The vertical and horizontal distribution of CO 2 densities in the upper mesosphere and lower thermosphere as measured by CRISTA

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 17, NO. D3, 818, doi:1.19/1jd74, The vertical and horizontal distribution of CO densities in the upper mesosphere and lower thermosphere as measured by CRISTA M. Kaufmann, O. A. Gusev, and K. U. Grossmann Department of Physics, University of Wuppertal, Wuppertal, Germany R. G. Roble, M. E. Hagan, and C. Hartsough High Altitude Observatory, National Center for Atmospheric Research, Boulder, Colorado, USA A. A. Kutepov Institute for Astronomy and Astrophysics, University of Munich, Munich, Germany Received April 1; revised 8 September 1; accepted 8 September 1; published 18 October. [1] The Cryogenic Infrared Spectrometers and Telescopes for the Atmosphere (CRISTA) experiment measured the global distribution of CO 4.3 mm infrared emissions in the mesosphere and lower thermosphere during two Space Shuttle missions in November 1994 and August The daytime radiances have been inverted to CO number densities in the 6 13 km range by using a nonlocal thermodynamic equilibrium model. A detailed sensitivity study of retrieved CO number densities was carried out. The O( 1 D) excitation mechanism and model parameters constitute the most important uncertainties of retrieved CO, typically 1 %. The inaccuracy due to uncertainties in other atmospheric parameters is usually less than 1%. The CO volume mixing ratio (VMR) deviates from being well mixed between 7 and 8 km, which is significantly lower than indicated by previous rocket-borne mass spectrometer data and model calculations but is in good agreement with the data obtained by other 4.3 mm emission and absorption experiments. The global distribution of CRISTA- CO density shows significant longitudinal and latitudinal structures. The zonal mean CO densities are increasing toward polar summer latitudes below 9 km and above 11 km. Between 9 and 11 km, the latitudinal gradient is reversed. At 1 km, the gradient is mostly pronounced, reaching up to % difference between low and high latitudes. These variations are compared with results obtained by the Thermosphere/Ionosphere/Mesosphere Electrodynamics General Circulation Model (TIME-GCM), showing very good agreement for the latitudinal distribution. Below 11 km, this variation is mostly due to the change in total density rather than to the CO VMR. INDEX TERMS: 3394 Meteorology and Atmospheric Dynamics: Instruments and techniques; 336 Meteorology and Atmospheric Dynamics: Remote sensing; 34 Atmospheric Composition and Structure: Middle atmosphere composition and chemistry; 3 Atmospheric Composition and Structure: Thermosphere composition and chemistry Citation: Kaufmann, M., O. A. Gusev, K. U. Grossmann, R. G. Roble, M. E. Hagan, C. Hartsough, and A. A. Kutepov, The vertical and horizontal distribution of CO densities in the upper mesosphere and lower thermosphere as measured by CRISTA, J. Geophys. Res., 17(D3), 818, doi:1.19/1jd74,. 1. Introduction [] Carbon dioxide is very important for the energy balance of the Earth s atmosphere. Its infrared emission at 1 mm is the dominant radiative cooling mechanism in the stratosphere, mesosphere and lower thermosphere. In the troposphere the absorption and reemission of thermal radiation from the Earth s surface implies that CO is one of the most important greenhouse gases. Its concentration in the Copyright by the American Geophysical Union //1JD74 atmosphere has increased greatly over recent decades (1 ppmv yr 1 )[Brasseur et al., 1999] due to anthropogenic influences, mainly by fossil fuel burning and changes in land use. The symptoms of this increase are the global warming of the Earth s surface caused by an increased greenhouse effect and an enhanced cooling of the middle and upper atmosphere due to the loss of CO infrared radiation to space [Roble and Dickinson, 1989]. The increase of CO near the surface propagates upwards reaching the upper stratosphere after 6 years [Bischof et al., 198]. [3] CO has a nearly constant volume mixing ratio (VMR) in the entire lower and middle atmosphere owing CRI 1-1

2 CRI 1 - KAUFMANN ET AL.: CRISTA CO DENSITIES to turbulent mixing and the absence of significant sources or sinks. The tropospheric source induces a small negative slope of CO VMR with altitude. In the upper mesosphere the CO VMR is depleted through molecular diffusion and destroyed by solar UV radiation and higher up photochemically by collisions with O + [Trinks and Fricke, 1978]. [4] The knowledge of the CO density in the upper mesosphere and lower thermosphere (UMLT) is not only important for the energy balance. The height of the turbopause and its global structure is an indicator of the mixing processes and transports in this region. Their knowledge is particularly important because the UMLT is often used as a boundary layer for atmospheric models (e.g., in the D model of Garcia and Solomon [198]). In addition, accurate CO density profiles are essential for the derivation of kinetic temperatures from the CO 1 mm emissions. [] The first measurements of CO densities in the upper atmosphere were performed by rocket-borne mass spectrometers [Offermann and Grossmann, 1973; Philbrick et al., 1973; Trinks and Fricke, 1978; Offermann et al., 1981]. Mass-spectrometric measurements are in situ measurements, which can only be done by means of sounding rockets. The data is therefore limited to a small number of single profiles. [6] A different method of deriving CO densities is the measurement of its infrared emissions or absorptions. This can be done with sensors mounted on sounding rockets or on satellites, which allow for global data coverage. Absorption or occultation measurements usually take the Sun as a light source during sunrise or sunset. The advantage of this technique is that the absorption depends only on the CO density, the kinetic temperature and the pressure and not on the excitation of the CO molecules. Solar occultation measurements of carbon dioxide were performed with the Grille spectrometer on board Spacelab 1 [Girard et al., 1988] and by the Atmospheric Trace Molecule Spectroscopy (ATMOS) instrument on Spacelab 3 [Rinsland et al., 199] and in the Atmospheric Laboratory for Applications and Science (ATLAS) 1, and 3 missions [Kaye and Miller, 1996]. For low Earth orbit altitudes of a few hundred km the number of sunrises and sunsets is about 16 each per day resulting in only 3 vertical profiles per measurement day. [7] Emission measurements overcome these limitations in the local time coverage and the comparative sparseness of occultation data. Rocket-borne emission measurements of the CO n 3 band at 4.3 mm were performed with the SPIRE (Spectral Infrared Rocket Experiment) [Wintersteiner et al., 199; Nebel et al., 1994] and SISSI (Spectroscopic Infrared Structure Signatures Investigation) [Vollmann and Grossmann, 1997] experiments. Several other rocket experiments [Stair et al., 197, and references therein] examined this emission due to its sensitivity to auroral energy deposition but without attempting to derive CO densities. The Stratospheric and Mesospheric Sounder (SAMS) on Nimbus 7 [López-Puertas and Taylor, 1989] measured for the first time the global distribution of CO 4.3 mm radiances up to 1 km from 1978 to Its successor, the Improved Stratospheric and Mesospheric Sounder (ISAMS) on the Upper Atmosphere Research Satellite (UARS) [López-Puertas et al., 1998b; Zaragoza et al., ] measured primarily the CO 4.3 mm second major isotope emissions up to 1 km. [8] The main difficulty of deriving CO from its infrared emissions is to model the occupancy of its vibrational states, Table 1. CRISTA Measurements in the UMLT Region Mode TH a, km Day time, UT Number of scans SAT b, km CRISTA M/T CRISTA- M T M T a Tangent heights. b Step along track. which are not in local thermodynamic equilibrium (LTE). The n 1 mm fundamental band departs appreciably from LTE at 8 9 km whereas non-lte effects in the n 3 bands must be taken into account down to the stratosphere. In case of LTE the excitation is solely determined by collisions with the ambient atmosphere and the vibrational excitation follows Boltzmann s distribution law for the local kinetic temperature. If the density decreases, nonlocal excitation mechanisms become more relevant. Above 9 km the excitation of n vibrational modes is primarily determined by three processes: (1) the absorption of radiation from the lower atmosphere, () the radiative loss of energy and (3) the collisional excitation and deexcitation by the major constituents of the atmosphere, in particular ground state atomic oxygen (O( 3 P)). The most important quantities in this context are the kinetic temperature, O( 3 P) densities and the quenching constant for collisions between CO and O( 3 P). A more complete description of the n excitation mechanisms can be found in the studies of Wintersteiner et al. [199], Edwards et al. [1993], and Ogibalov et al. [1998]. In contrast to the n mode, the excitation of the n 3 mode changes strongly from day to night. During nighttime, the departure of the n 3 levels from LTE is strongly influenced by the interaction of radiation from different atmospheric layers. Additionally, there is evidence that highly excited hydroxyl molecules stimulate the CO asymmetric stretch mode [Kumer et al., 1978]. During daytime n 3 vibrations are primarily excited by the absorption of solar radiation. The radiative interaction of different atmospheric layers and collisional processes especially with highly excited O( 1 D) molecules (via N ) are other important mechanisms [Nebel et al., 1994; Edwards et al., 1996]. The kinetic temperature and the O( 3 P) densities are only of minor importance. Therefore 4.3 mm daytime emissions can be modeled quite well and they are very suitable for deriving CO densities in the UMLT. [9] Cryogenic Infrared Spectrometers and Telescopes for the Atmosphere (CRISTA) is a limb-scanning satellite experiment [Offermann et al., 1999; Riese et al., 1999; Grossmann et al., ] and was flown twice aboard the Astronomical Shuttle Pallet Satellite (ASTRO-SPAS) freeflying platform on the Space Shuttle missions STS 66 (1994) and STS 8 (1997). The orbit altitude was 3 km, and the orbit inclination was 7 during the missions. Atmospheric limb measurements were taken from 4 1 November 1994 and 8 16 August CRISTA makes global measurements of trace-gas thermal emissions in the 4 71 mm region with a high horizontal and vertical resolution and with a

3 KAUFMANN ET AL.: CRISTA CO DENSITIES CRI Latitude [deg] CRISTA 1 CRISTA SZA [deg] Figure 1. SZA-latitude coverage of CRISTA-1 and CRISTA-. Shaded areas indicate twilight or nighttime data. medium spectral resolution (l/l ). The radiance is analyzed by means of Ebert-Fastie grating spectrometers in 13 different wavelength intervals. CO emissions in the UMLT are recorded in the 1 mm and 4.3 mm regime with good signal-to-noise ratio up to km. From the 4.3 mm data CO densities are derived during daytime for single profiles up to about 13 km. [1] A number of different measurement modes were performed, including scans of different altitude regions of the atmosphere. Table 1 gives an overview of the CRISTA measurements in the UMLT-region including the number of vertical profiles measured. The vertical step is.1.6 km in the UMLT during these observation modes. The width of the vertical field of view is about 1. km. To obtain a high horizontal resolution, three telescopes are used to sound the atmosphere simultaneously. However, the 4.3 mmco channel exists only in one of the telescopes. The latitudinal coverage of the center telescope is from S to 63 N for CRISTA-1 and from 71 Sto71 N for CRISTA- in the upper mesosphere. For a given latitude, observations are made at two different local solar times and at different solar zenith angles (SZA). Figure 1 shows the SZA-latitude coverage for both CRISTA missions. Daytime measurements (SZA < 8 ) cover S to 63 N for CRISTA-1 only in ascending nodes of the orbit. CRISTA- daytime measurements are available from 1 S to 7 N. South of 3 N daytime data was measured only in the ascending node of the orbit. North of 3 N the local time shift is up to 1 hours between the ascending and descending nodes of the orbit. Both CRISTA flights took place near the minimum of the solar cycle and under mostly geomagnetically undisturbed conditions.. The Non-LTE Model of CO [11] The vibrational states of CO can be excited or deexcited by collisional and radiative processes. Collisional processes imply the exchange of thermal and vibrational energy in vibrational-translational (V-T) or vibrationalvibrational (V-V) collisions or the transfer of electronic energy. The quenching of electronically excited O( 1 D) by N leads to an excitation of N vibrations. The N vibrational energy is then transferred to CO n 3 vibrations with high efficiency due to the closeness of its vibrational energies [López-Puertas et al., 1998a]. Radiative processes include the exchange of photons between different atmospheric layers and the absorption of solar radiation. [1] The modeling of the vibrational excitation separates into two major parts: (1) the solution of the radiative transfer equation, which relates the local radiation field to the excitation of CO in the entire atmosphere and () the coupling of all possible vibrational states assuming steady state (statistical equilibrium equations). Both parts are calculated very efficiently with the non-lte model of Kutepov et al. [1998]. The multilevel non-lte problem is solved using the Accelerated Lambda Iteration (ALI) technique [Rybicki and Hummer, 1991; Kutepov et al., 1998] and the radiative transfer is calculated line-by-line using the Feautrier method [Rybicki and Hummer, 1991]. It is assumed that the molecules are in rotational LTE. The size of the non-lte model is limited to vibrational states with energies below 1 cm 1 for the CO main isotope. The sets of vibrational states, which are considered for the CO minor isotopes, are usually smaller; Table lists the upper limits of vibrational states. For the CO main isotope our set is identical to the set used for the interpretation of the ISAMS 4.3 mm data [López-Puertas et al., 1998a]. This includes levels having one or two n 3 (asymmetric stretch) excitations and levels from the bandstretch manifold (n 1, n and n 1 n excitations) up to four n quanta. Radiative transfer and absorption of solar radiation is computed for all lines available in the 1996 HITRAN database [Rothman et al., 1996] for the selected set of vibrational states. For the solar flux at the top of the atmosphere blackbody radiation at Table. Upper Limits of Vibrational States for the Different CO Isotopes Isotope Level Energy, cm N O 1 16.

