Satellite observation of atmospheric methane : intercomparison between AIRS and GOSAT TANSO-FTS retrievals

Introduction Conclusions References Tables Figures


Introduction
As the third most important greenhouse gas after carbon dioxide (CO 2 ) and water vapor, atmospheric methane (CH 4 ) has a lifetime of about 12 years and is more effective in absorbing long-wave radiation, as its radiative forcing is about 26 times more than that of CO 2 on a 100-year time horizon and accounts for 32 % of the total anthropogenic well-mixed greenhouse gas radiative forcing (IPCC, 2013).Mainly due to the impact of human activities, the concentration of CH 4 in the atmosphere has increased from the pre-industrial levels of about 700 ppb to recent levels of about 1800-1900 ppb.
Ground-based networks, such as NOAA/ESRL/GMD (National Oceanic and Atmospheric Administration, Earth System Research Laboratory, Global Monitoring Division), provide measurements of CH 4 at the surface with a long temporal record but for a limited number of stations, primarily in the Northern Hemisphere.Aircraft measurements from NOAA/ESRL/GMD (Tans, 2009) and ARIES operated on UK FAAM aircraft (Illingworth et al., 2014), as well as some research campaigns, provide sparse, intermittent measurements of CH 4 vertical profiles.Because of a limited number of in situ measurements in time and space domain, the quantification of CH 4 emissions from different sources and in different regions still remains largely uncertain.In recent years, space-borne measurements of CH 4 from satellites have become available, such as the measurements using the thermal infrared (TIR) sensors, which include the Atmospheric InfraRed Sounder (AIRS) on NASA/Aqua (Aumann et al., 2003;Xiong et al., 2008Xiong et al., , 2010a, b), b), the Tropospheric Emission Spectrometer (TES) on NASA/Aura (Payne et al., 2009;Wecht et al., 2012;Worden et al., 2012), and the Infrared Atmospheric Sounding Interferometer (IASI) on METOP-A and METOP-B (Xiong et al., 2013;Crevoisiter et al., 2009Crevoisiter et al., , 2013;;Razavi et al., 2009).Measurements using the Near-Infrared (NIR) sensors include the SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) instrument onboard ENVISAT for 2003-2009(Frankenberg et al., 2008, 2011)), and the Thermal And Near infrared Sensor for carbon Observation (TANSO) onboard the Greenhouse gases Observation SATellite (GOSAT) from 2009 to present (Yokota et al., 2009;Paker et al., 2011;Schepers et al., 2012;Saitoh et al., 2012).These space-borne measurements provide complementary data sources to surface observations for monitoring atmospheric CH 4 with a large spatial and temporal coverage.
AIRS, GOSAT TANSO-FTS TIR and other thermal infrared sensors, including TES and IASI, have been used to retrieve atmospheric CH 4 , and these data have been used for analyzing the spatial and temporal variation of CH 4 , so an intercomparison of these two different products from AIRS and GOSAT will provide useful information to users to better understand the characteristics of these two products.Validation with AIRS V6 CH 4 data was recently made using ∼ 1000 aircraft profiles (Xiong et al., 2015), and the results show the mean biases of AIRS CH 4 at layers 343-441 and 441-575 hPa are −0.76 and −0.05 % and the RMSEs are 1.56 and 1.16 %, respectively.Some correlation of the retrieval error with degrees of freedom (DOFs) was also found, and the errors in the spring and in the high northern latitudes are larger than in other seasons or regions.A comparison between the GOSAT TIR methane retrievals with those of AIRS therefore represents an indirect validation of the GOSAT data with in situ measurements.Section 2 provides a brief introduction of these two instruments, their retrieval algorithms and the data used in this study.Section 3 shows the comparison results, which include the comparison of the retrieved profiles, the information content characterized by the DOFs and the averaging kernels from these two instruments.Both the retrieved CH 4 mixing ratios and the total column amounts in different seasons and different regions are compared.A summary and conclusion are given in Sect. 4. The cloud-cleared FOR radiance spectrum is then used to retrieve profiles with a spatial resolution of approximately 45 km (Aumann et al., 2003).The atmospheric temperature profiles, water vapor profiles, surface temperatures and surface emissivity are required as inputs to compute the radiances in the CH 4 absorption band.The differences between the computed radiances and the AIRS measured radiances for clear pixels or the derived cloud-cleared FOR radiances for partially cloudy pixels are used to derive CH 4 profiles based on optimal estimation method.A total of 50-60 CH 4 absorption channels near the 7.66 µm band are selected for the retrievals.The AIRS retrieval algorithm is a sequential retrieval method with multiple steps, in which the temperature and water vapor profiles are retrieved using appropriate channels in previous steps.Thus the quality of the CH 4 retrievals depends on the whole AIRS science team's efforts in improving the temperature and moisture profiles as well as surface temperature and emissivity products.More details of AIRS CH 4 retrievals in its most recent version, i.e., version 6 (V6), can be found in Xiong et al. (2015).

