Pressure broadening in the 2 ν 3 band of methane and its implication on atmospheric retrievals

N2-broadened half widths and pressure shifts were obtained for transitions in the 2 ν3 methane band. Laboratory measurements recorded at 0.011 cm −1 resolution with a Bruker 120 HR Fouriertransform spectrometer were analysed from 5860 to 6185 cm −1. A 140 cm gas cell was filled with methane at room temperature and N 2 as foreign gas at pressures ranging from 125 to 900 hPa. A multispectrum nonlinear constrained least squares approach based on Optimal Estimation was applied to derive the spectroscopic parameters by simultaneously fitting laboratory spectra at different ambient pressures assuming a Voigt line-shape. At room temperature, the half widths ranged between 0.030 and 0.071 cm−1 atm−1, and the pressure shifts varied from –0.002 to –0.025 cm−1 atm−1 for transitions up to J =10. Especially for higher rotational levels, we find systematically narrower lines than HITRAN predicts. The Q and R branch of the new set of spectroscopic parameters is further tested with ground based direct sun Fourier transform infrared (FTIR) measurements where systematic fit residuals reduce by about a factor of 3–4. We report the implication of those differences on atmospheric methane measurements using high-resolution ground based FTIR measurements as well as low-resolution spectra from the SCanning Imaging Absorption SpectroMeter for Atmospheric ChartographY (SCIAMACHY) instrument onboard ENVISAT. We find that for SCIAMACHY, a latitudinal and seasonally varying bias of about 1% can be introduced by erroneous broadening parameters. Correspondence to: C. Frankenberg (c.frankenberg@sron.nl)


Introduction
Methane (CH 4 ) is, after carbon dioxide, the second most important anthropogenic greenhouse gas, directly contributing 0.48 W/m 2 to the total anthropogenic radiative forcing of 2.63 W/m 2 by well-mixed greenhouse gases (IPCC, 2007). Although the global annual source strength of methane (550±50 Tg/year) is comparatively well constrained, considerable uncertainties still exist in regard to the partitioning amongst sources and their spatial and temporal distribution.
SCIAMACHY (SCanning Imaging Absorption SpectroMeter for Atmospheric ChartographY) onboard ENVISAT provides the unique opportunity to monitor methane globally with high sensitivity for the entire atmospheric column, including the surface layers where sources are located. Results obtained so far (Frankenberg et al., 2005a(Frankenberg et al., , 2006Bergamaschi et al., 2007;Buchwitz et al., 2006) underline the potential of SCIAMACHY to detect local and regional methane sources. However, there are indications for systematic biases as detailed in Frankenberg et al. (2006) and Bergamaschi et al. (2007). Hence, we want to investigate whether spectroscopic uncertainties can be the cause of observed retrieval biases.
Most research groups (Frankenberg et al., 2006;Buchwitz et al., 2006;Schneising et al., 2008) perform methane retrievals from SCIAMACHY in a microwindow in channel 6, ranging from 5983 to 6138 cm −1 (1629-1671 nm). This spectral range comprises the strong Q and R branches of the 2ν 3 methane band as well as numerous weaker lines for which the quantum assignments are unknown. Figure 1 depicts one of the recorded FTS (Fourier transform spectrometer) spectra as well as a SCIAMACHY spectrum over the Sahara. One can clearly identify the congested Q-branch at Published by Copernicus Publications on behalf of the European Geosciences Union. In addition, the attribution to Q and R-branch transitions in the 2ν 3 band is given. Please note that the SCIAMACHY spectrum is not absolute-radiometric calibrated, so only relative variations should be considered. The major absorber in the SCIAMACHY spectrum is CH 4 but smaller absorptions by CO 2 and H 2 O are also present.
