- Email: [email protected]
HITRAN spectroscopy evaluation using solar occultation FTIR spectra
Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
Contents lists available at ScienceDirect
Journal of Quantitative Spectroscopy & Radiative Transfer journal homepage: www.elsevier.com/locate/jqsrt
HITRAN spectroscopy evaluation using solar occultation FTIR spectra Geoffrey C. Toon a,n, Jean-Francois Blavier a, Keeyoon Sung a, Laurence S. Rothman b, Iouli E. Gordon b a b
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, United States Harvard-Smithsonian Center for Astrophysics, Cambridge, MA, United States
a r t i c l e in f o
abstract
Article history: Received 22 April 2016 Received in revised form 22 May 2016 Accepted 22 May 2016 Available online 11 June 2016
High resolution FTIR solar occultation spectra, acquired by the JPL MkIV Fourier transform spectrometer from balloon, covering 650–5650 cm 1 at 0.01 cm 1 resolution, are systematically analyzed using the last four versions of the HITRAN linelist (2000, 2004, 2008, 2012). The rms spectral fitting residuals are used to assess the quality and adequacy of the linelists as a function of wavenumber and altitude. Although there have been substantial overall improvements with each successive version of HITRAN, there are nevertheless a few spectral regions where the latest HITRAN version (2012) has regressed, or produces residuals that far exceed the noise level. A few of these instances are investigated further and their causes identified. We emphasize that fitting atmospheric spectra, in addition to laboratory spectra, should be part of the quality assurance for any new linelist before public release. & 2016 Elsevier Ltd. All rights reserved.
Keywords: Spectroscopy Infrared Atmosphere Remote Sensing HITRAN
1. Introduction The HITRAN linelist [19] is without doubt the most widely used spectroscopic database and underpins most atmospheric remote-sensing experiments. As our understanding of the terrestrial atmosphere improves, the accuracy required of new measurements becomes more stringent. Hence the requirements on the accuracy and completeness of the spectroscopy become inexorably more challenging. It is usually impossible to fit high-resolution Infra-Red solar spectra (high SNR) down to their noise level, at least below 25 km tangent altitude. Residuals are typically dominated by systematic errors arising from defects in the assumed atmospheric temperature (T), pressure (P), and volume mixing ratio (vmr) profiles, the observation geometry (pointing), the instrumental response (e.g., instrument line n
Corresponding author. E-mail address: [email protected] (G.C. Toon).
http://dx.doi.org/10.1016/j.jqsrt.2016.05.021 0022-4073/& 2016 Elsevier Ltd. All rights reserved.
shape, zero-level-offsets, channel fringes, phase errors, aliasing, ghosts), and spectroscopic inadequacies. For spectra measured with a well-calibrated instrument under wellknown atmospheric conditions, the first three types of systematic errors can usually be minimized, revealing the underlying spectroscopic problems. In the past we successfully used the Total Carbon Column Observing Network (TCCON) Fourier Transform Spectrometer (FTS) in Park Falls, WI for validating spectroscopic parameters of molecular oxygen in different spectral windows [5,6]. In the first paper not only the parameters of strong magnetic dipole lines in the 1.27 μm band were validated but the electric quadrupole lines were identified for the first time in this band. Previously they were thought to be orders of magnitude weaker. Similar evaluations were done in limited spectral regions of water, methane and carbon dioxide prior to the release of the HITRAN 2012 database. In this paper, we document the quality of fits to atmospheric spectra collected by the JPL MkIV interferometer and
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
fitted using the JPL Gas Fitting software (GFIT), to provide a benchmark for assessment of future linelists (e.g., HITRAN 2016) and to highlight gases and spectral regions where further spectroscopic work would be beneficial.
2. MkIV instrument The MkIV FTS is a double-passed FTIR spectrometer designed and built at JPL in 1984 for atmospheric observations [21]. It covers the entire 650–5650 cm 1 region simultaneously with two detectors: a HgCdTe photoconductor covering 650–1800 cm 1 and an InSb photodiode covering 1800–5650 cm 1. The MkIV instrument has flown 24 balloon flights since 1989. It has flown on over 40 flights of the NASA DC-8 aircraft as part of various campaigns during 1987–1992 studying high-latitude ozone loss. It has also made over 1100 days of ground-based observations since 1985 (covering three full solar cycles) from a dozen different sites, from Antarctica to the Arctic, from sea-level to 3.8 km altitude.
3. Balloon spectra A solar occultation of balloon spectra measured by the JPL MkIV FTS over Alaska in May 1997 was used for this work. This particular balloon flight was chosen because it was part of the Photochemistry of Ozone Loss in the Arctic Region In Summer (POLARIS) campaign [10], which aimed to study the summertime ozone depletion from NOx chemistry. In addition to balloon flights, the NASA ER-2 aircraft made dozens of flights over Alaska. There were also frequent ground-based remote-sensing observations (e.g. [22]). Thus the atmosphere over Alaska was exceptionally well characterized during this period. For example, the MkIV balloon profiles of various atmospheric gases were compared with coincident in situ measurements by instruments on board the NASA ER-2 aircraft [23]. The MkIV balloon spectra cover the 650–5650 cm 1 region at 0.01 cm 1 resolution and tangent altitudes from 10 to 38 km. The sunrise spectra used in this study were ratioed by a MkIV high-sun (35° solar zenith angle) spectrum measured around solar noon from 39 km altitude during a Sep 1996 balloon flight. Since the MkIV instrument was unchanged between these two flights, and the Mid Infra-Red (MIR) solar spectrum changes little over 8 months, the ratioing removes spectral features arising from the sun or the instrument.
4. Spectral fitting methodology The balloon spectra were fitted using the four latest versions of the HITRAN linelist: HITRAN 2000 [17], HITRAN 2004 [16], HITRAN 2008 [18], and HITRAN 2012 [19]. For gases not present in the main HITRAN linelist, but for which measured absorptions coefficients are provided on the HITRAN website (e.g., CFC-11, CFC-12, HCFC-22, etc.), pseudo-line-lists derived from those absorption spectra were used [http://mark4sun.jpl.nasa.gov/pseudo.html]. These derived pseudo-line-lists were unchanged between
325
the four HITRAN versions, as was everything else (spectra, software, atmospheric models, etc.). So the differences in the spectral fits result solely from changes to the line-byline portion of HITRAN. The spectral fitting was performed with the GFIT code, a non-linear least-squares algorithm developed at JPL that scales the atmospheric gas volume mixing ratios (vmrs) to fit calculated spectra to those measured. For balloon observations, the atmosphere was discretized into 100 layers of 1 km thickness. Absorption coefficients were computed line-by-line assuming a Voigt lineshape. The line mixing option was not used in this work. Sen et al. [20] provide a more detailed description of the use of the GFIT code for retrieval of vmr profiles from MkIV balloon spectra. GFIT was previously used for the Version 3 analysis [7] of spectra measured by the Atmospheric Trace Molecule Occultation Spectrometer (ATMOS), and is currently used for analysis of TCCON spectra [26]. Fig. 1 lists the 112 fitted windows, and the adjusted gases in each. Gases not listed (e.g. O2) are still included in the calculation, but their vmr profile is not adjusted. Isotopologues are given different a priori vmr profiles to account for any atmospheric fractionation. A frequency shift was also fitted for each window, which means that line position errors affecting an entire window will not impact the Root Mean Squared (rms) fitting residuals. However, line position errors specific to particular transitions, or which differ between overlapping bands of different gases, will increase the rms residual. We note that these windows are broader than those currently adopted by the Network for Detection of Atmospheric Change Infra-Red Working Group (NDACC IRWG) in the MIR, but similar to those used by TCCON in the Near Infra-Red (NIR). The MIR windows are generally narrow to avoid troublesome interfering features, but we expect them to be widened in the future as the spectroscopy improves.
5. Results Fig. 1 shows the rms spectral residuals (observed calculated) achieved for each window, plotted versus the central wavenumber of that window. Three sample tangent altitudes are illustrated: 10, 20 and 31 km. The size of the horizontal-bars represents the width of the fitted window. In total, 112 windows were fitted across the 670– 5600 cm 1 region, with widths ranging from 4 to 114 cm 1, the broader windows being necessary to bridge across regions, such as the v3 band of CO2 which is completely saturated at lower altitudes. The different colors represent the four different linelists. In windows where there was no change in the rms, intermediate colors are seen (e.g., orangeþgreen¼ brown). The top panel shows results from 31 km tangent altitude. Here the lowest rms values represent the measurement noise and are found in "window" regions where absorption is very weak e.g., 800–950, 1150–1250, and 2400–2700 cm 1. The bottom panel represents 10 km altitude. Here the smallest rms residuals are found in a few saturated regions (i.e., no photon flux), e.g. 660–740, 1500–1600, 2200–2400, and 3700–3900 cm 1. The largest residuals occur at altitudes
326
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
Fig. 1. RMS spectral residuals obtained from fitting MkIV balloon spectra with the four latest versions of HITRAN. 112 different windows were fitted covering the entire 670–5600 cm 1 region. Top panel shows results from 31 km tangent altitude spectrum, middle panel from 20 km, and bottom panel from 10 km. The different colors denote the different HITRAN versions. The noise level is about 0.2%, which hides spectroscopic problems smaller than this.
