Carbon dioxide clouds, which are speculated by models on solar and extra-solar planets, have been recently observed near the equator of Mars. The most comprehensive identification of Martian CO2 ice clouds has been obtained by the near-IR imaging spectrometer OMEGA. CRISM, a similar instrument with a higher spatial resolution, cannot detect these clouds with the same method due to its shorter wavelength range. Here we present a new method to detect CO2 clouds using near-IR data based on the comparison of H2O and CO2 ice spectral properties. The spatial and seasonal distributions of 54 CRISM observations containing CO2 clouds are reported, in addition to 17 new OMEGA observations. CRISM CO2 clouds are characterized by grain size in the 0.5–2 μm range and optical depths lower than 0.3. The distributions of CO2 clouds inferred from OMEGA and CRISM are consistent with each other and match at first order the distribution of high altitude (>60 km) clouds derived from previous studies. At second order, discrepancies are observed. We report the identification of H2O clouds extending up to 80 km altitude, which could explain part of these discrepancies: both CO2 and H2O clouds can exist at high, mesospheric altitudes. CRISM observations of afternoon CO2 clouds display morphologies resembling terrestrial cirrus, which generalizes a previous result to the whole equatorial clouds season. Finally, we show that morning OMEGA observations have been previously misinterpreted as evidence for cumuliform, and hence potentially convective, CO2 clouds.
 Low to midlatitudes clouds can be more easily observed and studied than polar clouds as they occur during daytime. Despite the large number of CO2 clouds observations acquired so far, several properties are still subject to debate as they differ depending on the data set, method or viewing geometry used for their estimation. For example, the four nighttime clouds observed in SPICAM limb data and expected to be CO2 clouds [Montmessin et al., 2006a] are thin (τ < 0.01), small-grained (reff ≃ 0.1 μm) and located for half of them in the 30°S–40°S latitude range, which is not consistent with daytime nadir direct detections of CO2 clouds by OMEGA [Määttänen et al., 2010]. The discrepancy could be due to day-night variability [González-Galindo et al., 2009] or to observational/methodological biases (nadir versus limb viewing geometry/confusion with H2O clouds). Similarly, clouds interpreted as probable CO2 clouds have been reported in the Ls 140°–170° range from TES/MOC limb data [Clancy et al., 2007] while CO2 cloud detections are almost never reported at these LS from OMEGA nadir data [Määttänen et al., 2010]. The cloud morphology is also subject to debate as it has been alternatively described as mostly “cumuliform” [Montmessin et al., 2007] or on the contrary mainly “ripple-like to filamentary” [Scholten et al., 2010]. This apparent disagreement could be due to observational biases such as spatial sampling or to the existence of different categories of clouds possibly related to local time or solar longitude [Määttänen et al., 2010]. The cumuliform morphology of CO2 clouds inferred from some observations has raised the question of a possible convective activity within these clouds [Montmessin et al., 2007; Määttänen et al., 2010; Colaprete et al., 2008]. Calculation have however concluded that the whole energy available in the cloud, in the form of latent heat, can form at maximum a convective cell 3 to 5 order of magnitude smaller than the size of the cloud [Määttänen et al., 2010]. Getting better constraints about the distribution and properties of CO2 clouds will help to understand their formation mechanism: on the one hand, ongoing studies try to assess which physical mechanisms can decrease mesospheric temperatures [Spiga et al., 2010; González-Galindo et al., 2010] as global circulation models do not indicate sufficiently cold temperatures to initiate CO2 cloud [Montmessin et al., 2007]; one the other hand, a better understanding of the microphysics of CO2 cloud formation is required [Määttänen et al., 2010], notably to understand why subfreezing temperatures are observed without clouds [Forget et al., 2009].
 OMEGA [Bibring et al., 2005] is a very powerful tool for the study of CO2 clouds as it is able to unambiguously identify the CO2 composition of a cloud based on a near-IR spectral feature located at 4.26 μm [Montmessin et al., 2007; Määttänen et al., 2010]. OMEGA is however not exempt of observational biases, such as a low spatial resolution of 0.3 to 5 km and gaps in the spatial/time coverage. Moreover, the “4.26 μm” method cannot detect potential low altitude or small-grained CO2 clouds [Montmessin et al., 2007]. The CRISM instrument is, with OMEGA, the second visible and near-IR imaging spectrometer currently orbiting Mars. CRISM offer a significantly higher spatial resolution (20 m) than OMEGA but its shorter wavelength range does not include the diagnostic “CO2 ice” 4.26 μm feature.
 In this paper, we present a new method that makes it possible to identify CO2 clouds in the CRISM data set, without ambiguity regarding H2O clouds. Results are compared to those established by previous studies, which make it possible to extend our knowledge of CO2 cloud occurrences and properties, and understand the different observational biases associated with various data sets. New nadir and limb observations by OMEGA are also presented.