about 6000 cm −1 as well as the more widely spaced transitions in the R branch at higher wavenumbers. The spectral resolution of SCIAMACHY in this channel is relatively poor with a FWHM (Full width at half maximum) of 1.33 nm (≈4.8 cm −1 ), sampling the entire Q-branch with a mere 3-4 detector pixels. Thus, individual lines are not resolved, so that incorrect line-shapes do not necessarily result in systematic retrieval residuals. However, if absorptions are strong, the line-shape and pressure shifts govern the sensitivity of the retrievals with respect to trace gas perturbations (Frankenberg et al., 2005b) as broader lines absorb more efficiently. This can cause systematic biases that change with the degree of saturation of the absorption lines, i.e. also with viewing geometry. Washenfelder et al. (2003) already found systematic biases in methane total columns retrieved from ground-based FTIR measurements in the P-branch of the 2ν 3 band.
For methane, the HITRAN 2004 linelist (Rothman et al., 2005;Brown et al., 2003) in this region is mainly based on empirical line positions and intensities of Margolis (1988Margolis ( , 1990. Measurements were performed at low pressures, thus only estimated broadening coefficients from limited measurements were adopted. The widths ranged from 0.054 to 0.066 cm −1 atm −1 at 296 K for J =0 to 10, and transitions of the same lower state energy were given the same widths regardless of rotational symmetry (A,F,E). All air-broadened pressure shifts were set to a constant -0.008 cm −1 atm −1 . Margolis (1990) originally provided calculated ground state energies for assigned 2ν 3 transitions and empirical lower state energies for many of the unassigned features, but in HI-TRAN 2004, all the empirical values were replaced by -1, leaving lower state energies only for the assigned transitions in the 2ν 3 band.
Methane exhibits a complex spectrum with each J rotational level split into tetrahedal components and an ordering index. At temperatures encountered in the atmosphere, the multiplets are strongly blended even if only Doppler broadened. This renders the accurate measurement of individual broadening coefficients and pressure shifts difficult. To date, only few transitions in the 2ν 3 band have been studied with respect to pressure broadening. Gharavi and Buckley (2005) report broadening coefficients and line strengths for the R3 and R4 multiplets, while Kapitanov et al. (2007) report broadening and pressure shifts for the R3 triplet. To complicate matters, numerous studies observe collisionalnarrowing and line-mixing effects in methane spectra. Dufour et al. (2003) report Dicke-narrowing for the 2ν 3 R0 transition and in addition line mixing effects for the R3 multiplet. Pine (1992Pine ( , 1997; Pine and Gabard (2003) and Predoi-Cross et al. (2007) report Dicke narrowing and line mixing effects in transitions at about 3000 and 4350 cm −1 . Mondelain et al. (2007) show how collisional narrowing and line-mixing in the P9 manifold of the ν 3 band impact atmospheric profile retrievals by up to 7% (see also Mondelain et al. (2008) for a discussion on the effect of the temperature dependence of the broadening coefficient). In this study, we neglect those intricacies of methane spectroscopy but follow a pragmatic approach to determine an updated set of effective methane spectroscopic parameters with focus on pressure-broadening.
The paper is structured as follows: Sect. 2 describes the experimental setup, followed by a description of the applied inverse method in Sect. 3. Derived spectroscopic parameters are presented and discussed in Sect. 4 and the impact on atmospheric retrievals is shown in Sect. 5. Conclusions are drawn in Sect. 6.

Experimental setup
Laboratory spectra were recorded over the 5600-6300 cm −1 spectral range with a Bruker IFS 120 HR FTS located at the Institute of Environmental Physics of the University of Bremen. All spectra were obtained using a tungsten lamp as the infrared source, a 1.7 mm input aperture at the entrance of the FTS, a CaF 2 beamsplitter and a liquid nitrogen cooled InSb detector. An optical filter was used to reduce the bandpass. This optical filter was not wedged, which resulted in a sinusoidal structure in the continuum. The gas mixtures were introduced in a 140 cm cell. For thermal stabilization and insulation the cell body is enclosed by 2 coaxial quartz jackets. The inner one is temperature stabilized by a flow of ethanol from a thermostatic bath. The outer one is evacuated for thermal insulation. Windows at the front and back end of the cell are made of CaF 2 . The cell was located behind the interferometer and the light passed twice through the cell before being detected.
