7.1. Laboratoire de Météorologie Dynamique General Circulation Model
 To help interpret the SPICAM observations, we used the Laboratoire de Météorologie Dynamique (LMD) general circulation model (GCM) [Forget et al., 1999] which has been extended to the thermosphere [Angelats i Coll et al., 2005]. The simulations used in this paper are based on a version of the model with 50 layers extending from about 5 m above the surface to 7 10−10 Pa (above 250 km) and an horizontal resolution of 3.75° in latitude by 5.675° in longitude. In the lower and middle atmosphere it includes the radiative heating due to the absorption of solar radiation by dust and CO2 (in the near infrared) as well as the cooling induced in the 15 μm thermal IR band of CO2. Nonlocal thermodynamic equilibrium (LTE) effects are taken into account at high altitude non-LTE effects are known to be important at these altitudes for the radiative balance [López-Puertas and López-Valverde, 1995], and are included in the GCM in a parameterized form [López-Valverde et al., 1998; López-Valverde and López-Puertas, 2001].
 Molecular viscosity and thermal conduction, which plays a significant role in the thermosphere, are parameterized. Chemical species have been added to the simulations as described by González-Galindo et al. . There are a total of 11 components in the carbon, oxygen and hydrogen families that are transported, diffused and modified through chemical reactions. Specifically, the chemical species included are CO2, CO, O2, O(3P), O(1D), H2, H, OH, H2O, HO2, H2O2. Molecular nitrogen and Argon are also included but are treated as inert and have no part in the chemical reaction process. In the upper atmosphere, the model accounts for 22 chemical reactions between the 11 chemically active species. Molecular diffusion is taken into account with a state of the art approach in which all species diffuse simultaneously using the exact theory of multicomponent molecular diffusion [Angelats i Coll et al., 2005]. The heating of the thermosphere by UV and EUV radiation is calculated using the model described by González-Galindo et al. . It includes the absorption by CO2, O2 and atomic oxygen, as well as by all the other minor species of the model in the spectral range between 0.1 and 337.7 nm.
 Below 50–70 km, the modeled thermal structure of the atmosphere is primarily forced by the amount of airborne dust, which is known to be highly variable seasonally and from year to year. Above 100–120 km, it also depends on the UV and extreme UV solar input which typically varies with the seasonal change of sun distance and with the 11 years solar cycle.
 In this paper, our “reference” simulation is based on a climatology of the dust as observed by Mars Global Surveyor TES between 1999 and June 2001 [M. D. Smith et al., 2001, Smith, 2004], a Martian year thought to be typical. It is topped by a thermosphere computed assuming solar averaged conditions regarding the EUV flux, which is a realistic proxy for the 2004–2005 period observed by SPICAM. In addition to this baseline scenario, we have used (1) a cold “low dust” scenario corresponding to an extremely clear atmosphere (visible dust opacity τ = 0.1 at 700 Pa), topped with a “solar minimum” thermosphere, and (2) a warm “dusty” extreme scenario corresponding to an atmosphere dustier than observed outside global dust storms (τ= 0.7 + 0.3cos(LS + 80°) at 700 Pa), topped with a “solar maximum” thermosphere. Statistical comparisons with three Martian years of TES data, radio occultation data and PFS data indicate that outside global dust storms period, the observed low atmosphere temperatures are always within a few kelvins from the GCM temperatures simulated with the reference scenario [Montabone et al., 2006; Millour et al., 2007], and systematically above and below the GCM cold and warm simulations.
7.2. Analysis of the Density Variations With the GCM
 Figure 11 shows a comparison of SPICAM observations plotted as a function of season (as in Figure 4) with GCM prediction at the same location and time. The GCM densities are interpolated from 30° Ls average values at the same local time. Therefore the effect of thermal tides is included whereas the transient day-to-day variability is completely smoothed out. To first order, the GCM densities roughly follow the observed seasonal evolution. The observations are more scattered than the modeled values, probably because of transient waves. The model underestimates densities between Ls = 140° and Ls = 180°, certainly because of the unusual dust loading observed in 2004 and not included in the 2000–2001 dust scenario used to force the GCM simulations. However, outside this period, the predicted densities are in most case higher than observed. The discrepancy is significant at 70 km, and increase at higher altitude. Could that be due to a lower atmosphere less dusty and less warm than expected? To assess this hypothesis, and better estimate the impact of dust variations on the high atmosphere densities, we first compared the observations with results from the “dusty” and “low dust” GCM simulations described above. Figure 12 presents a comparison of the SPICAM density data at 70 and 100 km with the corresponding GCM predictions obtained with the “reference”, “low dust” and “dusty” scenarios described above. It shows that variations in dust loading could explain a part of the discrepancy, but not all. In particular SPICAM density observations around Ls = 60–120° remains significantly lower than values obtained with the “low dust” extreme scenario (τ = 0.1 at 700 Pa). Was the Martian atmosphere clearer than that in 2004? Fortunately, this can be investigated using the TES observations which are available until 31 August 2004 (Ls = 81°). Figure 13 shows the TES mean temperature profile in the 0°S–40°S latitudinal belt observed by SPICAM, during Ls = 30–80° (northern fall). It is compared to similar temperature profiles obtained in the previous 2 years by TES and to the GCM prediction. The comparison shows that the GCM is in good agreement with the TES observations from 2000, as expected, and reveals that at the time of the SPICAM observations, the lower atmosphere was actually dustier and warmer than in the previous years! Within that context, the low densities observed by SPICAM as low as 70 km are difficult to reconcile with the GCM prediction and the TES observations. One possibility is that the middle atmosphere between 10 Pa and 0.1 Pa (i.e., between 40 and 70 km) was colder than predicted by the GCM. Using the hydrostatic equation, one can show that the 30% density difference observed at 70 km on Figure 4 would be explained by a 10 K temperature difference between 40 and 70 km (assuming equal density at 40 km, which may not be the case according to Figure 13, however).