where the problematic absorptions are strong, but not saturated. In the vast majority of cases, the smallest/best rms residuals are obtained with the 2012 linelist, and the worst with the HITRAN 2000 linelist. There are, however, some exceptions to this, which are investigated further in the remainder of the paper because they are instructive. Fig. 2 shows the rms residual from eight selected windows plotted versus tangent altitude. This uses the same data as shown in Fig. 1, but provides an orthogonal cut through it. Only in panels (d) and (g) does HITRAN 2012 provide unambiguously the best residuals at all altitudes. In panel (a) it provides the same residuals as all other editions. In panel (f) it provides the best residuals except above 28 km. In panel (b) it provides slightly worse residuals than HITRAN 2008, while in panels (c), (e) and (h) at least one of the previous editions provides significantly better residuals at all altitudes. Fig. 2a shows the altitude variation of the rms residual in the window centered at 1334 cm 1, where absorption is dominated by HNO3 above 12 km. There have been only minor changes to the spectroscopy here since HITRAN
2000. Fig. 3 shows an example of a spectral fit using HITRAN 2012 in this window. Residuals exceeding 20% can be seen in the 1316–1325 cm 1 region of the P-branch. The inset shows that these residuals are spaced by about 0.4 cm 1, which is characteristic of many HNO3 transitions, and are due to some of the calculated HNO3 absorptions being much larger than those measured. The offending lines have intensity uncertainty codes of 3 in HITRAN 2012, meaningZ20%. Fig. 2b shows the altitude variation of the rms residuals in the 2752 cm 1 window where O3 and CH4 are the main absorbers. At all altitudes, the HITRAN 2012 linelist produces slightly worse fits (larger rms residuals) than HITRAN 2008; this is most noticeable in the 25–30 km altitude range. The bulges in the HITRAN 2000/4 residuals around 20 km are due to O3 and HCl problems, which were largely fixed with HITRAN 2008. Fig. 4 shows a spectral fit in this region using HITRAN 2012, revealing large antisymmetrical residuals implying line position errors. The residual at 2742.3 cm 1 is due to a 0.001 cm 1 CH4 line position error and was not present in fits using HITRAN 2008. The position uncertainty code for this line is 4,
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
327
Fig. 2. RMS spectral fitting residuals plotted as a function of tangent altitude for eight selected windows. Different colored traces represent different versions of HITRAN linelists. Named gas is dominant absorber in each window. The variation of the rms residuals as a function of altitude contains information about their cause. Residuals peaking at 20 km, such as in panels (a,d) are caused by a stratospheric gas (O3, HNO3). Residuals that decrease steadily with altitude, such as in panels (b,c,e,f) indicate problems with the spectroscopy of a tropospheric gas (e.g., CH4 or N2O). Residuals that increase sharply below the 11 km tropopause, such as in panel (h), indicate spectroscopic problems with H2O.
meaning 0.0001–0.001 cm 1. The residual at 2761.4 cm 1 is due to an O3 line that is a resonance transition. A 0.002 cm 1 position error for this line is present in all
HITRAN versions after 2000, when the line had the correct position. This error exceeds the position error code of 4 (0.0001–0.001 cm 1).
328
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
Fig. 3. Spectral fit to a 20 km tangent altitude spectrum in the 1307–1360 cm 1 window containing the ν3 HNO3 band, whose Q-branch is located at 1325 cm 1. In the main panel, the black diamond symbols represent the measured spectrum, the black line the fitted calculation, and the green line the HNO3 contribution. The upper panel shows the residuals (measured-calculated). The inset shows more detail of the 1317.1 1318.5 cm 1 region where residuals exceed 25%. Although H2O, N2O and CH4 also absorb strongly in this window, they are not responsible for the large residuals, which peak at 20 km (Fig. 2), confirming HNO3 as the culprit.
Fig. 4. Spectral fit to the 2752 cm 1 window at 31 km tangent altitude shows two large anti-symmetrical residuals. The residual at 2742.3 cm 1 is due to a CH4 line with a 0.001 cm 1 position error and was not present in fits using HITRAN 2008. The residual at 2761.4 cm 1 was present in HITRAN 2004/8/12, but not HITRAN 2000, and is due to an O3 resonance transition with a 0.002 cm 1 position error.
Fig. 2c shows the altitude variation of the rms residuals in the 2827 cm 1 window, where residuals are dominated by CH4. HITRAN 2008 and 2012 give virtually identical fits, but not as good as HITRAN 2004. This situation of the 2004 linelist giving the best fits prevails over the 2500– 3000 cm 1 region at the lower altitudes. This explains why some NDACC–IRWG investigators continue using the earlier HITRAN linelists for this important region. Fig. 2d shows the altitude variation of the rms residuals of a window centered at 3540 cm 1. Large residuals peak at 20 km. This spectral region is characterized by strong
absorption by CO2 and HNO3. To the best of our knowledge no laboratory studies of the ν1 HNO3 band have yielded a linelist. Its absence from all HITRAN versions gives rise to large rms residuals at 20 km where the HNO3 concentration peaks. The improvement from HITRAN 2000 to 2012 is due to the CO2 spectroscopy, which has been relentlessly upgraded over the years. Fig. 5 shows spectral fits in this window using HITRAN 2012. The residuals clearly show the P, Q, and R-branch structure to be expected of HNO3. The missing manifolds can clearly be seen in the inset, and have a spacing of about 0.4 cm 1.
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
329
Fig. 5. Spectral fits at 23 km in the 3528–3574 cm 1 window. Although the main absorbers here appear to be CO2 and H2O, the entire HNO3 ν1 fundamental band is actually missing from HITRAN, which gives rise to the large downward residuals. The P-, Q-, and R-branch structure can be seen in the residuals, with the Q-branch at 3551.6 cm 1. Many of the HNO3 manifolds are hidden behind saturated CO2 and H2O lines, especially in the R-branch. The inset shows the seven strongest missing P-branch manifolds, the one at 3538.98 cm 1 being obscured by a saturated CO2 line.
Fig. 6. Spectral fit from 4050 to 4140 cm 1 at 10 km altitude. The main absorber here is CH4 and the huge residual at 4115.659 cm 1 is due to a missing CH4 line (see inset). This line was present in HITRAN 2008, so its omission in 2012 was probably a mistake.
Fig. 2e shows the altitude variation of the rms residuals in the 4096 cm 1 window, where absorption is dominated by CH4. The HITRAN 2000 linelist gave reasonable spectral fits at all altitudes, but the HITRAN 2004 and 2008 linelists, which were identical in this region, represented a large deterioration. HITRAN 2012 is much improved, but not quite as good as HITRAN 2000. Fig. 6 show spectral fits to this window using the HITRAN 2012 linelist. The large positive residual at 4115.659 cm 1 is due to a missing CH4 line, which was present in the 2008 linelist. Part of the motivation for this work is to be able in the future to quickly identify such avoidable mistakes. With this missing line inserted, HITRAN 2012 would give better spectral fits than any previous linelist. Fig. 2f shows the altitude dependence of the rms residual in the 4190 cm 1 window where absorption is dominated by CH4. This window has worsened since
HITRAN 2008 at altitudes above 28 km due to the poorer line positions (see Fig. 7). Below 20 km, however, HITRAN 2012 produces much-improved residuals due to the inclusion of 13CH4 lines based on Niederer et al., [11], which were absent from earlier HITRAN versions. Since the 13 CH4 lines are much weaker than the 12CH4 lines, their absorptions only become apparent (i.e., exceed the noise level) below 20 km. Fig. 2g shows the altitude variation of the rms residuals in the 4910 cm 1 window, where absorption is dominated by CO2. Indeed, part of this window overlaps the strong CO2 band used by GOSAT [27] and OCO-2 [28] . It is seen that a modest improvement occurred between HITRAN 2000 and 2004, which was due to the addition of the O3 band, then a huge improvement in HITRAN 2008 related to the CO2 line positions. A modest improvement was then effected in HITRAN 2012 relating mainly to improved CO2
330
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
Fig. 7. Spectral fit to the 4190 cm 1 region, which is dominated by CH4. At the higher altitudes (31 km shown above) the HITRAN 2012 fits are worse than those using HITRAN 2004 or 2008 due to a worsening of the line positions. This is particularly noticeable in the ν3 þ ν4 Q-branch around 4215–4219 cm 1 with peak residuals of 6% versus 4% with HITRAN 2004/8.
Fig. 8. Spectral fit to a 12 km altitude spectrum in the 5360–5420 cm 1 window where absorption is dominated by H2O. The large positive residual is due to a factor 4 intensity over-estimate of the line at 5390.339 cm 1 (see inset). This large residual was not present in fits using HITRAN 2004 or 2008.
line intensities. This window is typical of the vast majority, with improvements being observed with each successive HITRAN version. Fig. 2h shows the altitude variation of the rms residuals in the 5390 cm 1 window which is dominated by H2O. After a big improvement in HITRAN 2004, the entire 5200–5500 cm 1 region has regressed in the past two HITRAN versions. Fig. 8 shows a fit with the HITRAN 2012 linelist. The large residual at 5390.339 cm 1, which reaches 25% at 12 km altitude (due to a factor 4 intensity over-estimate), was not present in fits using HITRAN 2004 or 2008. In HITRAN 2012 the H2O line intensities were based on the ab initio calculations of [9]. Although these are generally an improvement, there are occasional lapses. The line at 5290.339 cm 1 is one of 14 "rogue" H2O lines
that have been identified in the 5000–7500 cm 1 region of HITRAN 2012. They probably relate to resonance transitions. If these rogue H2O lines were empirically corrected, the resulting linelist would be better than any previous HITRAN version.
6. RMS Summary Table 1 shows the rms spectral residual achieved for each window, averaged over all altitudes, for each of the four HITRAN linelists. The smallest rms value for each window is colored blue, versus black for the higher values. Out of 112 fitted windows the HITRAN 2000 linelist produces the best fits in two cases; the 2004 linelist in 25
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
331
Table 1 The 112 fitted windows: central wavenumbers, widths, adjusted gases, and vertically-averaged rms residuals (%) using the HITRAN 2000/04/08/12 linelists. For each window, the smallest rms value is colored blue. The HITRAN 2000 linelist produces the best fits in 2 cases; the 2004 linelist in 25 cases; the 2008 linelist in 20 cases, and the 2012 linelist in 65 cases.
