The Tharsis Rise is a large volcanic province in the tropics of Mars1 (latitude range: ±40° N, longitude range: 220–300° E). It is a broad topographic dome that rises about 5 km above the surrounding terrain and covers a region 5,000 km wide2. It contains some of the Solar System’s largest and tallest volcanoes3, such as Olympus Mons (21 km altitude), Arsia Mons (18 km), Ascraeus Mons (18 km) and Pavonis Mons (14 km), but also smaller shield volcanoes such as Ceraunius Tholus (9 km). Volcanic activity on Mars has been concentrated predominantly in this region throughout the planet’s geological history, persisting into current times, as evidenced by lava flows that are as recent as 2.4 million years old4. No current volcanic activity has been detected in Tharsis, although recent geophysical data show that Mars is still geodynamically active5,6,7.

Notable orographic water ice clouds and other atmospheric phenomena have been observed in Tharsis8,9,10,11,12. Water ice clouds play a fundamental role in cycling water on Mars, moving moisture for thousands of kilometres from polar regions to relatively dry equatorial areas13,14. In addition, Tharsis is situated along the route of an important cross-equatorial exchange of water vapour, where approximately 1012 kg of water is annually transferred between the northern and southern hemispheres through the solstitial Hadley cells15. Atmospheric observations16 have revealed a localized enrichment in water vapour above the Tharsis volcanoes, suggesting that an active exchange of water vapour between the regolith and the atmosphere may be ongoing, probably facilitated by desorption from the regolith and/or sublimation of frost. A subsequent study17 confirmed the water-vapour enrichment over these areas but hypothesized that the local circulation pattern typical of the volcanic region is possibly responsible for the enrichment as it may carry considerable amounts of water vapour upslope.

Apart from the polar regions, water ice manifests on the surface as seasonal frost in mid- and low-latitude locations. NASA’s (the National Aeronautics and Space Administration’s) Viking 2 lander detected water frost at ~48° N18,19,20. In addition, orbital observations from a variety of instruments revealed that water frost can occur up to 13° S in the southern hemisphere and as low as 32° N on shaded pole-facing slopes21,22,23. However, the presence of frost at the tropics (~0° N latitude) was not expected because of higher average surface temperatures24 and lower humidity25. Some studies predicted that on most of Mars’s surface, small amounts of H2O frost can condense nightly if radiative cooling is strong enough26,27. For example, extremely small amounts of water frost have been observed to condense near the equator on the high thermal emissivity calibration targets of NASA’s Opportunity rover28,29.

Most of the Martian atmosphere is composed of CO2 gas, and therefore CO2 frost can also form if surface temperatures are low enough30. On the basis of nightly surface temperatures and thermal modelling, it was shown that in the equatorial regions CO2 frost may condense diurnally29,30,31,32. Predictions30,31 indicated that putative CO2 frost deposits may persist for only a few minutes after sunrise (~6:00 Local Solar Time (lst)) before sublimating back into the atmosphere. Follow-up global surveys, utilizing early-morning colour observations from the Thermal Emission Imaging System (THEMIS33) were conducted to search for these frosts in the equatorial regions, but no evidence of morning CO2 frost was identified34.

Observations by the Colour and Stereo Surface Imaging System (CaSSIS35) on board the European Space Agency’s (ESA’s) Trace Gas Orbiter (TGO) provide strong evidence for morning frost deposition on the equatorial Tharsis volcanoes. We present here these observations coupled with supporting evidence from other instruments and modelling.

Observations of frost

Early-morning images (lst = 7:11; latitude = 18.5° N, longitude = –133.5° E; spatial resolution = 4.5 m pixel–1) of Olympus Mons caldera acquired by CaSSIS (at utc 2022 November 25) in the late northern winter (solar longitude (Ls) ~345°) on Mars year (MY) 36 first revealed bluish deposits (at ~500 nm) on sections of the caldera floor and rim (Fig. 1). The CaSSIS observation suggests a spatial correlation between the bluish deposits and topography (Fig. 1d). The deposits are concentrated on the caldera floor but are absent on well-illuminated warm slopes and farther north on the volcano flank. The finding was confirmed five days later with a High Resolution Stereo Camera (HRSC)36 observation acquired on 2022 November 30 (lst = 7:20; latitude = 18.2° N, longitude = −133.2° E; spatial resolution = 800 m pixel–1), which revealed that the diffuse bluish ‘halo’ deposit was ubiquitous on the entire caldera floor and rim (Fig. 1b,c). The halo is absent on the volcano flanks and is concentrated only at the mountain summit. During the CaSSIS detection, the Nadir and Occultation for Mars Discovery (NOMAD37) spectrometer was operating and acquired a ride-along observation (instantaneous field of view = 17.5 km × 0.5 km). The nadir spectral data acquired in the NOMAD limb nadir and solar occultation (LNO) channel revealed that the deposit is frost (Fig. 1e) as indicated by the elevated ice index values (more than 3σ confidence; Methods and Supplementary Figs. 1 and 2).