GOSAT TANSO-FTS TIR and the retrieval algorithm description
GOSAT was launched into a sun-synchronous orbit on 23 January 2009 by an H-IIA launch vehicle.GOSAT is on a 666 km orbit and has a 3-day revisit orbit cycle and a 12day operation cycle.The local solar time is 13:00 ± 15 min.GOSAT carries two sensors: the TANSO-FTS and the TANSO-CAI.The IFOV of the TANSO-FTS is 10.5 km in diameter, and that of the TANSO-CAI is 0.5-1.5 km in diameter.TANSO-FTS on board GOSAT makes global observations, including both nadir and off-nadir measurements, of approximately 56 000 ground points every 3 days.TANSO-FTS consists of four spectral bands: Band 1 (0.75-0.78 µm), Band 2 (1.56-1.72 µm), Band 3 (1.92-2.08 µm), and Band 4 (5.5-14.3µm).The spectra resolution of Band 4 is 0.2 cm −1 and its signal-to-noise ratio (SNR) averages approximately 300 at Band 4 for a blackbody temperature of 280 K. (Kuze et al., 2009(Kuze et al., , 2012;;Saitoh et al., 2009).More information on TANSO-FTS TIR and its calibration can be found in Kuze et al. (2012).In the TIR retrieval algorithm version 1.0, all the channels in 7.3-8.8µm, which include both the CH 4 and N 2 O absorption bands, are used for CH 4 retrieval.In the TANSO-FTS TIR V1.0 CH 4 retrieval processing, we simultaneously retrieve H 2 O, N 2 O, O 3 , and temperature other than CH 4 .We also simultaneously derive surface temperature and surface emissivity as a correction parameter of spectral bias inherent in TANSO-FTS TIR V161.160L1B spectra in the same manner as CO 2 retrieval (Saitoh et al., 2016).The retrieval algorithm is a non-linear maximum a posteriori method with linear mapping (Rodgers, 2000).The a priori CH 4 profiles used in the retrieval are taken from the National Institute for Environmental Studies (NIES) transport model (Maksyutov et al., 2008;Saeki et al., 2013).Profiles of temperature and water vapor required for the retrieval are taken from the Japan Meteorological Agency Grid Point Values (JMA-GPV) data set.Values of surface emissivities are estimated by a linear regression analysis using the Advanced Spaceborne Thermal Emission Reflection Radiometer (ASTER) spectral library (Baldridge et al., 2009), the information of land cover, sea ice, wind speed, and the vegetation index derived from the TANSO-Cloud and Aerosol Imager (CAI).Surface temperatures are estimated from the TANSO-FTS TIR spectra in the window region.The signalto-noise ratios (SNR) of TANSO-FTS at around the 7-8 µm band are estimated to be 70-100, and the measurement covariance matrix used in the retrieval is based on the SNR values.The footprint of GOSAT-TANSO is 10.5 km in diameter, and the number of scan points of GOSAT in cross-track direction is five before July 2010 and three thereafter.