332
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
Table 1 (continued )
cases; the 2008 linelist in 20 cases, and the 2012 linelist in 65 cases. It is perhaps surprising that the 2004 linelist is still best in so many windows. Table 1 also shows the central wavenumber of each window, its width, and the fitted gases. Gases not listed in Table 1 are still included in the fitting, but their vmr profiles are not scaled. Fig. 9 plots the information in Table 1: average rms versus central wavenumber of each window. By averaging over all altitudes, we ensure that an altitude is included where any spectroscopic errors are exposed: low altitudes for the weak absorbers, high altitudes for the strong absorbers.
Fig. 10 shows the average rms residual (over all windows) plotted versus altitude. It shows a significant overall improvement with each new HITRAN version. The general decrease with altitude reflects the weakening of the spectral absorption. The rapid increase in RMS below the tropopause (11 km) is mainly due to H2O. The slight bulge around 16–21 km is primarily due to HNO3.
7. Discussion Atmospheric solar absorption Fourier Transform Infra-Red (FTIR) spectra provide an excellent test of spectroscopy. The
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
333
Fig. 9. Vertically-averaged RMS spectral fitting residuals, plotted versus wavenumber. Upper and lower panels show same data, but lower panel is zoomed and clipped to reveal more detail in low RMS values.
Fig. 10. Wavenumber-averaged RMS spectral fitting residuals, plotted versus altitude, showing the relentless improvement at all altitudes with each new HITRAN version. The rapid increase below the tropopause (11 km) is mainly due to H2O. The slight bulge around 16–21 km is primarily due to HNO3. The general decrease with altitude reflects the weakening of the spectral lines.
fact that the sun is so bright allows the measurement of high resolution and high Signal-to-Noise Ratio (SNR) spectra covering a wide bandwidth. For example, the JPL MkIV FTS covers the 650 to 5650 cm 1 simultaneously at 0.01 cm 1 resolution with a SNR of 400:1, and covers the 5–40 km altitude range. These attributes facilitate identification and diagnosis of problems in regions with high rms residuals. They also allow assessment of the consistency of different bands (not shown in this work). It is perhaps surprising that a single transition with a 0.001 cm 1 position error (e.g., Fig. 4) produces a discernable degradation in the rms fitting residual over a 57 cm 1 wide window cluttered with absorption lines,
measured by a spectrometer of 0.01 cm 1 resolution (ten times the position error). From a balloon platform, a large range of atmospheric conditions can be observed in a relatively short time span (30 min) during which the instrument and atmosphere undergo little change. Observing sunset at 9 km (300 mbar) tangent altitude represents 2–3 orders of magnitude increase in atmospheric airmass compared with viewing the sun at low airmass from float altitude (38 km; 3 mbar). This allows absorption bands of widely different intensities to be checked in a consistent manner. Furthermore, the temperature of the observed airmass also varies from 200 to 250 K, depending on
334
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
altitude, providing a check on the lower state energies (E") of the observed transitions. In spectra covering such a wide range of conditions, observed at high resolution and SNR, it is difficult for spectroscopic problems to remain hidden. And the problems that do remain hidden are probably not important for remote sensing applications. Linelists based on theoretical models can occasionally produce very poor results for transitions that are in resonance. Although this affects only a tiny fraction of the lines, the resulting residuals dominate the spectral fits, making the new theoretical linelist appear poorer than its predecessor when in fact it is better for the vast majority of lines. Empirically adjusting these occasional lapses in the theoretical linelists is clearly the way forward, but who should be doing these adjustments? Clearly the spectroscopists that generated the linelist are the best placed to do so, understanding the strengths and weaknesses of their measurements and predictions. And having them doing it would avoid multiple, lessinformed, end-users from doing the same job. Yet many spectroscopists are reluctant to perform ad hoc empirical adjustments to their linelist, preferring instead to improve the underlying theory, which is the correct long-term solution. But this doesn't help the end-user faced with the short-term problem of what to do about a large systematic residual in their desired fitting window. Should they empirically adjust the linelist, or narrow their window to exclude the large residuals, or somehow modify their fitting software to deweight large residuals? Something has to be done because standard least-squares fitting techniques are highly sensitive to large residuals since they try to minimize the sum of their squares. Even though the large residual may not be due to the target gas of interest, or may not even overlap any target gas absorptions, it can still strongly influence the retrieval through cross-talk with other retrieved variables (non-target gases, continuum level, frequency shift). We therefore advocate a more rigorous checking of any new linelist, comparing it with its predecessors by fitting well-characterized atmospheric and laboratory spectra and investigating windows where the rms fitting residual worsened. This will catch the more important preventable errors, which can then be fixed by empirical adjustment, or by substitution from an earlier linelist version. To be clear, we do not advocate adjusting spectroscopic linelists based on fits to atmospheric spectra – that is better done with laboratory spectra. Atmospheric spectra are, however, invaluable for pointing out where the spectroscopy is inadequate, and which gases are likely responsible. Although not illustrated in this paper, when systematic residuals are seen in fits to MkIV atmospheric spectra, those same residuals are also seen in fits to laboratory spectra made under similar conditions (absorption depth, pressure, temperature) in over 90% of cases. This suggests that the HITRAN linelist could be significantly improved using existing laboratory spectra by empirical adjustment of the culpable parameters. Several of the largest fitting residuals can be attributed to inadequacies in the HNO3 spectroscopy. Besides the completely missing ν1 fundamental band centered at 3551 cm 1, and several obvious missing combination bands (e.g. 2645, 4107 cm 1), the ν3 and ν2 fundamentals at 1325 and 1725 cm 1 respectively, also have serious problems. Although the very recent work by Perrin et al. [13] will improve the
situation in the ν3 band if incorporated into HITRAN, HNO3 will still remain the dominant source of stratospheric fitting errors. Although it could be argued that the ν1, ν2, and ν3 bands of HNO3 are not needed for atmospheric HNO3 measurement, since the overlapping ν5 and 2ν9 HNO3 bands at 850–950 cm 1 are sufficient, this is a self-fulfilling prophecy. If the ν1, ν2, and ν3 spectroscopy were better, these bands would become more widely used. Since the ν2 and ν3 bands are stronger than the ν5 and 2ν9 bands, better spectroscopy would allow HNO3 profiles to be retrieved more accurately under low-HNO3 conditions, e.g. at high altitudes and in the tropics. In addition, the ν1 band centered at 3551 cm 1 is ideally situated for in situ measurements of atmospheric HNO3 by room-T tunable diode lasers. Unfortunately at this point very little reliable experimental information exists on the ν1 band in the literature, preventing derivation of the important spectroscopic constants that could be used to model this band in the atmosphere. More high-resolution laboratory experiments are desperately needed for this band. Poor HNO3 spectroscopy also thwarts measurement of trace gases whose main absorption bands overlap HNO3 bands, such as SO2 at 1350 cm 1, OH at 3570 cm 1, and HO2 at 3440 cm 1. The construction of the methane line list for HITRAN 2012 is described in Brown et al. [2]. While it represents some of the major advances in methane spectroscopy in recent times, it is clearly in need of very careful revision. This process is already underway within a dedicated task group of the HITRAN advisory committee. The most straightforward fix is to restore the lines that were inadvertently dropped from the database such as the line at 4115.659 cm 1 (Fig. 6). Most of the errors, associated with the line positions (including those described in Figs. 4 and 7) seem to be associated with the employment of two not entirely compatible software packages MIRS [12] and STDS [25] that were used to fit and predict transitions. This approach was employed in Albert et al. [1] and Daumont et al. [3], which are the origin of many HITRAN 2012 transitions involving Pentad and Octad states. A thorough investigation of this matter is planned for the next HITRAN release. Spectroscopy of water vapor will also be a subject of substantial update although as it was mentioned, with the exception of a few lines, HITRAN 2012 was already a large step forward. Apart from other improvements the addition of the missing lines of the deuterated isotopologues of water in the NIR region is planned. Considering that there will be an extensive combination of experimental and theoretical results, there is much scope for human error and therefore these validations will be most useful.
8. Future work The current analysis, being confined to altitudes above 9 km, emphasizes stratospheric gases and exposes errors in the positions and intensities of spectral transitions. The next step would be to analyze ground-based spectra where H2O plays a much more important role and where errors in line widths and shifts become more important. This would also allow the spectral coverage to be extended above 5700 cm 1
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
by using TCCON and Kitt Peak ground-based solar spectra in addition to MkIV spectra. Extending this analysis to lower altitudes will require inclusion of line mixing [15,4,8] effects which above 10 km are confined to relatively narrow spectral regions and therefore don't have much impact on the mean rms residual of a window tens of cm 1 wide. At lower altitudes, however, the impacts of line-mixing will be broader and stronger and will have to be included. Sub-dividing some of the windows, such that we would end up with more the 112 windows with a narrower average width, would provide more specificity in the identification of regions where the rms fits worsen. For example, in a wide window it is possible for improvements in the spectroscopy of gas A to mask a worsening of the gas B spectroscopy, such that the overall rms still improves. Using the sun as a source allows the MkIV balloon spectra to cover a very broad spectral region simultaneously with high resolution and SNR. The gas amounts retrieved from various windows provide information about the band-to-band consistency of the spectroscopy. Such information is often difficult to extract from laboratory spectra, where the dimmer sources often necessitate a narrower spectral bandwidth, preventing simultaneous measurement of widely separated bands. In a future work we will examine the band-to-band consistency of the retrieved gas amounts. Finally, we need to evaluate HITRAN 2016 when it comes out later this year. As a result of the work shown in this paper, this can now be done quickly (a few days) since the windows and spectra are already defined and the scripts for running the GFIT code and making the plots already exist.