4 CRI 1-4 KAUFMANN ET AL.: CRISTA CO DENSITIES CRISTA- 3 4 CRISTA-1 1e-9 1e-8 1e-7 1e-6 Intensity [W / cm sr] Figure. Integrated radiances (3 39 cm 1 ) for different SZAs. The profiles were averaged in specific SZA intervals; latitudinal and time coverages were not restricted. The number of profiles that enter the average is given in brackets; error bars represent the standard deviation. The random error of the instrument for single profiles is indicated by the vertical lines. (1) CRISTA-1, SZA: 48 (14). () CRISTA-1, SZA: 76 8 (). (3) CRISTA- 1, SZA > 11 (64). (4) CRISTA-, SZA: 48 (13). () CRISTA-, SZA: 76 8 (163). (6) CRISTA-, SZA > 11 (). 1 interval 3 39 cm 1. Typical differences are 3% above 1 km and % between 6 and 8 km. Below 6 km the differences decrease. Also included in Figure are two nighttime profiles (SZA > 11 ) from the two CRISTA missions. Radiances are lower by a factor of ten or more at all altitudes above 6 km as compared to the daytime data. Radiances are very similar for CRISTA-1 and CRISTA-. The dependence of the CO 4.3 mm limb radiances on the SZA is included in all non-lte models for the CO 4.3 mm daytime excitation [López-Puertas and Taylor, 1989; Nebel et al., 1994; Shved et al., 1998]. For the calculation of the non-lte limb radiances we developed a fast line-by-line radiation code. The equation of radiative transfer is solved using the integral approach [Goody and Yung, 1989]. The Voigt profile is used for line shape modeling. Source function and absorption coefficients are calculated for each line separately and without overlap. The wave number range for frequency integration depends on the tangent height and the line strength. The atmospheric layering is performed using Curtis-Godson absorberweighted mean values [Goody and Yung, 1989] assuming homogeneous illumination conditions along the line of sight. The radiation code has been tested against the GENLN transmittance and radiance model [Edwards et al., 199, 1993]. The differences are very small (<%) above 8 km. At altitudes below 9 km differences increase due to line overlapping, which is not implemented in our code. To estimate the magnitude of line overlapping, two GENLN runs were performed. The first run includes overlap between all lines and all bands and the second run assumes that all lines are completely isolated from each other as assumed by a line-independent line-by-line model. The results of this comparison are shown in Figure 3 (thick 76 K is assumed. The set of collisional rate constants was taken from the work of Shved et al. [1998]. For thermal collisions of CO (n ) with atomic oxygen we use a value of (T/3) 1/ cm 3 s 1 as recommended by López- Puertas et al. [1998a]. A detailed description of the individual excitation processes of vibrational levels is given in a number of papers [e.g., Shved et al., 1998; López-Puertas et al., 1998a]. 3. Limb Radiance in the CRISTA 4.3 Mm Channel [13] The CRISTA 4.3 mm channel (SCS ) covers the wave number interval cm 1. The spectral resolution (l/l) is 41 and the wave number increment between two subsequent spectral sampling points is about 1.4 cm 1. During daytime the wave number integrated 4.3 mm band exhibits a good signal-to-noise ratio (S/N > 1) up to about 1 km (14 km) for CRISTA-1 (CRISTA-) and up to 7 km (9 km) at night, respectively. Consequently, CO densities in the UMLT can only be derived during daytime as stated above. [14] At altitudes above km the daytime signal shows a strong variation with the solar zenith angle. Figure shows a comparison of radiance profiles measured by CRISTA-1 and CRISTA- for different solar zenith angles (SZA = 48 and SZA = 76 8 ). The radiance profiles are obtained by spectral integration of all intensities in the Difference [%] Figure 3. Decrease of simulated limb radiance (3 39 cm 1 ) due to line overlapping. (1) N, 177 W, SZA =. () Same as 1, but CO reduced by 1%. (3) Same as 1, but temperature increased by 1 K. (4) Same as 1, but SZA = 8.