2. CRISM Data Analysis
2.1. Identification of Clouds
 The CRISM web map interface (http://crism-map.jhuapl.edu/) provides a rapid overview of the spectral images acquired by CRISM. For each targeted observation, a visible RGB composite image based on wavelengths 0.592 μm, 0.533 μm and 0.492 μm and a near-IR brightness image at a wavelength of 1.33 μm are automatically processed [Murchie et al., 2007]. We have performed a global survey of all CRISM targeted observations obtained from the beginning of the mission to summer 2010 in the latitude range [60°S–60°N] using this interface. Both the visible and near-IR images have been scanned in search of cloud morphologies. About 200 observations have been found to show morphological features characteristic of atmospheric clouds (see examples in Figure 1). Cloudy features appear systematically in the RGB images but are frequently poorly discernable in the near-IR. Observational biases associated with this detection method are discussed in sections 2.2 and 2.4.
2.2. Cloud Spectral Properties
 We first analyze the spectral properties of these clouds to estimate their composition. Observations obtained above a cloud contain both the cloud and the surface signatures (the latter frequently dominates). To isolate the cloud impact on the radiance, we need to compare the spectral properties observed above a cloud to the spectral properties of a nearby surface assumed to be similar to the surface below the cloud. CRISM clouds are typically divided in small optically thicker areas which extent is lower than a few km for the north-south dimension (Figure 1). CRISM observations are obtained at about 15.00 local time, hence the sun is typically SW oriented with a solar zenith angle of about 45° near the equator. Considering that the altitude of clouds is greater than a few km, the surface below a given area of the cloud is mainly illuminated by sunlight that did not go through it. At first order, the reflectance Rcloud measured by CRISM above a cloud is thus of the form (see Figure 2):
where Rsurface is the reflectance measured above the nearby surface (assumed to be Lambertian), Tcloud is the cloud transmission factor (affecting the light coming from the ground and going through the cloud), and Scloud is the reflectance of the cloud (light scattered upward by the cloud without interaction with the surface). Note that Tcloud, which account for multiple scattering within cloud particles, can be greater than 1. This simple equation neglects the role of the light scattered downward by the cloud and then scattered upward by the surface through the cloud, a relevant approximation considering the small extent of the clouds compared to their altitudes. To first order, the ratio Rcloud/Rsurface is an estimate of Tcloud when Rsurface is sufficiently high, and the subtraction Rcloud - Rsurface is an estimate of Scloud when Tcloud is close to 1 and/or Rsurface is close to 0. As observations provide both Rcloud and Rsurface, we can therefore rapidly infer an estimate of Tcloud and Scloud by calculating the ratio and subtraction of the cloud spectrum versus the surface spectrum (see an example in Figure 3). These estimates will be compared to radiative transfer modeling of Tcloud and Scloud in section 2.3.
 We assume that only 3 components can mainly participate to the composition of present-day Martian clouds given our current comprehension of the physics and chemistry of the Martian atmosphere: mineralogical aerosols (“Martian dust”), water ice, and CO2 ice. Other species, such as H2SO4, have been hypothesized to condense in the Martian atmosphere during earlier epochs [Levine and Summers, 2011]. However, there is no observational or theoretical evidence for the formation of these aerosols on present-day Mars. Metastable particles of clathrate could theoretically exist [Chassefière, 2009], but have not been observed and are not expected to form localized clouds as they would originate from the subsurface where the pressure is high enough for their formation. Spectral signatures of such clathrate hydrate particles (methane clathrate, CO2 clathrate) will be dominated by their main component: water ice [Schmitt et al., 2003; Dartois and Schmitt, 2009]. Micrometeorites, which spectral properties will differ from Martian dust, are clearly not expected to form localized clouds of small particle size such as observed by CRISM [Court and Sephton, 2011; Toppani et al., 2001].
 Martian dust aerosols can be easily distinguished from ice particles in the visible wavelengths: both CO2 and H2O ices present no major absorptions (Tcloud is close to unity) while Martian dust is characterized by a strong, ubiquitous ferric absorption with Tcloud decreasing from 1 μm to 0.4 μm [Vincendon et al., 2009]. We have detected no significant contribution from dust for the clouds analyzed in this study. This is consistent with the method used to identify clouds (see section 2.1): ice is bright at visible wavelengths compared to the surface, while the spectral differences between surface and dust aerosols are smaller as they both contain a ferric absorption feature. We present in the next section a methodology to differentiate CO2 from H2O clouds on the basis of their apparent spectral properties.
