Atmospheric CO2 retrieved from ground-based near IR solar spectra



[1] The column-averaged volume mixing ratio (VMR) of CO2 over Kitt Peak, Arizona, has been retrieved from high-resolution solar absorption spectra obtained with the Fourier transform spectrometer on the McMath telescope. Simultaneous column measurements of CO2 at ∼6300 cm−1 and O2 at ∼7900 cm−1 were ratioed to minimize systematic errors. These column ratios were then scaled by the mean O2 VMR (0.2095) to yield column-averaged vmrs of CO2. These display similar behavior to the Mauna Loa in situ surface measurements. During the period 1977–1995, the column-averaged mixing ratio of CO2 increased at an average rate of 1.49 ± 0.04 ppmv/yr with seasonal variations of ∼7 ppmv peak-to-peak. Our retrievals demonstrate that this remote technique is capable of precisions better than 0.5%.

1. Introduction

[2] The atmospheric CO2 VMR has increased from about 280 ppmv in pre-industrial times (before 1800) to about 360 ppmv by the 1990s [IPCC, 1995]. Previous studies based on observations from the in situ network of surface stations indicate that the biosphere and ocean sinks are absorbing about half of the anthropogenic CO2 emission [Battle et al., 2000]. The nature and geographic distribution of these sinks, however, remains too uncertain to predict their responses to future climate or land use changes [IPCC, 1995]. Although in situ measurements of CO2 from the existing surface network are highly accurate (0.1 ppmv), their sparse spatial coverage (both vertical and horizontal) limits their usefulness in constraining models of CO2 sources and sinks.

[3] Rayner and O'Brien [2001] have shown that global satellite measurements of the CO2 column, even with poorer precision (2 ppmv, 0.5%), would provide a better constraint on geographic and temporal distribution of CO2 sources and sinks than the existing surface network. This calculation assumes, however, that such space-borne observations are bias-free. To test and eliminate bias, an extensive correlative measurement effort will be required. These measurements, additionally, can be important for constraining aspects of the CO2 behavior that will not be measurable from space in the foreseeable future (e.g. diurnal variations). In situ measurements of CO2 from the existing surface network cannot alone be used to validate space-borne measurements of column CO2 due to their much higher spatial resolution and their confinement to the planetary boundary layer.

[4] Ground-based spectral observations of the sun at high spectral resolution can, in principle, be used to determine the total CO2 column. The near IR is particularly attractive for this task because the sun is sufficiently bright that high signal-to-noise (S/N) at high spectral resolution can be obtained with minimal integration time and uncooled detectors.

2. Measurements

[5] Between 1979 and 1985, numerous high-resolution absorption spectra (Δν = 0.014 cm−1) of the Sun were obtained with the 1-m Fourier transform spectrometer (FTS) at the McMath telescope complex on Kitt Peak (31.9 N, 111.6 W, 2.07 km above sea level). Observations covering the spectral range from 6000 cm−1 to 8000 cm−1 were used to determine the column-averaged dry air VMR of CO2 and CH4 [Wallace and Livingston, 1990]. In that work, equivalent width (EQW) analyses were performed on 19 well isolated lines of the CO2 (21°2) − (00°0) band (vo = 6348 cm−1) and 14 lines of the O2 0–0 1Δg3Σg band (vo = 7882 cm−1) to get the column amount of both gases. Assuming the O2 VMR to be constant (0.20946), the column ratio (CO2/O2) was scaled to yield the column-averaged CO2 VMR in the atmosphere over Kitt Peak.

[6] In this letter, we report a reanalysis of the Kitt Peak observations using a more sophisticated spectral retrieval algorithm with updated spectroscopic linelists for CO2, O2 (including the collision induced absorption), H2O, and absorption in the solar atmosphere. This allows us to simultaneously fit the entire O2 band (containing more than 200 significant lines) and two CO2 bands centered at 6228 and 6348 cm−1. This allows us to include in our analysis spectra obtained at high airmasses, that Wallace and Livingston [1990] excluded due to limitations imposed by their EQW technique. In addition, we have analyzed spectra obtained more recently at the same facility.

3. Data Analysis

[7] Spectra are analyzed using a line-by-line algorithm developed at JPL for the analysis of solar absorption spectra. In the retrieval, least square analysis is performed over a prescribed spectral window to derive the slant column abundance of the target gases. The fitting residual is defined as:

display math

YiM is one measurement in one spectrum, YiC is the forward model calculation, x is the target gas scale factor, σi is the uncertainty of YiM. In our forward model the atmosphere is represented by 70 vertical levels. Pressure- and temperature-dependent absorption coefficients are computed line-by-line for each level and used in the forward model to produce spectra YiC(x) for comparison.

