Introduction
Atmospheric methane (CH4) is not only an important greenhouse gases
(GHGs) but also plays an important role in atmospheric chemistry. As a GHG,
it plays 25 times more effective on a per unit mass basis than CO2 in
absorbing long-wave radiation on a 100-year time horizon (IPCC, 2007). The
reaction of CH4 with hydroxyl radicals (OH) produces CH3 and
water, and the removal of OH via this reaction significantly impacts many
other oxidation processes related with OH in the atmosphere. It is found
that the concentration of CH4 in the atmosphere has increased from
the pre-industrial levels of about 700 parts per billion (ppb) to about
1800–1900 ppb, and this increase is mainly attributed to the impact of human
activities. However, the increase rate of CH4 is not stable, and after
remaining stable for over a decade, a rapid increase has been observed
beginning in 2007. The causes of this increase are the focus of many recent
studies (e.g., Bousquet et al., 2011; Bergamaschi et al., 2013; Bruhwiler et al.,
2014), and more observations are required.
Systematic high-precision measurements of CH4 mixing ratios and the
CH4 isotope ratios have been made for over 25 years by taking the air
samples near the ground and measuring the concentration in the laboratory
(e.g. Chen and Prinn, 2006; Bousquet et al., 2006, 2011), which include the
measurements at the sites of NOAA/ESRL/GMD (National Oceanic and Atmospheric
Administration, Earth System Research Laboratory, Global Monitoring
Division) networks and other sites under the umbrella of the WMO (World
Meteorological Organization) Global Atmosphere Watch (GAW) programme.
Another type of measurements with a good data record is the ground-based
remote sensing using solar Fourier-transform-spectrometry (FTS) instruments,
which provide measurements of the total column amount of CH4, and
these data are available from the Network for the Detection of Atmospheric
Composition Change (NDACC) (http://www.ndsc.ncep.noaa.gov/). The vertical
variation of CH4 in the atmosphere was measured in recent years using
aircrafts, and most of them were made especially over North America. One
usage of these data is to validate satellite remote sensing observations
(see Xiong et al., 2010, 2013a and references therein).
Recent improvements in satellite sensors, particularly the increase of
spectral resolution, made the space-borne measurements of CH4 from
satellites possible. One class of instrument uses thermal infrared (TIR)
sensors, such as the AIRS on NASA/AQUA (Aumamn et al., 2003; Xiong et al.,
2008, 2010), the Tropospheric Emission Spectrometer (TES) on NASA/Aura (Payne
et al., 2009; Wecht et al., 2012; Worden et al., 2012; Alvarado et al.,
2015), and the Infrared Atmospheric Sounding Interferometer (IASI) on METEOP-A
and METEOP-B (Crevoisier et al., 2009, 2013; Razavi et al., 2009; Xiong et
al., 2013a). Another class of instrument employs Near-Infrared (NIR) sensors,
such as the SCanning Imaging Absorption spectroMeter for Atmospheric
CHartographY (SCIAMACHY) instrument onboard ENVISAT for 2003–2009
(e.g. 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–present (Yokota et al., 2009; Park
et al., 2011; Schepers et al., 2012; Saitoh et al, 2012). These satellite
sensors provide a complementary measurement to surface and airborne
observations of atmospheric CH4 with a large spatial and temporal
coverage (Kirschke et al., 2013). Combining the surface and satellite
measurements allows inverse modeling to better constrain the quantification
of CH4 sources and sinks in different time and space domain (e.g.
Meirink et al., 2008; Bergamaschi et al., 2013; Houwelling et al., 2014;
Massart et al., 2014). For example, Massart et al. (2014) assimilated
SCIAMACHY, GOSAT/TANSO, IASI and a combination of TANSO and IASI CH4
products in the Monitoring Atmospheric Composition and Climate Interim
Implementation (MACC-II) system to produce the atmospheric CH4 analysis
in about 6 months behind real time. Using the four-dimensional variational
(4DVAR) inverse modeling system TM5-4DVAR and data of the surface
observations during 2000–2010 from NOAA/ESRL/GMD network
together with retrievals of column-averaged CH4 mole fractions from
SCIAMACHY, Bergamaschi et al. (2013) found that the global total emissions
for 2007–2010 are 16–20 Tg CH4 yr-1 higher compared to 2003–2005, and
that most of the inferred emission increase was located in the tropics
(9–14 Tg CH4 yr-1) and mid-latitudes of the northern hemisphere
(6–8 Tg CH4 yr-1). Using the CarbonTracker-CH4 assimilation system but the
surface observation data only, Bruhwiler et al. (2014) estimated emissions
from natural sources in 2007 were greater than the decadal average by 4.4±3.8 Tg CH4 yr-1.