Fig. 1: Frost detection on the Olympus Mons caldera by CaSSIS, HRSC and NOMAD.
figure 1

a, Global view of Mars with white box marking the location of Olympus Mons. b, HRSC wide-angle image of Olympus Mons acquired in the early morning (lst = 7:20, Ls = 346.7°, latitude = 18.2° N, longitude = −133.2° E). The black dashed line indicates the orbit of the TGO corresponding to the images in d and e. The white box highlights the close up in c. c, Zoomed-in view of the Olympus Mons caldera. The white and blue dashed rectangles show the footprints of the CaSSIS and NOMAD-LNO observations, respectively. d, High-resolution (4.5 m pixel–1) CaSSIS colour image of frost on the caldera floor and northern rim of Olympus Mons (lst = 7:11, Ls = 344.1°). Frost is absent on the well-lit steep slopes. The blue rectangle marks the footprint of the one NOMAD-LNO observation that falls within the frost-covered area. e, NOMAD-LNO channel observation of the Olympus Mons caldera. The ice index values (Methods) indicate the presence of frost over the caldera floor (>µ + 3σ). The coloured areas on the plot indicate the confidence intervals. HRSC image ID: hn889_0000 (b,c). CaSSIS colour image ID: MY36_022332_162_0_NPB (d). NOMAD-LNO observation ID: 20221125_082524 (e). Credit: b, ESA/DLR/FU Berlin; d, ESA/TGO/CaSSIS under a Creative Commons license CC-BY-SA 3.0 IGO.

Repeat imaging by HRSC shows that the frost deposits on top of Olympus Mons (Fig. 1b) appear only in the early Martian morning (lst = ~7:00–7:30; latitude = 18.2° N, longitude = −133.2° E) and are spatially correlated with a geological bright halo unit (Extended Data Fig. 1a,d). This unit may be dust that is relatively brighter than the surrounding material due to different grain size or texture38. This bright halo unit is also observed in Context Camera39 images (Extended Data Fig. 2a). Materials consisting of smaller particles may exhibit different thermophysical properties such as lower thermal conductivity40 and high thermal emissivity41. Surfaces with such properties cool down more at night and warm up more slowly in the morning, further enhancing the likelihood and duration of frost formation. This latter point is illustrated by CaSSIS observations of frost on dust deposits that have not been removed by winds (Extended Data Fig. 2b,c). As shown by CaSSIS, frost may also condense leeward of small craters where air-fall dust can accumulate and is perhaps less compact (Extended Data Fig. 2d). Porous and less-compact materials provide more nucleation sites for frost formation42. Outside of the bright halo, frost is found near the northern rim of Olympus Mons, but its emplacement is more localized (Extended Data Fig. 2e–i). In conclusion, the observed frost patterns on Olympus Mons, particularly in areas with geologically distinct bright dust deposits, underscore the importance of thermophysical properties such as low thermal conductivity and high thermal emissivity, as well as surface texture, in governing the formation, distribution and persistence of frost on Mars.

Within the CaSSIS database, 13 instances of frost have been found (Extended Data Fig. 3). These include detections not only on the largest Tharsis volcanoes of Olympus, Ascraeus and Arsia Montes but also on the smaller-sized Ceraunius Tholus shield volcano (Extended Data Fig. 4). In one case, the frost deposits on Arsia Mons (Fig. 2a) are observed in the early Martian morning and during southern winter solstice (lst = ~8:00, Ls = ~90°, latitude = −8.7° N, longitude = −121.1° E). The frost line dividing warm and shadowed slopes is, however, not observed in the repeat CaSSIS observations of this location, which were acquired during late southern spring (Fig. 2b–d). Photometric analysis shows that frost is associated with an increase in ratioed reflectance of up to 20% at wavelengths (Fig. 2e; CaSSIS blue (BLU) filter bandwidth is 390–570 nm (ref. 43)). The fact that frosty surfaces are sometimes brighter only at blue wavelengths, implying a lower spectral slope, can also be observed in a linearly stretched CaSSIS BLU filter image (Extended Data Fig. 5 and Methods) and average spectra from a k-means clustering analysis44,45 applied on a topographically corrected and photometrically normalized CaSSIS cube (Supplementary Fig. 3). The photometric and clustering analyses suggest that the frost deposits are probably very thin.

Fig. 2: Frost occurrence on a crater in the Arsia Mons caldera.
figure 2

a, Frost on the shadowed slope of the crater in an early-morning observation during southern winter in MY 35 (latitude = −8.74° N, longitude = −121.14° E). bd, No frost in an early-morning observation (b) and no frost in afternoon observations (c,d) during late southern spring in MY 36. The spectral profile along the black line in a is shown in e and reveals a marked increase in reflectance up to 20% in the BLU filter when frost is present. Errors are from the uncertainty in the absolute calibration of the instrument and are about ~3% (ref. 43). The illumination direction is indicated by the arrows in the bottom right corner of each image. North is up in all panels. The CaSSIS image IDs are shown in order (ad): MY35_008465_192_0_NPB, MY36_020297_350_3_NPB, MY36_020366_190_1_NPB and MY36_020478_190_3_NPB. Credit: a, ESA/TGO/CaSSIS under a Creative Commons license CC-BY-SA 3.0 IGO.