[8] Pressure and temperature profiles are obtained from NOAA's Climate Diagnostics Center (CDC), which are assimilated temperature data on 17 levels from 1000 to 10 mbar with 1° × 1° geographic resolution. Climatological temperature profiles are used for levels with pressure ≤10 mbar. A study of the temperature sensitivity of the retrieved CO2/O2 ratio revealed a T-dependence of 0.07%/K. So if the CDC profiles were systematically in error by 5 K at all levels, the CO2/O2 error would be 0.35%, or about 1 ppmv.

[9] The CO2 spectral parameters are derived from the latest HITRAN compilation [Rothman et al., 1998]. The background solar linelist is based on the work of Livingston and Wallace [1991]. For the O2 lines, the HITRAN strengths were updated with a recent recalculation [Goldman, private communication] that was found to give improved agreement with laboratory data [Newman et al., 2000]. In addition to the discrete O2 lines of the 1Δ band, there is also underlying O2 continuum absorption [Smith and Newnham, 2000, 2001]. Although this O2 collision induced absorption (CIA) was included in the line-by-line calculation to improve estimation of the continuum, only the discrete 1Δ O2 lines were used in the computation of the O2 column amount.

4. Results

[10] 414 spectra obtained at Kitt Peak over the period 1977–1995 have been analyzed. Due to other demands on the instrument, solar observations in the 6000–8000 cm−1 spectral region were generally made on only a few days each year. The spectra were measured at various times throughout the day from sunrise to sunset, weather permitting. The spectra analyzed are generally co-additions of 2–4 individual spectra, each taking 7 minutes to acquire. The effective solar zenith angle (SZA) was calculated as the average of the SZAs of the individual interferograms at their zero path difference (ZPD) times. At high airmasses (∼10), this was often significantly different (5% in airmass) from the mean SZA, due to the fact that the ZPDs were typically not in the middle of the double-sided interferograms, and due to the non-linear nature of the variation of the airmass with time.

[11] Examples of the spectral analysis are shown in Figure 1. Two spectral bands of CO2 and one of O2 were fitted as shown. We have excluded from further analysis the 71 observations (17%) that produce rms spectral residuals greater than 6.0% for O2 or 3.0% for CO2. The larger threshold for rejection of O2 spectra is due to the difficulty in fitting the O2 CIA at large SZAs. Most of these rejected spectra were not intended for use in the 6000–8000 cm−1 region. In remaining spectra, those measured at low SZA generally have rms residuals of about 0.4–0.8% for O2 and 0.7–0.8% for the CO2. These residuals are dominated by solar features and by the O2/CO2 lines themselves. For spectra measured at high SZA (>75 deg) the residuals are larger (up to 3% in CO2 and 5% in O2) due to the increasing strengths of the O2/CO2 lines.

Figure 1.

Examples of spectral fits to a Kitt Peak spectrum measured at 70.70° SZA on May 9, 1981. Diamonds are the measurements and black lines are the fitted transmittance. (a) The 6300 cm−1 region, containing the (14°1)–(00°0) and (21°2)–(00°0) bands of CO2; (b) The 7900 cm−1 region, containing the O2 0–0 1Δg3Σg band. The contributions from CO2/O2, H2O and solar features are shown by the red, green and blue lines respectively. The O2 CIA is represented by the grey line in (b). The residual traces (measured-calculated) are clearly dominated by systematic errors in both regions.

[12] The retrieved slant column amounts were divided by the calculated airmass to produce vertical column amounts. Account was taken of the 250 m optical path inside the telescope, which adds nearly 3% to the noontime O2 slant column. The daily-average value of CO2 and O2 vertical column amounts and their ratio (CO2/O2) are shown in Figure 2. The long-term trend in the CO2 column amounts (Panel A) is evident, but the seasonal variation is unclear. For the O2 column amount (Panel B), no long-term trend is observed. The retrieved dry VMRs (converted from column amount after dividing by 2.15 × 1022 times the surface pressure in mbar) of O2 fall within the range of 0.202 to 0.217, with a mean of about 0.209 — remarkably close to the actual O2 VMR for dry air (0.2095).

Figure 2.

Time series of retrieved vertical columns of (a) CO2 and (b) O2. (c) the column-averaged CO2 VMR over Kitt Peak determined from the ratio of the individual CO2 and O2 spectra (green dots) and the daily-averaged value (red diamonds) are shown. The column measurements in panel (c) have been scaled by 1.058 to match the absolute value of the Mauna Loa record (blue). Error in the spectroscopic database for CO2 is likely responsible for the discrepancy (see text).