2.3. Spectral Discrimination Between CO2 and H2O Clouds
 We isolated 2 spectral categories of clouds in our data set. A typical example is shown in Figure 3. The first category of clouds is characterized by a strong water ice absorption feature at 3 μm, frequently associated with water ice absorption features at 1.5 μm and 2 μm. The presence of these characteristics water ice features imply that these clouds are dominated by water ice. On the contrary, the second category of clouds shows no water ice features in their ratio spectrum (Figures 3a and 3b), even after performing averaging to enhance the signal-to-noise ratio. Two hypotheses can then explain these clouds: (1) they are composed of water ice, but due to their small particle size or near-IR optical depth, the water ice features are not detectable; (2) they are not primarily composed of water ice, which indicates a CO2 ice composition if we make the reasonable assumption that H2O and CO2 are the only condensable species.
 We performed Mie and radiative transfer modeling to assess the spectral properties of H2O and CO2 ice clouds of various particle size and optical depth. We computed the single scattering parameters of various Gamma size distributions of ice particles using a Mie radiative transfer code [Mishchenko and Travis, 1998]. We used the H2O ice optical constants gathered by Schmitt et al.  for a 145 K temperatures, which are composed of values from Grundy and Schmitt  and Trotta  for wavelengths greater than 1 μm and values from Warren  for wavelengths less than 1 μm. For CO2 ice, we used the CO2 optical constants measured by Schmitt et al.  for a temperature of 15 K. While 145 K is a relevant temperature for Martian atmospheric water ice [Smith, 2004], 15 K is too low as mesospheric CO2 clouds will typically have higher temperatures about 80 K [Montmessin et al., 2006a]. We compared our CO2 ice constants with measurements by Hansen [2005, 1997] acquired at 150 K and found no significant differences within our wavelength range given our spectral resolution. A Monte-Carlo multiple scattering radiative transfer code [Vincendon et al., 2007] is then used to simulated the impact of clouds with various optical depths and mean grain sizes on solar radiance. Results are presented in Figure 4.
 We explored a range of visible optical depths varying from 0.05 to 0.5 at 0.5 μm and 6 different particle size distributions (reff = 0.1 μm and νeff = 0.05, reff = 0.5 μm and νeff = 0.05, reff = 1 μm and νeff = 0.1, reff = 2 μm and νeff = 0.1, reff = 3 μm and νeff = 0.1, reff = 6 μm and νeff = 0.1). This range corresponds to detectable clouds using our approach (see section 2.1), as smaller cloud optical depths will not significantly scatter light at 0.5 or 1.3 μm (Figure 4). The spectral behavior of CO2 clouds is mainly featureless: the transmission factor smoothly varies about 0.99 for an optical depth of 0.1 (Figure 4b). On the contrary, H2O clouds transmission contains a strong 3 μm absorption feature (Figure 4a), and two shallower 1.5 μm and 2 μm bands which detection requires a high signal-to-noise ratio for small-grained clouds (Figure 4g). Small-grained CO2 clouds (0.1 μm) mainly scatter at shorter visible wavelengths, as we are close to the Rayleigh regime for this grain size (Figure 4d). Larger grain sizes significantly scatter at all near-IR wavelengths; the wavelength of the maximum is roughly equal to the mean grain size (Figure 4d). The scattering pattern of H2O ice is the combination of a similar wavelength slope and a strong absorption (large grains) or emission (small grains) feature near 3 μm (Figures 4c and 4e). This complex spectral behavior results from the increase of the attenuation coefficient of water ice from 0.01 at 2.8 μm/3.6 μm to ≈ 1 at 3.1 μm [Warren, 1984; Grundy and Schmitt, 1998]. As the attenuation coefficient increases near 3.1 μm, the refractive index also increases from 1.3 to 1.8. For small particles, this attribute results in a strong increase of the quantity of light scattered by water ice grains at 3 μm to 3.6 μm compared to surrounding wavelengths. A similar effect is observed for CO2 ice at wavelengths about 4.3 μm [Montmessin et al., 2007]. For both H2O and CO2 clouds, changing the optical depth between 0.05 and 0.5 mainly results in scaling figures by a constant factor (Figures 4d, 4f, and 4h).
