1. Introduction

The emission of CO₂ has increased remarkablyover the past century due to mass consumption of fossilfuels and the expansion of industrial activities. Anincrease of atmospheric CO₂ is expected to lead tosignificant future climate change because of its poweras a greenhouse gas. The reliable prediction of futureatmospheric CO₂ concentration and associated globalclimate changes requires an adequate underst and ing ofthe sources and sinks of CO₂; unfortunately,this underst and inghas significant gaps and uncertainties arelarge(Stephens et al., 2007). The persistence of largeuncertainties is in part due to the sparseness and inhomogeneousgeographical distribution of the approximately180 CO₂ measurement sites that constitute thecurrent observing system(Globalview-CO₂,2010).

Theoretical studies have shown that satellite measurementsof CO₂ in combination with atmospherictransport models could be used to significantly reduceuncertainties in surface CO₂ fluxes(Rayner and O’Brien,2001; Houweling et al., 2004). Satellite-basedremote sensing of CO₂ can be accomplished by measuringradiances in either the thermal infrared(IR)spectral region(Chédin et al., 2002,2003a,b; Engelen et al., 2004; Aumann et al., 2005; Strow et al., 2006)or the near-infrared(NIR)region(Tolton and Plouffe, 2001; O’Brien and Rayner, 2002; Mao and Kawa, 2004,Buchwitz et al., 2005a,b,2006,2007; Houweling et al., 2005; Bösch et al., 2006; Barkley et al., 2006a,b,c,2007). With respect to CO₂ column-average values,retrievals using measurements from the NIR regionslightly outperform those using measurements fromthe IR region(Christi and Stephens, 2004). TheSCanning Imaging Absorption spectroMeter for AtmosphericCHartographY(SCIAMACHY)is the firstsatellite instrument to measure reflected solar radiationin the NIR spectral region that contains theCO₂ absorption b and s. SCIAMACHY near-infrarednadir measurements have been used to retrieve verticalcolumns of CO₂ from the 1.58-μm absorptionb and (Schneising et al., 2008). With the aim of makingglobal space-based measurements of atmosphericCO₂ at the spatial resolution and accuracy needed tocharacterize surface sources and sinks,the NationalAeronautics and Space Administration(NASA)selectedthe Orbiting Carbon Observatory(OCO)asthe fifth mission in Earth System Science Pathfinder(ESSP)Program(Crisp et al., 2004). The OCO wasunfortunately lost due to launch failure in February2009; nevertheless,its successors and the GreenhouseGases Observatory Satellite(GOSAT)use the CO₂vibration-rotation transition b and s in the 2.06- and 1.61-μm(or alternatively 1.58-μm)regions to makespace-based measurements of CO₂ with the precision,resolution, and coverage needed to characterize CO₂sources and sinks on regional scales.

The High Resolution Transmission(HITRAN)database is a recognized international st and ard usedfor a vast array of applications including atmosphericradiation transmission simulations and atmosphericremote sensing. Feng et al.(2007) and Feng and Zhao(2009)have previously analyzed the effects of improvementsto the HITRAN 2004 database(Rothman et al., 2005)on radiative transfer and transmittance calculations.A large number of experimental and theoreticalinvestigations of CO₂ spectra have been undertakensince the release of HITRAN 2004. The HITRAN2008 CO₂ line lists in the NIR region from 4300 to7000 cm⁻¹ completely replace HITRAN 2004 data inthis spectral range(Rothman et al., 2009). This studyprovides a comparison of the simulated transmissionspectra and sensitivity to changes in CO₂ amount usingline parameters from the HITRAN 2004 and HITRAN2008 databases. The purpose of this comparisonis to investigate the effects of the modifications ofline parameter data on retrievals of CO₂ atmosphericamount by NIR reflected sunlight. Three spectral regionsare selected that cover the CO₂ b and s used bythe SCIAMACHY and GOSAT instruments,as wellas the eventual successor to the OCO.2. Outline of the calculation