The benefit of satellite data in inverse modeling is limited by the
uncertainty in both the satellite data and the transport model. As pointed
out by Houwelling et al. (2014), assimilating SCIAMACHY retrievals into TM5
4DVAR increases the estimated inter-annual variability of large-scale fluxes
by 22 % as compared to that resulting from assimilating only surface
observations, and systematic errors in the SCIAMACHY measurements are a main
factor limiting the performance of the inversions. Therefore validation of
the satellite retrieval products to quantify their accuracy and precision is
very important to facilitate their use in inverse modeling, and at the same
time, it is one essential step for further improvement of the retrieval
algorithms.
As a stable TIR sensor, AIRS has been used to retrieve temperature and water
vapor profiles as well as some trace gases since 2002. Xiong et al. (2008)
described the characterization and some validation of AIRS-V5 CH4
product. Some significant improvements are expected for AIRS-V6 temperature
and water vapor retrieval products as well as trace gases. This study is the first one to
systematically evaluate the quality of the AIRS-V6 CH4 product using
in-situ aircraft profiles. Section 2 provides a brief summary of the
CH4 retrieval improvements in AIRS-V6 and its sensitivity. Section 3
describes the validation data and method. Based on these validations, some
optimization of quality control (QC) is recommended. Section 4 provides the
validation results and error analysis. The summary of results and
conclusions are given in Sect. 5.
Improvements in CH4 retrieval in AIRS-V6 and averaging
kernels
AIRS and improvement in AIRS-V6 CH4 retrieval
AIRS was launched in polar orbit (13:30 local solar time, ascending node) on
the EOS/Aqua satellite in May 2002. It has 2378 channels covering 649–1136,
1217–1613 and 2169–2674 cm-1 at high spectral resolution (λ/Δλ=1200) (Aumann et al., 2003). The spatial resolution
of the AIRS field-of-view (FOV) is 13.5 km at nadir, and in a 24-h period
AIRS nominally observes the complete globe twice per day. In order to
retrieve CH4 in both clear and partial cloudy scenes, a 3 × 3 array of 9
AIRS FOVs within the footprint of the Advanced Microwave Sounding Unit
(AMSU) is used to derive a single cloud-cleared radiance spectrum in a
field-of-regard (FOR), which is then used for retrieving profiles with a
spatial resolution of about 45 km. The AIRS retrieval algorithm is a
sequential retrieval with multiple steps, in which the temperature and water
vapor profiles, surface temperature and surface emissivity are retrieved
first using channel subsets optimized for the component being retrieved. Thus the CH4 retrieval in version 6
benefited from the AIRS science team's efforts to improve the temperature
and moisture profiles and their quality control (e.g. Susskind et al., 2011;
Maddy et al., 2012). AIRS data used in this study were downloaded from the
NASA Goddard Earth Sciences Data and Information Services Center (DISC)
(http://disc.sci.gsfc.nasa.gov/AIRS/data-holdings/by-data-product-V6).
For CH4 retrieval, the upstream AIRS Level 2 retrieval products,
including atmospheric temperature profile, water vapor profile, surface
temperature and surface emissivity, and a CH4 first guess profile are
used as the initial atmospheric state inputs to the forward Radiative
Transfer Algorithm (RTA) (Strow et al., 2003) to compute the upwelling
radiance in the pre-selected CH4 absorption channels. The difference
between the computed radiance and the associated AIRS Level 2 cloud cleared
radiance product, ΔR, is represented in a linear
approximation to the change in the CH4 profile in percentage, ΔX, as
follows:
ΔRn=Sn,L⋅ΔXL+ε,
where Rn is the cloud cleared radiance (observed), and
ΔRn is Rn minus the RTA calculated
radiance in channel n, ΔXL is the difference of
CH4 from the first guess at layer L that will be derived, and
Sn,L is a vector of the sensitivities of radiance in channel n
to changes ΔXL in the CH4 profile at L
different levels, and ε is the error. For trace gas retrievals,
a percentage or logarithmic perturbation to the trace gas abundance is used
because Eq. (1) is more linear using this formulation than using an absolute
difference in gas abundance. The ΔXL is obtained by
solving the Eq. (1) using singular value decomposition (SVD), and
damping the least significant eigenfunctions of the SVD to constrain the
solution. Similar algorithm was also used for the retrieval of nitrous Oxide
(N2O), and more detail description of this algorithm for
trace gases retrieval can be found in Xiong et al. (2008, 2014).
