A unique feature of the Martian climate is the possibility for carbon dioxide, the main atmospheric constituent, to condense as ice. CO2 ice is usually detected as frost but is also known to exist as clouds. This paper presents the first unambiguous observation of CO2 ice clouds on Mars. These images were obtained by the visible and near-infrared imaging spectrometer OMEGA on board Mars Express. The data set encompasses 19 different occurrences. Compositional identification is based on the detection of a diagnostic spectral feature around 4.26 μm which is produced by resonant scattering of solar photons by mesospheric CO2 ice particles in a spectral interval otherwise dominated by saturated gaseous absorption. Observed clouds exhibit a strong seasonal and geographic dependence, concentrating in the near-equatorial regions during two periods before and after northern summer solstice (Ls 45° and 135°). Radiative transfer modeling indicates that the 4.26 μm feature is very sensitive to cloud altitude, opacity, and particle size, thereby explaining the variety of spectra associated with the cloud images. On two orbits, the simultaneous detection of clouds with their shadow provides straightforward and robust estimates of cloud properties. These images confirm the conclusions established from modeling: clouds are thick, with normal opacities greater than 0.2 in the near infrared, and are lofted in the mesosphere above 80 km. The mean radius of CO2 ice crystals is found to exceed 1 μm, an unexpected value considering this altitude range. This finding implies the existence of high-altitude atmospheric updrafts which are strong enough to counteract the rapid gravitational fall of particles. This statement is consistent with the cumuliform morphology of the clouds which may be linked to a moist convective origin generated by the latent heat released during CO2 condensation.
 The existence of CO2 ice on Mars is a salient climatic feature for a planet of the solar system since it is the only evidence collected to date for condensation of a bulk atmosphere. CO2 ice is present as clouds or as surface frost and its formation is initiated whenever temperature drops to sufficiently low values (∼145 K at 6 mbar) for local pressure to exceed saturation. Such cold conditions are only met in restricted areas: the winter polar night and the mesosphere (that we can define as the atmospheric layer located between 45 and 100 km of altitude). Due to their singular nature with respect to the terrestrial case, such CO2 ice clouds forming out of the main atmospheric component are very intriguing but their detection is a challenging task. Most of CO2 ice spectral features in the infrared overlap with those of surface mineralogical components or with those of CO2 gas itself [Bell et al., 1996]. As surface frost, CO2 ice can be readily identified and tracked from remote sensing since it is usually present with sufficient thickness and spatial extent to buffer surface thermal emission [Kieffer et al., 1976; Titus et al., 2001] and to produce deep absorption features in the near-infrared [Bibring et al., 2005; Langevin et al., 2006]. When aloft, however, CO2 ice is not as concentrated as in surface frost, particle residence time being limited by sedimentation and atmospheric variability, and is thus harder to detect.
 To date, identification of such clouds has remained ambiguous. The spikes of light reported from Mariner observations of the bright limb and presented as evidences of reflection by CO2 ice are suspicious because their altitude of emission was too low (25 km) for potential CO2 condensation [Herr and Pimentel, 1970]. The discovery, decades later, of ubiquitous CO2 fluorescence at comparable wavelength eventually discarded the validity of Mariner detections [Lellouch et al., 2000; Drossart et al., 2006]. The presence of low-lying (<20 km) clouds in the polar night has been indirectly inferred from anomalously cold surface temperature retrievals [Kieffer et al., 1976] and from the detection of optically thick layers interfering with the Mars Orbiter Lidar Altimeter echoes [Petengill and Ford, 2000]. Mars Pathfinder images showing bright, bluish morning clouds [Smith et al., 1997] have been interpreted by Schofield et al.  and Clancy and Sandor  as resulting from supercold atmospheric conditions at high altitude.