5 KAUFMANN ET AL.: CRISTA CO DENSITIES CRI 1-1e9 Total CO 66 FB CO Minor Isotopes CO Hotbands CO O3 N 1 km 1 km 1e1 Intensity [W / cm sr cm -1 ] 1e11 1e8 8 km 6 km 1e9 1e Wavenumber [cm -1 ] Figure 4. Simulated spectral contributions in the CRISTA 4.3 mm channel (SZA = 8 ) for tangent heights specified in the plot labels (1, 1, 8, and 6 km). Note the different intensity scaling of the top and the bottom panels. line). At km altitude, differences in radiance are about 1% and decrease rapidly to % at 6 km and % at 8 km. The dependence of line overlapping on the excitation of the molecules or on the radiative transfer is influenced by various atmospheric parameters and is illustrated by the additional profiles in Figure 3. Below 6 km it varies by 3% on a change of the SZA from to 8. Its dependence on temperature and CO density is less than 1% above 6 km tangent height, indicating that the effect is essentially independent of the atmosphere at least above 6 km. [1] The effects due to line overlapping are therefore included in our radiative transfer code by application of a correction function. This function depends only on the SZA as explained above. [16] Limb spectra measured in the cm 1 wave number interval are composed of the emissions from several atmospheric species and from different bands of these species. Simulated daytime limb emission spectra for various tangent heights convoluted by the instrument line shape function of CRISTA are shown in Figure 4. The atmospheric parameters are taken from the works of Wintersteiner et al. [199] (CO density), Manuilova and Shved [199] (O 3 density), Echle et al. [1994] (average CO profile), USSA76 [U.S. Standard Atmosphere, 1976] (N O), and MSIS [Hedin et al., 1991] (temperature, pressure, atomic oxygen; for equator in November 1994). The rovibrational excitation of all molecules was calculated simultaneously using the non-lte model of Kutepov et al. [1998]. The collisional models are described by Manuilova et al. [1998] (O 3 ), Shved and Gusev [1997] (N O), and Kutepov et al. [1997] (CO). The model of CO was described in section. [17] In the thermosphere highly excited NO + emits in the 4.3 mm wavelength region, too [Winick et al., 1987; Picard et al., 1987; Caledonia et al., 199; Dothe et al., 1996]. We estimate the intensity of NO + using limb line-of-sight densities modeled by the strategic high altitude radiance code (SHARC) for the CIRRIS 1A experiment [Dothe et al., 1996]. Because NO + depends strongly on the solar activity we used NO + calculations of the Thermosphere/ Ionosphere/Mesosphere Electrodynamics General Circulation Model (TIME-GCM) (Roble, personal communication) to scale the SHARC calculation (for medium solar activity) to the CRISTA conditions (solar minimum). [18] The CRISTA SCS channel is optimized for CO and CO emissions. The dominant spectral features (Figure 4) are from the CO main isotope (66) fundamental band (FB), various CO hotband and minor isotope emissions and from CO. In the 39 cm 1 region, there are predominantly emissions from CO. At 1 km at nonpolar latitudes NO + emissions contribute approximately 3 4% to the total radiance between 3 and 39 cm 1, but the contribution increases by a factor 3 toward polar latitudes. Between 79 and 18 cm 1 the main spectral feature is the CO(1-) transition. In the lower mesosphere O 3 (11-, and 1-1) emissions around 11 cm 1 give % of the signal in this wavelength region. The contribution of N O is noticeable between 1 and cm 1 only, reaching 3% of the radiance at 6 km at these wavelengths.

6 CRI 1-6 KAUFMANN ET AL.: CRISTA CO DENSITIES cm cm Cumulated Radiance [%] Figure. Portion of intensity that stems from the tangent layer and the atmospheric layers above. The vertical binning is 1 km. The calculation is for N, 179 E, SZA = (8 November 1994). Solid lines summarize emissions from the CO main isotope (3 39 cm 1 ) and dashed lines represent the part of the spectrum with strong CO main isotope FB emissions. [19] Figure shows the part of radiance that stems from atmospheric layers close to the tangent point. At 1 km tangent height the emissions of the CO 66 FB dominate the signal and 6% of the total radiance (3 39 cm 1 ) is emitted from a km thick layer above the tangent point and another % of the radiation originates from 1 to 13 km. At 11 km the fundamental band becomes optically thick, because the spectral lines are only Doppler broadened and very narrow. Emissions from highly excited hotbands become important and dominate the shape of the spectrum. Their contribution increases and is maximum at 8 km and implies that the optical thickness remains constant from 1 km to 8 km. At 8 km tangent height % of the radiation between 3 and 39 cm 1 originates from a km thick layer above the tangent point, whereas in the cm 1 wavelength region, which is solely determined by CO 66 FB emissions, only 1% of the radiance stems from the vicinity of the tangent point. By contrast, below 8 km the pressure is large enough for the Lorentzian wings of strong lines to contribute appreciably to the total radiance. At 6 km tangent height % of the radiance between 3 and 39 cm 1 is from km above the tangent point and in the cm 1 region as much as 3%. Therefore the contribution of the 66 FB to the spectral shape is increasing again. [] Although the optical thickness of the atmosphere is very high for 4.3 mm limb radiance, the vertical resolution of retrieved CO is not much affected by it, because the retrieval steps downward through the atmosphere, retrieving products at the tangent level using the radiances just from that level and values of the already retrieved products above the tangent level. The vertical resolution of CRISTA depends primarily on the tangent height step (.1.6 km). 4. Retrieval Technique [1] The vibrational excitation of CO depends only weakly on CO number densities. This allows the vibrational excitation modeling and the inversion of limb radiances to CO number densities to be split. [] Therefore the retrieval of CO density can be performed iteratively in two steps. In the first step we calculate the vibrational excitation of CO using the non-lte model. In the second step we freeze this excitation and simulate measured radiances with the forward radiance model and retrieve CO number densities using the onion-peeling technique [Russell and Drayson, 197].. Sensitivity Study for the CO Retrieval [3] In this section we discuss the effect of the most important atmospheric quantities on the derivation of CO densities. The non-lte model parameters that most affect the 4.3 mm radiances are discussed by Kutepov et al. (in preparation)..1. Variations of 4.3 Mm Limb Radiances [4] Due to the optical thickness of the 4.3 mm limb radiance in the UMLT, retrieved CO densities exhibit a strong nonlinear response to intensity variations. We examine the sensitivity k ¼ CO =CO I=I, which describes the propagation of uncertainties in intensity (I) into retrieved CO densities. Tangent Height [km] Sensitivity k Figure 6. Sensitivity k of retrieved CO to radiance shifts. (1) Response of CO density to a global radiance shift (4%). Error bars indicate the variation of k according to the magnitude of other shifts ( 8%). () Response of CO to radiance shifts (4%) only at the tangent height assuming a. km vertical sampling of the atmosphere. Error bars represent 8% shifts. 1