In practice,the retrieval of atmospheric CO₂ fromdifferential absorption of reflected sunlight is undertakenby using the ratio of measured radiances ata weak absorption channel(called an off-line) and a strong absorption channel(called an on-line)toestimate column CO₂ without retrieving surface reflectance(O’Brien and Rayner, 2002; Mao and Kawa, 2004). In a clear atmosphere,this ratio can be expressedas

where μ₀ is the cosine of the solar zenith angle,μ₁ isthe cosine of the satellite zenith angle(equal to 1.0 atnadir viewing), and τ is the optical thickness along avertical path from surface to space defined by

where k is the specific absorption coefficient derivedfrom laboratory measurements or calculated using aradiative transfer model and a molecular spectroscopicdatabase, and ρ is the concentration of the absorbingmaterial. The transmittance is defined as the exponentialfunction of optical thickness as follows:

To retrieve high-accuracy CO₂ column amounts fromhigh spectral resolution measurements,it is necessaryto precisely calculate the transmittance spectra. Reliablespectroscopic databases are absolutely essentialfor calculating accurate transmission spectra of radiativelyactive molecular species(Kratz,2008).

The Line-by-Line Radiative Transfer Model(LBLRTM),developed by Atmospheric and EnvironmentalResearch,Inc.(Clough et al., 1992,1995),isused to simulate the transmission spectra in this study.The LBLRTM is an accurate,efficient, and wellvalidated line-by-line code that can be used over thefull spectral range from the microwave to the ultraviolet.This model provides the foundation for manyradiative transfer applications. The model uses aVoigt line shape with a line cutoff of 25 cm⁻¹ fromthe line center at all atmospheric levels with an algorithmbased on a linear combination of approximatingfunctions. This approach allows the inclusion of bothcollision- and Doppler-broadening processes. The HITRANdatabase is used as input to LBLRTM. Themain features of LBLRTM are described in Clough etal.(2005).

The 1976 United States(U.S.)st and ard atmosphereis used for atmospheric profiles up to 120-km altitude.The model atmosphere is assumed to be planeparallel and divided into 50 total layers. The verticalresolutions of the layers of 0–25,25–50, and 50–120km are 1,2.5, and 5 km,respectively. Each layer isconsidered as a homogeneous path in the calculations.The spectral regions used in this study are “B and -1”(4810–4890 cm⁻¹),“B and -2”(6195–6255 cm⁻¹), and “B and -3”(6315–6375 cm⁻¹). These three b and s correspondto the CO₂ vibration-rotation transition b and sin the 2.06-,1.61-, and 1.58-μm regions,which are usedby current or near-future satellite instruments that retrieveCO₂ using nadir observations. Absorption byonly six major species(H₂O,CO₂,O₃,N₂,CO, and CH₄)is included in the calculations, and the atmosphericprofiles used are representative of the defaultvalues in the U.S. st and ard model atmosphere exceptfor CO₂. The influences of aerosols on the transmittancecalculations are not considered here.

The reference transmittance spectra are computedassuming a CO₂ concentration of 379 ppmv.This value is equal to the global mean atmosphericmixing ratio of CO₂ in 2005(Forster et al., 2007). Atriangle scanning function for simulating the instrumentslit function is used to convolve the line-by-linecalculated monochromatic transmittance and estimatethe effects of changes in spectral resolution on transmittance and transmittance sensitivity. The same calculationsare carried out at two spectral resolutions(0.2 and 1.0 cm⁻¹)at Full Width at Half-Maximum(FWHM)in B and -1. These simulated spectral resolutionsare approximately equal to the spectral resolutionsof GOSAT and OCO,respectively. One additionalspectral resolution(5.8 cm⁻¹)is applied atFWHM in B and -2 and B and -3,which is representativeof the spectral resolution realized by channel-6(5710–10000 cm⁻¹)of the SCIAMACHY spectrometer usedto retrieve CO₂. The output spectral sampling step is0.1 cm⁻¹ for all of the spectral resolutions.3. Results3.1 Sensitivity to changes in CO₂ concentrations