 The only spectroscopic identification of a CO2 ice cloud has been obtained very recently. The Planetary Fourier Spectrometer instrument on board Mars Express [Formisano et al., 2006] has detected solar light reflected around 85 km showing CO2 ice absorption signatures embedded in those of gas. At this altitude, layers of a likely condensate origin have been reported several times in a recent past. Stellar occultation at UV wavelength by the Mars Express SPICAM instrument revealed clouds lofted up to 110 km that were interpreted to be composed of CO2 ice since they were found to lie within supersaturated CO2 environment [Montmessin et al., 2006]. In the visible, longitudinally confined mesospheric bright structures distinctly detached from the lower dust haze have also been observed using combined MGS Mars Orbiter Camera images and Thermal Emission Spectrometer scans of the limb [Clancy et al., 2007]. In spite of this growing observational data set, our knowledge of CO2 ice clouds remains fragmentary.
 The need for CO2 ice cloud models is motivated by the importance of the CO2 cycle in Mars climate. The formation of CO2 ground frost produces >30% seasonal variations of the mean surface pressure. However, the actual partitioning between direct deposition of CO2 ice on the ground and precipitation subsequent to cloud formation is unknown, although it partly controls CO2 frost albedo and thus frost retreat. For this reason, a link has been made between the existence of a residual CO2 ice cover at the South Pole and stationary weather patterns favoring precipitation in the same area [Colaprete et al., 2005]. The formation of CO2 ice clouds is a major step in our understanding of Mars climate but it poses unusual problems for microphysicists: the growth of cloud particles from the condensation of a bulk atmosphere produces microscopic pressure gradients which deviate from the diffusion-driven growth model that prevails on Earth, Venus, Titan and Mars (in case of water ice clouds). Cloud formation is further complicated by the thermal variations subsequent to latent heat released and absorbed during phase changes, the relative amount of atmosphere affected by the process being necessarily substantial. As the temperature of an air parcel relaxes to CO2 saturation point, it may become more buoyant than surroundings, thereby triggering convective instabilities in a way similar to developing moist convection on Earth [Colaprete and Toon, 2003]. This implies that the formation of a CO2 ice cloud cannot be studied separately from its dynamical environment since feedback loops between dynamics and CO2 condensation may eventually dictate the very nature of a CO2 cloud formation. However, the complexity of the phenomenon and the current lack of observations make it difficult to establish a definitive model for this particular type of clouds although recent groundbreaking efforts have given a basis for its development [Colaprete and Toon, 2002, 2003].
 The data presented in this paper have been obtained with the Observatoire pour la Minéralogie, l'Eau, les Glaces et l'Activité (OMEGA) instrument on board the Mars Express mission. We describe the occurrences of mesospheric CO2 ice clouds in the equatorial region of Mars. Clouds have been detected by their diagnostic scattering emission feature in the infrared. From OMEGA hyperspectral images, we have been able to derive the main cloud properties: altitude, opacity and particle size. This work is intended to provide a complete observational set of information to help understand and represent CO2 cloud formation in the atmosphere of Mars.
2. OMEGA Data
 The OMEGA instrument on board Mars Express is a visible and near-infrared imaging spectrometer with coaligned channels working in the 0.38–1.05 μm visible and near-IR range (VNIR channel) and in the 0.93–5.1 μm short wavelength-IR range (SWIR channel). The VNIR channel operates in a pushbroom mode with the telescope focal plane providing instantaneously one cross-track line, the other image dimension being formed by the motion of the satellite. The SWIR channel operates in a whiskbroom mode. In this channel, only one spatial pixel is recorded at a time. The image is constructed the satellite motion and by a mirror located in font of the telescope scanning in the cross-track direction. The beam is split into two parts that are redirected to two optical subsystems covering two contiguous portions of the SWIR spectral interval. The SWIR integration time ranges from 2.5 to 5 ms, depending on the satellite ground track velocity; while the VNIR integration times is 50–200 ms. These integration times allow OMEGA to obtain a SNR > 100 over the entire spectral range (see Bibring et al.  for further details on the instrument).