7 KAUFMANN ET AL.: CRISTA CO DENSITIES CRI CRISTA-1 GARCIA HUNT SHIMAZAKI RODRIGO TIME-GCM O( 1 D) Density [cm -3 ] Figure 7. Various O( 1 D) profiles. The thick solid line shows the mean CRISTA-1 O( 1 D) profile; error bars indicate the standard deviation. The thin solid line shows the extension of the mean CRISTA-1 profile into the profile of Rodrigo et al. [1991] as it is used in the retrieval. Other profiles are from the works of Hunt [1971] (dashed line; for equinox, equator, noon), Shimazaki and Laird [197] (double-dot dashed line; for equinox, equator, noon and low solar activity), Rodrigo et al. [1991] (dot dashed line; for equinox, midlatitudes, noon and moderate solar activity), and TIME-GCM (crosses; for the CRISTA- time period; noon; latitude = 6. ). [] Figure 6 (solid line) shows the response of CO density to a radiance shift of 4% at all altitudes simultaneously. k is equal to 1 above 1 km as one expects for optically thin emissions. At 11 km the 4.3 mm FB emission becomes optically thick, which is expressed in larger values for k (. 3). Between 8 and 1 km the sensitivity drops and typical k values are 1.. Below 8 km k increases rapidly to 3 4. This behavior reflects the optical thickness of the various contributions to the spectrum according to Figure 4. At selected altitudes error bars illustrate the sensitivity to intensity variations by % or 8% respectively. Figure 6 (dashed line) also shows the response of retrieved CO to local radiance shifts and therefore we applied a 4% intensity shift at the tangent height only. Such an assumption would for instance simulate a noise peak in the detector signal at a given tangent height. The variation of k in the case of 8% shifts is again expressed in the error bars. For the calculation we assumed a. km vertical binning similar to the sampling of CRISTA. Corresponding k-factors are usually doubled compared to the global shift. This is due to the limb geometry and the interpolation of adjacent vertical layers (e.g., at 1 km only % of the radiance stems from the. km width tangent layer, i.e., CO density must be changed by x to change the limb intensity by x). The larger difference between the two k-values is in the lower mesosphere, where only a small fraction of the radiance comes from the tangent point... O( 1 D) [6] The excitation of CO n 3 vibrations by quenching of O( 1 D) (via N ) gives a significant enhancement of 4.3 mm limb radiances. O( 1 D) is produced by the photodissociation of (1) O (l < 176 nm) and () O 3 (l < 31 nm) and lost in (3) collisions with N or O. Below 8 km the fraction of reaction (1) of the total O( 1 D) production is less than 3% and can even be neglected below 7 km. The second process is not important above 9 km. The production of O( 1 D) is strongly dependent on local photochemical conditions. O( 1 D) is derived under the assumption of photochemical equilibrium [Brasseur and Solomon, 1986] from CRISTA O 3 densities. Photolytic rate constants for () are calculated by means of a photochemical model [Lary, 1991]. The constants for (1) are not implemented in the model and therefore are taken from the work of Allen et al. [1984] for noon conditions and for altitudes below 1 km. Quenching constants for (3) are compatible with the work of DeMore et al. [1997]. Above 1 km O( 1 D) is extended smoothly into the profile of Rodrigo et al. [1991]. [7] Figure 7 shows a comparison of the mean CRISTA-1 O( 1 D) profile and several model calculations. In the mesosphere CRISTA O( 1 D) is within the range of the model profiles, although CRISTA covers SZAs from 3 to 8, whereas the model calculations except TIME-GCM were performed for equinox noon conditions. In the lower thermosphere CRISTA O( 1 D) coincides with the calculation performed by Shimazaki and Laird [197], which is higher Difference [%].. Figure 8. Sensitivity of retrieved CO density to global shifts of O( 1 D). The reference profile was taken from the work of Rodrigo et al. [1991]; the background atmosphere and SZA are the same as in Figure. The dotted line shows a calculation without O( 1 D). For the other curves, O( 1 D) was scaled by the quoted factors. no

8 CRI 1-8 KAUFMANN ET AL.: CRISTA CO DENSITIES CO [%] Τ [Κ] Figure 9. Different temperature shifts (right panel) and the corresponding changes in retrieved CO density (left panel); SZA =. than other models by a factor. Differences between the different model calculations are quite large and a large uncertainty must therefore to be attributed to the O( 1 D) densities used for the CO retrieval (see below). [8] Figure 8 shows the response of retrieved CO densities to uncertainties in O( 1 D). Retrieved CO densities do not respond linearly to an enhancement or decrease in O( 1 D). Above 1 km a reduction of O( 1 D) by a factor of increases CO by 1%, whereas a similar enhancement decreases CO by 1 1%. The influence of O( 1 D) excitation is at all altitudes above 1 km strongly nonlocal due to radiative coupling. At 14 km the influence on CO is totally affected by the O( 1 D) quenching processes below 1 km. The strongest sensitivity appears between 1 km and 11 km, because the flux of solar photons, which can be absorbed in the CO FB, decreases rapidly in these altitudes and the excitation by O( 1 D) becomes relevant. In the middle mesosphere solar-pumped hotband emissions dominate the 4.3 mm signal and their dependence on O( 1 D) is smaller (<1%)..3. Kinetic Temperature and Pressure [9] The dependence of CO vibrational excitation on kinetic temperature variations is discussed in detail by Shved et al. [1998] and consists primarily of three mechanisms: (1) the shift of solar pumping due to density changes, () the (inverse) temperature dependence of vibrationalvibrational energy exchange between CO and N, and (3) the change of the rotational distribution within a vibrational band and its effect on the opacity. However, the relative population of excited states that emit 4.3 mm radiance is only weakly dependent on the kinetic temperature and therefore retrieved CO density versus altitude depends weakly on temperature. [3] Figure 9 shows the sensitivity of retrieved CO to different temperature shifts. The magnitudes of the shifts were chosen with regard to the temperature uncertainties in the retrieval and will be discussed later. Hydrostatic equilibrium is adjusted in all cases. Cases 1 assume constant temperature shifts of, 3, and ±1 K. Case 6 shifts the kinetic temperature only in the lower thermosphere to isolate net thermospheric temperature effects. Otherwise CO [%] (decrease of O( 3 P)) CO [%] (increase of O( 3 P)) Figure 1. Effect of different scalings of O( 3 P) on retrieved CO density; the magnitude is displayed in the legends. Arrows assign curves and scales.