Transmission spectra were calculated by dividing sample spectra (resolution=0.011 cm −1 ) by the spectra obtained for the evacuated cell (resolution=0.1 cm −1 ). Twenty-five interferograms were co-added for the calculation of the spectra, resulting in a signal-to-noise of about 200 for the transmission spectra.
To achieve a uniform and constant methane mixing ratio in the cell, it was first filled with N 2 , then with methane and finally with N 2 . During the measurements the pressure in the cell remained stable to within 0.6% or better. Table 1 gives an overview of measurement conditions and measured spectra applied in this study. All spectra were recorded consecutively on the same day, and we assumed the internal calibration of the FTS was unchanged during that time. The pressure in the absorption vessel was measured with two independent capacitive pressure transducers of 500 hPa and 1000 hPa maximum range (MKS Baratron).

Data analysis
We applied a nonlinear fitting approach to derive the spectral line parameters pressure-broadening coefficient and pressure-induced shift. Parameters for each individual line were fitted using multiple laboratory spectra simultaneously. Relative line intensities were also fitted but strictly constrained to the Margolis (1988) values given in HITRAN, permitting only small deviations.
Details about the multispectrum fitting technique and its advantages can be found in Benner et al. (1995). Even though the multispectrum approach greatly reduced uncertainty in the retrieval, the inverse problem remained underdetermined as the blended lines could not be fully separated. For this reason, we added additional constraints using the Optimal Estimation technique (Rodgers, 2000). This approach allowed us to attribute prior uncertainties to the target parameters, thereby minimising oscillations of parameters whose retrieval errors would be strongly correlated in an unconstrained least squares approach. The solution to this nonlinear constrained least squares approach which maximises the a posteriori probability density function of the state vector is given by Rodgers (2000): where x a = a priori state vector , x i = state vector at the ith iteration , S = (pixel)error covariance matrix , S a = a priori covariance matrix , F (x i ) = forward model evaluated at x i , K i = Jacobian of the forward model at x i , y = measurement vector .
In our particular case the state vector comprised relative line intensity, pressure broadening parameter and pressureinduced shift of each individual line as well as a low order polynomial (for each spectrum separately) to account for broad-band baseline structures. The forward model simulates the transmission spectra of the gas cells depending on the integrated column density of methane (fixed, equals concentration c times absorption path length l) and the spectral parameters to be retrieved: where f ILS is the instrumental line-shape (in our case a sinc convolved with a box function) and ⊗ denotes convolution. S i ,γ i and δ i are line intensity, N 2 -broadening coefficient and pressure shift, respectively, for each transition i. f voigt denotes a Voigt function and ν 0,i is the vacuum wave number of transition i. The entire forward model F consists of a concatenation of all j simulated transmission spectra. The entire state vector comprises S i ,γ i and δ i for each line i and 2 polynomial coefficients a for each spectrum j separately. Each line is treated separately, even for multiplets, and no cross-correlations between lines is assumed (i.e. S a is diagonal). A standard Voigt line-shape is applied and the Jacobian of the transmission with respect to shift and broadening coefficients computed analytically, as explained in Schreier (1992) and references therein. Self-broadening was neglected since the methane volume mixing ratio in the cell was 2% at most.
From a measurement perspective, no attempt was made to determine an accurate total methane content in the cell a priori. Hence, the integrated column density of methane was determined using a fit covering the isolated R0 and R1 transitions. The line strengths retrieved in this study are thereby linked to the R0 and R1 strengths given in HITRAN (Margolis, 1988). For the final determination of spectral parameters, cell column densities were kept fixed. Prior line strengths were taken from HITRAN and prior pressure shifts are all reset to -0.011 cm −1 atm −1 as Kapitanov et al. (2007) reports this pressure shift for the R3 triplet.