 OMEGA has successfully performed more than 3 years of operation in orbit and has mapped the surface with a resolution at pixel level ranging from 300 m to 4.8 km depending on spacecraft altitude. In the spectral range of the instrument, signal is shaped by surface mineralogical properties and by scattering on airborne dust and cloud particles. The spectral resolution (ranging from 7 to 20 nm) permits identification of several surface and atmospheric species through their distinct absorption/emission features [Bibring et al., 2005]. A typical OMEGA spectrum shows a steep increase of reflectance from 0.3 to 0.7 μm due to the presence of ferric oxide at the surface, a deep hydration band at 3 μm, absorption by gaseous CO2 at 2, 2.7, and 4.3 μm, and planetary thermal emission longward of 4 μm.
2.1. The 4.3 μm CO2 Gas Band
 The case of the 4.3 μm band is interesting as it varies dramatically with space and time. This band is shaped by the R-P branch system of the ν3 fundamental asymmetric stretching mode of CO2. At standard Martian pressure, this strong transition induces saturated absorption within the 4.2 to 4.5 μm spectral domain. Exceptions to this typical behavior nonetheless exist. For instance, when OMEGA viewing geometry is oriented at the limb, the band becomes dominated by non-LTE fluorescent emission of CO2 molecules, with most of the emission originating from layers located above 80 km [Drossart et al., 2006]. CO2 fluorescence exhibits a prominent double-peak structure with a central dip at 4.3 μm [López-Valverde et al., 2005] which, at OMEGA spectral resolution, merges into a single spike of emission. Another type of spectral distortion occurs during nadir observations over high-topography regions where the 4.3 μm band partially desaturates due to reduction of atmospheric mass. Figure 1 shows a multiwavelength sequence recorded by OMEGA during Orbit 3913, when the distance of the spacecraft from the planet allowed it to obtain a global nadir-to-limb view of Mars. This image shows the only spatial features that can be identified in the 4.3 μm range: elevated landmarks (in this case, the top of a volcano) and fluorescent atmospheric layers at the limb.
2.2. The 4.26 μm Anomaly
 The third case of anomalous behavior is only found at particular locations and seasons and does not show any correlation with topography. Only 19 occurrences of this spectral anomaly have been recorded to date among 1,000 observation sequences. The best illustration is given by the multiwavelength image displayed in Figure 2 corresponding to an OMEGA nadir pass over Meridiani Planum during mid northern spring. In this image, an area of unexpected emission at 4.24–4.26 μm is detected right at the equator on the prime meridian. Outside this area, photons are absorbed by gas and the signal is dominated by noise. Figure 2 shows that the spatial pattern observed at 4.26 μm is also apparent at 0.5 μm, less pronounced at 2.7 μm but is markedly absent in the image collected at 1.3 μm. Analysis of the full spectrum indicates a substantial increase of brightness only in specific wavelength sections: shortward of 0.7 μm and in the middle of the 2.7 and 4.3 μm-CO2 band system (Figure 3). In the case of the 2.7 μm, however, the spatial pattern of emission is not easily separated from surface albedo variations and a cross-correlation analysis was needed to confirm the existence of the same pattern at 4.26 and 2.7 μm (see Figure 4).
2.3. CO2 Ice Spectral Features
 The 2.7 and 4.26 μm spectral features are the result of airborne CO2 ice cloud particles resonantly scattering photons back to space. As shown by Figure 5, these wavelengths are characteristic of the most prominent CO2 ice bands in the infrared [Warren, 1986]. The 2.7 μm band is two orders of magnitude weaker than the one at 4.26 μm and is too narrow to be properly resolved at OMEGA spectral sampling. Clearly, the 4.26 μm band is far more discriminating as it exhibits a well-resolved reflectance peak (see Figure 3). As previously stated [Fink and Sill, 1982], the fundamental ν3 band of CO2 ice at 4.26 μm is probably the strongest infrared band inventoried to date for a molecule. Associated with the dramatic increase of CO2 ice imaginary index, the real part of the refractive index also exhibits large fluctuations around 4.26 μm, yielding the Fresnel reflectance peaks observed by OMEGA. Similar reflectance spike around 4.24 μm was also reported from a spectroscopic study of CO2 ice in laboratory using a spare model of the IRS instrument that flew on Mariner 6 and 7 missions [Herr and Pimentel, 1970].