9 KAUFMANN ET AL.: CRISTA CO DENSITIES CRI 1-9 Table 3. Systematic Errors of Retrieved CO Densities z, km O( 1 D) O( 3 P) T KIN NLTE model Retrieval Instrument a bias, % bias, % bias, % tides b, % bias, % bias, % bias, % c c d e 6 4 a CRISTA-1. For CRISTA-, 1% larger. b Tidal perturbations vary a factor of two or more depending on latitude and local solar time. c Due to the neglect of NO +. d For polar summer profiles appreciably higher. e Due to the normalization of CRISTA radiances corresponding to a CO VMR of 36 ppmv. upwelling radiation would mix temperature dependencies in the thermosphere with those in the mesosphere. The dependence of the CO density retrieval on the kinetic temperature varies strongly with altitude. In the thermosphere retrieved CO densities depend very weakly on local temperature variations (<%, Case 6). Between 9 and 1 km and below 7 km the retrieval does strongly depend on mesospheric temperature shifts; a 1 K temperature shift in the mesosphere affects retrieved CO up to 4% at 1 km. Between 7 and 9 km the sensitivity on temperature variations is significantly lower (<%). Depending on the trend of temperature shifts in CO, VMR errors are increased or decreased in comparison to errors in CO number densities..4. Atomic Oxygen [31] Atomic oxygen plays a significant role in quenching CO vibrational modes in the UMLT [Houghton, 1969; Sharma and Wintersteiner, 199], which can affect the 4.3 mm radiance in different ways. An enhancement of O( 3 P) reduces the population of the CO (11) vibrational mode, which is the main emitter of 4.3 mm radiance in the lower thermosphere. For CO n modes in the lower thermosphere the effect is reversed: An enhancement of O( 3 P) couples the CO n modes closer to the neutral atmosphere and for this reason the vibrational excitation of the n modes is enhanced. This affects the 4.3 mm radiance in two ways: (1) The CO (1111) vibrational mode, which is also an important emitter of 4.3 mm radiance and which is radiatively pumped from the first n vibration, is also enhanced. () The vibrational partition sum (Q v ), which is primarily determined by the population of the n modes, is enhanced. For optically thin conditions the limb radiance depends on the inverse of Q v in that part of the atmosphere that is responsible for the emission [Edwards et al., 1998] and an enhancement of O( 3 P) decreases 4.3 mm limb radiances. This effect is important above 1 km and becomes dominant above 1 km. [3] The net effect of all three parts is a decrease of 4.3 mm limb radiance and therefore retrieved CO will increase if atomic oxygen is enhanced. Figure 1 shows the response of retrieved CO to variations of atomic oxygen. Assuming an uncertainty factor of in O( 3 P), retrieved CO changes less than 1% below 9 km, but increases rapidly up to 8 13% at 11 km and decreases to 4% at 1 km. 6. Error Analysis [33] The uncertainty components of retrieved CO are grouped in two categories: 1. Random effects (imprecision): random errors are primarily detector noise and fluctuations of the atmosphere, which are not implemented in the model atmospheres. It is assumed that the random error decreases with the number of measurements obtained under similar conditions.. Systematic effects (inaccuracy): systematic errors (or bias) of retrieved CO result primarily from (1) calibration uncertainties and instrument effects, () uncertainties in various parameters used in the non-lte model and approximations in the retrieval algorithm, and (3) uncertainties of atmospheric parameters, which may not vanish on average (e.g., tides). Table 3 summarizes the systematic Table 4. Random Errors of Retrieved CO Densities z, km O( 1 D), % O( 3 P), % T KIN, % Instrument, % Total, % CR1 CR CR1 CR a a >1 a > >1 a > a Given for low SZA. For high SZAs, the error is approximately % larger.