The major improvements in AIRS-V6 CH4 retrieval algorithm include an
increase in the number of retrieval layers, selection of more optimum
channels, setting of damping parameter to constrain the retrieval, and small
changes in the RTA regarding to the tuning of CH4 absorption
coefficients in the peak absorption channels near 7.66 µm. Compared to
AIRS-V5 with seven retrieval layers, 10 vertically overlapping trapezoidal
functions are used in AIRS-V6, as shown in Fig. 1, and the pressure levels
are listed in Table 1. There are about 200 AIRS channels spanning the 7.66 µm
CH4 absorption band, of which 63 used in the AIRS-V6 CH4
retrieval. In AIRS-V5, a 2 % increase in methane absorption coefficients
for strong absorption channels near 1306 cm-1 was implemented (Xiong et
al., 2008), and in V6 this adjustment was updated to 1–2 % for these peak
channels. The damping parameter was then optimized after the increase of
retrieval layers and re-selection of channels. The CH4 first guess
profile (“a priori”) is similar to that in Version 5, which is a function of
latitude and pressure (to capture its strong latitudinal and vertical
gradients), and was generated through a non-linear polynomial fitting (Xiong
et al., 2008).
AIRS sensitivity and the averaging kernels
The averaging kernels are defined to provide a simple characterization of
the relationship between the retrieval and the true state, and the retrieval
sensitivity can be illustrated from the sum of the columns of the averaging
kernel matrix, which is also referred to as “the area of the averaging
kernel” (Rodgers, 2000). As an example, Fig. 2 shows the averaging kernels for
the 10 trapezoidal functions in the high northern latitude and in the
tropics in September 2009. The averaging kernels corresponding to these 10
functions are broad and exhibit significant overlap, indicating that the
retrieved amounts of CH4 at different layers are not independent. The
degrees of freedom (DOF), defined as the fractional number of significant
eigenfunctions used in the retrieval process and computed as the trace of
the averaging kernel matrix, is about 1.1 in the tropics and 0.9 in the high
northern hemisphere in this case. In general, the DOF in the tropics is
higher than in the high latitudes, and the summer DOF is higher than the
winter DOF.
To better see the retrieval sensitivity and its variation as a function of
latitude and season, Fig. 3 shows the zonal averaged peak sensitive layer
using two days' global data on 30 March 2010 and 1 July 2010. The peak
sensitive layer in the tropics is between 100–500 hPa and has little
seasonal variation. In the mid-latitude regions, the peak sensitive layer is
between 200–600 hPa in March and rises to higher altitude at 150–600 hPa in
July due to the enhanced water vapor in the atmosphere in summer time. The
sensitive layer also depends on the thermal contrast. In the high northern hemisphere (60–90∘ N), the peak sensitive
layer in March is between 300–600 hPa and rises to a slightly higher
altitude in July as the water vapor is less than in the tropics.
Validation to AIRS-V6 CH4 product
Data of aircraft measurement profiles used for validation
Table 2 lists the in-situ aircraft campaigns whose measured CH4
profiles are used for validations in this paper. Figure 4 shows their
spatial locations on the globe. Except for the HIPPO campaigns, measurements
are mostly located over North America. More details of these campaigns
are described as follows:
The Intercontinental Chemical Transport Experiment (INTEX) part A field
mission was conducted in the summer of 2004 (1 July–15 August 2004) over
North America (NA) and the Atlantic. This effort had a broad scope to investigate
the transport and chemistry of long-lived greenhouse gases, oxidants and their
precursors, aerosols and their precursors, as well their relationship with
radiation and climate. NASA's DC-8 and J-31 were joined by aircraft from a large
number of European and North American partners to explore the composition of the
troposphere over NA and the Atlantic as well as radiative properties and effects
of clouds and aerosols in a coordinated manner (Simpson et al., 2002). An air
sample is collected in different altitudes using a conditioned, evacuated 2-L
stainless steel canister equipped with a bellows valve, and is returned to the
UC-Irvine laboratory for CH4 analysis using gas chromatography (GC, HP-5890A)
with flame ionization detection. The use of the primary CH4 calibration
standards dating back to late 1977 ensures that these measurements are internally
consistent. The measurement accuracy is ±1 % and the analytical precision
at atmospheric mixing ratios is about 1 ppbv (Simpson et al., 2002, 2006).