Data used
This study is made using the standard products of both sensors.GOSAT TANSO-FTS  the data from the ascending mode of AIRS, with quality flag equal to 0 or 1, are used for comparison with GOSAT-TIR.
Note that the number of retrieval profiles from AIRS is much denser than GOSAT-TIR, as shown in Fig. 1.In this case, the number of profiles from GOSAT-TIR is 1479, while the number of the AIRS profiles from ascending node with QC = 0 and 1 is 164 355.The AIRS retrievals within 1 • from each GOSAT-TIR measurement and in the same day were averaged to match up with each GOSAT-TIR measurement.The errors resulting from the time difference of about 4 h between AIRS and GOSAT-TIR observations were not accounted for in this study given that CH 4 is a long-lived and well-mixed gas.For simplification, in the comparison of the total column CH 4 , the AIRS gridded products from NASA DISC in 1 • × 1 • were used directly, and the GOSAT-TIR data were interpolated to the same geographical grid as AIRS.
Comparisons between GOSAT-TIR and AIRS CH 4 products include (1) CH 4 profile comparison, (2) comparison of CH 4 mixing ratios with and without using the averaging kernels, and (3) comparison of the column-averaged CH 4 and the total column abundance in different latitude zones and different times.The CH 4 total column abundance from AIRS ascending mode whose unit is molecules cm −2 was used.Since the unit of GOSAT-TIR CH 4 profile was ppb, pressure profile P and surface pressure P 0 are used to convert the unit of GOSAT-TIR CH 4 profile.First, pressure gradient is calculated as where i denotes layer number; the formula to calculate Tc is where P T is the top-layer pressure.Pressure profile is included in the GOSAT-TIR product.As a GOSAT TANSO-FTS TIR CH 4 profile consists of 22 layers, and AIRS-V6 CH 4 profile contains 100 layers in the supporting product and 10 layers in the standard product, interpolation of AIRS CH 4 profile from 100 layers to 22 layers was made using the pressure data included in both CH 4 products.Later in this paper the "differences" between GOSAT-TIR and AIRS CH 4 are calculated as where X G denotes the CH 4 mixing ratios or column amounts from GOSAT-TIR, and X A from AIRS.

Profile comparison
Figure 2 shows a simple comparison of the GOSAT-TIR profiles and the coincident AIRS CH 4 profiles from 1 day of global data on 4 September 2010.They are in a good agreement above 100 hPa and below 400 hPa, with mean difference no more than 50 ppb, but at 200-300 hPa AIRS CH 4 spans a large range and tends to be smaller than GOSAT TIR on average.
The averaging kernels are defined to provide a characterization of the relationship between the retrieval and the true state.The retrieval sensitivity can be obtained from the sum of the rows of the averaging kernel matrix, which is also referred to as "the area of the averaging kernel" (Rodgers, 2000).To better demonstrate the differences in sensitivities between AIRS and GOSAT-TIR retrievals, Fig. 3 shows an example of the averaging kernels using data at a randomly selected location (38 • S, 180 • W) on 4 September 2010.There are 10 retrieval layers for AIRS and 22 for GOSAT-TIR.The area of the averaging kernels, which is computed as the sum of all individual kernels, from AIRS is larger than that from GOSAT-TIR, as shown in Fig. 3. To demonstrate the sensitivity variations in latitudes, Fig. 4 shows a curtain plot of the area of averaging kernels using 1 day of global data on 4 September 2010.Overall, the patterns from AIRS are similar to GOSAT TIR, with both peak sensitivities located in the 300-600 hPa range in the high latitudes and 200-600 hPa in the tropics.The sensitivities below 800 hPa are small for both, which reflects the major limitation of TIR in measuring the change of CH 4 in the lower troposphere.
The information content, which is usually represented as the DOF, is computed as the trace of the averaging kernel matrix (Rodgers and Connor, 2003).Figure 5 shows the variation of DOFs at different latitudes, and on average the DOF of AIRS CH 4 is approximately 1.1, whereas the mean DOF for the GOSAT-TIR retrieval of CH 4 is approximately 0.61.