 CO2 ice possesses other diagnostic absorption bands in the near-infrared [Bell et al., 1996] as shown in Figure 5, in particular several features around 2 and 3 μm which are routinely used for the detection of hoar frost [Langevin et al., 2006]. One can actually notice in Figure 4 the presence of a small peak of correlation around 2 μm which may point to another spectral region affected by the cloud. However, OMEGA does not permit the use of the 2, 2.7 and 3 μm bands for the discrimination of these CO2 ice clouds since spectral sampling does not allow us to resolve these features when they are so weak. The quasi-absence of detection at these wavelengths is indicative of a relatively short path length of photons inside these clouds compared to what is usually observed in case of frost.
3. CO2 Ice Cloud Detection and Mapping
 The hyperspectral image shown in Figure 2 is the first picture of a cloud in the Martian atmosphere spectrally identified as CO2 ice. Photons at 4.26 μm originate dominantly from reflection of the solar beam with a marginal contribution from surface thermal emission and possibly from non-LTE fluorescent emission due to solar pumping of CO2 vibrational upper levels with re-emission in the 4.3 μm band. However, major affect of fluorescence can be ruled out by the facts that (1) the 4.26 μm cloud reflectance spike is shifted 40 nm (i.e., twice OMEGA spectral sampling) shortward of fluorescence peak and (2) the background fluorescence emission remains at or below noise level when OMEGA is pointing nadir, i.e., an order of magnitude less than the intensity of the signal observed at 4.26 μm. Although fluorescence could be enhanced via cloud backscattering like O2 dayglow emission on Venus [Connes et al., 1979], this enhancement cannot exceed a factor of 2 as it corresponds to concentrating the spherically isotropic fluorescent emission in the half-space above the clouds.
3.1. Spectroscopic Modeling
 Interestingly, reflection at 4.26 μm is occasionally (about 40% of all cases) complemented by a secondary spike of light at 4.32–4.34 μm (Figure 6). This disparity of spectral signature has been investigated in further details by means of a radiative transfer model. The goal of this study was to determine the spectral modulations implied by changes in cloud opacity, height and particle size.
3.1.1. Model Description
 To this end, we have used the Spherical Harmonic Discrete Ordinate Model (SHDOM) in the 1D plane-parallel version [Evans, 1998] to conduct several photometric simulations of the 4 to 4.6 μm spectral domain. Computations were made in a line-by-line mode with a spectral sampling of 0.004 cm−1. Surface albedo was set at 0.15 uniformly within the spectral interval. CO2 gas absorption coefficients were computed from the HITRAN 2004 database [Rothman et al., 2005] using a Voigt line profile. Line-by-line results are then convolved by OMEGA instrumental function which can be represented by a Gaussian peak with a FWHM of 23 nm. Atmospheric structure was constrained using the European Martian Climate Database (EMCD) [Lewis et al., 1999] and CO2 ice refractive index values were extracted from a compilation of laboratory experiments [Warren, 1986; Hansen, 1997]. Due to the complexity associated with the representation of non-LTE emission, CO2 fluorescence was ignored as well as thermal emission from the atmosphere and the surface. Despite the effectiveness of these two processes around 4.3 μm, observations indicate that in the absence of CO2 ice clouds, the 4.2 to 4.5 μm interval exhibits marginal departures from the background noise level. This implies that thermal emission and fluorescence remain well below the cloud signal level that we aim at simulating in this study.