 To summarize, water ice clouds are always characterized by a diagnostic 3 μm water ice feature, frequently accompanied by 1.5 μm and 2 μm features (see also Langevin et al. ). The 3 μm feature is already several % deep in the transmission spectrum of thin clouds with normal optical depth of 0.05–0.1 corresponding to our limit of detection. All clouds detected using our approach (section 2.1) will therefore show diagnostic features of water ice in their transmission spectra (∼ratio) if they are composed of water ice. On the contrary, CO2 ice is characterized by a constant transmission spectrum. As a consequence, the spectral properties of clouds near 1.5, 2 and 3 μm are diagnostic of their H2O versus CO2 composition: clouds that lack water ice features in their ratio spectrum are CO2 clouds (Figure 3). On this basis, we have inspected the spectral properties of all CRISM cloudy observations and we have classified clouds as CO2 (Table 1) or H2O (Table 2) accordingly. We can see in Figure 3c that a shallow but wide feature near 3 μm is observed in the “subtraction” spectrum derived for CO2 clouds. This feature is not expected for CO2 ice clouds scattered reflectance (Figure 4). The origin of this feature is unclear: as explained in section 2.2, the “subtraction spectrum” is only a first order estimate of the cloud scattered reflectance, it still contains part of the surface reflectance; as the surface of Mars is characterized by a strong wide 3 μm absorption band, the shallow and wide 3 μm band observed in the subtraction spectrum could be a remnant of this surface feature. However, our modeling indicates that Tcloud is slightly lower than 1 (Figure 4b), which would results in a weak contribution of an upturned surface 3 μm feature, while we observe an actual 3 μm feature. A change in the assumptions regarding the single scattering properties of CO2 ice particles could result in Tcloud > 1, which would explain the stronger water ice feature observed above the cloud. Alternatively, this feature could betray the presence of a few water ice particles within the CO2 cloud, or even within CO2 ice grains as water ice grains could act as condensation nuclei for CO2 ice.
Table 1. List and Properties of CRISM Observations Containing CO2 Cloudsa
Observations used as examples in Figures 1, 3, 13, 14, 15 and 16 are indicated in the last column. Here “app. morph.” refers to the apparent morphology classification performed in Figure 14.
Table 2. List and Properties of CRISM Observations Containing H2O Clouds
2.4. Observational Biases
 We provide in this section information about the various observational biases associated with CRISM observations of clouds. This will help comparing cloud properties retrieved with CRISM, such as apparent morphology or spatial distribution, with information provided by previous studies based on other instruments.
 First, at high resolution, CRISM observes a given surface scene of about 11 × 11 km with a varying emergence angle: the pointing changes to compensate for the spacecraft motion in order to increase the integration time. The orbit of MRO is nearly polar: the emergence (emission angle) varies from about −35° in the south part of a given image to 0° (nadir viewing) in the center of the image to +35° in the north part. As the field of view of CRISM remains constant, this procedure creates the hourglass shape seen in CRISM image once projected on the ground [Murchie et al., 2007]. As a consequence, at 75 km altitude, the typical altitude of CO2 clouds [Scholten et al., 2010; Määttänen et al., 2010], a surface of 11 × 116 km is intercepted in the field of view: high altitude clouds are therefore compressed by a factor of about 10 in the nearly north-south direction of the MRO orbit, without being changed in the east-west direction. This effect is illustrated in Figure 5: a CO2 cloud with NW-SE stripes observed by HRSC is idealized as linear cloud features at 75 km altitude and projected in the CRISM observing geometry to estimate biases resulting from this specific CRISM viewing geometry. The altitude of known CO2 ice clouds typically ranges from 50 to 100 km [Määttänen et al., 2010; Scholten et al., 2010; Montmessin et al., 2007, 2006a; Clancy et al., 2007]. This altitude range corresponds to North-South compression factors between 7 and 14.
 Second, the CRISM data set tends to be biased toward surface sites that have been intensively studied for their mineralogical diversity, such as Mawrth Vallis, Terra Meridiani, Valles Marineris and Nili Fossae. We can see in Figure 6 that the distribution of CRISM observations is to first order well-scattered through the whole Martian planisphere. However, we can see that observational biases are responsible for small shape details of CO2 clouds clusters.
 Third, there are also some observational biases which depend on the solar longitude:
 1. The distribution of CRISM observations as a function of LS is slightly irregular, notably due to coverage gaps related to MRO safe mode or instrumental problem occurrences in 2009 and 2010. However, the distribution of equatorial CO2 cloud occurrence is not correlated to these variations (Figure 6c).
 2. A very limited number of observations have been acquired by CRISM at mid northern latitudes (40°N–60°N) during northern autumn (Figure 6b). This will prevent us from doing comparison with cloud detections previously reported in this area (Figure 7).
 Finally, as our cloud detection method is based on the visual observation of cloudy features in CRISM high resolution images, we will not detect clouds that are uniform over large areas greater than the extent of CRISM images (i.e., larger than 11 km in the east-west direction and up to 110 km in the north-south direction; see Figure 5).