Comparison of CH 4 with and without using the averaging kernels
Since intercomparison is made between two space-based sensors, it is necessary to take account of the different characteristics of the observing systems, particularly their averaging kernels, which is usually applied to the "truth" based on the following equation (Rodgers and Connor, 2003): where A is the averaging kernel.Here we use this equation to calculate the difference between GOSAT-TIR and AIRS CH 4 .So X represents the true state of CH 4 profile; X c is the mean of comparison ensemble of CH 4 profile, and it can be calculated using a regression-based function of latitude and longitude; X is the retrieved quantity related to the true profile X and x is the error.The computed value of X is referred to as the convolved data later in this paper, which is usually compared with the retrieved CH 4 mixing ratio in validation studies.Considering that the AIRS retrieval layers are coarser than those of GOSAT-TIR, we used the AIRS averaging kernels, A, to convert the GOSAT-TIR CH 4 profiles, X, and the convolved (or smoothed) GOSAT-TIR profiles ( X) are then used to derive CH 4 total column for comparison.This calculation is based on the Eq. ( 25) from Rodgers and Connor (2003).where a 1 is the total column averaging kernel of AIRS CH 4 ; c c is the total column derived from X c ; and ĉ 12 is the related total column using convolved GOSAT-TIR profiles.As the AIRS averaging kernel is a 10 by 10 matrix, the GOSAT-TIR CH 4 profile and AIRS first-guess profile are interpolated onto the 10 pressure layers of the AIRS retrieval grid.
Figure 6 shows the distribution of the absolute differences between GOSAT-TIR CH 4 and AIRS CH 4 column mixing ratios.The upper panel gives the statistical histogram of AIRS CH 4 difference to smoothed GOSAT-TIR CH 4 , while the lower panel shows that to unsmoothed GOSAT-TIR CH 4 .According to Fig. 6, the number of matched pairs with small differences increases after smoothing.A comparison of the column-averaged mixing ratio, X CH 4 , in Fig. 7 also shows that the correlation coefficient between AIRS and GOSAT TANSO-FTS TIR retrievals increases from 0.88 to 0.91 after using the smoothed data, and the mean difference decreases from −21.32 to −2.78 ppb.The mean difference and standard deviation between AIRS and the smoothed GOSAT-TIR data are smaller than those without smoothing using the averaging kernels, demonstrating that applying the averaging kernels helps achieve better agreements in the intercomparison between two different measurements, as suggested by Rogers and Connor (2003).
To show the impact of using averaging kernels in the intercomparison, Fig. 8 shows the scatter plot of AIRS versus GOSAT-TIR CH 4 mixing ratios in four retrieval layers of 272-343, 343-441, 441-575 and 575-777 hPa.The correlation coefficients between AIRS and the smoothed GOSAT-TIR values are 0.70, 0.70, 0.79 and 0.87 in these four layers respectively, while the correlation coefficients between AIRS and GOSAT-TIR without smoothing are 0.36, 0.45, 0.57 and 0.75 respectively.
In next sections, we will focus on the comparison of the total abundance between AIRS and GOSAT-TIR retrievals without applying averaging kernel for smoothing.