Prior broadening coefficients were not taken from HI-TRAN but adapted from measurements in the fundamental by Pine (1992Pine ( , 1997, as done by Washenfelder et al. (2003). The actual fitting was performed in 10 cm −1 bins for the sake of computational efficiency. In the inversion, four N 2 broadened spectra at different pressures at room temperature were used (see Table 1). As we were mainly interested in broadening coefficients, we applied relatively strict prior constraints for line intensity and pressure shift. We then performed retrievals with increasingly stricter constraints on pressure broadening and found that only for prior covariances below about (0.002 cm −1 atm −1 ) 2 , fit residuals started getting worse. Hence we chose this value as optimal constraint on pressure broadening. Depending on the spectral microwindow, the fit typically converged within 3-10 iterations. Table 2 lists the employed prior uncertainties for the individual parameters.

Results and discussion
Figures 2, 3 and 4 show the fit results for the P, Q and Rbranch, respectively. Only measured FTS spectra and fitting residuals are displayed. As can be seen in the systematic red residuals, applying the HITRAN database yields wrong line- Table 2. Fit parameters used in the inversion.
widths and partly shifted lines. While there only seems to be a shift for low rotational levels in the Q branch, higher levels show systematic residuals typical of too high broadening coefficients (wings too strong, line center too weak). The P and R-branches behave similarly and a zoom on the R6 multiplet in Fig. 5 shows the improvements in the fits more clearly. Looking closely at the R6 multiplet, some systematic residuals at intermediate pressures (125 and 250 hPa) exist, hinting at deviations from the Voigt line-shape, perhaps due to Dicke-narrowing. Line-mixing as observed in the ν 3 fundamental can also play a role. The same holds true for other transitions. However, those residuals are typically below 1%, being very small compared to residuals induced when applying the current HITRAN broadening coefficients. Overall, the final residuals after fitting line parameters show far less systematic features and are dominated by the instrumental artifact of channeling due to the applied spectral filter. The fitted N 2 broadening coefficients and pressure shifts for strong transitions in the spectral region under investigation are shown in Fig. 6, along with the ratio of present and HITRAN line strengths. To estimate possible systematic errors and the impact of prior values in the retrieved parameters, additional fits were performed with varying assumptions as a sensitivity study: -2 fits with different low frequency correction terms for the channeling; -1 fit where methane abundances in the cell are an additional fit parameter, i.e. not fixed; -1 fit with HITRAN broadening and pressure shift parameters as prior; -2 fits with universal prior values for pressurebroadening: 0.058 and 0.05 cm −1 atm −1 .

Fig. 2.
Fit results of the multi-spectra inversion for the P-branch. The upper panel depicts the N 2 broadened FTS spectrum at 900.5 hPa. The lower panels show the fitting residuals for each FTS spectrum (decreasing pressure from top to bottom) separately in black (measuredmodelled times 100). The red lines indicate residuals if no changes to the HITRAN database were applied. The gray lines show residuals using the HITRAN database but replacing broadening coefficients with results by Pine (1992Pine ( , 1997 in the fundamental.   Table 3 gives an overview of the results for the transitions of the 2ν 3 band that are assigned in HITRAN (please note that some upper state rotational levels are not yet attributed to specific transitions). The outcome of the sensitivity study is used as an error estimate for retrieved broadening coefficients and pressure shifts.