 In our model, the atmosphere is divided into 40 layers of equal thickness (5 km). The amount of absorber inside each layer is calculated using a total column pressure of 6 mbar that is vertically partitioned according to the EMCD temperature profile. A cloud is added by setting a uniform value of the extinction coefficient within two adjacent layers so as to match a user-specified cloud optical depth (τc). Doing so, we limit the vertical extent of the cloud to 10 km, a value in agreement with the cloud profiles retrieved from SPICAM stellar occultations [Montmessin et al., 2006]. The asymmetry parameter 〈g〉 and the single-scattering ϖ of the layer are computed with Mie theory assuming a lognormal distribution of cloud particles with an effective variance νeff of 0.15. ϖ is also weighted by the relative amount of the absorbing gaseous component (i.e., CO2) of the layer. The cloud phase function is of Henyey-Greenstein type, which we acknowledge is a poor representation of the cloud particle scattering function especially at backscattering angles, the common geometry for nadir sensing. Scattering phase functions are extremely dependent on the particle shape. Ice crystals adopt structural habits which vary with composition and temperature. In the case of CO2 ice, the most likely crystal shape is cubic or octahedral [Wergin et al., 1997]. Determining the exact angular distribution of scattering with such particle shape is beyond the level of accuracy required here. More importance is given to correctly reproduce the spectral shape of the cloud feature, a characteristic that is more directly tied to the intrinsic properties of CO2 ice. The use of an approximate phase function remains appropriate considering the fact that available models for realistic shapes (e.g., spherical or spheroidal geometries) do not warrant a higher degree of consistency, being themselves a distant representation of the expected cubical shape. In addition, the Henyey-Greenstein function can be easily expanded along spherical harmonics and is thus of convenient use.
3.1.2. Model Results
 A typical example of a simulated OMEGA spectrum showing the effect of a CO2 ice cloud is displayed in Figure 7a. In this model configuration, a thick (τc = 1) cloud has been added at 90 km. Cloud optical properties were computed for a particle radius of 0.8 μm. Two curves are shown; one obtained after convolution at the actual OMEGA resolution and one obtained with a resolution 10 times greater. In the latter, the 4.26 μm spike is significantly emphasized since the width of the absorption band is much narrower than OMEGA spectral sampling. At the coarser OMEGA resolution, the simulated intensity is found maximum at 4.28 μm, i.e., 20 nm longward of the observed peak position. However, the refractive indices that we use in this wavelength range were selected by Warren  among a set of experimental data showing significant scatter of the band position and width. Both characteristics are temperature-dependent, which may factor in the small disagreement existing between model and observations. Additionally, the reliability of the experimental data is questionable; some reinterpretation work was performed by Warren  to account for the contribution of reflection in the determination of the laboratory sample transmission. Finally, the position of the modeled peak is very sensitive to the spectral registration accuracy of the optical constants. A 5 nm shift of the cloud optical properties, which corresponds to the wavelength resolution of the optical constants, is sufficient to obtain a peak correctly positioned.
 It is intriguing to note that this band of CO2 ice is not detected as an absorption but as a reflecting feature. In fact, CO2 ice is so absorbing at this wavelength that it adopts a metallic behavior and sustains important skin effect limiting the penetration of photons inside the material, thereby leaving some potential for scattering. This explains why the single scattering albedo values of CO2 ice particles at 4.26 μm are not strictly equal to zero, increasing from 0.04 for a radius of 0.1 μm to a maximum of 0.5 for particles larger than 5 μm (see Figure 8 for the particle optical properties as a function of wavelength). The dark CO2 ice particles appear bright because they contrast with the much darker and purely absorbing gaseous environment. Should the band be located in a bright continuum region of the spectrum, it would then appear as an absorption feature.
3.1.3. Double-Peak Feature
 Several cases were run to investigate the sensitivity of the spectral shape to the cloud properties. The information extracted from these simulations can be summarized as follows:
 1. Clouds must be located at altitudes higher than 40 km to reduce photon path through the 4.3 μm absorbing medium and to produce reflectance spikes comparable to what is observed (see Figure 7b). This lower limit of altitude is marginally affected by the approximate representation of the cloud phase function as it is first constrained by the amount of absorption sustained by photons along their travel through gas.
 2. Only particles with a radius greater than 0.2 μm can efficiently scatter light at 4.26 μm. Smaller sizes exhibit single scattering albedo close to zero and are thus only absorbing (Figure 7c).
 3. Transitioning from a single (4.26 μm) to a double-peak spectral structure (4.26 + 4.32 μm) occurs when the product τc × ϖ is greater than 0.5 or alternatively when the particle radius is larger than 1 μm (Figures 7c and 7d). Separation of these two possibilities is thus difficult to make from consideration of the 4.3 μm band alone.