[13] The ratio of the CO2 and O2 column amounts, scaled by the standard atmospheric O2 fraction of 0.2095, gives the column-averaged dry volume-mixing ratio of CO2 in the atmosphere (Figure 2, Panel C). Clearly, the precision of the CO2/O2 ratio is improved over that of the CO2 column. The high degree of correlation between error in CO2 and O2 suggests that much of the scatter in the unratioed observations is common to both spectral regions. We speculate that these errors arise in the spectra (e.g. uncertainties in instrumental line shape) or in the calculated airmass (e.g. uncertainty in surface pressure or SZA). In Figure 2c, we have scaled the CO2 column average dry mixing ratio by 1.058. This scaling was empirically determined to minimize bias between the Kitt Peak data and those from Mauna Loa, and probably reflects an error in the CO2 line strengths and/or pressure broadening coefficients. Also shown in Figure 2 (Panel C) are the Mauna Loa CO2 in situ measurements. Their trend and seasonal cycles are remarkably similar to the Kitt Peak data. The average CO2 increase between 1977 and 1995 is 1.49 ± 0.04 ppm/yr, the same as that measured from Mauna Loa (∼1.47 ppm/yr) over the same period.

[14] Details of the seasonal cycles for CO2/O2 in the period 1979–1985 are expanded in Figure 3. We compared the Kitt Peak data with two sets of in situ measurements: Mauna Loa (19.5N, 155.6W, 3397m above sea level) and Niwot Ridge (40.1N, 105.6W, 3475m above sea level). The phase of Kitt Peak data is nearly the same as Mauna Loa measurements but has a lag of about one month compared with the Niwot Ridge record, despite the latter site being much closer to Kitt Peak. The seasonal amplitude of Kitt Peak data (∼7 ppm peak-to-peak) is also more similar to Mauna Loa record than to the Niwot Ridge record. We do not know the reason for the different behavior of the Niwot Ridge data, but we speculate that the Mauna Loa and Kitt Peak records might better represent the free tropospheric CO2, while the Niwot Ridge record is more heavily influenced by local sources and sinks of CO2.

Figure 3.

Time series of daily averaged retrieved values of the Kitt Peak column CO2 VMRs (red open diamonds) over period 1979–1985, compared with the Mauna Loa (blue) and the Niwot Ridge (pink) in situ measurements.

[15] To understand the diurnal variation and to estimate the precision of the retrieval, 231 column results of 40 different days for which more than 4 column measurements exist were analyzed. For each day, the O2 and CO2 vertical column amounts were converted into VMRs and averaged to get the daily mean. The variation of one measurement from the daily mean is defined as

display math

where x is one measurement and 〈x〉 is the mean of the day. The diurnal variations for O2 and CO2 obtained in this manner are partly correlated and therefore the CO2/O2 ratios exhibit less error than CO2 VMRs obtained by dividing the CO2 column amount by pressure. The range for O2 variation is ±2.0%, whereas for CO2 it is ±1.2%.

[16] The diurnal variations in the CO2/O2 ratios are shown in Figure 4. Most of the deviation is in the range of −0.5 ∼0.5%, except for a few measurements made at high airmasses (>10). Assuming the CO2 VMR over Kitt Peak is essentially constant during one day, the scatter in Figure 4 is a measure of the precision of the column measurements. The origin of the increase of CO2/O2 at high airmass is likely non-geophysical. We speculate that the different behaviors of the continuum signal in the CO2 and O2 spectral regions as a function of airmass contributes to this anomalous diurnal behavior in CO2/O2. Since the integrating periods of the Kitt Peak measurements are quite long (about 30 minutes at 5 airmasses and 15 minutes at 10 airmasses), it is impossible to un-ambiguously assign an average airmass to these spectra.

Figure 4.

Diurnal variations of the CO2/O2 ratio (see text for details).

5. Conclusions

[17] Reanalysis of near IR spectra obtained at Kitt Peak National Solar Observatory have demonstrated that the column-averaged CO2 VMR can be retrieved with ∼0.5% precision. By simultaneously fitting all lines in one band and using new spectroscopic linelists, we decreased the temperature sensitivity of data retrieval and greatly improved the precision compared with Wallace and Livingston's analysis.

[18] We believe that remaining errors are dominated by deficiencies in the spectroscopic linelists, and by airmass ambiguities exacerbated by the long scan duration of the Kitt Peak spectrometer. With improvements to the spectroscopic linelists and with modern, faster-scanning FTIR spectrometers dedicated to the measurement of CO2, substantially better results can be anticipated. We are confident that with these improvements, this technique will be more than adequate for ground-truthing future space-based CO2 column observations.


[19] We thank Yuk Yung, Linda Brown, Jim Randerson, Lloyd Wallace, Bill Livingston, and Mike Dulick for many helpful discussions. Part of this work was performed at the Jet Propulsion Laboratory, California Institute of Technology, under contract with NASA. Paul Wennberg and Geoffrey Toon acknowledge support from NASA's upper Atmosphere Research Program. Zhonghua Yang is supported with a graduate research assistantship funded by NASA's ACMAP program (NAG5-11739).