9. Concluding remarks Currently, some atmospheric remote-sensing scientists maintain their own linelists, founded on HITRAN, but perhaps not the latest version for every gas of every band. When a new HITRAN version is released, these scientists test it on the spectral regions they care about. If it is better, they adopt it for those regions where it is better, giving rise to a hybrid linelist. Sometimes they will do a partial adoption, including some gases from the latest HITRAN, but not others. Often a new linelist will be worse initially than its predecessor due to a few isolated large residuals (e.g. H2O: 5000–7500 cm 1). But after empirical correction of these, based on fits to laboratory spectra, significant improvements are seen. The linelist used by the TCCON community [24] is a good example of this empirical, "greatest hits", approach to spectroscopic database management. In this particular case, the criteria for deciding whether a new linelist is better or worse is two-fold: the rms fitting residuals, and the consistency of gas amounts retrieved from different windows. Since this quality control work will be re-done anyway after release of HITRAN 2016, to decide which parts to adopt, we thought that it would be more efficient to do it before the public release, and hence address the easy-to-fix problems before the linelist becomes publically available, sparing the end-users the hassle of dealing with regions where their fits
335
get worse. If the new HITRAN versions were more dependably superior than their predecessors, most end-users would abandon maintaining their own hybrid linelists. We advocate assessing new linelists by fitting wellcharacterized atmospheric spectra (in addition to laboratory spectra). This allows identification of regions where the new linelist performs worse than its predecessor and usually reveals the cause of the deterioration. The thornier issue is what to do about these problems, once identified. There are cases where the need to fix the linelist is obvious and uncontroversial. For example the resonance O3 line at 2671.4 cm 1 can be fixed by empirically adjusting its position based on fits to laboratory spectra. And the missing CH4 line at 4115.659 cm 1 can be fixed by substitution from the previous HITRAN version. But this type of work is error prone, hence the need for performance benchmarks against which to test any empirical adjustments. Spectroscopic issues such as the missing HNO3 bands will require longer-term solutions. Finally we recognize that the Earth's atmosphere is not the sole use for the HITRAN linelist. It is also used for radiativetransfer calculations in other environments, such as the atmospheres of other planets, or industrial processes. It is possible, for example, that a linelist update that makes fits slightly worse for Earth spectra, much improves fits to spectra of Titan's atmosphere. So a balance has to be struck between the needs of the diverse end-users of HITRAN.
Acknowledgements We thank NASA's Upper Atmosphere Research Program who funded the JPL MkIV instrument through Grant NNH12ZDA001N-UACO. The Columbia Scientific Balloon Facility (CSBF) who launched the balloon and recovered the MkIV payload. The HITRAN team gratefully acknowledges support from the NASA AURA program Grant NNX14AI55G. We also thank the Worldwide spectroscopy community whose work is encapsulated in the HITRAN linelists. Part of this research was performed at the Jet Propulsion Laboratory, California of Technology, under contract with NASA.
References [1] Albert S, Bauerecker S, Boudon V, Brown LR, Champion J-P, Loëte M, et al. Global analysis of the high resolution infrared spectrum of methane 12CH4 in the region from 0 to 4800 cm 1. Chem Phys 2009;356:131–46. [2] Brown LR, Sung K, Benner DC, Devi VM, Boudon V, Gabard T, et al. Methane line parameters in the HITRAN2012 database. J Quant Spectrosc Radiat Transf 2013;130:201–19. http://dx.doi.org/10.1016 /j.jqsrt.2013.06.020. [3] Daumont L, Nikitin AV, Thomas X, Régalia L, Von der Heyden P, Tyuterev VG, et al. New assignments in the 2 μm transparency window of the 12CH4 Octad band system. J Quant Spectrosc Radiat Transf 2013;116:101–9. http://dx.doi.org/10.1016/j.jqsrt.2012.08.025. [4] Edwards David P, Strow L Larrabee. Spectral line shape considerations for limb temperature sounders. 96.D11. J Geophys Res Atmos 1991:20859–68. [5] Gordon IE, Kassi S, Campargue A, Toon GC. First identification of the electric quadrupole transitions of oxygen in solar and laboratory spectra. J Quant Spectrosc Radiat Transf 2010;111: 1174–83. http://dx.doi.org/10.1016/j.jqsrt.2010.01.008.
336
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
[6] Gordon IE, Rothman LS, Toon GC. Revision of spectral parameters for the B- and γ-bands of oxygen and their validation against atmospheric spectra. J Quant Spectrosc Radiat Transf 2011;112: 2310–22. http://dx.doi.org/10.1016/j.jqsrt.2011.05.007. [7] Irion FW, Gunson MR, Toon GC, Chang AY, Eldering A, Mahieu E, et al. Atmospheric trace molecule spectroscopy (ATMOS) experiment version 3 data retrievals. Appl Opt 2002;41(33):6968–79. [8] Levy A, Lacome N, Chackerian C. Collisional line mixing. In: Rao KN, Weber A, editors. Spectroscopy of the earth's atmosphere and interstellar medium. Boston: Academic Press; 1992. p. 261–337. [9] Lodi L, Tennyson J, Polyansky OL. A global, high accuracy ab initio dipole moment surface for the electronic ground state of the water molecule. J Chem Phys 2011;135 034113-10. [10] Newman PA, Fahey DW, Brune WH, Kurylo MJ, Kawa SR. Preface [to special section on photochemistry of Ozone loss in the arctic region in summer (POLARIS)]. J Geophys Res 1999;104(D21): 26481–95. http://dx.doi.org/10.1029/1999JD900832. [11] Niederer HM, Wang XG, Carrington T, Albert S, Bauerecker S, Boudon V, et al. Analysis of the rovibrational spectrum of 13CH4 in the Octad range. J Mol Spectrosc 2013;291:33–47. http://dx.doi.org/ 10.1016/j.jms.2013.06.003. [12] Nikitin A, Champion JP, Tyuterev VG. The MIRS computer package for modeling the rovibrational spectra of polyatomic molecules. J Quant Spectrosc Radiat Transf 2003;82:239–49. http://dx.doi.org/ 10.1016/S0022-4073(03)00156-0. [13] Perrin A, Flaud J-M, Ridolfi M, Vander Auwera J, Carlotti M. MIPAS database: new HNO3 line parameters at 7.6 μm validated with MIPAS satellite measurements. Atmos Meas Tech Discuss 2015;8: 11643–71. http://dx.doi.org/10.5194/amtd-8-11643-2015. www.atmosmeas-techdiscuss.net/8/11643/2015/. [15] Rosenkranz PW. IEEE Trans Antennas Propag 1975;AP 23:498. [16] Rothman LS, Jacquemart D, Barbe A, Chris Benner D, Birk M, Brown LR, et al. The HITRAN 2004 molecular spectroscopic database. J Quant Spectrosc Radiat Transf 2005;96:139–204. [17] Rothman LS, Barbe A, Chris Benner D, Brown LR, Camy-Peyret C, Carleer MR, et al. The HITRAN molecular spectroscopic database: edition of 2000 including updates through 2001. J Quant Spectrosc Radiat Transf 2003;82:5–44.
[18] Rothman LS, Gordon IE, Barbe A, Chris Benner D, Bernath PF, Birk M, et al. The HITRAN 2008 molecular spectroscopic database. J Quant Spectrosc Radiat Transf 2009;110:533–72. [19] Rothman LS, Gordon IE, Babikov Y, Barbe A, Chris Benner D, Bernath PF, et al. The HITRAN2012 molecular spectroscopic database. J Quant Spectrosc Radiat Transf 2013;130:4–50. [20] Sen B, Toon GC, Blavier J-F, Fleming EL, Jackman CH. Balloon-borne observations of mid-latitude fluorine abundance. J Geophys Res 1996;101(D4):9045–54. [21] Toon GC. The JPL MkIV interferometer. Opt Photonics News 1991;2: 19–21. [22] Toon GC, J.-F. Blavier B Sen, Salawitch RJ, Osterman GB, Notholt J, Rex M, et al. Ground-based observations of Arctic O3 loss during spring and summer 1997. J Geophys Res 1999;104(D21): 26497–510. http://dx.doi.org/10.1029/1999JD900745. [23] Toon GC, J.-F. Blavier B Sen, Margitan JJ, Webster CR, May RD, Fahey DW, et al. Comparison of MkIV balloon and ER-2 aircraft profiles of atmospheric trace gases. J Geophys Res 1999;104(26):779–26,790. [24] Toon GC. Telluric line list for GGG2014. TCCON data archive, hosted by the Carbon dioxide information analysis center. U.S.A: Oak Ridge National Laboratory; http://dx.doi.org/10.14291/tccon.ggg2014.atm. R0/XXXXXXX. [25] Wenger C, Champion JP. Spherical top data system (STDS) software for the simulation of spherical top spectra. J Quant Spectrosc Radiat Transf 1998;59:471–80. http://dx.doi.org/10.1016/S0022-4073(97) 00106-4. [26] Wunch D, Toon GC, Blavier J-FL, Washenfelder RA, Notholt J, Connor BJ, et al. The total carbon column observing network, Philosophical Transactions of the Royal Society. Ser A Math Phys Eng Sci 2011;369 (1943):2087–112. http://dx.doi.org/10.1098/rsta.2010.0240. [27] Kuze A, Suto H, Nakajima M, Hamazaki T. Thermal and near infrared sensor for carbon observation Fourier-transform spectrometer on the Greenhouse Gases Observing Satellite for greenhouse gases monitoring. Appl. Opt. 2009;48(35):6716–33. [28] Crisp D, Miller CE, DeCola PL. NASA orbiting carbon observatory: Measuring the column averaged carbon dioxide mole fraction from space. J. Appl. Remote Sens. 2008;2:23–508.