INTEX-B was a major NASA led multi-partner atmospheric field campaign
completed in the spring of 2006 (http://cloud1.arc.nasa.gov/intex-b/).
INTEX-B was performed in two phases. In its first phase (1–21 March),
INTEX-B operated as part of the Megacity Initiative: Local And Global Research
Observations (MILAGRO) campaign with a focus on observations over Mexico and
the Gulf of Mexico. In the second phase (17 April–15 May), the main INTEX-B
focus was on trans-Pacific Asian pollution transport. Multiple airborne platforms
carrying state of the art chemistry and radiation payloads were flown in concert
with satellites and ground stations during the two phases of INTEX-B (Singh et al., 2009).
The CH4 aircraft measurements in INTEX-B are similar to INTEX-A.
The Arctic Research of the Composition of the Troposphere from Aircraft
and Satellites (ARCTAS) mission was conducted in April and June–July 2008 by the
Global Tropospheric Chemistry Program and the Radiation Sciences Program of NASA.
Its objective was to better understand the factors driving current changes in
Arctic atmospheric composition and climate. Three research aircrafts (DC-8, P-3, B-200)
were used and a total of 24 research flights had been made. The aircraft were based in
Alaska in April (ARCTAS-A) and in western Canada in June–July (ARCTAS-B).
The DACOM instrument used is an infrared tunable diode laser absorption spectrometer
which makes measurements of CH4 (as well as CO and N2O) at a 1 Hz sample rate.
The CH4 accuracy is tied to NOAA's Earth System Research Laboratory,
Global Monitoring Division (NOAA/ESRL/GMD) carbon cycle group standards and
is nominally 1 %, and the precision is 0.1 % (1 sec, 1σ).
CH4 observations during ARCTAS-A showed little variability and no
indication of significant April emissions from Arctic ecosystems. The July
observations in ARCTAS-B over the Hudson Bay Lowlands revealed higher wetland
emissions of methane than previously recognized (Jacob et al., 2010).
Aircraft measurements of the CH4 vertical profiles were also made by
the NOAA/ESRL Alaska Coast Guard (ACG) flight, in which CH4 was measured
with a Cavity Ringdown Spectroscopy (CRDS) analyzer at 0.4 Hz frequency with
an overall measurement uncertainty of 2 ppb (Karion et al., 2012 and the
references therein). ACG data in 2009, 2010 and 2011 were used and these data were provided by NOAA/ESRL/GMD.
Aircraft measurements of the CH4 vertical profiles by the HIAPER
Pole-to-Pole Observations (HIPPO) program over the Pacific Ocean (Wofsy et al., 2011)
provide a unique dataset for validation over a wide latitudinal range (67∘ S–85∘ N).
The National Science Foundation's Gulfstream V (GV) were used during all the five
HIPPO missions. The GV transected the Pacific Ocean from 85∘ N to 67∘ S,
performing in-progress vertical profiles every 220 km or 20 min
(Wofsy et al., 2011, 2012). CH4 was measured with a Quantum Cascade Laser
Spectrometer (QCLS) at 1 Hz frequency with accuracy of 1.0 ppb and precision
of 0.5 ppb (Kort et al., 2012). HIPPO methane data are reported on the NOAA04
calibration scale. The NOAA04 scale was designated as the official calibration
scale, and consists of 16 gravimetrically prepared primary standards covering
the nominal range of 300–2600 nmol mol-1. This makes it suitable for
use in calibrating standards for the measurement of air extracted from ice
cores and contemporary measurements from GAW sites. This new scale results in CH4
mole fractions that are a factor of 1.0124 greater than the previous scale
(now designated CMDL83) (Dlugokencky et al., 2005). HIPPO data was downloaded from (http://hippo.ornl.gov/dataaccess).