3. Results and Discussion
3.1. Spatial Distribution
 We have identified 54 observations of CO2 clouds in the CRISM data set (Table 1). All of these observations are located in the [13°S, 9°N] latitude range (we have searched for CO2 clouds between 60°S and 60°N). The spatial and time distributions of these clouds are compared to the distribution of H2O clouds observed by CRISM in the [25°S, 25°N] range in Figure 6. We can see that clouds classified as H2O or CO2 on the basis of their spectral properties have very distinct spatial and seasonal patterns.
 We compare the spatial and LS distributions of CRISM CO2 clouds to previously published detections of CO2 or high altitude clouds (>60 km) in Figures 7, 8, and 9. OMEGA observations refer to the 50 Mars Express orbits previously published for Mars Years (MY) 27 to 29 [Määttänen et al., 2010], plus 17 new Mars Express orbits for MY 30 (Table 3) where we have detected CO2 clouds using the same approach as originally developed by Montmessin et al. . Thirteen of these seventeen new detections are scattered (in a relatively homogeneous manner) between LS 10° and LS 80°, three have been obtained at LS 115–123° and one at LS 151°. A given Mars Express orbit can contain widespread CO2 clouds extending over several tens of latitude degrees: we have sampled these OMEGA clouds every 5° latitude and obtain 150 OMEGA cloud occurrences that can be compared to CRISM observations. High altitudes clouds refer to TES/MOC “MEM” clouds [Clancy et al., 2007], HRSC “CO2” clouds [Scholten et al., 2010; Määttänen et al., 2010], SPICAM “CO2” clouds [Montmessin et al., 2006a] and THEMIS high altitude clouds (>60 km) [McConnochie et al., 2010].
Table 3. List and Properties of the New OMEGA Observations With CO2 Clouds
 CRISM CO2 clouds are exclusively observed near Valles Marineris (between 113°W and 53°W) and near Terra Meridiani (between 10°W and 26°E), with the exception of one detection at 155°E. All CRISM detections except the 155°E one fall within previously known areas of CO2 cloud activity as derived from the composition-based OMEGA CO2 cloud detections. The 155°E detection occur in an area of known sporadic CO2 cloud activity (one OMEGA observation at 120°E and two HRSC [Scholten et al., 2010] “high altitude” cloud observed in the neighborhood). The OMEGA+CRISM pattern of CO2 cloud activity is generally in agreement with TES/MOC and HRSC observations of mesospheric clouds, except for a few TES/MOC observations near 85°W and 10°N (Figure 7). We do not observe with daytime CRISM and OMEGA observations any cloud with a CO2 spectroscopic composition in the latitude range 30–40°S where SPICAM has observed 2 nighttime high altitude bright clouds referred as “CO2” clouds. This latitude range corresponds to an area of frequent cold “CO2 ice compatible” mesospheric temperatures at night based upon SPICAM mesospheric temperature retrievals [Forget et al., 2009]. These two SPICAM clouds are also very distinct from other clouds in latitude/season and longitude/season diagrams (Figure 9). No CO2 clouds are seen by CRISM at mid latitudes, while CO2 (OMEGA) or high altitude (THEMIS) clouds are reported between 40°N and 50°N for LS in the [215°–275°] range (Figure 7). As discussed in section 2.4, only a few CRISM observations have been acquired for these latitudes at appropriate LS. Hence, this mismatch is not significant.
3.2. Time Distribution
 All CRISM CO2 clouds are observed between LS 0° and LS 138° (Figure 8), with a main maximum at LS 20° and a secondary maximum at LS 130°. Between LS 0° and LS 40° (main maximum), about 30% of CRISM equatorial (±10° latitude) observations near Valles Marineris and Terra Meridiani contain CO2 clouds. CRISM LS distribution is in good agreement with OMEGA detections, as well as with most TES/MOC and HRSC mesospheric cloud occurrences. There are a few differences between CRISM and OMEGA after LS 40°: the LS 20° peak ends at LS 40° for CRISM versus LS 60° for OMEGA; the frequency of clouds increases from LS 100° to LS 135° in CRISM data while it decreases in OMEGA data. However, these differences may be biased by the low number (12) of clouds observed by CRISM between LS 40° and LS 140°. A distinct peak activity in mesospheric clouds is reported by TES/MOC between Ls 140° and Ls 166° (Figures 8 and 9), which is not observed by either OMEGA or CRISM (OMEGA clouds are observed up to Ls 135°, + one cloud at Ls 151°). The pathfinder cloud (located at 19°N and 33°W and observed at night before dawn), interpreted as a CO2 cloud [Clancy and Sandor, 1998], also occur at later Ls of 162°. We note that the Martian years of OMEGA and CRISM observations differ from that of TES/MOC and Pathfinder observations. As explained in section 2.4, northern mid latitudes are infrequently observed by CRISM during northern fall, which prevents comparisons with the THEMIS and OMEGA cloud detections for the corresponding LS range (LS > 200°, see Figure 9).