Contents lists available at ScienceDirect
Journal of Quantitative Spectroscopy & Radiative Transfer journal homepage: www.elsevier.com/locate/jqsrt
HITRAN spectroscopy evaluation using solar occultation FTIR spectra Geoffrey C. Toon a,n, Jean-Francois Blavier a, Keeyoon Sung a, Laurence S. Rothman b, Iouli E. Gordon b a b
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, United States Harvard-Smithsonian Center for Astrophysics, Cambridge, MA, United States
a r t i c l e in f o
abstract
Article history: Received 22 April 2016 Received in revised form 22 May 2016 Accepted 22 May 2016 Available online 11 June 2016
High resolution FTIR solar occultation spectra, acquired by the JPL MkIV Fourier transform spectrometer from balloon, covering 650–5650 cm 1 at 0.01 cm 1 resolution, are systematically analyzed using the last four versions of the HITRAN linelist (2000, 2004, 2008, 2012). The rms spectral fitting residuals are used to assess the quality and adequacy of the linelists as a function of wavenumber and altitude. Although there have been substantial overall improvements with each successive version of HITRAN, there are nevertheless a few spectral regions where the latest HITRAN version (2012) has regressed, or produces residuals that far exceed the noise level. A few of these instances are investigated further and their causes identified. We emphasize that fitting atmospheric spectra, in addition to laboratory spectra, should be part of the quality assurance for any new linelist before public release. & 2016 Elsevier Ltd. All rights reserved.
Keywords: Spectroscopy Infrared Atmosphere Remote Sensing HITRAN
1. Introduction The HITRAN linelist [19] is without doubt the most widely used spectroscopic database and underpins most atmospheric remote-sensing experiments. As our understanding of the terrestrial atmosphere improves, the accuracy required of new measurements becomes more stringent. Hence the requirements on the accuracy and completeness of the spectroscopy become inexorably more challenging. It is usually impossible to fit high-resolution Infra-Red solar spectra (high SNR) down to their noise level, at least below 25 km tangent altitude. Residuals are typically dominated by systematic errors arising from defects in the assumed atmospheric temperature (T), pressure (P), and volume mixing ratio (vmr) profiles, the observation geometry (pointing), the instrumental response (e.g., instrument line n
Corresponding author. E-mail address: [email protected] (G.C. Toon).
http://dx.doi.org/10.1016/j.jqsrt.2016.05.021 0022-4073/& 2016 Elsevier Ltd. All rights reserved.
shape, zero-level-offsets, channel fringes, phase errors, aliasing, ghosts), and spectroscopic inadequacies. For spectra measured with a well-calibrated instrument under wellknown atmospheric conditions, the first three types of systematic errors can usually be minimized, revealing the underlying spectroscopic problems. In the past we successfully used the Total Carbon Column Observing Network (TCCON) Fourier Transform Spectrometer (FTS) in Park Falls, WI for validating spectroscopic parameters of molecular oxygen in different spectral windows [5,6]. In the first paper not only the parameters of strong magnetic dipole lines in the 1.27 μm band were validated but the electric quadrupole lines were identified for the first time in this band. Previously they were thought to be orders of magnitude weaker. Similar evaluations were done in limited spectral regions of water, methane and carbon dioxide prior to the release of the HITRAN 2012 database. In this paper, we document the quality of fits to atmospheric spectra collected by the JPL MkIV interferometer and
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
fitted using the JPL Gas Fitting software (GFIT), to provide a benchmark for assessment of future linelists (e.g., HITRAN 2016) and to highlight gases and spectral regions where further spectroscopic work would be beneficial.
2. MkIV instrument The MkIV FTS is a double-passed FTIR spectrometer designed and built at JPL in 1984 for atmospheric observations [21]. It covers the entire 650–5650 cm 1 region simultaneously with two detectors: a HgCdTe photoconductor covering 650–1800 cm 1 and an InSb photodiode covering 1800–5650 cm 1. The MkIV instrument has flown 24 balloon flights since 1989. It has flown on over 40 flights of the NASA DC-8 aircraft as part of various campaigns during 1987–1992 studying high-latitude ozone loss. It has also made over 1100 days of ground-based observations since 1985 (covering three full solar cycles) from a dozen different sites, from Antarctica to the Arctic, from sea-level to 3.8 km altitude.
3. Balloon spectra A solar occultation of balloon spectra measured by the JPL MkIV FTS over Alaska in May 1997 was used for this work. This particular balloon flight was chosen because it was part of the Photochemistry of Ozone Loss in the Arctic Region In Summer (POLARIS) campaign [10], which aimed to study the summertime ozone depletion from NOx chemistry. In addition to balloon flights, the NASA ER-2 aircraft made dozens of flights over Alaska. There were also frequent ground-based remote-sensing observations (e.g. [22]). Thus the atmosphere over Alaska was exceptionally well characterized during this period. For example, the MkIV balloon profiles of various atmospheric gases were compared with coincident in situ measurements by instruments on board the NASA ER-2 aircraft [23]. The MkIV balloon spectra cover the 650–5650 cm 1 region at 0.01 cm 1 resolution and tangent altitudes from 10 to 38 km. The sunrise spectra used in this study were ratioed by a MkIV high-sun (35° solar zenith angle) spectrum measured around solar noon from 39 km altitude during a Sep 1996 balloon flight. Since the MkIV instrument was unchanged between these two flights, and the Mid Infra-Red (MIR) solar spectrum changes little over 8 months, the ratioing removes spectral features arising from the sun or the instrument.
4. Spectral fitting methodology The balloon spectra were fitted using the four latest versions of the HITRAN linelist: HITRAN 2000 [17], HITRAN 2004 [16], HITRAN 2008 [18], and HITRAN 2012 [19]. For gases not present in the main HITRAN linelist, but for which measured absorptions coefficients are provided on the HITRAN website (e.g., CFC-11, CFC-12, HCFC-22, etc.), pseudo-line-lists derived from those absorption spectra were used [http://mark4sun.jpl.nasa.gov/pseudo.html]. These derived pseudo-line-lists were unchanged between
325
the four HITRAN versions, as was everything else (spectra, software, atmospheric models, etc.). So the differences in the spectral fits result solely from changes to the line-byline portion of HITRAN. The spectral fitting was performed with the GFIT code, a non-linear least-squares algorithm developed at JPL that scales the atmospheric gas volume mixing ratios (vmrs) to fit calculated spectra to those measured. For balloon observations, the atmosphere was discretized into 100 layers of 1 km thickness. Absorption coefficients were computed line-by-line assuming a Voigt lineshape. The line mixing option was not used in this work. Sen et al. [20] provide a more detailed description of the use of the GFIT code for retrieval of vmr profiles from MkIV balloon spectra. GFIT was previously used for the Version 3 analysis [7] of spectra measured by the Atmospheric Trace Molecule Occultation Spectrometer (ATMOS), and is currently used for analysis of TCCON spectra [26]. Fig. 1 lists the 112 fitted windows, and the adjusted gases in each. Gases not listed (e.g. O2) are still included in the calculation, but their vmr profile is not adjusted. Isotopologues are given different a priori vmr profiles to account for any atmospheric fractionation. A frequency shift was also fitted for each window, which means that line position errors affecting an entire window will not impact the Root Mean Squared (rms) fitting residuals. However, line position errors specific to particular transitions, or which differ between overlapping bands of different gases, will increase the rms residual. We note that these windows are broader than those currently adopted by the Network for Detection of Atmospheric Change Infra-Red Working Group (NDACC IRWG) in the MIR, but similar to those used by TCCON in the Near Infra-Red (NIR). The MIR windows are generally narrow to avoid troublesome interfering features, but we expect them to be widened in the future as the spectroscopy improves.
5. Results Fig. 1 shows the rms spectral residuals (observed calculated) achieved for each window, plotted versus the central wavenumber of that window. Three sample tangent altitudes are illustrated: 10, 20 and 31 km. The size of the horizontal-bars represents the width of the fitted window. In total, 112 windows were fitted across the 670– 5600 cm 1 region, with widths ranging from 4 to 114 cm 1, the broader windows being necessary to bridge across regions, such as the v3 band of CO2 which is completely saturated at lower altitudes. The different colors represent the four different linelists. In windows where there was no change in the rms, intermediate colors are seen (e.g., orangeþgreen¼ brown). The top panel shows results from 31 km tangent altitude. Here the lowest rms values represent the measurement noise and are found in "window" regions where absorption is very weak e.g., 800–950, 1150–1250, and 2400–2700 cm 1. The bottom panel represents 10 km altitude. Here the smallest rms residuals are found in a few saturated regions (i.e., no photon flux), e.g. 660–740, 1500–1600, 2200–2400, and 3700–3900 cm 1. The largest residuals occur at altitudes
326
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
Fig. 1. RMS spectral residuals obtained from fitting MkIV balloon spectra with the four latest versions of HITRAN. 112 different windows were fitted covering the entire 670–5600 cm 1 region. Top panel shows results from 31 km tangent altitude spectrum, middle panel from 20 km, and bottom panel from 10 km. The different colors denote the different HITRAN versions. The noise level is about 0.2%, which hides spectroscopic problems smaller than this.