For isolated lines, such as R0 and R1, the retrieval is very stable and insensitive to prior assumptions; accuracies are expected to be about 1%. Larger variations occur when lines are strongly blended as spectral parameters within multiplets are underdetermined, deteriorating accuracies to about 5-10%. However, even for those transitions we did not find   Fig. 3 but for the R6 manifold with six strong transitions. The sum of the measured intensities in the manifold is greater by only 1.2% compared to the Margolis (1988) values. In contrast, the average of measured widths is 19% smaller than the average of default widths applied in HITRAN. The residuals seen in the fit of the lowest pressure spectrum could arise from combined inaccuracies of the retrieved positions, relative strengths, widths, the presence of weak transitions not included in the database or effects of non-Voigt line-shapes.
a systematic dependence on the prior. Figure 7 shows the retrieved broadening parameters as a function of |m|, where |m| corresponds to the rotational state J (upper state J for R branch lines, lower state for the Q and P-branch). The |m| dependence is very similar for all branches, slightly increasing in the beginning with decreasing widths for higher rotational levels. As for the comparison with results from the ν 3 band (Pine, 1992(Pine, , 1997, which are displayed in the lower panel, we see a slightly different |m| dependence, especially at higher rotational levels our measurements show systematically lower broadening coefficients. Pine (1992Pine ( , 1997 included collisional narrowing in their analysis and broadening parameters derived with a non-Voigt profile cannot be easily compared with Voigt-only retrievals, as can be seen in Dufour et al. (2003). However, Predoi-Cross et al. (2006) noted similar differences at higher J in the air-broadened widths at 4300 cm −1 . Finally, at the end of the present study, new measurements of N 2 -broadening in the ν 4 fundamental were reported by Martin and Lepère (2008). Comparison for 31 transitions with the same rotational quantum assignments produced an average ratio (ν 4 to 2ν 3 widths) of 1.01±13%. Given the experimental precisions of the studies, the large root-mean-square is interpreted as evidence of the vibrational dependences of the widths, as discussed by Predoi-Cross et al. (2006). However, such conclusions must await theoretical validation through the calculations such as those produced by Antony et al. (2008).
One should mention that it is virtually impossible to derive independent spectral parameters within a multiplet, especially in the congested Q-branch. Errors in the retrieved parameters are therefore strongly correlated and differences in retrieved broadening coefficients within a multiplet should be interpreted with care as many solutions can lead to identical line-shapes. For example, different retrieved pressure shifts within a multiplet might accommodate line-mixing or Dicke narrowing features, thereby not representing the pressure shift in its physical sense any more. However, the main objective of this work is to create a set of empirical spectroscopic parameters that allows a best possible modelling of methane cross sections at typical pressures encountered in the Earth's atmosphere applying the simple and widely-used Voigt line-shape. There are only a few studies of broadening parameters in the 2ν 3 band to compare with and a comparison for multiplets is not straightforward as the results depend on a variety of assumptions that might differ from study to study. However, for the fully isolated R0 and R1 lines a comparison is possible with work from Darnton and Margolis (1973) and also with the results from Pine (1997) for the fundamental. The comparison is shown in Table 4 and shows very good agreement between the different results, both in absolute terms and also concerning the difference of the broadening coefficient between the R0 and R1 line.

Open issues
Studies of methane pressure broadening are greatly hindered by the lack of basic characterization of the spectrum in this region. The spectral interval between 4800 and 6300 cm −1 contains 14 vibrational bands with some 60 sub-vibrational components. A complete linelist would likely contain more than 20,000 features, but at present no global analysis has been published to provide a reliable calculation of line positions, strengths and lower state energies. Similarly, there is no theoretical model that predicts pressure broadening coefficients with the accuracies required for state-of-the-art remote sensing. Hence, more laboratory measurements and development of theoretical models are needed to achieve 1% accuracies for line intensities and broadening parameters as well as to resolve line-mixing and collisional-narrowing effects.
