., 2009), since its satellite platform orbit will likely be similar to the one planned for the MERLIN instrument.The level 2 product, "CAL_LID_L2_05kmCLay", provides cloud optical depths for a 5 km field of view (FOV) along the satellite flight track.A total column CH 4 from our forward model is spatially interpolated at each 5 km FOV.If the cloud optical depth is below 0.5, the 5 km CH 4 amount is included in a 70 km average; 5 km FOV values with higher cloud optical depths are rejected.If all FOVs are clear, 15 adjacent 5 km values are included in the 70 km average.The result is a ∼8000 element vector of total column CH 4 averages.We consider two types of instruments: one assumes a very low noise level of 0.47 ppb per 5 km FOV measurement and a second assumes noise levels consistent with the proposed instruments, 47 ppb.The results are vectors of simulated CH 4 column amounts that the instrument would observe over a 20 d period.These are termed observables and are shown in Fig. 8.Separately, we input the simulated column CH 4 observations to a flux retrieval algorithm similar to the one depicted in Fig. 3.As before, we use the standard linear inverse method (chapter 3 of Rodgers, 2000).For the first guess (a priori) we use a homogeneous field of very low source strengths, 0.5 mg day −1 m −2 .We set the uncertainty in the observations consistent with the added noise.We assume no spatial covariance so the off-diagonal elements of the forward model error covariance matrix are zero.The results are retrieved surface  7. The three left-hand panels are from a control experiment that starts with our best estimate of the actual surface fluxes as shown in Fig. 5 for 1 July-14 August (recall these were inverted from data sampled at the 9 European CH 4 monitoring stations).The color of the panel perimeter (black, red and blue) refers to the color of the boxes in Fig. 7.The value for G&N (Germany and the Netherlands) is the cumulative emission for Germany and the Netherlands in tera-grams (Tg) per year.The three right-hand panels are for the perturbation experiment where the cumulative emissions for Germany and Netherlands is reduced by half.
CH 4 emissions at the 262 grid boxes; they are shown in the black and red boxes of Fig. 7.
We run two cases.The control case starts with our best estimate of the actual surface fluxes from 1 July to 14 August.During this period the cumulative G&N emissions are 6.4 Tg yr −1 .The perturbation case reduces the G&N emissions 50 % to 3.2 Tg yr −1 .The top panels of Fig. 9 are the "true" emissions used for these two cases.The grid boxes over G&N are outlined in black on the right-hand panels.When we only account for biases from instrument sampling and rejection of 5 km FOV observations for cloud conditions, we retrieve surface fluxes shown in the middle (red panel outline).Here we assume a very low instrument noise level of 0.47 ppb per 5 km FOV measurement.Despite the reduction in observed information due to data rejection for cloud conditions, the perturbation retrievals are still similar to the truth.The retrieved emissions over G&N are 5.8/2.8 for the control/perturbation case, respectively.Also note that both cases maintain the strong emissions over Finland, Ireland and northern Italy, as they should.When the instrument noise is increased to the level expected for the proposed instruments, we retrieve the emissions shown at the bottom of Fig. 9 (blue outlined panels).As expected, there is further degradation from the truth but the retrieved emissions over G&N are still lower for the perturbation vs. control case (2.5 vs. 4.7).We repeated this sensitivity experiment for a 20 day period in winter using retrieved fluxes from 15 November to 31 December.More persistent cloudy conditions during winter vs. summer result in retrieved fluxes that do not compare well with the truth, especially when realistic noise levels are assumed.Increasing the time duration of the observable to more than 20 days will improve the retrievals.Unfortunately, available computational resources precluded inversion of a larger matrix.