Comparison of CH 4 total column abundance in different latitude zones
As the sensitivity of TIR measurements is impacted by the surface thermal contrast and the water vapor content in the atmosphere (Deeter et al., 2007;Xiong et al., 2010b), the sensitivity varies with latitudes and seasons.Below we compare the differences between AIRS and GOSAT TANSO-FTS TIR retrieved total column abundance in six latitude zones from south to north with an interval of 30 • .As shown in Figs. 9 and 10, the correlations between AIRS and GOSAT-TIR are reasonably good, and the correlation coefficient for the least correlated case is 0.83 in zone 30-60 • S. The split of CH 4 daily comparison is due to these data being located in the high mountains between Chile and Bolivia in South America.This reflects a larger uncertainty in the mountain or coastline regions for AIRS and/or GOSAT.To show the change of their differences with time, Figs. 9 and 10 also show the monthly means of the differences from August 2010 to June 2012.In the tropics their differences are less than 1 % in all seasons, but in the mid to high latitudes in the Northern Hemisphere, GOSAT-TIR is ∼ 1-2 % lower than AIRS, with the largest bias occurring in September.At high latitudes in the Southern Hemisphere (60-90 • S) the differences of GOSAT from AIRS show a large variation with time, i.e., from −3 % in October to +2 % in July.This large difference in the high attitudes in the Southern Hemisphere is related to the very low DOFs, particularly in GOSAT-TIR retrievals (see Fig. 5), and the large uncertainties in the retrieval of atmospheric states when there is snow/ice coverage over the ocean during October to July.
To better show the difference between GOSAT-TIR and AIRS in different latitudes, we computed the mean difference over a 2-year period from 1 August 2010 to 30 June 2012 and in each 15 • zone.As shown in Fig. 11, the standard deviations in the Southern Hemisphere high latitudes are much larger than in the other latitudes, and the mean differences are smaller from 60 • S to 10 • N but increase in the Northern Hemisphere to −1.5 % at 60 • N.

Comparison of seasonal cycles from AIRS and GOSAT
Using the monthly averaged total column density of CH 4 from AIRS and GOSAT products, we compared the seasonal cycles of CH 4 from 1 August 2010 to 30 June 2012.The left panels in Fig. 12 are the comparisons in the Northern Hemisphere, and the right panels are the comparisons in the Southern Hemisphere.Again, GOSAT-TIR agrees with AIRS to within 1 % in the mid-latitude regions of the Southern Hemi-  sphere and in the tropics.However, the seasonal variation in the tropics from AIRS observations is larger than that from GOSAT.In the mid to high latitudes in the Northern Hemisphere, GOSAT-TIR is ∼ 1-2 % lower than AIRS, but the seasonal variations agree well.In the high-latitude regions in the Southern Hemisphere, the seasonal variation of the total column of CH 4 is large, which is due to a lot of data points with very low total column of CH 4 observed during October-January from both AIRS and GOSAT-TIR.However, AIRS and GOSAT agree well in capturing the variation even though their difference is relatively larger than in other regions (see Fig. 10).

Summary and conclusions
A thorough comparison of AIRS V6 and GOSAT TANSO-FTS TIR V1.0 CH 4 products using 2 years of data (1 August 2010 to 30 June 2012) has been made.In this comparison, AIRS measurements within a collocation window of 1 • by 1 • from each GOSAT-TIR measurement in the same day were used.Both the CH 4 mixing ratios and total column amounts have been compared.To understand the differences in the retrievals from these two different instruments, we also compared the differences in the averaging kernels and the DOFs and examined the use of averaging kernels on the comparison results.
The peak sensitive layers of AIRS and GOSAT-TIR are at similar height, which is at 200-600 hPa in the tropics and 300-600 hPa in the high-latitude regions.However, due to the lower SNR of GOSAT TANSO-FTS spectra in the 7-8 µm CH 4 band, or over-constraint in the GOSAT retrieval algorithm, the DOF of GOSAT-TIR V1.0 retrievals is lower than AIRS.
The comparisons of the profiles showed that the AIRS CH 4 is similar to GOSAT-TIR CH 4 , except that the AIRS values tend to be lower than GOSAT-TIR at 200-300 hPa.At 300 hPa, the CH 4 mixing ratios from GOSAT are 10.3 ± 31.8 ppbv higher than AIRS, and at 600 hPa, the GOSAT-TIR CH 4 is −16.2 ± 25.7 ppb lower than AIRS.Between 300 and 600 hPa, where they have peak sensitivities, AIRS and GOSAT-TIR agree very well.As expected, applying the averaging kernels to smooth the GOSAT-TIR retrievals results in a better agreement between GOSAT with AIRS products.
The comparison of the total column amounts of CH 4 shows that the correlation coefficients between AIRS and GOSAT TANSO-FTS TIR are more than 0.8 in all cases, and the GOSAT-TIR CH 4 agrees with AIRS to within 1 % in the mid-latitudes of the Southern Hemisphere and tropics, but in the mid to high latitudes of the Northern Hemisphere, GOSAT-TIR is ∼ 1-2 % lower than AIRS depending on different seasons.In the high latitudes of the Southern Hemisphere the bias varies from −3 % in October to +2 % in July.This large difference in high-latitude regions is associated with the low information content (or DOFs) and larger uncertainties in the retrievals of both AIRS (Xiong et al., 2015) and GOSAT.Thus a much stricter quality control should be used as suggested by Xiong et al. (2015).We also found AIRS and GOSAT-TIR have a good agreement in capturing the monthly variation of CH 4 density.
In this study, the time difference between AIRS and GOSAT-TIR measurements has not been taken into account.So, the differences, if they could have been measured at the same time, could be slightly smaller than what we presented here.These results demonstrate that the thermal infrared sensors such as AIRS and GOSAT TANSO-FTS TIR can provide valuable consistent information of CH 4 in the mid-upper troposphere.Further comparisons using more recent data as well as direct comparison with aircraft measurements are ongoing.