3.3. Composition of High Altitude Clouds From OMEGA Limb
 This extensive comparison of various data sets highlights a few significant differences between spectroscopy-based identification of CO2 clouds (OMEGA and CRISM) and mesospheric clouds interpreted as probable CO2 clouds (TES/MOC, SPICAM, THEMIS, HRSC). It is difficult to draw conclusions from comparison only, as many factors could contribute to these discrepancies:
 1. Interannual variability: OMEGA and CRISM observations barely overlap with TES/MOC observations (MY 26–29 for CRISM-OMEGA versus MY24–26 for TES). They however overlap with SPICAM, HRSC and THEMIS observations.
 2. Local time: OMEGA observations are restricted to daytime and CRISM observes about 15.00 local time. On the contrary, SPICAM limb profiles are acquired during nighttime, THEMIS observations of mid latitudes clouds was obtained at twilight and the Pathfinder cloud was observed before sunrise. TES/MOC observations of mesospheric clouds have however also been acquired during daytime.
 3. Viewing geometry: thin clouds could be undetectable in nadir viewing geometries (OMEGA, CRISM) while seen in limb viewing geometries (TES/MOC, SPICAM). However, THEMIS also observes in nadir and the 3 OMEGA limb detections of CO2 clouds reported so far are within OMEGA nadir clouds areas [Gondet et al., 2008].
 4. “CO2 composition” detection limit of both OMEGA and CRISM. A very distinct technique has however been used to detect CO2 clouds with CRISM compared to OMEGA.
 5. H2O clouds could be misinterpreted as CO2 clouds when no composition measurements are available. However, H2O clouds are usually found at altitudes up to 60 km only [Heavens et al., 2011].
 To bring new constraints about this last hypothesis, we have analyzed OMEGA limb observations acquired at time and places where discrepancies are noticed. One OMEGA limb observation showing high altitude bright layers extending up to 75–80 km has been obtained at LS 150° above latitudes 30–37°S for a longitude of 18°E (Figure 10). This observation can be compared to the high altitude clouds detected by SPICAM at the same latitudes a little earlier in the season (LS 135°) and by TES/MOC at the same LS a little closer to the equator (<20°S).
 At visible and shorter near-IR wavelengths, bright atmospheric layers are observed at altitude extending up to 75–80 km (Figure 11). The spectral properties of these high altitude aerosols are diagnostic of H2O ice (Figure 12): as discussed in the modeling section (2.3), small-grained (0.1–0.5 μm) water ice aerosols present a scattering maximum at 3.1–3.3 μm. On the contrary, no CO2 ice signatures are observed at 4.26 μm. The observation of water ice clouds at such high altitudes seems consistent with preliminary results derived from MCS data [Kass et al., 2011], and with haze top altitude inferred from SPICAM stellar occultations [Montmessin et al., 2006b].
 This observation demonstrates that water ice aerosols can be found at very high altitudes where detached aerosols layers have been previously interpreted as CO2 ice layers. In particular, the 2 “detached layers” observed by SPICAM UV sensor at similar latitudes (∼35°S) and solar longitude (∼135°) at 75–95 km [Montmessin et al., 2006a] could in fact correspond to water ice aerosols, and not CO2 ice aerosols as claimed in the title of this publication. We can see in Figure 11 that the top altitude of the water ice layer slightly increases as wavelength decreases: this is consistent with a vertical gradient of particle size, with smaller particles at higher altitude. As a consequence, at UV wavelengths, the top of this water ice layer will be observed at even higher altitudes. This is also consistent with the fact that water ice clouds are usually observed at slightly lower altitudes at far-IR wavelengths [Heavens et al., 2011]. Moreover, SPICAM observes at night while OMEGA observes during the day, and water ice clouds are expected to be observed at even higher altitude at night [Heavens et al., 2010]. The simultaneous presence of temperature below the CO2 frost point was a major reason to interpret SPICAM clouds as CO2 ice clouds. However, the cold, CO2 ice compatible temperatures are systematically found by SPICAM at altitudes above the top of the detached layers [Montmessin et al., 2006a], and numerous, about 8%, SPICAM observations show such subfreezing temperatures, without detectable nearby ice particles for a large majority of them [Montmessin et al., 2011]. Particles size also matches, as both OMEGA high altitude water ice clouds and SPICAM clouds correspond to small, submicron grains. On the contrary, CO2 clouds are usually larger-grained (see section 2.4). As a consequence, we believe that H2O ice is an interpretation at least as valid as CO2 ice for these high altitude bright layers that are observed in areas where OMEGA and CRISM do not observe CO2 clouds, but where OMEGA does observe mesospheric H2O clouds. Similarly, the peak in equatorial mesospheric cloud activity observed by TES/MOC in late northern summer could partly result from these high altitude late northern summer water ice clouds observed at low latitudes.