 The secondary peak at 4.32 μm corresponds to a local minimum of imaginary index yielding a local maximum of ϖ. As the spectral behavior of CO2 ice is nonresonant at this wavelength, particle scattering efficiency is dictated first by the size of the particle relative to the wavelength; i.e., submicron particles are weakly scattering in comparison with micron-sized particles whose size is comparable to wavelength. This explains why the secondary peak of intensity appears when particle radius exceeds 1 μm. The rise of the secondary peak with increasing opacity is due to multiple scattering. For low cloud opacities (τc < 1), scattered intensity scales linearly with τc × ϖ and saturates near τc ∼ 1. The resonant behavior of CO2 ice near 4.26 μm subsequently yields a maximum of cloud opacity around this wavelength and thus a maximum of intensity. For greater opacities, scattered intensity does not increase linearly with τc × ϖ. Photons can be reflected several times, the number of scattering events depending on ϖ. Since ϖ is greater at 4.32 μm than at 4.26 μm, multiple-scattering is far more effective in the former case and this wavelength will exhibit greater intensities.
3.2. Analysis of Cloud Shadows: Estimation of Cloud Altitude, Opacity, and Particle Size
 In the case of Orbit 501, we can state that the double-peak structure is produced by reflection on particles with a radius exceeding 1 μm. In this OMEGA image (Figure 9), the shadow projected by the cloud on the surface can be observed 100 km southeast. Interestingly, while the cloud itself can no longer be spotted at 1.3 μm, its shadow still appears distinctly on the surface at this wavelength. The cloud is viewed in a side scattering geometry, which does not produce enough brightness enhancement when the cloud opacity is low to moderate and the surface is bright. In the shadow area, however, solar flux attenuation is the result of cloud particles scattering light in all directions, thereby maximizing the flux of light intercepted by the cloud and creating larger contrast with surroundings.
 Simple trigonometry indicates that this shadow is cast by a cloud lofted at an altitude of 80 km. These results support the lower limit of 40 km deduced from modeling. Furthermore, contrast of brightness between the shadowed area and its surroundings (as plotted in Figure 10) is related to the total attenuation of light by the cloud and therefore obeys Beer-Lambert's law. Contrast at any wavelength can be expressed by
where Rish and Rosh are the radiances respectively inside and outside the shadow, τc is the cloud opacity, θ is the solar zenithal angle and Fdif/Fdir is the ratio of the diffuse over the direct component (assumed to be solely generated by dust) of the solar flux outside the shadow. Using a value of 0.2 representative of the background dust opacity in equatorial regions at the season of the detection [Bell et al., 2006], the Fdif/Fdir ratio in the visible is estimated around 0.3 by the SHDOM model. At 0.5 μm, the contrast ratio is 0.83, implying τc of 0.23. The Rish/Rosh ratio is minimum at 1.4 μm (with τc = 0.5, assuming the same value for Fdif/Fdir) and gradually increases up to 1 between 1.4 and 4 μm (see Figure 10). According to Mie theory, this spectral variation is indicative of cloud particles with an effective radius of 1.5 μm (and a more loosely constrained effective variance of 0.1). Contrast ratio is found to decrease again in the thermal part of the spectrum. Thermal emission is reduced in the shadow area due to a surface temperature 10 K lower than in the surroundings (221.5 K versus 232 K) as inferred from fitting the 5 μm spectral range with a Planck function. This reduction of thermal emission matches the reduction of energy caused by the cloud masking visible flux and thus implies a soil in thermal equilibrium. As the cloud shadow could also be spotted on the spectral images of Orbit 551, the same procedure has been used, yielding a cloud altitude of 83 km and cloud particles with a similar size.