Validation method
For each aircraft measurement profile, we calculated the mean location
(latitude and longitude) and time. All AIRS retrievals (with quality flag
equal to 0, 1) coincident with each aircraft profile within 200 km and from
the same day were used to compute the mean retrieved profile, which is then
compared with the corresponding aircraft profile smoothed after applying the
averaging kernels as follows:
x^=Ax+I-Axa,
where I is the identity matrix, A is the averaging
kernel matrix, xa is the first guess profile (unit: part
per billion, ppb), x is the in situ aircraft measurement
profile, and the computed value x^, referred to as the convolved
data later in this paper, will be compared with the retrieved CH4
mixing ratio. As the aircraft profiles do not span the entire vertical range
defined by the averaging kernels, extension of the aircraft profiles is
required when using Eq. (2). This can be done using output from a chemistry
model or climatology data to represent CH4 mixing ratios in the upper troposphere and higher levels.
In this paper we used the monthly averaged CH4 data in 2007 from an
Atmospheric General Circulation Model (AGCM)-based chemistry transport model
(hereinafter ACTM) (Patra et al., 2011) to extrapolate from the ceiling of
the aircraft profile to the top of atmosphere and from the lowest
measurement height to the bottom of the atmosphere. The profile is then
mapped to the 100 levels grid of RTA (Strow et al., 2003). The aircraft
profiles with their ceilings beneath the 350 hPa pressure level were not
used in validation.
The averaging kernels for each retrieved profile are applied to the same
collocated aircraft data, and the mean of these convolved profiles is
compared with the mean of the collocated retrieved profiles.
Optimization of quality control
The AIRS retrievals with quality flag equal to 0, 1 are usually counted as
“good quality” retrievals and recommended to use. However, from the
comparison with aircraft measurements we found some “good quality”
retrieval profiles (with QC =0,1) show obvious oscillation. For example,
Fig. 5 shows AIRS retrievals collocated with HIPPO-2 measurement at the
location of (82.43∘ N, 150.4∘ W) in 21 November 2009. All the retrievals
with quality flag as 0 and 1 within 150 km from aircraft measurements are
shown in Fig. 5. However, two blue profiles show a big “bump” of
over 1950 ppb at 400 hPa but a low value of about 1800 ppb near the surface.
Considering CH4 is well mixed in the mid-lower troposphere, these two
profiles show obvious oscillation as opposed to the aircraft measurement and
the first-guess. These two profiles are not as good as we expected, and we
think the qualities of these profiles should be marked as “not good” and
need to reset their quality flags to 2.
From the experiences we learned from this validation study, the quality flag
for a retrieved profile that has the quality flag as 0 or 1 but does not pass
the following two tests needs to reset to 2:
Oscillation test: for CH4 mixing ratio between 350 and 50 hPa above
the surface, if (the maximum–the mean) and (the minimum–the mean) are in
opposite signs and its difference (the maximum–the minimum) is 5 % larger than the mean.
Strong inversion: the maximum CH4 mixing ratio at ∼ 400 hPa is
150 ppb (< 65∘ N) or 250 ppb (> 65∘ N) higher than the mixing ratio near the surface.
The profiles (with quality flag as 0 or 1) failed in either test are not
recommended for use. However, we found the number of these profiles is only
a small portion (less than 5 % in the cases we examined) of the total
profiles with quality flag equal to 0 or 1, so for statistic analysis of computing mean
CH4 using more than hundreds of profiles in a large region, the error
resulted from these bad quality profiles is estimated to be insignificant.
In this paper, the above optimized quality control was used to filter out
some bad quality profiles.
Results and discussion
Validation results
Figure 6a shows the mean bias and RMS error of the retrieval in 100 levels
using 941 aircraft profiles from nine campaigns. For comparison the error of
the firstguess profile is also plotted. Below 400 hPa the bias is less than
0.5 % and RMS error is less than 1.5 %, and, as expected, the retrieval
error as compared to the smoothed aircraft measurements is smaller than the
direct comparison without applying the averaging kernels. From the
information of the number of the aircraft measurements (Fig. 6b) we can
see that the number of samples above 300 hPa is much less, so most profiles
need to be extrapolated to the top of atmosphere using model data. This is
why the retrieval bias and RMS error at above 300 hPa are even larger than
the error of the first guess. This result also suggests that the first guess
used in AIRS-V6 retrievals is good to represent the mean of climatology.