Temperature dependence of pressure broadening is so far unknown for most transitions and therefore requires further research. While HITRAN attributes a temperature exponent of 0.75 universally in the spectral range considered here, Darnton and Margolis (1973) and Gharavi and Buckley (2005) find values ranging between 0.83 and 0.93 for R0-R4 in the 2ν 3 branch. At 243 K, this might already lead to discrepancies in the broadening coefficient of 3% (temperature exponent 0.75 vs 0.9). The theoretical prediction in a recent study by Antony et al. (2008) may provide some guidance about the variation of the temperature dependence as a function of the quantum numbers. In any case, atmospheric scientists using the HITRAN database should thus be aware of its uncertainties as provided in error codes and referenced original literature. Table 3. Retrieved broadening coefficients and pressure shifts in the 2ν 3 band for 12 CH 4 listed for lower rotational level J , symmetry species C and ordering index n . For broadening coefficients and pressure shifts, the standard deviation of the fitted parameters in the sensitivity study is provided as error estimate. The differences to HITRAN and to Pine (1992Pine ( , 1997 values are listed separately, calculated as ((γ this study − γ reference )/γ reference * 100). Broadening parameters are given for 296 K, assuming a temperature dependence exponent of 0.75. A more detailed list as well as retrieved parameters for all lines are provided in the supplementary material. http://www.atmos-chem-phys. net/8/5061/2008/acp-8-5061-2008-supplement.zip J C n a line position [10 −3 cm −1 atm −1 ] -HITRAN - Pine(1992Pine( ,1997  a Transitions for which the n are not shown are thought to be perturbed and may in fact be a different assignment altogether. In addition, some of the assigned transitions may be blended with transitions from other bands. A full analysis of the bands in this region is needed before an accurate interpretation of the measurements can be made.    Darnton andMargolis (1973) Pine (1997) (ν 3 ) R0 59.1 (±0.2(2σ )) 59.0 (±06.7(1σ )) 58.7 (±0.17(1σ )) R1 64.4 (±0.4(2σ )) 63.7 (±10.2(1σ )) 64.7 (±0.08(1σ ))

Additional information
Along with the results described in the paper, we provide two files with supplementary information http://www.atmos-chem-phys.net/8/5061/2008/ acp-8-5061-2008-supplement.zip. One file lists retrieval results (line strengths, broadening coefficients and pressure induced shifts) and associated standard deviations from the sensitivity studies for about 1300 lines in the 5860 to 6184.5 cm −1 range. A second file provides these results in a HITRAN-format (2001) file for easy application. In this file, the following changes were made: -retrieved N 2 -broadening coefficients are scaled with 0.985 to give an estimate of air-broadening; -empirical lower state energies measured by Margolis (1990) are restored; -exponent of the temperature dependence of broadening has been reset to 0.85 for all lines; -default value for shifts are changed from -0.008 to -0.011 cm −1 atm −1 at 296 K.
Even though the results of this study are most reliable for almost 150 strong lines of the 2ν 3 band, about 1300 transitions in the considered spectral range have been fitted and the results incorporated in the supplementary files http://www.atmos-chem-phys.net/8/5061/ 2008/acp-8-5061-2008-supplement.zip.

Impact on atmospheric methane measurements
Erroneous broadening coefficients can have a profound impact on the retrieval of gases (Washenfelder et al., 2003). Mondelain et al. (2007) found deviations of up to 7% due to line-mixing and pressure broadening effects in atmospheric profile retrievals. In this study we consider total column retrievals which should be less susceptible to errors in lineshapes as compared to profile retrievals that actually rely on an accurate knowledge of the line-shape to differentiate trace gas perturbations at different height layers. However, possible errors depend on a variety of factors like spectral resolution, retrieval method and viewing geometry.
Here we discuss two kinds of measurements, first a high resolution ground based direct sun measurement using an FTIR instrument located in Bremen, Germany, and second This work / Pine(1992Pine( ,1997 Fig. 7. Measured N 2 -broadening coefficients retrieved in the P, Q and R-branch (depicted as black square, red triangle, blue diamond, respectively) as a function of |m|, where |m| is the lower state J for P and Q transitions and the upper state J for R branch lines. For easier identification, branches are slightly shifted with respect to |m|. The lower panels depicts the ratio of the retrievals with results from Pine (1992Pine ( , 1997 in the ν 3 band. Values greater than one indicate that broadening coefficients retrieved in this study are larger than those from Pine (1992Pine ( , 1997 retrievals from the SCIAMACHY instrument (Bovensmann et al., 1997) onboard the European research satellite EN-VISAT. As SCIAMACHY retrievals are the main driver for this study, we focus only on the Q and R-branch in both cases.