where the problematic absorptions are strong, but not saturated. In the vast majority of cases, the smallest/best rms residuals are obtained with the 2012 linelist, and the worst with the HITRAN 2000 linelist. There are, however, some exceptions to this, which are investigated further in the remainder of the paper because they are instructive. Fig. 2 shows the rms residual from eight selected windows plotted versus tangent altitude. This uses the same data as shown in Fig. 1, but provides an orthogonal cut through it. Only in panels (d) and (g) does HITRAN 2012 provide unambiguously the best residuals at all altitudes. In panel (a) it provides the same residuals as all other editions. In panel (f) it provides the best residuals except above 28 km. In panel (b) it provides slightly worse residuals than HITRAN 2008, while in panels (c), (e) and (h) at least one of the previous editions provides significantly better residuals at all altitudes. Fig. 2a shows the altitude variation of the rms residual in the window centered at 1334 cm 1, where absorption is dominated by HNO3 above 12 km. There have been only minor changes to the spectroscopy here since HITRAN
2000. Fig. 3 shows an example of a spectral fit using HITRAN 2012 in this window. Residuals exceeding 20% can be seen in the 1316–1325 cm 1 region of the P-branch. The inset shows that these residuals are spaced by about 0.4 cm 1, which is characteristic of many HNO3 transitions, and are due to some of the calculated HNO3 absorptions being much larger than those measured. The offending lines have intensity uncertainty codes of 3 in HITRAN 2012, meaningZ20%. Fig. 2b shows the altitude variation of the rms residuals in the 2752 cm 1 window where O3 and CH4 are the main absorbers. At all altitudes, the HITRAN 2012 linelist produces slightly worse fits (larger rms residuals) than HITRAN 2008; this is most noticeable in the 25–30 km altitude range. The bulges in the HITRAN 2000/4 residuals around 20 km are due to O3 and HCl problems, which were largely fixed with HITRAN 2008. Fig. 4 shows a spectral fit in this region using HITRAN 2012, revealing large antisymmetrical residuals implying line position errors. The residual at 2742.3 cm 1 is due to a 0.001 cm 1 CH4 line position error and was not present in fits using HITRAN 2008. The position uncertainty code for this line is 4,
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
327
Fig. 2. RMS spectral fitting residuals plotted as a function of tangent altitude for eight selected windows. Different colored traces represent different versions of HITRAN linelists. Named gas is dominant absorber in each window. The variation of the rms residuals as a function of altitude contains information about their cause. Residuals peaking at 20 km, such as in panels (a,d) are caused by a stratospheric gas (O3, HNO3). Residuals that decrease steadily with altitude, such as in panels (b,c,e,f) indicate problems with the spectroscopy of a tropospheric gas (e.g., CH4 or N2O). Residuals that increase sharply below the 11 km tropopause, such as in panel (h), indicate spectroscopic problems with H2O.
meaning 0.0001–0.001 cm 1. The residual at 2761.4 cm 1 is due to an O3 line that is a resonance transition. A 0.002 cm 1 position error for this line is present in all
HITRAN versions after 2000, when the line had the correct position. This error exceeds the position error code of 4 (0.0001–0.001 cm 1).
328
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
Fig. 3. Spectral fit to a 20 km tangent altitude spectrum in the 1307–1360 cm 1 window containing the ν3 HNO3 band, whose Q-branch is located at 1325 cm 1. In the main panel, the black diamond symbols represent the measured spectrum, the black line the fitted calculation, and the green line the HNO3 contribution. The upper panel shows the residuals (measured-calculated). The inset shows more detail of the 1317.1 1318.5 cm 1 region where residuals exceed 25%. Although H2O, N2O and CH4 also absorb strongly in this window, they are not responsible for the large residuals, which peak at 20 km (Fig. 2), confirming HNO3 as the culprit.
Fig. 4. Spectral fit to the 2752 cm 1 window at 31 km tangent altitude shows two large anti-symmetrical residuals. The residual at 2742.3 cm 1 is due to a CH4 line with a 0.001 cm 1 position error and was not present in fits using HITRAN 2008. The residual at 2761.4 cm 1 was present in HITRAN 2004/8/12, but not HITRAN 2000, and is due to an O3 resonance transition with a 0.002 cm 1 position error.
Fig. 2c shows the altitude variation of the rms residuals in the 2827 cm 1 window, where residuals are dominated by CH4. HITRAN 2008 and 2012 give virtually identical fits, but not as good as HITRAN 2004. This situation of the 2004 linelist giving the best fits prevails over the 2500– 3000 cm 1 region at the lower altitudes. This explains why some NDACC–IRWG investigators continue using the earlier HITRAN linelists for this important region. Fig. 2d shows the altitude variation of the rms residuals of a window centered at 3540 cm 1. Large residuals peak at 20 km. This spectral region is characterized by strong
absorption by CO2 and HNO3. To the best of our knowledge no laboratory studies of the ν1 HNO3 band have yielded a linelist. Its absence from all HITRAN versions gives rise to large rms residuals at 20 km where the HNO3 concentration peaks. The improvement from HITRAN 2000 to 2012 is due to the CO2 spectroscopy, which has been relentlessly upgraded over the years. Fig. 5 shows spectral fits in this window using HITRAN 2012. The residuals clearly show the P, Q, and R-branch structure to be expected of HNO3. The missing manifolds can clearly be seen in the inset, and have a spacing of about 0.4 cm 1.
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
329
Fig. 5. Spectral fits at 23 km in the 3528–3574 cm 1 window. Although the main absorbers here appear to be CO2 and H2O, the entire HNO3 ν1 fundamental band is actually missing from HITRAN, which gives rise to the large downward residuals. The P-, Q-, and R-branch structure can be seen in the residuals, with the Q-branch at 3551.6 cm 1. Many of the HNO3 manifolds are hidden behind saturated CO2 and H2O lines, especially in the R-branch. The inset shows the seven strongest missing P-branch manifolds, the one at 3538.98 cm 1 being obscured by a saturated CO2 line.
Fig. 6. Spectral fit from 4050 to 4140 cm 1 at 10 km altitude. The main absorber here is CH4 and the huge residual at 4115.659 cm 1 is due to a missing CH4 line (see inset). This line was present in HITRAN 2008, so its omission in 2012 was probably a mistake.
Fig. 2e shows the altitude variation of the rms residuals in the 4096 cm 1 window, where absorption is dominated by CH4. The HITRAN 2000 linelist gave reasonable spectral fits at all altitudes, but the HITRAN 2004 and 2008 linelists, which were identical in this region, represented a large deterioration. HITRAN 2012 is much improved, but not quite as good as HITRAN 2000. Fig. 6 show spectral fits to this window using the HITRAN 2012 linelist. The large positive residual at 4115.659 cm 1 is due to a missing CH4 line, which was present in the 2008 linelist. Part of the motivation for this work is to be able in the future to quickly identify such avoidable mistakes. With this missing line inserted, HITRAN 2012 would give better spectral fits than any previous linelist. Fig. 2f shows the altitude dependence of the rms residual in the 4190 cm 1 window where absorption is dominated by CH4. This window has worsened since
HITRAN 2008 at altitudes above 28 km due to the poorer line positions (see Fig. 7). Below 20 km, however, HITRAN 2012 produces much-improved residuals due to the inclusion of 13CH4 lines based on Niederer et al., [11], which were absent from earlier HITRAN versions. Since the 13 CH4 lines are much weaker than the 12CH4 lines, their absorptions only become apparent (i.e., exceed the noise level) below 20 km. Fig. 2g shows the altitude variation of the rms residuals in the 4910 cm 1 window, where absorption is dominated by CO2. Indeed, part of this window overlaps the strong CO2 band used by GOSAT [27] and OCO-2 [28] . It is seen that a modest improvement occurred between HITRAN 2000 and 2004, which was due to the addition of the O3 band, then a huge improvement in HITRAN 2008 related to the CO2 line positions. A modest improvement was then effected in HITRAN 2012 relating mainly to improved CO2
330
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
Fig. 7. Spectral fit to the 4190 cm 1 region, which is dominated by CH4. At the higher altitudes (31 km shown above) the HITRAN 2012 fits are worse than those using HITRAN 2004 or 2008 due to a worsening of the line positions. This is particularly noticeable in the ν3 þ ν4 Q-branch around 4215–4219 cm 1 with peak residuals of 6% versus 4% with HITRAN 2004/8.
Fig. 8. Spectral fit to a 12 km altitude spectrum in the 5360–5420 cm 1 window where absorption is dominated by H2O. The large positive residual is due to a factor 4 intensity over-estimate of the line at 5390.339 cm 1 (see inset). This large residual was not present in fits using HITRAN 2004 or 2008.
line intensities. This window is typical of the vast majority, with improvements being observed with each successive HITRAN version. Fig. 2h shows the altitude variation of the rms residuals in the 5390 cm 1 window which is dominated by H2O. After a big improvement in HITRAN 2004, the entire 5200–5500 cm 1 region has regressed in the past two HITRAN versions. Fig. 8 shows a fit with the HITRAN 2012 linelist. The large residual at 5390.339 cm 1, which reaches 25% at 12 km altitude (due to a factor 4 intensity over-estimate), was not present in fits using HITRAN 2004 or 2008. In HITRAN 2012 the H2O line intensities were based on the ab initio calculations of [9]. Although these are generally an improvement, there are occasional lapses. The line at 5290.339 cm 1 is one of 14 "rogue" H2O lines
that have been identified in the 5000–7500 cm 1 region of HITRAN 2012. They probably relate to resonance transitions. If these rogue H2O lines were empirically corrected, the resulting linelist would be better than any previous HITRAN version.
6. RMS Summary Table 1 shows the rms spectral residual achieved for each window, averaged over all altitudes, for each of the four HITRAN linelists. The smallest rms value for each window is colored blue, versus black for the higher values. Out of 112 fitted windows the HITRAN 2000 linelist produces the best fits in two cases; the 2004 linelist in 25
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
331
Table 1 The 112 fitted windows: central wavenumbers, widths, adjusted gases, and vertically-averaged rms residuals (%) using the HITRAN 2000/04/08/12 linelists. For each window, the smallest rms value is colored blue. The HITRAN 2000 linelist produces the best fits in 2 cases; the 2004 linelist in 25 cases; the 2008 linelist in 20 cases, and the 2012 linelist in 65 cases.
