Understanding the present and past water cycle on Mars requires an accurate knowledge of the distribution and amount of H2O available near the surface. In this article, we present a map of the distribution of the surface material exposed between 87°S and 70°S in the summer (e.g., CO2 and H2O ices, dust) based on temperature measurements made by Thermal Emission Imaging System (THEMIS). Our compositional map (100 m per pixel) is in good agreement with spectral mapping returned by Observatoire pour la minéralogie, l'Eau, la Glace et l'activité (OMEGA) and Compact Reconnaissance Imaging Spectrometer for Mars (CRISM). Exposed water ice covers a total surface area of approximately 40,000 km2. A large fraction of the water ice is exposed at the periphery of the CO2 cap. An approximately 25,000-km2 large patch centered at 83.5°S and 345°E is discovered and represents the largest exposure of water ice in the southern hemisphere. It is not located on the south polar layered deposits but on the surrounding mantled terrains. THEMIS VIS, MOC and HiRISE images indicate that the surface roughness of exposed water ice terrains is typically lower than that of the surrounding dust. Polygonal patterns are observed on the water ice but not exclusively. There is a strong correlation between the surface albedo, the composition of the exposed material, and the timing of the initial seasonal CO2 frost deposition and final removal. These exposed water ice outcrops are not stable in the present environment and lose a vertical layer of tens to hundreds of micrometer a year to the atmosphere. The spike of water vapor above the south pole during the southern summer occurs while the water ice is still covered by a layer of seasonal CO2 frost, indicating that the sublimation of the exposed water ice is not the main contributor of vapor for the southern atmosphere. As water ice is not stable, it may indicate the past location of part of the CO2 perennial cap that has been eroded.
 The south polar layered deposits (SPLD) of Mars and the surrounding mantled terrains are of high interest because they contain the largest body of water ice near the surface of Mars [Zuber et al., 2007a], they have recorded the past climate and global catastrophic events modifying the surface [Clifford et al., 2000], and because they concentrate some of the most dynamic processes shaping the surface within a single region. These dynamic processes include the waning and waxing of the seasonal cap, basal sublimation and venting of the seasonal CO2 frost leading to the formation of spiders and etched polygons [Kieffer et al., 2006; Kieffer, 2007; Piqueux and Christensen, 2008b], expansion and contraction of the perennial cap over many years [Piqueux and Christensen, 2008a], possible initiation of local dust storms due to high atmospheric thermal gradients [Toigo et al., 2002 and references therein], and trapping and release of volatiles in the atmosphere during climate shifts [Ingersoll, 1974; Toon et al., 1980; Fanale and Jakosky, 1982; Jakosky and Carr, 1985; Mellon and Jakosky, 1995; Mellon et al., 2004]. The ices exposed at the surface form the interface between the solid water and the atmospheric reservoirs. The details of the exchange of volatiles between these two reservoirs must be well constrained to understand the current and past climates of Mars. This article contributes to this effort by presenting a thermal analysis of the near surface of the SPLD using the Mars Odyssey Thermal Emission Imaging System (THEMIS), allowing us to map the distribution of the surface materials, namely CO2 and H2O ices and particulated material (dust and sand).
 Several instruments, beginning with the Mariner 9 cameras in the 1970s, have provided albedo, morphological, thermal, spectral, topographic and chemical data of the south seasonal and perennial caps, the SPLD, and the surrounding mantled terrains. They have helped to identify the different constituents (e.g., CO2 and H2O ices, particulated materials), their relative proportions, distribution, and seasonal behavior. Kieffer  has shown that the south perennial cap is covered with CO2 frost and Kieffer  demonstrated that north polar cap is composed of water ice using temperature measurements made by the Infrared Thermal Mapper (IRTM). These data confirmed earlier discoveries based on telescopic observations and laboratory work [Herr and Pimentel, 1969; Kieffer, 1970; Larson and Fink, 1972]. The Mars Orbiter Camera (MOC), onboard Mars Global Surveyor (MGS), returned meters-scale images of the seasonal and perennial caps and helped to characterize the south polar environment. Among the most significant discoveries was the identification of erosive features (spiders, etched polygons) associated with what was subsequently hypothesized to be the basal sublimation of the translucent seasonal cap [Piqueux et al., 2003; Kieffer et al., 2006; Kieffer, 2007; Piqueux and Christensen, 2008b], and the identification of erosive depressions on the perennial cap [Thomas et al., 2000; Malin and Edgett, 2001; Malin et al., 2001; Thomas et al., 2005]. These observations, together with THEMIS IR measurements of water ice temperatures at their floors [Titus et al., 2003, 2004] were combined with thermal modeling [Byrne and Ingersoll, 2003], and showed that the southern perennial cap has a binary composition of CO2 ice (8–10 m thin veneer) on a wider body of mostly buried frozen water.