4. Cloud Distribution and Mapping
4.1. Seasonal and Spatial Distributions
 The 19 CO2 ice clouds detected by OMEGA exhibit strong ties to seasons and geography. Clouds are confined to the 15°S/15°N latitudinal belt (Figure 11), with the notable exception of Orbit 567 which is found around 50°S. This latitudinal distribution is consistent with the mesospheric layers detected by SPICAM [Montmessin et al., 2006] (which also show the presence of clouds around 30°S) and TES [Clancy et al., 2007]. The 19 occurrences are found within a restricted longitudinal corridor, i.e., between 270°E and the prime meridian, with ten of them located above Terra Meridiani. This zone coincides with one of the two sectors reported by Clancy et al.  for the occurrence of Mars Equatorial Mesospheric (MEM) clouds. The absence of detection by OMEGA in the western hemisphere where the second MEM zone is located may be related to a difference of local time, since OMEGA samples essentially around 8 AM while TES samples at 2 PM. Cloud lifetime is difficult to evaluate. However, it was possible to detect a cloud occurrence at every pass above Terra Meridiani during a period of 20 days in northern mid-spring. It is unlikely that the same cloud was observed throughout that period since the residence time of particles at such altitudes is limited by sedimentation (particle fall speed > 10 m s−1). We rather speculate that clouds last a few hours and are diurnally reactivated by the vertical propagation of thermotidal waves. This explanation is consistent with the fact that CO2 ice clouds are observed near the equator where nonmigrating thermal tides produce large diurnal temperature variations in the mesosphere [Wilson, 2002].
 Again consistently with SPICAM and TES, OMEGA clouds are observed only during two short seasons bracketing northern summer solstice, i.e., between Ls 21° and Ls 55° and then between Ls 100° and Ls 134° (referred as to Group 1 and Group 2 clouds in Figure 11). The colder near-aphelion climate is statistically more prone to experience CO2 condensation events at high altitude. Then the absence of clouds right at aphelion is puzzling. Still, the LMD model [Forget et al., 1999] predicts that the Martian mesosphere is annually and globally the coldest at these two seasons and near the equator and that a short period of warming occurs near aphelion (Figure 12). Clancy and Sandor  indicate that even colder (>10 K) minimum temperatures appear at the same periods in a submillimeter 12CO thermal profiling survey of the mesosphere.
4.2. Cloud Morphology
 With the imaging capability of OMEGA, it is possible to study the horizontal morphology of CO2 ice clouds for the first time. In many cases, clouds stretch throughout the cross-track direction of the swath, showing that CO2 cloud cover can be significant, exceeding 500 km as in Orbit 485. In some other cases (Orbit 501 and 551), cloud cover is more restricted and does not exceed 20 km.
 In Figure 13, we display a panel of images showing the cloud signature at 4.26 μm superimposed on corresponding images at 5.1 μm at which wavelength clouds and dust are nearly transparent to surface infrared emission, allowing clear discrimination of topography. These images illustrate the variety of CO2 ice cloud morphologies observed with OMEGA. Clouds can be found to exhibit regular rippled pattern (d) but appear predominantly as rounded isolated masses irregularly arranged (a-b-c-e-f) resembling terrestrial alto- or cirrocumulus. As observed on the images of Orbits 501 and 551, individual cloudlets possess sharp contours that project well-defined shadows. Differences of less than 10% in width and length are noted between the comma-shaped cloud of Orbit 501 and its shadow (Figure 9). This is in the range of the theoretical enlargement caused by the apparent diameter of the sun (20 arcmin). A large vertical extent of the cloud can be therefore excluded as, given the solar incidence of 53°, this would project a shadow beyond the outline of the cloudlet. We can thus estimate that the cloud vertical thickness cannot exceed 10 km in its central portion and 2 km at the edges.
4.3. A Moist Convection Origin?
 We speculate that cloud cumuliform appearance is the result of a “moist” convection process generated by the release of latent heat. Latent heat provides additional buoyancy and amplifies instability, potentially triggering deep convection events that are suspected to exist in the Martian polar night [Colaprete et al., 2003]. This mechanism, usually associated with the condensation of water inside rising air parcels on Earth, is probably reinforced on Mars with CO2 because the latter constitutes the bulk atmospheric component. The same process responsible for the existence of convective CO2 clouds in the polar night is also probably at work in the mesosphere. The elementary structure of the OMEGA clouds is generally of the order of 5 km, implying the existence of convective cells of similar size with a depth limited to a few kilometers as deduced from the cloud projection.