10 CRI 1-1 KAUFMANN ET AL.: CRISTA CO DENSITIES errors and Table 4 the random errors of retrieved CO. Sources of error and their magnitude are discussed in the sections that follow Measurement Uncertainties [34] The calibration of the CRISTA 4.3 mm channel contains several classes of uncertainties. Raw data corrections and the radiometric and wavelength calibrations are described in detail by Riese et al. [1999]. The most significant error in the radiometric calibration of the 4.3 mm channel is stray-light in the spectrometer. To eliminate this effect, we developed a correction model, which is compatible with the assumption of 36 ppmv CO VMR at 6 km tangent height. For CRISTA-1 this correction model changes the calibration performed in the laboratory by 3 ± 1.% (6 km) to 1 ± % (1 km); for CRISTA- this correction is much larger and uncertain by %. However, the further analysis of CO density data will show, that the systematic error of this correction is noticeably lower for CRISTA-, because retrieved CO densities agree usually better than 1% for the two CRISTA missions. Another disturbing effect, which must be taken into account, is the nonlinear relaxation character of the detectors. For CRISTA-1 this effect can be neglected but for CRISTA- it gives a 3% bias on the radiances which vary in time as a function of the detector temperature. Resulting uncertainties in derived CO densities were inferred using curve 1 in Figure 6 and are assigned to the systematic error of the instrument (Table 3, Instrument ) and range between 4 and 1% for CRISTA-1 and are approximately % larger for CRISTA-. [3] The random error of the instrument contains the following sources: During the mission the detector sensitivity varied by 1 % for CRISTA-1 (Si:Ga detector) and noticeably more for CRISTA- (Si:As BIB detector). Due to the strong sensitivity of retrieved CO on such intensity variations, it is not possible to perform reliable CO retrievals below 8 km for CRISTA-. Minor disturbances such as electrical interference are removed in the adjustment of the spectral region for the retrieval. The noise equivalent spectral radiance (NESR) is 1 1 W/(cm sr cm 1 )for CRISTA-1 and W/(cm sr cm 1 ) for CRISTA-. Additionally the SCS channels are contaminated with long period disturbances in the same order of magnitude. All these effects result in an imprecision of retrieved CO of 1 3% above 1 km for CRISTA- and of 7% or more for CRISTA-1 (Table 4, Instrument ) Difference [%] Figure 11. Change of retrieved CO density on a % enhancement of the quenching constant for the intermolecular V-V exchange between N and CO (n 3 ) (solid line) and the constant for intramolecular V-V exchange splitting a n 3 quantum into 1 4 n quanta (dashed line). 6.. Model Uncertainties [36] The uncertainties in the modeling of CO n 3 vibrational modes are discussed in detail by Kutepov et al. (in preparation) and consist of the following: 1. The radiative transfer, because the 4.3 mm FBofCO is the most optically thick rovibrational band in the middle atmosphere.. Quenching constants and the model of collisional processes (for transitions which were not measured in the laboratory) and other excitation mechanisms, e.g., the solar spectrum or the efficiency of transferring electronic energy from O( 1 D) via N to CO. [37] The radiative transfer is solved line by line for all transitions in the 1996 HITRAN database without significant approximations. The inaccuracy of spectroscopic parameters is negligible for CO in comparison with (). The quenching constants and the model of collisional processes constitute the most relevant uncertainties in our non-lte model. [38] The transfer of electronic energy from highly excited O( 1 D) to CO via N (1) poses three important questions: (1) With what efficiency does O( 1 D) excite the various vibrational states of N? () What is the accuracy of the quenching constant for the intermolecular V-V exchange between N and CO (n 3 )? (3) Is the V-V exchange during N -N collisions as fast as the N vibrational energy evolved appears only in N (1)? The uncertainty induced by (1) is considered in the systematic error of the O( 1 D) profile and will be discussed later. The rate constant () was measured by Inoue and Tsuchiya [197] for atmospheric temperatures and their values are in agreement with previous measurements [Taylor and Bittermann, 1969; Moore, 1973]. Sharma and Brau [1969] have published a first principles calculation that reproduces the experimentally measured rate to within a few percent. However, the analysis of the SAMS 4.3 mm measurements seems to support a rate constant twice that measured by Inoue and Tsuchiya [López-Puertas and Taylor, 1989]. Therefore we assume and inaccuracy of %. [39] This induces a systematic error in retrieved CO (Figure 11) of 3% at 6 km, it vanishes at the mesopause, increases to % at 11 km and decreases to a constant value of 1% above 1 km. [4] Another significant source of error in CO n 3 modeling is the splitting a n 3 quantum into 1 4 n quanta. The quenching constants are based on the measurements of Inoue and Tsuchiya [197] and Siddles et al. [1994]. The

11 KAUFMANN ET AL.: CRISTA CO DENSITIES CRI 1-11 branching ratio depends on the temperature and was inferred by Shved et al. [1998] from the experiments of Taine and Lepoutre [198]. For the rate constants we assume an inaccuracy of %, similar to the work of Clarmann et al. [1998]. Its effect on CO can be neglected above 8 km and increases to % at 6 km. No information exists on the quenching of CO (n 3 ) modes by collisions with atomic oxygen. We utilized the hypothesis of Shved et al. [1998] to estimate the quenching constants, which are compatible with the values used by López-Puertas et al. [1998a] and Nebel et al. [1994]. The uncertainty of this constant is considered in the inaccuracy of the atomic oxygen density (v. section 6.3). [41] Kutepov et al. (in preparation) calculate that uncertainties in other collisional constants affect CRISTA 4.3 mm radiances less than %. [4] The solar flux differs typically 3% assuming blackbody radiation in comparison with the solar irradiance computed by Kurucz [199]. The solar irradiance measurements reported by Thekaekara [1973, 1976] are 1% larger, but they are questioned for several reasons [Fröhlich et al., 1983] and therefore we assume, that the 4.3 mm radiance is affected by the choice of the solar spectrum typically by less than 3%. To summarize, uncertainties in the non-lte model parameters contribute an inaccuracy in the retrieved CO densities of typically 1 1% above 8 km. Due to the normalization of CRISTA radiances at 6 km altitude using the same non-lte model as for the retrieval, the systematic error in the non-lte model was reduced to less than % at 6 km O( 1 D) and O( 3 P) [43] It is difficult to estimate the error of the O( 1 D) profiles. Model calculations vary by a factor of 1 in the mesopause region (cf. Figure 7) and more than a factor of two in the mesosphere, which mainly depends on the rate coefficients and on the parameters related to the photodissociation coefficients. The dependence of O( 1 D) on the SZA and its effect on retrieved CO will be discussed in section 7.1. Above 11 km we assume an expanded uncertainty (level of confidence 67%) by a factor of Around 1 km we estimate. 