 The Thermal Emission Spectrometer (TES) measured the temperature, albedo and the spectra of the polar ices [Kieffer et al., 2000]. TES data suggested that the translucent cap region is due to CO2 crystals that are large enough to be transparent at the visible wavelengths [Kieffer et al., 2000]. The translucent ice is associated with the basal sublimation of the seasonal cap. Albedo and temperature measurements helped Kieffer et al.  to determine the ice mass budget of the CO2 caps. Topographic data returned by the Mars Orbiter Laser Altimeter (MOLA) [Smith et al., 1999] have been used to constrain the mechanical properties of the SPLD, and show that they are likely composed of a mix of water ice and dust [Nye et al., 2000]. In addition, MOLA has shown that the thickness of the seasonal CO2 frost deposits to be 1–2 m [Smith et al., 2001]. The integration of MOLA topography, Mars Express Mars Advance Radar Sounding and Ionosphere Analysis (MARSIS) mapping of the base of the SPLD, and high resolution gravity data made possible by the Mars Reconnaissance Orbiter flight configuration [Zuber et al., 2007b] have been used by Zuber et al. [2007a] to calculate the density of the SPLD and determine that they are made of 85% water with ∼15% dust by weight. MARSIS data confirmed that the SPLD are composed of almost pure water ice [Plaut et al., 2007]. The Mars Odyssey Gamma Ray Spectrometer instrument suite has mapped the distribution of the subsurface water ice, only observed at the high latitudes [Mitrofanov et al., 2002; Boynton et al., 2002; Feldman et al., 2004]. The first detection of exposed water ice near the south pole of Mars was made by Titus et al.  using THEMIS data. Bandfield  mapped the relative depth of the water ice table at the high latitudes using THEMIS measurements and thermal modeling. Observations of the seasonal cap by THEMIS and OMEGA also facilitated the understanding of the nature of the dusty and sandy spots and fans covering the seasonal cap on the translucent cap [Kieffer et al., 2006; Langevin et al., 2006; Kieffer, 2007]. Finally, THEMIS has mapped the changes of the distribution of the CO2 ice composing the perennial cap [Piqueux and Christensen, 2008a; Winfree and Titus, 2007]. Putzig and Mellon [2007a] have mapped the thermal inertia of the south polar region and found an extensive region between 87°S and ∼80°S with a very high thermal inertia (e.g., >1000 J m−1 K−1 s−0.5, also called “I units” in this article) consistent with exposed or near surface water ice. OMEGA, onboard Mars Express, has provided the first spectral detection of water ice near the south pole [Bibring et al., 2004] and constraints on the H2O crystal sizes [Douté et al., 2007]. The exposed water ice outcrops mostly near the CO2 perennial cap. The map produced by Bibring et al.  has a resolution of ∼2 km and a partial coverage of the polar latitudes but most of the terrains pole-ward of 87°S are observed, which is not always the case for other instruments because of their polar orbit. CRISM has also identified the spectra of water ice near the region where THEMIS has first discovered water ice [Titus et al., 2008].
 Most of the existing work has focused on the characterization of the ice volume in the SPLD. In this article, we extend the work of Titus et al.  and map at 100 m per pixel, the highest resolution to date, the distribution of the ices near the south pole of Mars using THEMIS measurements.
3.1. Data Selection
 The SPLD are free of seasonal frost during the summer, between Ls = 320° and 360° [Kieffer, 1976; Kieffer et al., 2000; Piqueux and Christensen, 2008a]. We have selected THEMIS IR data taken during this time period (Table 1). Approximately 90% of the data selected for this study are taken between 325° < Ls < 345° and most are near Ls = 335°. The SPLD range from 90°S to 75°S. Because of the inclination of Mars Odyssey polar orbit, there are no data available between the south pole and 87°S. This study applies to the latitudes southward of 70°S but most of our figures and discussion do not further discuss the case of the terrains between 70°S and 80°S because we have determined that they are only covered with dust. We have only selected the “noon” side of the THEMIS images, i.e., the portion of the images with the highest sun elevation. In the early afternoon, the surface temperatures are higher than near midnight and the signal-to-noise ratio is higher. Typically, the local times of the THEMIS images selected for this work are composed of between 12.5 H and 16.0 H (24 H equals 1 Martian day), when the thermal contrast between the different surfaces is high (Table 1).
 Polar materials (e.g., dust/sand, H2O and CO2 ices) are well characterized by their temperature and albedo [Titus et al., 2003; Piqueux and Christensen, 2008a]. In this study, the identification of polar materials is based on the fit of temperature measurements made by THEMIS with the surface temperature prediction of a thermal model. The surface temperatures are found after solving for:
with ɛ the emissivity of the surface, σ the Stefan–Boltzmann constant, Ts the surface temperature, SM the insolation at the surface of Mars after atmospheric correction, A the albedo, i the solar incidence angle, FIR the downwelling atmospheric radiation, L the latent heat of H2O sublimation with m the mass ice to sublime, t the time, I the thermal inertia, P the diurnal period, T the subsurface temperature field, and Z the depth. The portion of the numerical code that models the atmospheric contribution are from H.H. Kieffer's KRC model [Fergason et al., 2006]. The thermal inertia I is related to the thermophysical properties of the Martian materials by:
with ρ the density and Cp the specific heat. The model determines the surface temperature Ts from the thermal inertia. I is then interpreted in terms of composition or particle size when appropriate. The parameters for the thermal modeling are detailed in Table 2. They are either published values (e.g., water ice density, specific heat, thermal conductivity, etc.) or straightforward data provided by TES (surface albedo and emissivity, average atmospheric dust opacity, etc.).