Table 1 summarizes the updates that have beenmade to the near-infrared spectral portion(4300–7000cm⁻¹)of the HITRAN 2004 and HITRAN 2008 databases. These updates include changes to the totalnumber of line transitions,total line intensities,averageline intensities, and average air-broadened linehalf-widths for the species H₂O,CO₂,O₃,N₂,CO, and CH₄. The line parameters for CO₂ and O₃ haveundergone major updates,the line parameters for H₂O and CH₄ have undergone minor updates, and the lineparameters for N₂ and CO remain the same as in HITRAN2004. The number of line transitions compiledfor CO₂ has increased by a factor of 7.9 from 9237transitions in the 2004 database to 72886 transitionsin the 2008 database. The sum of the CO₂ line intensitiesin HITRAN 2008 is less than that in HITRAN2004,indicating that most of the added CO₂ lines areweak absorption lines.

Figure 1 shows the calculated transmittancespectra in B and -1,B and -2, and B and -3 at differentspectral resolutions using the HITRAN 2008database. The CO₂ absorption in B and -1 is considerablystronger than that in B and -2 or B and -3. Inaddition,the transmission spectra in B and -2 are verysimilar to those in B and -3.

Transmittance sensitivity describes the effects ofperturbations to atmospheric composition on the atmospherictransmittance calculation. In this paper,itis defined as

where T is the transmittance and x is the concentrationof atmospheric gases. Figure 2 shows relativechanges in the transmittance spectra that correspondto a 1-ppmv CO₂ column mixing ratio increase in thewhole atmosphere. These relative changes are calculatedat different spectral resolutions over the chosenthree b and s using line parameters from the HITRAN2008 database. When the atmospheric CO₂ mixingratio increases uniformly by 1 ppmv throughout thewhole column,the resulting change in transmittance isnegative because an increase in the amount of CO₂ inthe atmosphere increases absorption. Transmittancenear the absorption lines at a higher spectral resolutionindicates a higher sensitivity to changes in CO₂concentration. In addition,the transmittance sensitivity to atmospheric CO₂ change in B and -1 is aboutfour times higher than that in B and -2 or B and -3,becausethe transmission change is proportional to absorptionby CO₂. The transmittance in B and -1 decreasesup to 0.45% at 0.2-cm⁻¹ spectral resolution;when the spectral resolution is 1.0 cm⁻¹,the maximumdecrease of the transmittance is around 0.11%,as shown in Fig. 2a. In addition,the spectral dependenceof the sensitivity to changes in CO₂ concentrationsis similar to that of the transmittance,withpeaks near the centers of the absorption b and s. Thetransmittance changes in B and -2 and B and -3 appearvery similar since these vibration-rotation transitionshave very similar spectral structures and integratedintensities(Predoi-Cross et al., 2007). The transmittancein B and -2 decreases by up to 0.10% at 0.2-cm⁻¹spectral resolution; when the spectral resolutions is 1.0cm⁻¹,the maximum decrease of the transmittance isaround 0.02%. For the 5.8-cm⁻¹ spectral resolution,the maximum decrease of the transmittance is around0.02%(Fig. 2b).