Data availability
The data set of AIRS V6 CH 4 product used in this study is available at Goddard Earth Sciences Data and Information Services Center (DISC) (http: //mirador.gsfc.nasa.gov/cgi-bin/mirador/presentNavigation.pl?tree=project&project=AIRS).

Figure 1 .
Figure 1.Much larger coverage of AIRS retrievals (a) as compared to GOSAT TANSO-FTS (b) as shown from the global CH 4 total column density on 4 September 2010.AIRS data from ascending mode with QC = 0, 1 are plotted.

Figure 2 .
Figure 2. Comparison of the matched-up AIRS CH 4 profiles versus the GOSAT-TIR profiles using 1 day of global data on 4 September 2010.N = 1182.Lower panel is the mean difference of AIRS minus GOSAT profiles.

Figure 3 .
Figure 3.The averaging kernels of AIRS and GOSAT TANSO-FTS TIR V1.0 CH 4 retrievals on 4 September 2010.There are 10 dashed lines for AIRS retrievals and 22 dashed lines for GOSAT retrievals, corresponding to the retrieval layers used in each of them.The black solid line is the area of kernels divided by 4.

Figure 4 .
Figure 4.The area of the averaging kernels of CH 4 retrievals from (a) AIRS and (b) GOSAT TANSO-FTS V1.0 TIR observations at different latitudes on 4 September 2010.

Figure 6 .
Figure 6.Comparison of AIRS and GOSAT-TIR CH 4 matched-pair difference.The upper panel shows the statistical histogram of AIRS CH 4 difference to smoothed GOSAT-TIR CH 4 and the lower panel shows that to unsmoothed GOSAT-TIR CH 4 .

Figure 7 .
Figure 7.Comparison of X CH 4 between AIRS and (a) unsmoothed GOSAT TANSO-FTS TIR X CH 4 and (b) smoothed GOSAT-TIR X CH 4 using AIRS averaging kernel.Global data on 4 September 2010 are used.

Figure 9 .
Figure 9. Scatter plot of AIRS versus GOSAT TANSO-FTS TIR CH 4 total column density over three latitude zones in the Northern Hemisphere using data from 1 August 2010 to 30 June 2012 (left panels).Right panels show the variation of the mean difference in each month, and the bars are the standard deviation.

Figure 10 .
Figure 10.Same as Fig. 8 but for three latitude zones in the Southern Hemisphere.

Figure 11 .
Figure 11.(a) Means of relative errors of GOSAT-TIR total column CH 4 relative to AIRS total column CH 4 in every 15 • zonal using data from 1 August 2010 to 30 June 2012.(b) 15 • zonal means of AIRS CH 4 total column.

Figure 12 .
Figure 12.Trends of CH 4 monthly averaged total column amounts in different latitudes using AIRS and GOSAT-TIR products from 1 August 2010 to 30 June 2012.The left panels are the comparison in the Northern Hemisphere and the right panels are the comparison in the Southern Hemisphere.