Table 2. Parameters Used for the Thermal Modeling (Figure 1)
 Our mapping technique does not allow us to detect water ice on the carbon dioxide perennial cap. Water crystals mixed within the perennial cap are at CO2 ice temperature and cannot be detected based on their temperatures. In this specific case, the temperature of the water ice, and any other constituent of the perennial cap, is forced to remain at the buffered CO2 temperature (148 K) and is not deterministic of the composition of the minor ice phase. In this article, we do not map water ice that is exposed at the surface of the CO2 perennial cap; in all other cases, water ice is characterized by its temperature, even in the case of phase change. The diurnal temperature cycle of an ice, as well as its optical constants, are unique physical characteristics appropriate for unambiguous detection.
3.2.1. Apparent Temperature of Carbon Dioxide
 At a pressure of 7 millibars, the equilibrium temperature of CO2 with respect to phase change is 148 K. At such temperatures, THEMIS accuracy is ∼1 K [Christensen et al., 2004]. The atmospheric correction algorithm we used [Bandfield and Smith, 2003] introduces a few K of error depending on the state of the atmosphere at the time of acquisition. In addition, the perennial cap is punctured by circular erosive depressions (“Swiss Cheese”) whose floors are at water ice temperature (e.g., typically ∼25 K warmer than CO2 ice at Ls = 325°). As a result, subpixel mixing between carbon dioxide ice and warmer material occurs, and the apparent temperature of CO2 ice of the cap is 140 < < 158 K.
3.2.2. Apparent Temperature of Water Ice
Titus et al.  have shown that the sublimation of water ice near the south pole in the late summer does not absorb a significant fraction of the solar radiation (e.g., ∼0.2%) and does not significantly lower the temperature of the ice (e.g., ∼0.2 K). For this reason, the latent energy of sublimation of water ice is not considered and equation (1) becomes:
 Water ice has the highest thermal conductivity and inertia among geological materials observed at the surface of Mars [Titus et al., 2003, Table 2]. As a result, its diurnal temperature variation is characteristically small. Massive bedrock may have a somewhat similar value of thermal inertia, e.g., above 1200 I units [Kieffer et al., 1973] versus >2000 I units for H2O, occasionally resulting in similar temperatures, but the SPLD and the surrounding mantled terrains, composed of hundreds of meters to kilometers of dust and ice deposits [Murray et al., 1972; Soderblom et al., 1973; Cutts, 1973; Blasius et al., 1982; Milkovich and Head, 2005], are not expected to display bedrock exposures. We have determined, using TES data, that polar terrains with such high thermal inertias have a typical albedo of 0.25. This is the albedo we assign to water ice in the rest of this article and for the thermal modeling. At Ls = 335°, and using the parameters given in Table 2, we find that exposed water ice temperature is always close to 177 K and always significantly higher than the temperature of CO2 ice (Figure 1A). Earlier and later in the season, at Ls = 325° and Ls = 345°, water ice still remains >20 K warmer than CO2 ice (Figures 1B and 1C). To account for the latitudinal variations within the study area, we have calculated the temperature of exposed water ice at different latitudes and show two examples at 87°S and 80°S (Figures 1D and 1E). The shift of temperature due to the variation of latitude is ±2 K, with temperatures ranging from 176 K to 180 K at noon. We have also investigated the effect of different surface albedos because some variations may be expected as a function of the dust content of the ice. For surface albedo changes of up to 15% (e.g., 0.23 to 0.27), Figures 1F and 1G show that the effect on the surface temperature is very small (<1 K). Finally, we have computed the effect of sloping surfaces on the thermal diurnal cycle. Near the south pole and during the summer, the peak temperature of sloping surfaces occurs at a different local time as a function of the azimuth and slope. The thermal diurnal amplitude does not fundamentally change. Beyond 10° of slope, the thermal diurnal cycle gets significantly delayed or advanced and afternoon water ice temperatures are ∼175 K, which is close to the temperature of dust (Figure 1H). Therefore the surface material identification may not be accurate where slopes exceed 10°. Beyond 10° of slope, our mapping technique is not appropriate and a model taking the topography into account is required. These terrains correspond to the edges of the SPLD and the walls of the polar troughs, and together, they represent ∼1% of the total surface area between 70°S and the pole.
3.2.3. Apparent Temperature of Particulate Material
 Fine dust (I ∼ 50) is also expected at the surface of the SPLD [Titus et al., 2003]. SPLD dust albedo is 0.23 [Titus et al., 2003]. Dust does not conduct heat as efficiently as ice or sand [Presley and Christensen, 1997] and large surface diurnal thermal amplitudes are expected, with hotter temperatures than water ice near noon (189 K, Figure 1A) and cooler near midnight (168 K). We have calculated the temperature of dusty surfaces in a variety of conditions, between Ls = 325° and Ls = 345°, for latitudes ranging from 87°S and 80°S, with surface albedo of 0.21 to 0.25 and with sloping surfaces (Figure 1). Under this range of conditions, dusty surfaces in the afternoon are always significantly hotter than CO2 ice (by at least 15 K) and temperatures alone are sufficient to distinguish dust and CO2 ice.
 Near 7 am and 6 pm (for flat surfaces), the temperature of the dusty surfaces are similar to those of exposed water ice surfaces, typically near 177 K (Figure 1A). At these local times, thermal data are not appropriate to distinguish between dust and water ice. We have eliminated the portions of the THEMIS images near these local times in this study and only use the noon side of the images. Under these conditions, dust is typically hotter than 180 K which is significantly warmer than both water and CO2 ices.