Conclusion
We have developed a simple forward model using the FLEX-PART trajectory model to simulate ground-based in situ and spaceborne (total column) measurements from proposed CH 4 lidar instruments.This forward model is used in conjunction with a retrieval algorithm to obtain estimates of surface emissions over Europe that are constrained by 3 h surface observations sampled at nine European ground stations.The model is generally able to simulate the daily variability in surface CH 4 concentrations observed at the ground stations.This suggests the model correctly simulates the filaments that transport CH 4 from their sources to the groundbased stations.
The model is then used to determine if the detection limits and measurement precision of the proposed instruments are low enough to detect significant changes in CH 4 surface emissions.We have applied our model to a future scenario where the emissions from Germany and the Netherlands are reduced by 50 %.In summer, this perturbation can be detected to within 22 % using 20 days of sampled column CH 4 amounts using the error specifications of the proposed instruments.

Fig. 1 .
Fig. 1.The 262 tiles used in this study are 1.5 • in latitude, and 1.5 to 7 • in longitude, from south to north, in order to obtain a constant area of approximately 7000 km 2 .

Fig. 2 .
Fig. 2. (a) The forward model used to calculate surface methane concentrations.For simplicity this schematic only shows equations for two geographic locations.(b) Forward model used to calculate total column amounts.

Fig. 3 .
Fig. 3. Schematic of the retrieval algorithm used to obtain source strengths.

Fig. 4 .
Fig. 4. Observed (black) and analyzed (red) hourly methane concentrations (ppm) at nine European locations for the year 2010.Contribution from the retrieved background concentration is shown in green.Please note that different y-axis scales are used for the different stations.

(Fig. 5 .
Fig. 5. Retrieved surface source strengths (mg day −1 m −2 ) for five 45 d periods in 2010.Each tile with significant emission has value shown in black and estimated uncertainty in white.
a similar study, Vermulen et al., (2006) used the FLEXPART trajectory model to simulate a time series of hourly CH 4 surface concentrations at Mace Head and at the Cabauw tower in the Netherlands during 2002.They used prescribed CH 4 surface fluxes from the METDAT (METhane DATabase) and the EDGAR databases.They were able to simulate ∼75 % of the variability of the Cabauw observations and their simulation for Mace Head looks very similar to ours.Another similar study used CO 2 concentration observations at three ground-based mountain stations: Plateau Rosa, Monte Cimone and Zugspitze along with the FLEX-PART trajectory model to determine CO 2 source and sink regions

Fig. 6 .
Fig. 6.(a) Monthly retrieved emission strengths from the European Monitoring Atmospheric Composition and Climate (MACC) collaborative project for July 2010.(b) Annual anthropogenic surface emissions from the Emission Database for Global Atmospheric Research (EDGAR) for 2008.(c) Average retrieved emission strengths from this study May-Dec 2010.(d) EDGAR 2008 integrated over our 262 surface tiles.(a), (c) and (d) show the upper Silesia mining district with dark pentagon.Each tile with significant emission has value shown in black

Fig. 8 .
Fig. 8. (a) Simulated total column CH 4 values observed by a proposed laser instrument averaged to a 70 km spatial resolution, assuming clear sky conditions.(b) Same as (a) but rejecting CH 4 values co-located with CALIPSO-observed cloud optical depths greater than 0.5.(c) Same as (a) and (b) except plotted by sample number index.

Fig. 9 .
Fig. 9. Surface emission of CH 4 from the sensitivity experiments outlined in Fig.7.The three left-hand panels are from a control experiment that starts with our best estimate of the actual surface fluxes as shown in Fig.5for 1 July-14 August (recall these were inverted from data sampled at the 9 European CH 4 monitoring stations).The color of the panel perimeter (black, red and blue) refers to the color of the boxes in Fig.7.The value for G&N (Germany and the Netherlands) is the cumulative emission for Germany and the Netherlands in tera-grams (Tg) per year.The three right-hand panels are for the perturbation experiment where the cumulative emissions for Germany and Netherlands is reduced by half.

Table 1 .
Ground stations sampling hourly methane concentrations used in this study.Elevation is in meters.CRDS is Cavity Ring Down Spectrometry and GC-FID is Gas Chromatography with Flame Ionization Detector.