332
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
Table 1 (continued )
cases; the 2008 linelist in 20 cases, and the 2012 linelist in 65 cases. It is perhaps surprising that the 2004 linelist is still best in so many windows. Table 1 also shows the central wavenumber of each window, its width, and the fitted gases. Gases not listed in Table 1 are still included in the fitting, but their vmr profiles are not scaled. Fig. 9 plots the information in Table 1: average rms versus central wavenumber of each window. By averaging over all altitudes, we ensure that an altitude is included where any spectroscopic errors are exposed: low altitudes for the weak absorbers, high altitudes for the strong absorbers.
Fig. 10 shows the average rms residual (over all windows) plotted versus altitude. It shows a significant overall improvement with each new HITRAN version. The general decrease with altitude reflects the weakening of the spectral absorption. The rapid increase in RMS below the tropopause (11 km) is mainly due to H2O. The slight bulge around 16–21 km is primarily due to HNO3.
7. Discussion Atmospheric solar absorption Fourier Transform Infra-Red (FTIR) spectra provide an excellent test of spectroscopy. The
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
333
Fig. 9. Vertically-averaged RMS spectral fitting residuals, plotted versus wavenumber. Upper and lower panels show same data, but lower panel is zoomed and clipped to reveal more detail in low RMS values.
Fig. 10. Wavenumber-averaged RMS spectral fitting residuals, plotted versus altitude, showing the relentless improvement at all altitudes with each new HITRAN version. The rapid increase below the tropopause (11 km) is mainly due to H2O. The slight bulge around 16–21 km is primarily due to HNO3. The general decrease with altitude reflects the weakening of the spectral lines.
fact that the sun is so bright allows the measurement of high resolution and high Signal-to-Noise Ratio (SNR) spectra covering a wide bandwidth. For example, the JPL MkIV FTS covers the 650 to 5650 cm 1 simultaneously at 0.01 cm 1 resolution with a SNR of 400:1, and covers the 5–40 km altitude range. These attributes facilitate identification and diagnosis of problems in regions with high rms residuals. They also allow assessment of the consistency of different bands (not shown in this work). It is perhaps surprising that a single transition with a 0.001 cm 1 position error (e.g., Fig. 4) produces a discernable degradation in the rms fitting residual over a 57 cm 1 wide window cluttered with absorption lines,
measured by a spectrometer of 0.01 cm 1 resolution (ten times the position error). From a balloon platform, a large range of atmospheric conditions can be observed in a relatively short time span (30 min) during which the instrument and atmosphere undergo little change. Observing sunset at 9 km (300 mbar) tangent altitude represents 2–3 orders of magnitude increase in atmospheric airmass compared with viewing the sun at low airmass from float altitude (38 km; 3 mbar). This allows absorption bands of widely different intensities to be checked in a consistent manner. Furthermore, the temperature of the observed airmass also varies from 200 to 250 K, depending on
334
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
altitude, providing a check on the lower state energies (E") of the observed transitions. In spectra covering such a wide range of conditions, observed at high resolution and SNR, it is difficult for spectroscopic problems to remain hidden. And the problems that do remain hidden are probably not important for remote sensing applications. Linelists based on theoretical models can occasionally produce very poor results for transitions that are in resonance. Although this affects only a tiny fraction of the lines, the resulting residuals dominate the spectral fits, making the new theoretical linelist appear poorer than its predecessor when in fact it is better for the vast majority of lines. Empirically adjusting these occasional lapses in the theoretical linelists is clearly the way forward, but who should be doing these adjustments? Clearly the spectroscopists that generated the linelist are the best placed to do so, understanding the strengths and weaknesses of their measurements and predictions. And having them doing it would avoid multiple, lessinformed, end-users from doing the same job. Yet many spectroscopists are reluctant to perform ad hoc empirical adjustments to their linelist, preferring instead to improve the underlying theory, which is the correct long-term solution. But this doesn't help the end-user faced with the short-term problem of what to do about a large systematic residual in their desired fitting window. Should they empirically adjust the linelist, or narrow their window to exclude the large residuals, or somehow modify their fitting software to deweight large residuals? Something has to be done because standard least-squares fitting techniques are highly sensitive to large residuals since they try to minimize the sum of their squares. Even though the large residual may not be due to the target gas of interest, or may not even overlap any target gas absorptions, it can still strongly influence the retrieval through cross-talk with other retrieved variables (non-target gases, continuum level, frequency shift). We therefore advocate a more rigorous checking of any new linelist, comparing it with its predecessors by fitting well-characterized atmospheric and laboratory spectra and investigating windows where the rms fitting residual worsened. This will catch the more important preventable errors, which can then be fixed by empirical adjustment, or by substitution from an earlier linelist version. To be clear, we do not advocate adjusting spectroscopic linelists based on fits to atmospheric spectra – that is better done with laboratory spectra. Atmospheric spectra are, however, invaluable for pointing out where the spectroscopy is inadequate, and which gases are likely responsible. Although not illustrated in this paper, when systematic residuals are seen in fits to MkIV atmospheric spectra, those same residuals are also seen in fits to laboratory spectra made under similar conditions (absorption depth, pressure, temperature) in over 90% of cases. This suggests that the HITRAN linelist could be significantly improved using existing laboratory spectra by empirical adjustment of the culpable parameters. Several of the largest fitting residuals can be attributed to inadequacies in the HNO3 spectroscopy. Besides the completely missing ν1 fundamental band centered at 3551 cm 1, and several obvious missing combination bands (e.g. 2645, 4107 cm 1), the ν3 and ν2 fundamentals at 1325 and 1725 cm 1 respectively, also have serious problems. Although the very recent work by Perrin et al. [13] will improve the
situation in the ν3 band if incorporated into HITRAN, HNO3 will still remain the dominant source of stratospheric fitting errors. Although it could be argued that the ν1, ν2, and ν3 bands of HNO3 are not needed for atmospheric HNO3 measurement, since the overlapping ν5 and 2ν9 HNO3 bands at 850–950 cm 1 are sufficient, this is a self-fulfilling prophecy. If the ν1, ν2, and ν3 spectroscopy were better, these bands would become more widely used. Since the ν2 and ν3 bands are stronger than the ν5 and 2ν9 bands, better spectroscopy would allow HNO3 profiles to be retrieved more accurately under low-HNO3 conditions, e.g. at high altitudes and in the tropics. In addition, the ν1 band centered at 3551 cm 1 is ideally situated for in situ measurements of atmospheric HNO3 by room-T tunable diode lasers. Unfortunately at this point very little reliable experimental information exists on the ν1 band in the literature, preventing derivation of the important spectroscopic constants that could be used to model this band in the atmosphere. More high-resolution laboratory experiments are desperately needed for this band. Poor HNO3 spectroscopy also thwarts measurement of trace gases whose main absorption bands overlap HNO3 bands, such as SO2 at 1350 cm 1, OH at 3570 cm 1, and HO2 at 3440 cm 1. The construction of the methane line list for HITRAN 2012 is described in Brown et al. [2]. While it represents some of the major advances in methane spectroscopy in recent times, it is clearly in need of very careful revision. This process is already underway within a dedicated task group of the HITRAN advisory committee. The most straightforward fix is to restore the lines that were inadvertently dropped from the database such as the line at 4115.659 cm 1 (Fig. 6). Most of the errors, associated with the line positions (including those described in Figs. 4 and 7) seem to be associated with the employment of two not entirely compatible software packages MIRS [12] and STDS [25] that were used to fit and predict transitions. This approach was employed in Albert et al. [1] and Daumont et al. [3], which are the origin of many HITRAN 2012 transitions involving Pentad and Octad states. A thorough investigation of this matter is planned for the next HITRAN release. Spectroscopy of water vapor will also be a subject of substantial update although as it was mentioned, with the exception of a few lines, HITRAN 2012 was already a large step forward. Apart from other improvements the addition of the missing lines of the deuterated isotopologues of water in the NIR region is planned. Considering that there will be an extensive combination of experimental and theoretical results, there is much scope for human error and therefore these validations will be most useful.
8. Future work The current analysis, being confined to altitudes above 9 km, emphasizes stratospheric gases and exposes errors in the positions and intensities of spectral transitions. The next step would be to analyze ground-based spectra where H2O plays a much more important role and where errors in line widths and shifts become more important. This would also allow the spectral coverage to be extended above 5700 cm 1
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
by using TCCON and Kitt Peak ground-based solar spectra in addition to MkIV spectra. Extending this analysis to lower altitudes will require inclusion of line mixing [15,4,8] effects which above 10 km are confined to relatively narrow spectral regions and therefore don't have much impact on the mean rms residual of a window tens of cm 1 wide. At lower altitudes, however, the impacts of line-mixing will be broader and stronger and will have to be included. Sub-dividing some of the windows, such that we would end up with more the 112 windows with a narrower average width, would provide more specificity in the identification of regions where the rms fits worsen. For example, in a wide window it is possible for improvements in the spectroscopy of gas A to mask a worsening of the gas B spectroscopy, such that the overall rms still improves. Using the sun as a source allows the MkIV balloon spectra to cover a very broad spectral region simultaneously with high resolution and SNR. The gas amounts retrieved from various windows provide information about the band-to-band consistency of the spectroscopy. Such information is often difficult to extract from laboratory spectra, where the dimmer sources often necessitate a narrower spectral bandwidth, preventing simultaneous measurement of widely separated bands. In a future work we will examine the band-to-band consistency of the retrieved gas amounts. Finally, we need to evaluate HITRAN 2016 when it comes out later this year. As a result of the work shown in this paper, this can now be done quickly (a few days) since the windows and spectra are already defined and the scripts for running the GFIT code and making the plots already exist.