3.4. Optical Depth and Particle Size Estimate
 We have derived the mean optical depth and particle size of a sample of CRISM CO2 clouds using the modeling approach detailed in section 2.2. We show an example of this optical depth and particle size retrieval in Figure 13. Typically, CO2 clouds particle size as derived from CRISM observations range from 0.5 μm to 2 μm. This compares very well with particle size estimate based on OMEGA data [Montmessin et al., 2007; Määttänen et al., 2010]. Visible optical depths (0.5 μm) are typically of 0.2 (Figure 13), in agreement with Montmessin et al. , and present spatial variability within a given cloud filament, similarly to what is shown by Määttänen et al. . Normal optical depth retrieval based on limb observations [Clancy et al., 2007; Montmessin et al., 2006a] where several order of magnitude smaller. However, these retrievals assume a spatially homogeneous and widespread cloud, which is not relevant as CO2 clouds are localized with small scale structures.
 The morphology of equatorial CO2 clouds has been first described as mostly “cumuliform” on the basis of a limited sample of OMEGA observations, suggesting (with other theoretical considerations) a potential convective formation mechanism [Montmessin et al., 2007]. A revised analysis of an extended set of OMEGA data had lead to the conclusion that only a minor fraction of OMEGA observation, mostly obtained during morning hours, can be classified as cumuliform, while most OMEGA observations (mostly obtained during mid-afternoon) show either cirrus-type morphologies or undefined morphologies. A lower limit of the cumuliform percentage has been estimated to 15% [Määttänen et al., 2010]. This reanalysis was supported by the higher spatial resolution HRSC observations which showed that “the prevailing morphologies are ripple-like or filamentary forms” [Scholten et al., 2010], and that a given cloud can appear “clumpy” at low spatial resolution while “high spatial resolution reveals a mainly filamentary interior structure of the cloud” [Scholten et al., 2010]. The orientation of the filaments in HRSC data is “roughly E-W, sometimes NW-SE.” The HRSC observations reported in these publications are limited in local time (afternoon hours) and LS range (0–36°), while “the round-shaped OMEGA clouds have been observed in the morning except for one early afternoon cases in the northern mid-latitudes” [Määttänen et al., 2010]. Further observations were therefore required at that time to determine to which extent the results derived from the analysis of HRSC data can be or not extended to all CO2 clouds.
 We have classified the apparent CRISM cloud morphologies in Figure 14. There is a strong correlation between apparent morphology and composition: all CO2 clouds are characterized by more or less straight filaments, while H2O clouds never show “straight filament” apparent morphologies but more often present blurred substructures (Figure 14). The apparent morphology of the clouds is impacted by cloud altitude due to the specific CRISM viewing geometry (Figure 5). However, H2O and CO2 daytime cloud altitude distributions overlap: 0–80 km and 50–100 km respectively. As a consequence, this dichotomy also reflects a variation of the morphology depending on composition. We have compared a pair of simultaneous observations of a CO2 cloud area by OMEGA and CRISM in Figure 15. This comparison show how the spatial resolution can impact our apprehension of CO2 cloud morphology: while the OMEGA observation would have been classified as “cumuliform” or “undefined” [Määttänen et al., 2010], CRISM clearly show fine filament-like structures in this cloud. To conclude, morphologies in CRISM afternoon observations are consistent with HRSC results [Scholten et al., 2010], with clouds composed of filaments “roughly E–W” oriented. This extends the result of Scholten et al.  outside the LS 0–36° range as CRISM equatorial CO2 cloud observations extend from LS 0° to 140°.
 CRISM observations are restricted to afternoon hours, while most of the “round-shaped” OMEGA clouds classified as cumuliform are observed at morning hours [Määttänen et al., 2010]. We have looked for additional observational constraints corresponding to morning hours. In Figure 16, we compare a CRISM afternoon CO2 cloud to the pre-dawn cloud observes from the surface by Pathfinder and interpreted as a probable CO2 cloud [Clancy and Sandor, 1998]. This cloud is described by these authors as a “bluish wave cloud.” While taken in different conditions, Pathfinder (morning, from the surface) and CRISM (afternoon, from orbit) slanted observations show a remarkable agreement in terms of apparent morphology. Although there is no direct spectroscopic evidence for the Pathfinder cloud to be composed of CO2 ice, this observation may indicate that morning CO2 clouds are also composed of filaments. Similarly, the pathfinder cloud apparent morphology has been shown to match THEMIS equatorial high altitude clouds morphology [McConnochie et al., 2010]. In Figure 17 we present a pair of HRSC/OMEGA observation of the same morning CO2 cloud. The cloud, as seen with the low spatial resolution typical of OMEGA morning observations of CO2 clouds, appear as a “rounded and clumpy, cumuliform cloud.” However, the exact same cloud observed at the same time by HRSC, also onboard Mars Express, is similar to HRSC afternoon clouds with “cirrus-type” morphologies composed of filaments. As a consequence, our analysis does not support the conclusions of Montmessin et al.  and Määttänen et al.  that OMEGA observations reveal a cumuliform morphology for at least 15% of the clouds (mostly morning clouds): OMEGA data cannot be systematically used for cloud morphology interpretation due to its varying spatial resolution.