 Because this work focuses on the water ice distribution, and because modeling indicates that the temperatures of sand and dust on one hand, and the temperature of H2O ice on the other hand, are always significantly different, we do not differentiate dusty and sandy surfaces in the remaining of the text and in the figures. We chose to name “dust” any bare surface, free of CO2 and H2O ices.
3.2.4. Layered Material
Titus et al.  have shown that the temperature of the particulated surface of an area of the SPLD is best modeled with a layer of dust lying on a water ice substrate. Data from the Mars Odyssey neutron spectrometers are consistent with large fractions of water ice in the near subsurface [Mitrofanov et al., 2002; Boynton et al., 2002; Feldman et al., 2004] and a thermal study by Bandfield and Feldman  indicates that water ice is only covered with a shallow layer of particulate material. For completeness, we have calculated the diurnal temperature cycle of layered surfaces (e.g., dust on water ice) for a variety of conditions. Dust on water ice results in thermal diurnal cycles that are significantly different from exposed water ice but similar to pure dust [Putzig and Mellon, 2007b]. Figure 1I shows the diurnal cycle of pure dust versus 7 mm of dust on water ice. Titus et al.  found that this is the configuration that best models the thermal seasonal cycle on the area they studied. Again, at 7–8 am and 5–7 pm, layered, dusty/sandy and water ice surfaces display similar temperatures, and data acquired near these local times have been avoided.
3.2.5. Subpixel Mixing
 In some cases, measured temperatures cannot be matched with modeled temperatures. We assume that subpixel mixing between CO2, water ice, or particulate material is responsible for this behavior. These pixels are typically located at the margin of the CO2 cap or at the periphery of troughs and depressions in the perennial cap, and are therefore consistent with subpixel mixing of CO2 ice and other hotter materials. In most cases, these pixels are also located near exposed water ice, suggesting that intermediate temperatures correspond to the subpixel mixing of CO2 and H2O ice with no dust involved.
 In theory, the right mixture of dust and CO2 ice can result in temperatures similar to that of H2O ice. However, mid-summer adjacent CO2 ice and dusty surfaces are not stable because of their large temperature differences: 2–3 days after a particulated surface has been freed from its cover of seasonal frost, its temperature has increased by 30–80 K (mostly depending on the slope orientation), making any stable close contact with CO2 ice impossible. We have only selected data acquired in the mid-summer, many days after the CROCUS date, in order to only observe seasonally stable surfaces, e.g., surfaces where the CO2 ice is stable and not retreating. As a consequence, CO2 ice and dust subpixel mixing, leading to a possible erroneous identification of water ice is unlikely. Thanks to its high thermal inertia resulting in small thermal gradients, H2O ice is form transition areas between the perennial CO2 cap and the dust, resulting in areas where dust, H2O and CO2 ice are present in a single THEMIS pixel. These pixels have been mapped as “mixed” materials. In addition, the south polar region is imaged 12 times a day, every day, allowing us to observe each location at different local times. This wealth of data was used for any given location to verify the consistency of our original mapping with data acquired at other local times, reducing the possibility of mapping errors. Erroneous identifications of exposed water ice cannot be repeated several times a day because subpixel mixing results in diurnal (and seasonal) thermal variations that are very different from thermal variations of pure H2O ice. Subpixel mixing of various materials can result in the same apparent temperature as exposed water ice only at specific local times, but not all day long. We take advantage of the uniqueness of the diurnal curves to prevent misidentifications of exposed water ice.
3.2.6. Mapping Example
 For each temperature measurement (i.e., THEMIS pixel), we have assigned a surface composition (i.e., CO2 ice, H2O ice and dust) based on the thermal modeling presented above. Four types of surfaces have been identified, i.e., CO2 ice, H2O ice, dust, and mixed surface materials whose temperatures cannot be assigned to a single material. An example of mapping for a single THEMIS image is provided on Figure 2. Thermal modeling of the surface materials indicates that 145 K < < 158 K, 166 K < < 178 K, and 178 K < Tdust. The histogram of this image displays these three domains, centered at 152 K, 172 K and 187 K (Figure 2). The domain between 158 K and 166 K corresponds to a mix of different surface materials, resulting in intermediate temperatures. Typically, these intermediate temperatures are measured on extremely small fractions of the images, e.g., 5% on Figure 2. Only 0.3% of the terrains between 87°S and 80°S display mixed materials temperatures (and none southward of 80°S). The lower part of Figure 2 represents the composition map generated from the thermal and histogram analysis. By extending this process to the latitudes between 87°S and 70°S, we have divided every image's histogram in four thermal domains, assigned a composition to each (i.e., CO2 ice, mixed material, H2O ice and dust), and constructed the map of the distribution of the surface material (Figure 3).
4. Results and Discussion
4.1. Water Ice Distribution
 A map of the surface material near the south pole based on THEMIS surface temperature measurements is presented on Figure 3. 40,000 km2 of exposed water ice are identified near the south pole. The exposed water ice is observed: (1) on the SPLD, at the margin of the perennial cap, forming a region of transition between the dusty surfaces and the CO2 ice of the perennial cap; and (2) forming a 300 km by 80 km patch centered at 345°E and 83.5°S on the southern high latitude mantled terrains. This patch represents the largest continuous exposure of water ice on Mars other than the north polar cap. Near 335°E, 83.5°S the exposure of H2O is fragmented in ∼10 patches covering ∼700 km2 each. The equator-most outcrop of water ice is visible in the floor of a shallow unnamed crater centered at 80.2°S and 5°E.