As pure nitrogen was used as foreign gas in the laboratory measurements, retrieved broadening coefficients need to be converted for atmospheric applications. Pine (1992) finds a relatively constant mean pressure broadening ratio of 0.937 for O 2 compared to N 2 . Hence we scale the retrieved N 2 broadening coefficients with 0.985 (weighted average according to atmospheric abundances) to derive air-broadening coefficients needed for the analysis of atmospheric spectra used in this section.  Fig. 8. Pressure, temperature and water vapour profiles were taken from NCEP (National Centers for Environmental Prediction), the initial CH 4 profile from a TM5 model run (similar to the S1 scenario described in Bergamaschi et al., 2007) and the CO 2 profile set to a constant 380 ppm (parts per million). We used a least squares approach and scaled the entire initial profiles with factors for the respective trace gases. A disk center solar spectrum was calculated from a linelist provided by Geoffrey Toon, Jet Propulsion Laboratory, Pasadena, USA. CO 2 spectroscopic parameters from a recent study by Toth et al. (2007) are applied. The lower panel in Fig. 8 shows the residuals using the methane spectroscopic parameters derived in this study and the parameters given by HITRAN. It is striking that the new methane spectroscopy largely reduces systematic residual structures, not only for the methane lines itself but also for the CO 2 lines which exhibit a slight shift when HITRAN methane spectroscopy is applied. This is most likely due to the updated pressure shifts. Table 5 shows the retrieval results for the two different FTIR spectra applying updated and HITRAN methane spectroscopic parameters. As the time difference between the measurements was short (below 3 hours), the methane vertical column is expected to remain rather constant. This is true for the retrievals using the updated spectroscopy, where the difference between the two measurements is only 0.13%, underlining the precision that can be achieved in this spectral range. However, applying HITRAN spectroscopic parameters, the difference is 2.3%, rendering the retrievals virtually useless for the detection of small changes in atmospheric methane abundances. A similar systematic error depending on solar zenith angle was already found by Washenfelder et al. (2003) for the P-branch of the 2ν 3 band.

Low resolution space borne measurements by SCIA-MACHY onboard ENVISAT
Previous studies suggest systematic biases in the SCIA-MACHY methane retrievals (Frankenberg et al., 2006;Bergamaschi et al., 2007). Here we analyse the impact of the new spectroscopy using the retrieval approach described in Frankenberg et al. (2005b) for the year 2004. Two retrieval runs were performed that only differed in methane spectroscopy, all other retrieval parameters being identical. Figure 9 shows the average differences for the entire globe averaged through a year as well as latitudinal averages for individual months. The new retrieval gives in general higher methane columns, ranging from 0 to 1.5%. This can be easily explained by narrower lines that absorb less efficient as they are not resolved by SCIAMACHY (in contrast to the FTIR retrievals). Over highly elevated regions, such as the Himalayas or Andes, the differences are negligible. This can be explained by two reasons: First, the atmosphere is optically thinner, leading to less saturation effects that would be governed by line-widths. Second, pressure broadening and shift are less dominant. The differences for North America, for example, over the Rocky Mountains are smaller compared to the eastern regions. In that particular case, wrong spectroscopic parameters would create an artificial east-west gradient. The right panel of Fig. 9 depicts the dependence of the mean latitudinal bias on season, exemplified by 4 different months. Differences are minimal in tropical regions where the solar zenith angle is low. However, at higher latitudes the solar zenith angle and thus the bias changes during the year. Artificial seasonal structures with a peak-to-peak amplitude of about 1% could thereby be generated, which is crucial for source inversions as it cannot easily be discerned from seasonally varying methane source.