Here we did not use the subset of samples with aircraft measurements above
300 hPa to compute the error separately considering: (1) the match-up was based on
the mean latitude/longitude of the whole profile of aircraft measurement; (2) some profiles have
measurements above 300 hPa but do not have enough measurements at lower
levels; (3) the impact of stratospheric intrusion could exist to some profiles (Xiong et
al., 2013b).
Further comparison between AIRS retrievals with collocated aircraft
measurements in four trapezoid layers of 272–343, 343–441, 441–575
and 575–777 hPa are given in Fig. 7. Overall, the correlation between
the AIRS retrievals and the aircraft measurements is very good (R=0.73-0.95). A larger negative bias was found in upper layers,
particularly when the CH4 mixing ratio is lower than 1780 ppb.
Compared with first guess, the retrieval error in the layers of 272–343 hPa
(the upper left panel) is even larger, but the retrieval errors in other
layers are less than first guess and the correlations are improved,
demonstrating the right work did by the retrievals.
To check the performance of the retrievals in different time and location,
the retrieval error of CH4 at layer 441–575 hPa is analyzed using the
validation data in different seasons (Fig. 8) and different latitude zones
(Fig. 9). It is evident that a larger negative bias of the retrievals
occurs in the spring (March–May) (Fig. 8) and in the high northern
hemisphere (60–90∘ N) (Fig. 9). While the CH4 mixing ratio
is lower than 1800 ppb, there is a negative bias during the summer (June–August)
(Fig. 8) and mainly in the tropics (Fig. 9). From the
correlation coefficient between AIRS retrievals and aircraft measurements,
we found that, overall, a little worse correlation occurs in the mid-latitude region (30–60∘ N)
(lower left panel in Fig. 9). The
aircraft measurements used for validation in the mid-latitude region were mostly made over the
North America, so the latitudinal gradient of CH4 mixing ratio in this latitude zone
(30–60∘ N) is small as compared to that in other latitude
zones with an interval of 60∘, which is part of reason why the
correlation between the retrievals and aircraft measurements is relatively
smaller.
Error analysis
There are quite a few sources contributing to the retrieval errors. First is
the error from the radiative transfer model, particularly the uncertainty in
spectroscopy of CH4 absorption and line mixing. Recent study by
Alvarado et al. (2015) showed the updates to the spectroscopic parameters
for CH4 resulted in a substantially smaller mean bias in the retrieved
CH4 when compared with HIPPO observations. Some correction of 1–2 %
to the absorption coefficients was implemented for the strong CH4
absorption channels which are near 1306 micron and mostly sensitive to high
altitude (Xiong et al., 2008), but this correction is empirical and will need
improvement in the RTA. Next is the error propagation since the CH4
retrieval is based on the retrievals of atmospheric temperature and water
moisture profiles, surface temperature and emissivity, so good retrievals in
the upstream products will significantly impact to the retrieval of
CH4. Some estimation of error propagation can be found from Xiong et
al. (2008) and will not repeat here again. The third one is the lack of
in-situ observations in layers above 300 hPa and the miss-match of AIRS observation
with aircraft data in time and space domain. Even though the error due to
the time difference is expected to be small over ocean (Wecht et al., 2012),
but over land or in regions close to emission sources, this error will be
much larger. So, better collocated aircraft measurements with satellite observations will be greatly
helpful for satellite validations.
Here we just examined the relationships of the retrieval errors with
latitude, DOFs and cloud fraction. As shown in Fig. 10, the mean bias is
small and decreases from the southern hemisphere to mid-latitude of northern
hemisphere, but the bias in the high northern hemisphere above 60∘ N is
obviously larger than in the mid-latitude below 50∘ N. Also, the
retrieval biases in the high latitude regions have a larger variability than
in the tropics. From the upper right panel we can see the DOFs are mostly
between 1.2–1.4 in the tropics, and 0.6–0.8 in the high northern latitude
(above 60∘ N), and for DOFs > 1.2 the biases are mostly
positive. The biases are well correlated with the DOFs in the tropics
(30∘ S–30∘ N, R=0.74), but this correlation is much smaller in the
mid-high latitude regions (R=0.36, upper right panel). We also found that
among different seasons the best correlation occurred in the summer. From
the lower left panel, it is evident that most retrievals have a positive
bias under clear sky, and in the tropics the biases have a
negative-correlation with the cloud fractions (30∘ S–30∘ N, R=-0.6). We also found these positive biases under clear sky are largely found
in the summer and fall, and the correlation of the bias with the cloud
fraction in the summer is better than in other seasons. The correlation
between the bias and cloud cover fraction is small in the mid-high latitude
regions.