4.2. Correlation of Water Ice Occurrence and Surface Albedo
Titus et al.  first noticed a correlation between the nature of the surface material and the albedo. Bibring et al.  also observed this correlation for a larger surface area. The distribution of water and CO2 ices (Figure 3) closely matches Herkenhoff's  Unit Af of the Martian south polar region geologic map based on albedo and color images from Viking. Herkenhoff interpreted this Unit (Af) to be patches of partially defrosted terrains. We have compared the distribution of the ices and dust with regional albedo data and we confirm the correlation reported by Titus et al. . Figure 4 is an example of correlation between surface composition and albedo. The fact that water ice has a remarkably constant albedo indicates that it has a homogeneous crystal size and dust fraction. Douté et al.  found a typical H2O crystal size of 150 μm using OMEGA and Titus et al.  calculated that the typical south polar albedo of water ice may be due to a combination of 8 to 80 μm ice crystals and a dust fraction of 0.01% to 10%. In addition, the albedo boundaries observed on the THEMIS VIS and MOC images are similar to those recorded by Viking (resolution variable, typically near 90 to 900 m per pixel) and Mariner 9 (resolution of a few tens of meters per pixel, incomplete coverage) indicating that the distribution of water ice has not significantly changed since 1972.
4.3. Comparison With OMEGA
 In those regions of the south pole where OMEGA data were available, Bibring et al.  used spectroscopic methods to identify H2O ice. Where the coverage overlaps, our map is consistent with the H2O ice detection by OMEGA [Bibring et al., 2004]. A detailed comparison is limited by the difference of coverage (incomplete mapping by OMEGA between 87°S and 70°S and absence of coverage by THEMIS between 90°S and 87°S). Where both instruments have mapped the distribution of the ices, we note the following:
 1. The water ice exposures detected by OMEGA on the SPLD and the surrounding mantled terrains are also present on Figure 3 suggesting a good agreement between the 2 different mapping techniques.
 2. More small patches of water ice are mapped with thermal data on Figure 3 (in particular near 343°E 84.9°S). The non-detection of some outcrops of water ice can be due to the lower spatial resolution of OMEGA at the time of the mapping phase compared with THEMIS IR (1 OMEGA pixel corresponds to ∼400 THEMIS pixels) inducing some dilution of the signal. Alternatively, it is possible that these regions were covered with a very thin dust veneer. Figure 5 shows two of these regions, having the characteristic temperature of water ice but no spectral feature on OMEGA data. These patches have an albedo higher than the surrounding dust (Figure 3) consistent with that of water ice.
 3. None of the water ice on the CO2 perennial cap imaged by OMEGA is present on Figure 3. Water ice on the perennial cap is at CO2 ice temperature (e.g., ∼148 K) and is forced to stay at this temperature, without diurnal variations. In this case, the temperature of the water ice is not a function of its physical characteristics as defined in equations (2) and (3) (i.e., albedo, density, specific heat and thermal conductivity) but is maintained at the buffered temperature of CO2 ice.
4.4. Ground Patterns
Mellon  has shown that polygonal landforms should form at the high latitudes of Mars because of the seasonal thermal expansion and contraction of the water ice rich regolith. Mangold  has mapped the distribution of these polygonal features and Mangold et al. [2004a] have shown that polygons occur mainly on water ice rich terrains as detected by the Neutron Spectrometer onboard Mars Odyssey. Bibring et al.  observed the presence of polygonal features on exposed water ice. Mangold et al. [2004a], Mangold  and van Gasselt et al.  have previously mapped or classified the polygonal patterns based on their morphologies.
 We have mapped the distribution of the small-scale polygons (e.g., 50–200 m in size, Figure 6) near the south pole using MOC images and compared it with the occurrence of exposed water ice (Figure 7). We have observed a correlation between the occurrence of a specific type of polygons and exposed water ice. These polygons are outlined by ∼50 m to subpixel resolution (i.e., a few meters) wide sides, and do not show any apparent topography (Figure 6). These polygons are located near the margin of the exposed water ice, near the transition with dust covered surfaces, and are outlined by a material having the albedo of dust. The dark outlines of the polygons may correspond to troughs caused by the preferential sublimation of H2O ice along thermally stressed induced fractures [Marchant et al., 2002] and filled up with loose and mobile material (dust, sand, pebbles).
 Other types of polygonal features do not correlate with the exposure of water ice (Figure 7). Mellon's  model of formation of the high latitude polygons requires the regolith to contain some fraction of water ice, but does not require the water ice to be exposed. Therefore the lack of systematic correlation between the distribution of the polygonal ground patterns and the exposed water ice is not inconsistent with the theoretical model of formation. In addition, Piqueux and Christensen [2008b] have shown that a class of patterned ground (“etched polygons”) are formed (or at least widened) by the scouring power of the migrating CO2 gas that originates from the basal sublimation of the seasonal cap on the region where the carbon dioxide ice is translucent [Kieffer et al., 2006; Kieffer, 2007]. Since these ground patterns are associated with the seasonal CO2 ice, they are not expected to correlate with water ice, exposed or buried.