Figure 11. Comparison of SPICAM observations plotted as a function of season as in Figure 4 with GCM predictions at the same location and time (GCM data are from our “reference” 1999–2001 TES dust scenario). Except during the unexpected dusty period between Ls = 140° and Ls = 180°, the GCM tends to overestimate the density observed by SPICAM.
Download figure to PowerPoint
Figure 12. Same as Figure 11 for two altitudes and with GCM predictions obtained with three different dust scenarios in order to illustrate the variability of density at high altitude resulting from fluctuations in the low atmosphere dust loading.
Download figure to PowerPoint
Figure 13. Comparison of low atmosphere temperature profiles obtained by TES from Mars Global Surveyor [Smith, 2004] at 2am (Martian solar local time) in the past 3 Martian years with GCM predictions obtained using the “reference” dust scenario based on observation of dust loading during Martian year 25 (MY25). For each profile, data were averaged over all longitudes, latitude 0°S–40°S and over Ls = 30–80°. This corresponds to SPICAM observations where density significantly lower than predicted by the GCM were observed in the atmosphere above. Surprisingly, however, the lower atmosphere observed by TES was then dustier and warmer than assumed by the GCM, not colder. Thus the thermal structure of the atmosphere below 40 km cannot explain the discrepancy between the GCM and SPICAM observations. Martian years (MY) are numbered according to the calendar proposed by R. T. Clancy [Clancy et al., 2000] which starts on 11 April 1955 (Ls = 0°). The curve shown above thus correspond to the periods 3 August 2000 to 24 November 2000 (MY25), 26 June 2002 to 13 October 2002 (MY26), and 8 May 2004 to 29 August 2004 (MY27).
Download figure to PowerPoint
 Latitudinal variations of density can also be detected by SPICAM and compared to GCM predictions. The observed density at 100 km for the periods Ls = 30–120°, Ls = 130–180° and Ls = 240–300° are plotted as a function of latitude in Figure 14. The density measured around southern winter solstice (Ls = 30–120°) decreases by about 1 order of magnitude between the tropics and the polar night, certainly because of the decrease of temperature in the atmosphere below. This is correctly predicted by the GCM. At other seasons, no clear structure can be identified (unfortunately no SPICAM data are available in the northern winter polar night). Nevertheless, the GCM does a reasonable job to predict the spread and the possible variations seen in the observations. One exception are the northern winter midlatitudes around Ls = 240–300°: Above 30°N, the GCM often overestimates the density.
Figure 14. Comparison of SPICAM density at 100 km plotted as a function of latitude with GCM predictions at the same location and time. Three seasons are shown. GCM data are from our “reference” 1999–2001 TES dust scenario.
Download figure to PowerPoint
7.3. Unexpected Low Temperature at the Homopause
 The overestimation of the observed densities by the GCM strongly increases with altitude. This suggests that the modeled scale heights of the atmosphere are generally larger than in reality, and that the GCM temperatures above 70 km are overestimated. This is confirmed in Figures 15 and 16 which present comparisons between SPICAM retrieved temperatures and GCM predictions. Figure 15 shows a comparison for all data between 50°S and 50°N plotted as a function of season at four pressure levels. At 0.1 Pa (around 70–80 km), the GCM temperatures agree well with the observations (given the absolute error bars) and somewhat predict the subtle seasonal evolution described before. In particular, the GCM does simulate the observed increase of temperature between Ls = 100° and Ls = 150° in spite of the fact that as discussed above, the GCM could not account for the unusual increase of airborne dust observed at that time in 2004. We can thus conclude that to first order, the Ls = 130° dust storm did not directly affect the temperature above the 0.1 Pa level. Between Ls = 180° and Ls = 360° (northern fall and winter, the dusty season), a few observations exhibit temperatures 20 to 30 K larger than predicted, possibly because high altitude aerosols affected the temperature retrieval.
Figure 15. Comparison of the SPICAM temperature observations shown in Figure 8 with GCM predictions at the same location and time. Data obtained between 50°S and 50°N are plotted as a function of season (solar longitude Ls). GCM predictions are from our “reference” 1999–2001 TES dust scenario simulation. Depending on the season, 0.1 Pa corresponds to altitudes around 70–80 km, 0.01 Pa to 85–100 km, 0.001 Pa to 100–115 km, and 0.0001 Pa to 115–130 km.