Figure 3 illustrates the relative differences betweenthe calculated transmittance sensitivity to a 1-ppmv change in CO₂ mixing ratio using the HITRAN2004 and HITRAN 2008 databases. The HITRAN2008 database results are used as the reference. Whenthe spectral resolution in B and -1 is 0.2 cm⁻¹,the maximumrelative difference is about 54.6% and the meanrelative difference is about 7.0%; when the spectralresolution is 1.0 cm⁻¹,the maximum relative differenceis about 10.4% and the mean relative difference isaround 3.1%. When the spectral resolution in B and -2is 0.2 cm⁻¹,the maximum relative difference is around18.8% and the mean difference is around 2.1%; whenthe spectral resolution is 1.0 cm⁻¹,the maximum relativedifference is around 13.3% and the mean differenceis about 2.3%,while a spectral resolution of 5.8 cm⁻¹yields a maximum relative difference of around 11.9% and a mean difference of about 2.2%. When the spectralresolution in B and -3 is 0.2 cm⁻¹,the maximumrelative difference is about 9.1%, and the mean relativedifference is around –1.7%; when the spectral resolutionis 1.0 cm⁻¹,the maximum relative difference isabout 4% and the mean difference is about –1.0%, and when it is 5.8 cm⁻¹,the maximum relative differenceis around 3.0% and the mean difference is around–1.0%. These results indicate that using the HITRAN2004 database in B and -1 or B and -2 could result inoverestimates of the sensitivity of atmospheric transmissionto atmospheric CO₂ change,while using theHITRAN 2004 database could lead to underestimatesof the transmission sensitivity in B and -3.

3.2 Transmittance calculations

Figure 4 shows relative differences between calculatedtransmittance using the HITRAN 2004 and HITRAN 2008 databases in the three NIR b and s atdifferent spectral resolutions. As above,the HITRAN2008 database is used as the reference.

The transmittance differences in B and -1 are negativefor all the cases,indicating that use of theHITRAN 2004 database leads to underestimates oftransmittance and overestimates of CO₂ absorption.The most prominent differences between the calculatedtransmission spectra occur in the absorptionbranch centers. The maximum decrease is approximately22% at the 0.2-cm⁻¹ spectral resolution(Fig. 4a). This decrease is approximately equivalent to a49-ppmv change in the CO₂ column mixing ratio ifthe transmittance sensitivity to the CO₂ concentrationchange is taken as that shown in Fig. 2 and transmittance changes are assumed to be a linear functionof CO₂ concentration changes. The maximumdecrease is about 5% when the spectral resolution is1.0 cm⁻¹; this decrease is roughly equivalent to a 45-ppmv change in the CO₂ column mixing ratio. Themost prominent differences in B and -2(Fig. 4b)occurin the spectral region 6235–6250 cm⁻¹. As withB and -1,the differences are negative. When the spectralresolution is 0.2 cm⁻¹,the maximum decrease isapproximately 2.6%,roughly equivalent to a 26-ppmvchange in the CO₂ column mixing ratio. The maximumdecreases are 0.6% and 0.5% when the spectralresolutions are 1.0 and 5.8 cm⁻¹,respectively. Thesechanges are approximately equivalent to a 25-ppmvchange in the CO₂ column mixing ratio. The updatesto the HITRAN database have the smallest effecton transmittance in B and -3,relative to the effectson transmittance in B and -1 and B and -2. Thedifferences in B and -3 are only relatively large in thespectral region 6330–6340 cm⁻¹. The differences arepositive for most of B and -3,rather than negative asfor B and -1 and B and -2. When the spectral resolutionis 0.2 cm⁻¹,the maximum change is about 1.5%.This change is approximately equivalent to a 15-ppmvchange in CO₂ column mixing ratio. The maximumchange is about 0.3% for the 1.0-cm⁻¹ spectral resolution and about 0.2% for the 5.8-cm⁻¹ spectral resolution.These changes are roughly equivalent to a10-ppmv change in CO₂ column mixing ratio.3.3 Simulated satellite retrievals

Column CO₂ abundance is typically retrieved byusing the radiance ratio between an off-line channel and an on-line channel. This ratio is often taken tobe the transmittance ratio,eliminating the need toperform complicated retrievals of surface reflectance.Accordingly,the accuracy of off-line to on-line transmittanceratio calculations has important effects onthe retrieval results. In this paper,the off-line frequenciesare chosen to be 4810.5,6196.4, and 6315.5cm⁻¹ in B and -1,B and -2, and B and -3,respectively.Absorption is relatively weak at these frequencies.