4.5. Timing Variability of the Condensation and Sublimation of the Seasonal CO2 Ice
 The CROCUS date is defined by Kieffer et al.  for a given location as the date when the last seasonal CO2 frost sublimes, unveiling the underlying surface material. The terrains near the polar cap are typically free of CO2 ice by Ls = 300° [Kieffer et al., 2000]. However, for the next ∼10°Ls, a layer of seasonal CO2 frost does remain on surfaces where water ice is exposed later in the summer (Figure 8). There is a clear correlation between the distribution of water ice and the CROCUS date, i.e., the date at which the last seasonal frost is removed. The time of the earliest frost deposit, and therefore the thickness of the CO2 deposit, may be a function of the substrate properties [Haberle et al., 2008]. The earlier the surface gets frosted, the more CO2 ice will accumulate during the fall. THEMIS thermal observations acquired a few Martian days before the beginning of the polar night (∼Ls = 0°), indicate that the exposed water ice is at or near CO2 condensation temperature whereas dusty surfaces are not (Figure 9). As a result, exposed water ice reaches the CO2 condensation temperature before dusty surfaces do, and water ice surfaces start to accumulate frost a few days earlier. Data acquired near Ls = 10° indicate that all surfaces are covered with CO2 frost at this date. After the sunset or at extremely low solar illumination angles, the advantage of THEMIS temperature measurements is crucial over other techniques (reflected solar light remote sensing) in order to identify and map the composition of the polar surface materials. We have determined that near the beginning of the polar night, exposed water ice is first covered with CO2 frost and should accumulate more frost than dusty surfaces. This is consistent with the observation of the delay of the CROCUS date over terrains where water ice is exposed during the summer.
James et al.  compared telescopic data of the retreat of the south seasonal cap acquired throughout the 20th century. They found some variability, possibly associated with dust storms, but the overall result of this study was that the retreat of the seasonal cap has followed the same longitudinal and latitudinal assymetrical pattern since 1903 (year of the oldest set of observations used by James et al.). Piqueux and Christensen [2008a] have also shown that the retreat of the cap has been almost identical in 1894 (observations by Lowell ) when compared to TES observations of the 1999 retreat. Because the timing and the shape of the retreating seasonal cap are strongly controlled by the occurrence of exposed water ice (among other parameters such as latitude, local slope, wind field, occurrence of dust storms, [Kieffer et al., 2000]), it is likely that the similarity of the observations since 1894 indicate that water ice has been exposed in the same locations since at least the end of the 19th century.
4.6. Water Ice Thickness
 Our thermal models assume that the water ice layer is at least several diurnal skin depths thick. The diurnal skin depth δ is defined by:
giving δ > 25 cm for H2O ice on Mars. If the water ice layer was several times thinner than δ, the surface temperature measured would be detectably warmer than the outputs of our thermal models. The agreement between the measurements and the models indicate that the layer of exposed water ice mapped on Figure 3 is approximately a skin depth thick or more.
 THEMIS VIS (18 m per pixel) and MOC (a few meters per pixel) images of the exposed water ice terrains indicate very uniform surface characteristics. The water ice patches show little surface roughness at these scales, in contrast with the adjacent dusty surface (Figures 4, 5, 6, 10; also see Mangold et al. [2004b]). However, at higher spatial resolution, HiRISE images indicate that water ice surfaces are locally shaped by very small-scale (meter) polygonal landforms. Figure 11 is a HiRISE view of the water ice floor of a “Swiss Cheese” feature. It shows a dense network of polygons outlined by CO2 seasonal frost remaining in the troughs. It is not known at this time if these features are common because such small formations are only accessible on HiRISE images. On Figure 5, the low albedo smooth unit (e.g., H2O) is punctured by a series of modest dark buttes, rising above the surrounding surface, suggesting that the water ice deposit is thin compared to the local surface topography (a few meters at most). This is consistent with the prediction of Montmessin et al.  that the southern exposed water ice is a meters-thick layer at most. We have generated a series of MOLA topographic profiles in numerous places at the contact between the exposed water ice and the surrounding dust and we have not detected a difference of elevation between the two units (Figure 12) indicating that the thickness of the water ice unit is at or below the detection limit of MOLA (e.g., a few meters).
 Depending on the location, the boundary between water ice and dust can be either gradual or sharp (Figure 10). Where the compositional boundary is gradual, sub-circular patches of dark albedo material surrounded by water ice are observed. This dark material has an albedo that is consistent with dust, possibly indicating that the water ice layer is thin and punctured in places, forming windows where the underlying dust-rich terrain is visible. Alternatively, these dusty patches could be depressions where particulate material can accumulate on the water ice. Available data are not sufficient to distinguish these hypotheses, and it will be important to determine the thickness of the water ice layer exposed at the surface because it can provide fundamental constraints on the recent past and present polar environment.