Further analysis (lower right panel) indicates the DOFs is
negative-correlated with cloud fraction, and on average the correlation
coefficient is R=-0.7 in the tropics and R=-0.5 in other regions.
We also found that among different seasons the largest one is in the summer
with R=-0.8 and -0.7 in the tropics and other regions respectively (not
shown).
It has been a concern that the retrievals may be impacted by clouds since
the cloud-cleared radiances are used in the retrievals. Our analysis found the correlation of
the retrieval error with the DOFs is greater than the correlation with cloud
cover fraction, suggesting that the dominant factor to the retrieval is the
DOF. Examination to the correlation between DOFs and retrieval errors for
different cloud fractions shows that under clear sky, the correlation
coefficient is R=0.7 (for cloud amount < 0.1), and it decreases to R=0.28 for cloud amount > 0.5.
The correlation of retrieval error with cloud amount suggests the error in
the cloud-cleared radiance is one important error source, and further
improvement to the cloud-clearing algorithm is required in the future. These
results also imply that the “observed” spatiotemporal variation by AIRS
not only reflects the real change of CH4 in the atmosphere, but also
include some artificial impact from sensitivity, or DOF, and/or
contamination by cloud. So, for the analysis of CH4 distribution and/or
seasonal variation using the retrieval products from AIRS (same for other
thermal infrared sensors), some filtering of the data based on the DOFs
and/or cloud amounts will help to remove part of this artificial variation;
however, it is impossible to completely remove their impact to get the real
spatiotemporal variation of CH4 accurately based on the retrieved CH4 mixing ratios only.
Summary and conclusion
Significant improvements in the CH4 retrieval algorithm in AIRS-V6 was
made, which include the increase of the retrieval layers from 6 to 10,
reselect of channels and the adjustment of damping parameter. As a result,
the peak sensitive layer near the tropics is between 100–500 and 200–600 hPa
in the mid-high latitude regions, and the DOFs are mostly between
1.2–1.4 in the tropics, and 0.6–0.8 in the high northern latitude (above
60∘ N). In this paper, a thorough validation to AIRS-V6 CH4 using
about 1000 aircraft profiles from different campaigns was presented. In our
validation the mean of AIRS retrievals within 200 km from each aircraft
measurement and in the same day was compared to the aircraft measurement
after applying the averaging kernels. From these comparisons we found some
optimization to the quality filtering of AIRS retrievals is desirable. Even
though the population of these profiles is a small fraction (less than
5 %) of the total number retrieval profiles from the cases we examined,
and its impact is estimated to be small for statistical analysis using hundreds of profiles in a large region, it is
better to double-check the fraction of these profiles with inappropriate
qualities and decide whether it is necessary to use the optimized quality
flag suggested in this paper.
Validation results show that, on average, at layers 343–441 and 441–575 hPa
the retrieval biases are -0.76 and -0.05 % with the RMS errors of
1.56 and 1.16 %, respectively. The bias of AIRS CH4 is negative
in high altitude and is much larger than in the lower altitude. The mean
error in layer above 300 hPa is even larger than first guess, which is
mainly due to the extrapolation of aircraft measurements to the top of
atmosphere.
Further analysis of the retrieval errors with cloud fraction and the DOFs
show the retrieval bias is well correlated with the DOF, especially when DOF
is greater than 0.8. The correlation coefficient in the tropics is larger
than in other regions, and in the summer is larger than in other seasons. We
found the retrieval error is correlated with cloud cover when the cloud
cover is less than 0.5. From the correlation between the DOF and cloud cover
fraction, we conclude that the retrieval error is largely impacted by the
DOF. This finding implies that the “observed” spatiotemporal variations by
AIRS and/or other thermal infrared sensors not only reflect the real change
of CH4 in the atmosphere, but also include some artificial impact from
the sensors and the retrieval sensitivity. Considering the change of DOF in
different latitudes and different seasons (Fig. 3), we suggest that for
data analysis some filtering of the data based on the DOFs and/or cloud
amounts will help to remove part of this artificial variation.