9. Concluding remarks Currently, some atmospheric remote-sensing scientists maintain their own linelists, founded on HITRAN, but perhaps not the latest version for every gas of every band. When a new HITRAN version is released, these scientists test it on the spectral regions they care about. If it is better, they adopt it for those regions where it is better, giving rise to a hybrid linelist. Sometimes they will do a partial adoption, including some gases from the latest HITRAN, but not others. Often a new linelist will be worse initially than its predecessor due to a few isolated large residuals (e.g. H2O: 5000–7500 cm 1). But after empirical correction of these, based on fits to laboratory spectra, significant improvements are seen. The linelist used by the TCCON community [24] is a good example of this empirical, "greatest hits", approach to spectroscopic database management. In this particular case, the criteria for deciding whether a new linelist is better or worse is two-fold: the rms fitting residuals, and the consistency of gas amounts retrieved from different windows. Since this quality control work will be re-done anyway after release of HITRAN 2016, to decide which parts to adopt, we thought that it would be more efficient to do it before the public release, and hence address the easy-to-fix problems before the linelist becomes publically available, sparing the end-users the hassle of dealing with regions where their fits
335
get worse. If the new HITRAN versions were more dependably superior than their predecessors, most end-users would abandon maintaining their own hybrid linelists. We advocate assessing new linelists by fitting wellcharacterized atmospheric spectra (in addition to laboratory spectra). This allows identification of regions where the new linelist performs worse than its predecessor and usually reveals the cause of the deterioration. The thornier issue is what to do about these problems, once identified. There are cases where the need to fix the linelist is obvious and uncontroversial. For example the resonance O3 line at 2671.4 cm 1 can be fixed by empirically adjusting its position based on fits to laboratory spectra. And the missing CH4 line at 4115.659 cm 1 can be fixed by substitution from the previous HITRAN version. But this type of work is error prone, hence the need for performance benchmarks against which to test any empirical adjustments. Spectroscopic issues such as the missing HNO3 bands will require longer-term solutions. Finally we recognize that the Earth's atmosphere is not the sole use for the HITRAN linelist. It is also used for radiativetransfer calculations in other environments, such as the atmospheres of other planets, or industrial processes. It is possible, for example, that a linelist update that makes fits slightly worse for Earth spectra, much improves fits to spectra of Titan's atmosphere. So a balance has to be struck between the needs of the diverse end-users of HITRAN.
Acknowledgements We thank NASA's Upper Atmosphere Research Program who funded the JPL MkIV instrument through Grant NNH12ZDA001N-UACO. The Columbia Scientific Balloon Facility (CSBF) who launched the balloon and recovered the MkIV payload. The HITRAN team gratefully acknowledges support from the NASA AURA program Grant NNX14AI55G. We also thank the Worldwide spectroscopy community whose work is encapsulated in the HITRAN linelists. Part of this research was performed at the Jet Propulsion Laboratory, California of Technology, under contract with NASA.
References [1] Albert S, Bauerecker S, Boudon V, Brown LR, Champion J-P, Loëte M, et al. Global analysis of the high resolution infrared spectrum of methane 12CH4 in the region from 0 to 4800 cm 1. Chem Phys 2009;356:131–46. [2] Brown LR, Sung K, Benner DC, Devi VM, Boudon V, Gabard T, et al. Methane line parameters in the HITRAN2012 database. J Quant Spectrosc Radiat Transf 2013;130:201–19. http://dx.doi.org/10.1016 /j.jqsrt.2013.06.020. [3] Daumont L, Nikitin AV, Thomas X, Régalia L, Von der Heyden P, Tyuterev VG, et al. New assignments in the 2 μm transparency window of the 12CH4 Octad band system. J Quant Spectrosc Radiat Transf 2013;116:101–9. http://dx.doi.org/10.1016/j.jqsrt.2012.08.025. [4] Edwards David P, Strow L Larrabee. Spectral line shape considerations for limb temperature sounders. 96.D11. J Geophys Res Atmos 1991:20859–68. [5] Gordon IE, Kassi S, Campargue A, Toon GC. First identification of the electric quadrupole transitions of oxygen in solar and laboratory spectra. J Quant Spectrosc Radiat Transf 2010;111: 1174–83. http://dx.doi.org/10.1016/j.jqsrt.2010.01.008.
336
G.C. Toon et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 182 (2016) 324–336
[6] Gordon IE, Rothman LS, Toon GC. Revision of spectral parameters for the B- and γ-bands of oxygen and their validation against atmospheric spectra. J Quant Spectrosc Radiat Transf 2011;112: 2310–22. http://dx.doi.org/10.1016/j.jqsrt.2011.05.007. [7] Irion FW, Gunson MR, Toon GC, Chang AY, Eldering A, Mahieu E, et al. Atmospheric trace molecule spectroscopy (ATMOS) experiment version 3 data retrievals. Appl Opt 2002;41(33):6968–79. [8] Levy A, Lacome N, Chackerian C. Collisional line mixing. In: Rao KN, Weber A, editors. Spectroscopy of the earth's atmosphere and interstellar medium. Boston: Academic Press; 1992. p. 261–337. [9] Lodi L, Tennyson J, Polyansky OL. A global, high accuracy ab initio dipole moment surface for the electronic ground state of the water molecule. J Chem Phys 2011;135 034113-10. [10] Newman PA, Fahey DW, Brune WH, Kurylo MJ, Kawa SR. Preface [to special section on photochemistry of Ozone loss in the arctic region in summer (POLARIS)]. J Geophys Res 1999;104(D21): 26481–95. http://dx.doi.org/10.1029/1999JD900832. [11] Niederer HM, Wang XG, Carrington T, Albert S, Bauerecker S, Boudon V, et al. Analysis of the rovibrational spectrum of 13CH4 in the Octad range. J Mol Spectrosc 2013;291:33–47. http://dx.doi.org/ 10.1016/j.jms.2013.06.003. [12] Nikitin A, Champion JP, Tyuterev VG. The MIRS computer package for modeling the rovibrational spectra of polyatomic molecules. J Quant Spectrosc Radiat Transf 2003;82:239–49. http://dx.doi.org/ 10.1016/S0022-4073(03)00156-0. [13] Perrin A, Flaud J-M, Ridolfi M, Vander Auwera J, Carlotti M. MIPAS database: new HNO3 line parameters at 7.6 μm validated with MIPAS satellite measurements. Atmos Meas Tech Discuss 2015;8: 11643–71. http://dx.doi.org/10.5194/amtd-8-11643-2015. www.atmosmeas-techdiscuss.net/8/11643/2015/. [15] Rosenkranz PW. IEEE Trans Antennas Propag 1975;AP 23:498. [16] Rothman LS, Jacquemart D, Barbe A, Chris Benner D, Birk M, Brown LR, et al. The HITRAN 2004 molecular spectroscopic database. J Quant Spectrosc Radiat Transf 2005;96:139–204. [17] Rothman LS, Barbe A, Chris Benner D, Brown LR, Camy-Peyret C, Carleer MR, et al. The HITRAN molecular spectroscopic database: edition of 2000 including updates through 2001. J Quant Spectrosc Radiat Transf 2003;82:5–44.
[18] Rothman LS, Gordon IE, Barbe A, Chris Benner D, Bernath PF, Birk M, et al. The HITRAN 2008 molecular spectroscopic database. J Quant Spectrosc Radiat Transf 2009;110:533–72. [19] Rothman LS, Gordon IE, Babikov Y, Barbe A, Chris Benner D, Bernath PF, et al. The HITRAN2012 molecular spectroscopic database. J Quant Spectrosc Radiat Transf 2013;130:4–50. [20] Sen B, Toon GC, Blavier J-F, Fleming EL, Jackman CH. Balloon-borne observations of mid-latitude fluorine abundance. J Geophys Res 1996;101(D4):9045–54. [21] Toon GC. The JPL MkIV interferometer. Opt Photonics News 1991;2: 19–21. [22] Toon GC, J.-F. Blavier B Sen, Salawitch RJ, Osterman GB, Notholt J, Rex M, et al. Ground-based observations of Arctic O3 loss during spring and summer 1997. J Geophys Res 1999;104(D21): 26497–510. http://dx.doi.org/10.1029/1999JD900745. [23] Toon GC, J.-F. Blavier B Sen, Margitan JJ, Webster CR, May RD, Fahey DW, et al. Comparison of MkIV balloon and ER-2 aircraft profiles of atmospheric trace gases. J Geophys Res 1999;104(26):779–26,790. [24] Toon GC. Telluric line list for GGG2014. TCCON data archive, hosted by the Carbon dioxide information analysis center. U.S.A: Oak Ridge National Laboratory; http://dx.doi.org/10.14291/tccon.ggg2014.atm. R0/XXXXXXX. [25] Wenger C, Champion JP. Spherical top data system (STDS) software for the simulation of spherical top spectra. J Quant Spectrosc Radiat Transf 1998;59:471–80. http://dx.doi.org/10.1016/S0022-4073(97) 00106-4. [26] Wunch D, Toon GC, Blavier J-FL, Washenfelder RA, Notholt J, Connor BJ, et al. The total carbon column observing network, Philosophical Transactions of the Royal Society. Ser A Math Phys Eng Sci 2011;369 (1943):2087–112. http://dx.doi.org/10.1098/rsta.2010.0240. [27] Kuze A, Suto H, Nakajima M, Hamazaki T. Thermal and near infrared sensor for carbon observation Fourier-transform spectrometer on the Greenhouse Gases Observing Satellite for greenhouse gases monitoring. Appl. Opt. 2009;48(35):6716–33. [28] Crisp D, Miller CE, DeCola PL. NASA orbiting carbon observatory: Measuring the column averaged carbon dioxide mole fraction from space. J. Appl. Remote Sens. 2008;2:23–508.