The southern hemisphere, devoid of large methane sources, should exhibit a relatively small seasonal cycle and is therefore an ideal test case for the retrieval, especially since Frankenberg et al. (2006) found a distinct but unexplainable seasonal cycle in SCIAMACHY retrievals. Figure 10 shows the 30-day running mean (box filter) of SCIAMACHY retrievals over Australia in 2004. The two retrieval versions differing in spectroscopy are shown. In addition, TM5 model values optimised with accurate NOAA (National Oceanic and Atmospheric Administration) ground based flask measurements are shown (similar to model fields of scenaro S1 in Bergamaschi et al., 2007). The new version (scaled by 1.015 to match the absolute model values) corresponds very well with the model, deviating by at most by 5 parts per billion (about 0.3%) while Frankenberg et al. (2006) found systematic seasonal biases of about 40 ppb peak to peak. Most of this previous bias is actually caused by spectroscopic interferences with water vapor, as shown by Frankenberg et al. (2008). These findings are already taken into account in this study, eliminating a large part of the seasonal bias. Using the old retrieval version (HITRAN methane spectroscopic parameters, scaled by 1.02), however, still shows a distinct deviation by about 10 ppb from April through September, corresponding with high air mass factors.

Conclusions
In this study, we apply a multi-spectrum fitting technique to derive N 2 -broadening coefficients and pressure shifts of the methane molecule in the 5860 to 6185 cm −1 range which is used in methane retrievals using the SCIAMACHY spectrometer onboard ENVISAT. This range covers the strong 2ν 3 band for which, until now, only few transitions have been analysed with respect to pressure broadening.  Fig. 10. Timeseries of SCIAMACHY methane column averaged mixing ratios over Australia in 2004. Concurrent CO 2 retrievals have been used as proxy for the light path (Frankenberg et al., 2005a(Frankenberg et al., , 2006, using CarbonTracker by NOAA for model CO 2 profiles to minimise the impact of CO 2 variations. The air mass factor is shown in grey and a decontamination period (±10 days) of the satellite is overlaid as grey box. Time series data have been smoothed with a 30-day running time-average (box filter) and spatially averaged over the entire australian continent between -20 and -30 degree latitude. Black: retrieval with new CH 4 spectroscopic parameters (this study); red: retrieval with HITRAN CH 4 spectroscopic parameters; green: modelled CH 4 mixing ratio.
Compared to the HITRAN spectroscopic database, we found systematically lower broadening parameters, especially at higher rotational levels. These results were confirmed using ground based solar absorption FTIR measurements where we found systematic residuals and biases depending on solar zenith angle when applying HITRAN spectroscopic parameters. The results of this study largely reduced residuals and the bias. Some small systematic inconsistencies between modelled and measured methane spectra still exist, probably owing to line-mixing and Dickenarrowing effects. However, compared to the improvements achieved in this study, deviations from the Voigt line-shape are small.
We further show that SCIAMACHY retrievals based on the spectroscopy derived in this work are systematically different from a retrieval version using the HITRAN database. Differences for monthly and yearly averages are between 0 and 1.5%, depending on solar zenith angle and surface elevation. Thus, artificial seasonal structures with a peak-topeak amplitude of about 1% can be introduced in our particular retrieval setup. With the new spectroscopy, we find only very small differences (max. 5 ppb) as compared to an atmospheric model over Australia for monthly averages.
The main purpose of this study was to provide an effective set of spectroscopic parameters for methane retrievals in the Earth's atmosphere, minimising systematic seasonal and latitudinal biases. The new spectroscopy will not only be helpful for ground-based FTIR and SCIAMACHY retrievals but also be for future space missions such as the GOSAT (Greenhouse gases Observing Satellite) mission by the Japan Aerospace Exploration Agency covering a similar spectral range for retrieval.