1. due to the strong SZA dependence of the photolysis rate constant and in the mesosphere we expect ±33%. Derived CO densities are affected by O( 1 D) uncertainties by 1% in the mesosphere, 3 % around 11 km, and 1% above. [44] The random error of O( 1 D) is estimated to be 3%, which results in an imprecision of % in retrieved CO depending on altitude. [4] For O( 3 P) we assume a precision and accuracy factor of ; CO densities are affected less than % at most altitudes Kinetic Temperature [46] The kinetic temperature is derived operationally from CRISTA CO 1 mm measurements up to about 8 km. The standard retrieval, however, assumes LTE at all altitudes. First results obtained from a non-lte model retrieval of temperatures are given by Grossmann et al. []. For the CO density retrieval the operational (LTE) CRISTA temperatures are used up to 7 9 km. Between 7 and 9 km Table. Uncertainties of Kinetic Temperatures and Induced CO Density Errors z, km Bias Tides a Random T, K CO,% T, K CO,% T, K CO,% 13 1 b b c b a Tidal perturbations vary by a factor of two or more depending on latitude and local time. b For CRISTA temperatures. For GRAM temperatures, a factor of two higher. c For polar summer profiles appreciably higher. CRISTA temperatures are merged smoothly into GRAM temperatures [Justus and Johnson, 199] and above 9 km only GRAM temperatures are used. Below 7 km the CRISTA temperature error is less than K [Riese et al., 1999]. The temperature retrieval was not performed for measurement mode M/T during CRISTA-1 and only for 1% of mesospheric data during CRISTA-. For the remaining profiles, we use GRAM temperatures in the whole altitude region instead. [47] The deviation between CRISTA and GRAM temperature data is taken as an estimate for the inaccuracy of GRAM temperatures. [48] The major difference between GRAM and CRISTA temperatures is due to atmospheric tides, which are not explicitly implemented in GRAM. At altitudes where CRISTA temperatures are not available, the mean deviation between MSIS and GRAM temperatures is taken as an estimate for the systematic temperature error. Tidal effects are attributed to the systematic error but they are listed separately, because the strength of tidal perturbations depends strongly on the latitude and the local time of the observation. Mean amplitudes for the diurnal and semidiurnal tide for the mesosphere are taken from the Global Scale Wave Model (GSWM) [Hagan et al., 1999], which was run for the CRISTA time periods [Oberheide et al., ] and for the thermosphere from the work of Forbes [199]. Table lists temperature uncertainties for selected altitudes. Uncertainties of high latitude summer mesopause temperature profiles are discussed in detail by Grossmann et al. []. The neglect of tides in the temperature data affects retrieved CO densities appreciably only around 1 km (1%). [49] The random temperature error is estimated from the variance between CRISTA or MSIS and GRAM temperatures at high latitudes, where tidal effects can be neglected and varies between 7 K at 9 km to K at 11 km. The random error of retrieved CO due to the imprecision of temperature data is typically less than 3%, at specific altitudes 8%. 6.. Total Error [] The accuracy of retrieved CO densities is dominated by two error sources: the uncertainty in the O( 1 D) profile and by the uncertainties of collisional rate constants used in the non-lte model. Relatively small systematic errors appear in the mesosphere and also above 1 km with typical uncertainties of 1%. The largest systematic errors

12 CRI 1-1 KAUFMANN ET AL.: CRISTA CO DENSITIES a) b) Tangent Height [km] CRISTA- CRISTA-1 CRISTA CRISTA- 181 CRISTA e-8 1e-7 1e-6 1e+7 1e+9 1e+11 Radiance [W / cm sr] CO Density [cm-3] Figure 1. (a) CRISTA-1 and CRISTA- wave number integrated radiance measurements (3 39 cm 1 ). CRISTA-1 scan 164 was taken at 1 S, 66 E on 9 November 1994, CRISTA- scan 181 at 63 N, 164 W on 1 August 1997, and CRISTA- scan 1399 at 3 N, 9 E on 1 August The random error of the instrument is indicated by the vertical lines. (b) Retrieved CO densities for these scans. appear between 1 and 11 km reaching up an inaccuracy of 4 %. [1] The random errors of the retrieved CO densities are dominated by detector noise at all altitudes for CRISTA-1 and for CRISTA- above 1 km. CRISTA-1 CO densities fluctuate more than a factor above 11 km and typically 1% in the mesosphere. CRISTA- CO random errors are 1 3% between 1 and 1 km. Below 11 km the imprecision of O( 1 D) affects CRISTA- CO significantly. 7. Results [] CO densities are retrieved from single profile intensities between 6 and 1 km for CRISTA-1 and between 8 and 13 km for CRISTA-. [3] Figure 1a shows three wave number integrated radiance profiles from the two CRISTA missions. CRISTA-1 scan 164 and CRISTA- scan 1399 are mid latitude profiles whereas CRISTA- scan 181 was measured at high northern latitudes in the polar summer. The CO densities in Figure 1b are derived from these radiance profiles which are fitted to 1% or better at all altitudes by model calculations using these CO profiles. [4] The immediate CRISTA CO data product is density whereas mixing ratios require the adoption of a background atmosphere. Because CRISTA temperature data have not yet been derived operationally in the UMLT, we converted CRISTA CO densities to VMRs using temperature and pressure from the CRISTA 1 mm channel below 8 km and from the GRAM model above that. Average CO VMR profiles for the two CRISTA missions are displayed in Figures 13 and 14 in comparison with other measurements and calculations performed by the TIME-GCM. [] The TIME-GCM is a self-consistent 3D timedependent middle-atmosphere GCM extending from 3 to km. It generates the circulation, temperature, electrodynamics, and compositional structure internally. TIME- GCM has a by horizontal resolution, grid points per scale height in the vertical and a 4-minute time step. A detailed description of the TIME-GCM is given by Roble and Ridley [1994] and Roble [199, 1996, ] and references cited therein. The TIME-GCM CRISTA- simulation is described in detail by Hagan et al. []. Briefly, the TIME-GCM inputs for this study were consistent with the conditions that prevailed during 7 16 August That is, the F1.7 cm solar radio flux values and the Ap indices and the TIME-GCM parameterizations related to these proxies [Roble, 199; Roble and Ridley, 1987] were used to specify the time-dependent solar radiative, auroral, and cross-cap potential inputs. We used the NCAR globalscale wave model (GSWM) to specify migrating diurnal and semidiurnal tidal fields [e.g., Hagan et al., 1999] and NCAR National Center for Environmental Predictions (NCEP) reanalysis data [ accounted for the prevailing planetary wave activity at the 1 mb lower boundary. We refer the reader to Hagan et al. [] for additional details. The TIME-GCM data was sampled along the CRISTA orbital track for 14 August 1997 to get the same latitude-local time coverage and to sample wave characteristics (e.g., atmospheric tides) in the same manner as CRISTA does. [6] To estimate seasonal variations, mean TIME-GCM CO profiles for the CRISTA- period are plotted separately