Figure 5 shows relative differences in off-lineto on-line transmittance ratio when the HITRAN2004 database is used instead of the HITRAN 2008database. The results are presented for all three b and sat different spectral resolutions. The updates to theHITRAN database have the largest impact on thetransmittance ratio in B and -1. The ratio differencesare positive for most cases in B and -1. When the spectralresolution is 0.2 cm⁻¹,the maximum ratio differencebetween the two HITRAN databases is approximately28%; when this spectral resolution is reducedto 1.0 cm⁻¹,the maximum difference is approximately5%. The discrepancy is relatively large in B and -2 inthe spectral region 6235–6250 cm⁻¹. When the spectralresolution is 0.2 cm⁻¹,the maximum relative differences are around 2.6%. The maximum difference isabout 0.57% for the 1.0-cm⁻¹ spectral resolution and 0.44% for the 5.8-cm⁻¹ spectral resolution. The effectsof the updates to the HITRAN database on the transmittanceratios in B and -3 are relatively small,withthe largest discrepancy in the spectral region 6330–6340 cm⁻¹. The ratio differences in B and -3 are negativefor most of the cases,which is opposite in sign tothe differences in B and -1 and B and -2. The maximumdecreases in the ratio are 1.5%,0.39%, and 0.24% atthe 0.2-,1.0-, and 5.8-cm⁻¹ spectral resolutions,respectively.

Figures 6a–6d show the relative changes in offlineto on-line transmittance ratio in B and -1 for increasesof 5,15,25, and 50 ppmv in CO₂ columnmixing ratio. The results are presented for each ofthe different spectral resolutions using the HITRAN2008 database. The ratios for points far from the absorptionlines are almost unchanged by perturbationsin the CO₂ mixing ratio. Changes in the ratios atpoints near the absorption lines are positive becausean increase in the amount of CO₂ in the atmosphereacts to decrease atmospheric transmittance near theselines. When the spectral resolution is 0.2 cm⁻¹,themaximum ratio changes are 2.24% for a 5-ppmv increasein CO₂ column mixing ratio,6.81% for a 15-ppmv increase,11.56% for a 25-ppmv increase, and 24.15% for a 50-ppmv increase. When the spectralresolution is 1.0 cm⁻¹,these maximum ratio changesare approximately 0.47%,1.41%,2.35%, and 4.72%,respectively. A comparison of Figs. 5 and 6 indicatesthat the updates to the HITRAN database have animpact on transmittance ratios in the narrow spectralregion 4855–4880 cm⁻¹ that is roughly equivalent toa 50-ppmv CO₂ increase. The effect in the spectralregion 4820–4850 cm⁻¹ is equivalent to an increaseof more than 25 ppmv. In addition,the results indicatethat use of the HITRAN 2004 database in B and -1could result in underestimates of the CO₂ amount.

The results presented in Fig. 7 reveal the effect ofCO₂ increases on the off-line to on-line ratio in B and -2.When the spectral resolution is 0.2 cm⁻¹,the maximumratio changes are 0.46% for a 5-ppmv increase inCO₂ column mixing ratio,1.40% for a 15-ppmv increase,2.33% for a 25-ppmv increase, and 4.67% for a50-ppmv increase. When the spectral resolution is 1.0cm⁻¹,the maximum ratio changes are approximately0.11%,0.32%,0.53%, and 1.06%,respectively. Whenthe spectral resolution is 5.8 cm⁻¹,the maximum ratiochanges are approximately 0.07%,0.21%,0.35%, and 0.69%,respectively. A comparison of Figs. 5 and 7 indicates that the differences in the line parametersbetween the HITRAN 2004 and HITRAN 2008databases have an effect on the ratios in the narrowspectral region 6195–6225 cm⁻¹ that is roughly equivalentto a 10-ppmv increase in CO₂ column mixing ratio.The effect is equivalent to an increase of morethan 25 ppmv in the spectral region bounded by 6235 and 6255 cm⁻¹. As with B and -1,use of the HITRAN2004 database would likely lead to an underestimateof the CO₂ amount.