Download figure to PowerPoint
Figure 16. Average SPICAM temperature profiles as a function of pressure for selected season and latitudinal range, compared to GCM temperature profiles obtained at exactly the same location and time and similarly averaged. In each cases, three mean SPICAM profiles obtained assuming top temperatures of 100, 175, and 250 K. are shown. The three GCM profiles correspond to various dust and solar EUV fluxes.
Download figure to PowerPoint
 At higher altitude, the observed temperatures are lower than predicted, often by 40 to 60 K! This is better illustrated in Figure 16 which shows a selection of 12 averaged profiles representing all the latitudes and seasons observed by SPICAM, compared to GCM predictions obtained with our various dust and solar EUV scenarios. In many cases, the GCM strongly overestimates the observed temperatures around 0.001 Pa (85–100 km). The shape of the temperature profiles suggest that the differences result from the fact that (1) mesopause temperatures are overestimated by the model and (2) mesopause altitudes tend to be underestimated by the model.
 What could be the origin of such modeling errors? Radiative processes are thought to control the global mean temperature and altitude of the mesopause [Bougher et al., 1994]. Other processes (molecular conduction, chemical and dynamical processes) can influence the temperature structure, allowing individual temperature profiles to be far from radiative equilibrium [States and Gardner, 2000]. However, the systematic nature of the differences observed in the temperature of the mesopause seems to point to the IR radiative balance as the most likely responsible. This hypothesis is reinforced by the intercomparisons between the LMD GCM and the Mars Thermospheric General Circulation Model (MTGCM) (see, for example, Bougher et al.  for a description of this thermospheric GCM). These two different models use the same parameterization for the IR emissions by CO2, based on the work by López-Valverde and López-Puertas , and both predict similar temperatures in the mesopause, well above those observed by SPICAM. One of the most uncertain approximations included in this parameterization is the use of a constant atomic oxygen profile (which has a cooling effect on CO2, see below), instead of the “actual” atomic oxygen distribution predicted by the GCM. The reason of this approximation is twofold. First, little is known about the precise distribution of atomic oxygen in the upper mesosphere/lower thermosphere. The available data show an important variability [Stewart et al., 1992]. In the absence of a climatology of atomic oxygen in the upper atmosphere, an abundance appropriate for average conditions was taken from Nair et al. . Second, although the GCM now includes a photochemical model, that was not the case at the time this parameterization was developed. No variations in the concentrations of the different species could be predicted and it was difficult to account for a variable atomic oxygen.
 However, the atomic oxygen is known to have an important effect on the 15 μm cooling. The collisions with atomic oxygen excite the vibrational states of the CO2 molecule, enhancing the emission rate and the cooling [Bougher et al., 1994]. This can lead to variations of up to a factor 5 in the cooling when modifying the atomic oxygen profile, as shown by López-Puertas and López-Valverde, . Our most recent results obtained with the GCM photochemical scheme [González-Galindo et al., 2005] suggest that the fixed atomic oxygen profile used in the GCM could be significantly below the actual level. An underestimation of the atomic oxygen content implies an overestimation of the temperatures, which could explain the differences with the observations.
 To check this hypothesis, we have used the 1-D model of the Martian atmosphere developed at the Instituto de Astrofísica de Andalucía [López-Valverde et al., 2006]. This model can simulate the 15 μm cooling with both the parameterization included in the GCM (with the fixed atomic oxygen concentration) and the detailed NLTE scheme [López-Puertas and López-Valverde, 1995] on which the parameterization was based. This full model was able to use an enhanced atomic oxygen profile: mixing ratio of 2 10−2 below the homopause at 120 km and following the scale height of O above this level. This corresponds to a factor of 3 (at 200 km) to 10 (at 120 km) more than in the Nair et al.  profile originally used for the GCM.
 We made two 10-day simulation (a steady state for the temperature is reached after 2–3 days of evolution) with both schemes. The final temperatures at LT = 0 are represented in Figure 17. As can be seen, when using the detailed scheme with the improved atmospheric atomic oxygen concentration, the mesopause is at higher altitude and slightly cooler, and the temperatures in the lower thermosphere are significantly lower, improving the comparisons with SPICAM. Further analysis show that most of the difference is due to the change in atomic oxygen rather than in the model themselves. We conclude that a significant fraction of the differences between model and observations can be attributed to the fixed atomic oxygen in the 15 μm cooling parameterization. We are currently working on a new 15 μm cooling parameterization for the GCM that will include variable atomic oxygen following the photochemical model included in the GCM, as well as other improvements.
Figure 17. The impact of atomic oxygen on the temperature profile (simulated using the López-Valverde et al.  1-D model). Collision between atomic oxygen O and CO2 molecules can strongly enhance the CO2 infrared cooling and change the temperature profiles. Using a realistically enhanced profile of atomic oxygen concentration [O] (black line) rather than the [O] profile derived from Nair et al.  assumed in the GCM (green line) allows reduction of the discrepancy between the model and the SPICAM observations shown in Figure 16 (see text for details).
Download figure to PowerPoint