4.7. Water Ice Stability and Long Term Evolution
4.7.1. Sublimation Rates
 It is difficult to calculate the amount of water that sublimes every summer because the state of the atmosphere near the surface is not known. Smith [2002, 2004] observed that the polar spike of the TES derived water vapor concentration (30–40 μm) starts well before Ls = 300°, when the water ice is not yet exposed (Figure 8), indicating that the contribution of the exposed water ice to the atmospheric vapor content is minor. After Ls = 310°, when the H2O ice is exposed and sublimates, the vapor concentration above the cap has already significantly dropped (20–30 μm), further indicating that the exposed water ice is not a significant contributor to the atmospheric vapor, and that other sources are more important, e.g., the release of water trapped in the retreating southern CO2 seasonal cap [Pimentel et al., 1974] and the diffusion of vapor throughout the regolith [Jakosky and Farmer, 1982; Jakosky, 1983a]. In the late summer, near Ls = 325°, the exposed water ice surface temperature is maximal (T > 180 K) and the sublimation rate is also likely reaching its maximum value. At this time, the concentration of vapor in the atmosphere has dropped to ∼10 μm (or less) near the pole, which is a common southern hemisphere vapor concentration above surfaces where water ice sublimation may not contribute, further indicating that the exposed water ice is a minor source of atmospheric vapor.
 Assuming that most of the sublimation occurs during the 10 Martian days when the exposed water ice is the hottest, i.e., near Ls = 325° when T > 180 K (Figure 1), that this water vapor is evenly diluted between 80°S and 90°S and represents 10 precipitable μm as measured by TES [Smith, 2002, 2004], and that the water column is refilled every day [Jakosky and Farmer, 1982; Smith et al., 2002; Titus et al., 2003], 1 × 108 m3 or 1 × 1011 kg of water are lost every summer and injected in the atmosphere. For comparison, this is two orders of magnitude less water than the total amount in the atmosphere. Now let us assume that the atmospheric vapor only originates from the sublimation of surface water ice. The vertical erosion rate would be ∼200 μm per day or ∼2 mm per season. This is significantly higher than rates previously calculated for both polar caps (i.e., 100–800 μm per season by [Ingersoll, 1970; Toon et al., 1980; Haberle and Jakosky, 1990; 10 μm per day for Titus et al., 2003]) and not realistic since we have shown that the exposed water ice is not the main source of the southern atmospheric vapor.
4.7.2. Origin of the Water
Montmessin et al.  discuss in detail the possible origin of the water ice deposits observed by Titus et al.  and Bibring et al.  near the south pole of Mars. These authors favor a scenario where a meters-thick layer of H2O was deposited ∼21,500 years ago when the perihelion was synchronized with the northern summer. Under these conditions, the north H2O cap was unstable and released massive amounts of vapor in the atmosphere. Global atmospheric circulation modeling indicates that the vapor was transported to and deposited near the south pole at a rates as high as 1 precipitable mm per year, able to quickly form a thick cover of water ice. Since then, the perihelion has switched to the current values and the mantle of exposed water ice is no longer stable and sublimes. The original meters-thick deposits have thinned and retreated to the extent observed today.
4.7.3. Fate of the Water
 Once transferred to the atmosphere, water vapor may have several destinations. First, it can be cold-trapped on the south perennial cap. At 148 K, the partial pressure of water vapor in equilibrium with the frost is 10−7 mbar, several orders of magnitude less than the vapor content of the atmosphere, indicating that H2O will condense on the CO2 cap and stay until the cap is removed, during periods of high obliquity for example. The spectral signature of water ice on the south perennial cap has been identified by OMEGA [Bibring et al., 2004] and CRISM [Titus et al., 2008], confirming that the south perennial cap is presently a trap for water vapor.
 The regolith may act as a source of vapor during the spring, resulting in the spike of vapor in the atmosphere that was observed by TES [Smith, 2002, 2004] before Ls = 300°, when water ice is not yet exposed. In addition, the regolith may also act as a sink. The adsorption of H2O molecules is controlled by the temperature and the water vapor partial pressure in the pore [Jakosky, 1985]. After Ls = 315–325°, as the surface and subsurface temperatures decrease, water vapor can be adsorbed back by the regolith [Zent et al., 1986]. A fraction of the vapor originating from the exposed water ice may be adsorbed during the fall in the high latitude regolith.
 Finally, a fraction of the water vapor may remain in the atmosphere for longer periods of time and follow global winds. However, latitudinal vapor abundance during the southern spring indicates that little diffusion through the atmosphere occurs and that little H2O is transported to latitudes lower than 40°S [Smith, 2002, 2004]. Conflicting results have been published regarding the net flux of H2O of the south polar region. In the current Martian environment, there should be a net flux of water originating from the northern H2O perennial cap, to the south polar region, due to the existence of the austral perennial CO2 ice and the colder temperatures associated with it [Jakosky, 1983a, 1983b]. Over half an obliquity oscillation cycle (e.g., 5 × 104 years), 15 m of ice may be lost by the north polar cap and gained by the south perennial cap and the SPLD [Jakosky and Haberle, 1992]. However, the details of the incorporation of H2O ice in the SPLD is not known. At higher obliquity, the polar regions experience warmer summer seasons and perennial CO2 caps are not likely to exist [Toon et al., 1980]. Without an efficient cold trap at either of the poles, the amount of water vapor in the atmosphere rises significantly and is transported to the equatorial region and midlatitudes [Jakosky and Carr, 1985; Mellon and Jakosky, 1995; Mellon et al., 1997]. At low obliquity, both poles are cooler than at present and accumulate water by cold-trapping atmospheric vapor. More recent general circulation models [Richardson and Wilson, 2002] suggest that for most orbital configurations, the south cap is experiencing a net loss of H2O in favor of the north cap. Fenton and Richardson  and Richardson and Wilson  propose that the topographic dichotomy between the north and south hemispheres is responsible for a assymetrical Hadley circulation depleting the southern H2O reservoir.