Figure 8 presents the results of a similar analysisfor B and -3. When the spectral resolution is 0.2 cm⁻¹,the maximum ratio changes are 0.47% for a 5-ppmvincrease,1.40% for a 15-ppmv increase,2.33% for a25-ppmv increase, and 4.67% for a 50-ppmv increase.When the spectral resolution is 1.0 cm⁻¹,the maximumratio changes are approximately 0.09%,0.28%,0.47%, and 0.92%,respectively. When the spectralresolution is 5.8 cm⁻¹,the maximum ratio changesare approximately 0.08%,0.23%,0.39%, and 0.77%,respectively. A comparison of Figs. 5 and 8 indicatesthat the differences in the line parameters between theHITRAN 2004 and HITRAN 2008 databases have aneffect on the ratios in the narrow spectral region 6325–6345 cm⁻¹ that is roughly equivalent to a 25-ppmvdecrease in CO₂. The effect in the spectral region6350–6375 cm⁻¹ is equivalent to a change of less than10 ppmv. These results indicate that the amount ofCO₂ may be overestimated when the HITRAN 2004database is used to perform retrievals in B and -3.

4. Conclusions

The line parameters included in the latest editionof the HITRAN database have been modified since theprevious release in 2004. The HITRAN 2008 databaseprovides a complete replacement of the HITRAN 2004CO₂ line parameters in the NIR region from 4300 to7000 cm⁻¹. A series of sensitivity studies have beenperformed to evaluate the effects of these updates inthe HITRAN database on retrievals of atmosphericCO₂ from near-infrared spectra.

Our comparisons of transmittance sensitivity toa change of 1 ppmv in CO₂ mixing ratio using theHITRAN 2004 and HITRAN 2008 databases indicatemaximum relative differences of up to 54.6% at a spectralresolution of 0.2 cm⁻¹. Use of the HITRAN 2004database in the 2.06- and 1.61-μm regions could resultin an overestimate of the transmittance sensitivity toatmospheric CO₂ changes; use of the HITRAN 2004database in the 1.58-μm region could result in an underestimateof the transmittance sensitivity.

The updates to the HITRAN database have thelargest impact on transmittance and the off-line to onlinetransmittance ratio in the 2.06-μm spectral region and the smallest impact on these parameters in the1.58-μm spectral region. The influence of the updatesto the HITRAN database on the off-line to on-line ratiocalculation in the narrow spectral region 4855–4880cm⁻¹ in B and -1 appears to be roughly equivalent to achange in CO₂ of more than 50 ppmv. This influenceis approximately three times larger than that of cirrusclouds and bright maritime aerosols,both of whichare considered influential factors in the retrieval ofatmospheric CO₂ using NIR reflected sunlight(Mao and Kawa, 2004). Using the HITRAN 2004 databasein B and -1 or B and -2 leads to an underestimate ofthe column CO₂ abundance,whereas using the samedatabase in B and -3 leads to an overestimate of thecolumn CO₂ abundance. These findings suggest thatinadequacies in the forward model could be a majorsource of error in measurements of CO₂ from space.Additional studies are needed to properly evaluatethe impact of uncertainties in the forward radiativetransfer model on the accuracy of CO₂ retrievals. Simulatedretrievals have been performed using optimalestimation theory to examine the effects of the updatesto the HITRAN database on retrievals of CO₂vertical columns; the results have been reported inanother paper(Dai et al., 2012).

These results highlight that careful attentionmust be paid to changes in spectroscopic parametersto minimize systematic errors and improve theaccuracy of retrievals of atmospheric CO₂ using NIRreflected sunlight.

Acknowledgments. We thank the relevant PIs and their staff for establishing and maintaining theHITRAN database used in this investigation. We alsoacknowledge the useful comments by two anonymousreviewers that helped to improve this paper.