 Observed cloud morphologies provide clues regarding potential cloud formation and evolution mechanisms. Our analysis shows that equatorial CO2 cloud observations acquired so far either show cirrus-type morphologies or do not have sufficient spatial resolution to estimate cloud morphology. This conclusion is valid for morning and afternoon clouds during northern spring and summer. Calculations of the convective available potential energy performed by Määttänen et al.  show that the energy available in a 3 × 3 km CO2 cloud will create a convective cell smaller than 10 × 10 m to 100 × 100 m. The probability to observe such a convective cell with CRISM is low, first because it represents 1/1000 to 1/100 000 of the surface of the cloud, and second because it is at the limit of CRISM high spatial resolution (20 m). As we do not observe such a convective cell, our observations are consistent with this calculation implying a minor contribution of convection to these clouds. Martian equatorial CO2 clouds are composed of filaments with a preferential East-West orientation. This orientation is similar to wind orientation at these altitudes [Angelats i Coll et al., 2004; Scholten et al., 2010]. CO2 clouds could be initiated locally, e.g., by gravity waves [Spiga et al., 2010], and then stretched by the easterly mesospheric winds.
4. Summary and Conclusions
 In this paper, we developed a new method to detect CO2 clouds with near-IR orbital data acquired by CRISM. The method is based on radiative transfer modeling results showing that CO2 clouds have distinct spectral properties compared to other potential aerosols. These CRISM observations were compared to previously published OMEGA CO2 cloud detections [Montmessin et al., 2007; Määttänen et al., 2010] and to new OMEGA observations acquired in 2009 and 2010. An extensive comparison with mesospheric ice clouds (altitudes greater than 60 km) observed in various data set (TES/MOC [Clancy et al., 2007], HRSC [Scholten et al., 2010], SPICAM [Montmessin et al., 2006a] and THEMIS [McConnochie et al., 2010]) was also implemented.
 The new CRISM measurements reveal 54 CO2 clouds located between 13°S and 9°N: 24 near Valles Marineris (250°E–310°E), 29 near Terra Meridiani (350°E–30°E), and one at 155°E. These clouds occur within areas and periods of previously known CO2 cloud activity from OMEGA data, except for the 155°E cloud. CRISM CO2 clouds are observed between LS 0° and 140°, with a maximum activity at LS 20°. Their effective radius typically ranges from about 0.5 to 2 μm and their optical depths are up to 0.3 for the thickest clouds.
 At low to mid latitudes, the distribution of spectroscopic identifications of CO2 clouds is at first order consistent with the distribution of high altitude (>60 km) ice clouds, which shows that CO2 clouds dominate the mesosphere of Mars. Differences are observed at second order; they are notably due to infrequent mesospheric water ice clouds, as revealed by a water ice cloud layer extending up to 80 km observed by OMEGA.
 CO2 clouds in CRISM afternoon observations are composed of filaments and resemble terrestrial cirrus clouds; filaments are frequently East-West oriented. This extends the result of Scholten et al.  based on HRSC afternoon observations to the full LS range of equatorial CO2 cloud activity. A combined OMEGA/HRSC morning observation suggests that OMEGA morning observations have been previously misinterpreted as evidence for the existence of cumuliform, and hence potentially convective, CO2 clouds.
 The nature and origin of cloud condensation nuclei has not been addressed in this study. With the help of radiative transfer models of radially inhomogeneous particles, near-IR spectroscopy may hold information about the composition of condensation nuclei. In addition to nanophase ferric oxides coming from the surface and micrometeorites coming from space, water ice grains, present at CO2 clouds altitudes according to our limb observation, could provide efficient cores on which to condense. Interestingly, spectral properties derived above CO2 clouds contain a shallow feature at 3 μm resembling water-related features. Further modeling is however required to understand the origin (surface or cloud) of this spectral feature.
 The authors would like to thank R. T. Clancy, Y. Langevin, F. Montmessin, J. Mustard, B. Schmitt and A. Spiga for helpful discussions during the preparation of this paper.