 Without re-deposition, the lifetime of the exposed water ice that is not on the CO2 perennial cap is limited to short geologic periods because the sublimation rate likely corresponds to one of the highest erosive rates presently active on Mars. After Ls = 350°, the polar terrains are only a few K above the temperature of CO2 frost and a fraction of the atmospheric water vapor will condense on the surface. Because exposed water ice surfaces are cooler than particulate surfaces at any time of the day before the polar night, H2O vapor condenses preferentially on exposed water ice. The preferential condensation of vapor on exposed water ice at the end of the summer may compensate for part of the summer loss by sublimation, but it is not possible to quantify this flux because the vapor content of the atmosphere and other critical parameters are not known near the pole during the fall (see vapor retrieval technique limitations in the study by Smith [2002, 2004]).
 The presence of water ice adjacent to the south perennial cap is consistent with the hypothesis of the condensation of the atmospheric vapor originating from the north cap and high latitudes. The exposed water ice observed hundreds of kilometers away from the perennial CO2 is not stable, and may be the relic of a thicker and wider H2O ice sheet that formed 21,500 years ago as modeled by Montmessin et al. . It could also represent the relict of a former stable layer underlying portions of the perennial CO2 cap [Titus et al., 2003] that have been removed during a recent geologic time (e.g., centuries to millennia).
 We have mapped the distribution of carbon dioxide and exposed water ices near the south pole of Mars during the summer, when all of the seasonal CO2 frost is removed, using THEMIS temperature measurements (Figure 3). Thermal mapping provides results in agreement with near infrared spectral data, and demonstrates that not only is water ice present, but that it must be the dominant phase in order to control the temperature.
 1. Exposed water ice covers a surface area of ∼40,000 km2, mostly at the periphery of the perennial CO2 cap, and on the mantled terrains adjacent to the SPLD. Most of the exposed water ice is not observed on the SPLD, which is the largest body of water ice on Mars, but at lower elevation, near 345°E and 83.5°S, where a 300 by 80 km continuous patch is entirely mapped for the first time.
 2. Water ice on the CO2 perennial cap is buffered at ∼148 K and cannot be mapped with thermal data. OMEGA and CRISM observations are most suited for mapping H2O ice on the CO2 perennial cap.
 3. At the meter to tens of meters scale, exposed water ice terrains are very uniform in term of roughness, albedo and surface features. The surface composition correlates extremely well with the albedo, which is remarkably homogeneous in the case of the exposed water ice (Figure 4) and has been calculated to be resulting from 8–80 μm ice crystals with 0.01 to 10% dust [Titus et al., 2003] or 150 μm crystals with 30–35% dust [Douté et al., 2007]. This suggests that the size of the H2O crystals and the water-to-dust ratio are similar throughout the polar region.
 4. When comparing the map of the distribution of the surface material (Figure 3) with Mariner 9 and Viking images, the present compositional and past albedo boundaries correlate extremely well. Assuming that the albedo/composition correlation observed today was also true in the recent past, we deduce that the distribution of the exposed water ice has been identical during the last four decades at the kilometer scale.
 5. Because the retreat of the southern seasonal cap has followed the same pattern during the 20th century, and because it is strongly controlled by the occurrence of exposed water ice, it is likely that the distribution of exposed water ice has remained the same since at least the end of the 19th century.
 6. Water ice surfaces accumulate more seasonal CO2 frost during the fall and winter because they reach the carbon dioxide condensation temperature a few Martian days earlier (∼10°Ls) than dusty surfaces. As a result, they are defrosted later in the summer, confirming that the CROCUS date is also partially controlled by the composition of the surface material (Figure 9).
 7. The amount of vapor injected in the atmosphere by the sublimation of 40,000 km2 of ice is not sufficient to account for the concentrations of vapor observed by TES during the southern summer. This confirms that other sources (e.g., regolith vapor diffusion, release of water trapped in the seasonal cap) are more important processes for the transfer of volatiles.
 8. Exposed water ice is not stable in the current environment and sublimes at vertical rates ranging from a few tens to hundreds of microns per season. As a result, the lifetime of surface water ice is short, indicating that either re-deposition occurs (preferential condensation of atmospheric vapor on the cold exposed water ice surfaces in the late summer), that ice has been insulated from the atmosphere and uncovered recently (indicating the past location of the CO2 residual cap), or that we are observing the remains of a larger water ice body deposited under other climate conditions and soon to be removed. Part of the current exposed water ice is also inherited from past and present cold-trapping of water vapor at the very edges and on the CO2 perennial cap.
 9. A specific morphology of polygonal landforms (<200 m in size, dark outlines, with width ranging from MOC sub-pixel resolution to ∼50 m, Figure 6) seems to be exclusively associated with exposed water ice (Figure 7). However, they only represent a small fraction of the ground patterns identified and mapped indicating that there is no general correlation between the distributions of ground patterns and exposed water ice (Figure 7).
 We would like to thank T. Titus for his comments on a preliminary version of this article, N. Bridges and N. Mangold for their helpful reviews. This work was supported by the NASA 2001 Mars Odyssey THEMIS Project.