 The unexpected presence of micron-sized particles at altitudes of 80 km, where particles fall faster than 10 m s−1, would thus be explained by the existence of mesospheric updrafts. If unstable conditions with respect to the “wet” adiabatic CO2 profile do exist in the Martian mesosphere, intense convective events are also to be expected during CO2 ice cloud formation. Such conditions are indeed observed, since highly supersaturated environments have been reported in the mesosphere from SPICAM stellar occultation [Montmessin et al., 2006, 2007]. The observation of super-saturation levels >1,000, which has no counterpart elsewhere in the solar system, indicates that enough convection potential is available in the mesosphere to generate moist convection phenomena.
5. Conclusion and Perspectives
 In this paper, we have reported the detection and mapping of CO2 ice clouds in the equatorial region of Mars. Below is a summary of the findings:
 1. CO2 ice clouds have been identified through the diagnostic occurrence of Fresnel reflectance peaks at 4.26 μm where CO2 ice and gas possess their major absorption band.
 2. Spectroscopic modeling indicates that clouds are lofted in the mesosphere (>40 km) and that their particle radius is >0.2 μm.
 3. The simultaneous detection of the shadows allows for robust and straightforward cloud properties for two orbits: (1) Clouds exhibit visible and near-infrared opacity >0.2, (2) particle radius is estimated to range between 1 and 2 μm, and (3) clouds are located at 80 km.
 4. Clouds are confined within the equatorial belt and appear seasonally during two short periods before and after aphelion. These distributions are consistent with SPICAM high-altitude layers and with the MEM occurrences reported by Clancy et al.  from MOC-TES observations.
 5. The clumpy nature of the cloud morphology is consistent with convective regimes probably forced by the release of latent heat.
 Although the various Mars Global Surveyor and Mars Express observations of mesospheric CO2 ice clouds exhibit an overall agreement in terms of temporal and spatial distribution, significant differences can be noted between data sets with regards to cloud properties. SPICAM observations [Montmessin et al., 2006] reveal CO2 ice crystals in the submicron range (0.08 to 0.13 μm), consistent with the bluish nature of the Pathfinder predawn cloud [Smith et al., 1997]. From the lack of radiance in the infrared, Clancy et al.  could determine that MEM clouds cannot be composed of particles larger than 1.5 μm, this upper limit being equal to the one derived from our OMEGA images. The fact that both SPICAM and Pathfinder clouds were observed at night while OMEGA and TES observations refer to daylight conditions suggest significant cloud changes over the course of a day. The smaller particle size derived at night may correspond to the relics of a daytime cloud formation process followed by a progressive decrease of convection and the subsequent fall of the larger particles. Further characterization of cloud diurnal variation will require detailed coupled dynamics-radiative-microphysics modeling at meso-scale.
 The understanding of CO2 ice cloud formation is a major goal for Mars climate research for both current and past conditions. Billions of years ago, in the pristine and thick Martian atmosphere, CO2 ice clouds may have been a major greenhouse actor and may have helped to raise surface temperature above water melting point [Forget and Pierrehumbert, 1997; Mischna et al., 2000]. In light of our observations, however, it appears that CO2 clouds on Mars are essentially convective in nature and that a nonunity fractional cloud cover could be expected in such context. This aspect may have been a severe limitation for the cloud greenhouse effect since radiative transfer models predict that a 75% CO2 cloud cover produces 20 to 30 K less warming than a full cover [Forget and Pierrehumbert, 1997]. However, if the observed concentration of clouds in specific areas was also a feature of CO2 clouds in ancient Mars, greenhouse warming may have been very effective in selected places, allowing liquid water to run there. The notion of Mars being habitable globally may then be scaled down to the regions where cloud formation was favored by circulation patterns as on Earth in the Inter-Tropical Convection Zone.
 The authors express their sincere thanks to all the members of the Mars Express team for permitting the success of the mission. F.M. is indebted to A. Fedorova, who provided spectroscopic expertise in the early stages of the study. The authors also thank R. Todd Clancy for his constructive review of the manuscript and Robert Carlson for additional comments. Both have significantly helped improve the overall quality of the manuscript.