CaSSIS observations of Olympus and Arsia Montes indicate diurnal and possibly seasonal trends in frost deposition (Fig. 3). The four detections in Olympus Mons (Fig. 3a,b) are clustered around the early-morning hours (lst= ~7:00–7:30) and northern spring equinox (Ls = ~320–40°). Similarly, the four detections in Arsia Mons (Fig. 3c–d) fall within a slightly wider time range (lst = ~ 7:00–8:30) but around the southern winter solstice (Ls = ~45–145°). The early-morning non-detections in Arsia Mons fall within the southern summer period, which suggests seasonality (Extended Data Fig. 6b), but the lack of early-morning observations in the northern summer precludes us from making the same conclusion for the detections in Olympus Mons (Extended Data Fig. 6a). We removed observations at extremely high solar incidence angles (>85°) because of low image signal-to-noise ratio (SNR), and therefore there is an observational bias towards lsts at about 6:00 (Methods). Collectively, CaSSIS observations suggest that the frost cycle over Martian volcanoes is ephemeral and exhibits variability on multiple timescales. It appears to be influenced by diurnal patterns, probably reflecting daily temperature fluctuations. In addition, there is a probable control by the Martian seasons, indicating a longer-term variation in the frost cycle. On the basis of the CaSSIS observations, while there are indications of diurnal and seasonal influences on frost deposition on Martian volcanoes, these observations alone cannot definitively determine the composition of the frost. Therefore, we use simulations of surface temperatures as a proxy for frost composition.

Fig. 3: Seasonal and diurnal trends of frost on Olympus and Arsia Montes.
figure 3

ad, Rose diagrams showing the seasonal (a,c) and diurnal (b,d) frost detections by CaSSIS over Olympus Mons (a,b) and Arsia Mons (c,d). The width of each bin reflects the number of CaSSIS observations. Frost is detected around northern spring equinox (Ls = ~0°) on Olympus Mons and around southern winter solstice (Ls = ~90°) on Arsia Mons. Frost is detected only in the early-morning hours (~7:00–8:00 lst). The negative detections in the early morning bins correspond to observations that were acquired in warmer seasons.

Surface temperatures indicate water frost

At the time of CaSSIS frost detections in Olympus Mons (Fig. 1) and Arsia Mons (Fig. 2), the surface temperatures calculated by the general circulation model (GCM46) via the Mars Weather Research and Forecasting (WRF47) model are inconsistent with CO2 frost. The stability of CO2 frost at higher altitudes necessitates exceptionally low temperatures, specifically below 140 K, to maintain its solid state30. The predicted surface temperatures (at ~150 km model resolution) are ~150 K and 185 K at ~7:00 lst in Olympus Mons and at ~8:00 lst in Arsia Mons, respectively (Fig. 4). In addition, advanced high-resolution mesoscale modelling, with a model resolution of 5.47 km, reveals a substantial temperature difference between the surface temperature and the local CO2 frost point at the locations of CaSSIS observations, with a difference of approximately 10 K at Olympus Mons (Fig. 5d) and over 55 K at Arsia Mons (Extended Data Fig. 7d). In fact, the surface temperatures predicted at each CaSSIS frost location (Table 1) consistently exceed the CO2 frost point, corresponding to the mean surface temperature ~162 K (excluding C3 and C6). The stratification of water vapour in Mars’s atmosphere, especially near the surface, is not well understood48, making the determination of the H2O frost point challenging due to its considerable variability; however, it is generally accepted that this point occurs at around 180 K (ref. 14). Since the predicted surface temperatures at the time of CaSSIS, HRSC and NOMAD observations are too warm, this suggests that CO2 frost is unlikely, hence providing support for the presence of water frost. At these seasons (Ls = 346.7° for Olympus Mons and Ls = 93.8° for Arsia Mons), CO2 frost was also not observed by the Thermal Emission Imaging System29 or by the Emirates Mars InfraRed Spectrometer32. Interestingly, the GCM also predicts that some CO2 frost may be present at Ls = ~0–150° and at around sunrise (5:00–6:00 lst) in Arsia Mons (Fig. 4b). This result is consistent with previous studies indicating CO2 frost formation from minutes to tens of minutes after sunrise in the equatorial regions29,30,31. However, such potential CO2 frost deposits would sublime very quickly and would be difficult to detect by cameras and spectrometers due to low SNR34. In addition, we investigated the possible role of CO2 frost in regolith gardening and slope streak formation on Mars30,34,49. We found no slope streaks on the calderas of the largest Tharsis volcanoes or any obvious differences in talus boulder shapes and sizes (Methods and Extended Data Fig. 8). These results suggest that the diurnal CO2 or H2O frost cycle plays a minor (if any) role in landscape evolution at these sites.

Fig. 4: Modelled surface temperatures at Olympus Mons and Arsia Mons.
figure 4

a,b, Annual surface temperatures at four different local mean solar times (LMST). c,d, Diurnal surface temperatures at Olympus Mons (Ls = 350°) (c) and Arsia Mons (Ls = 90°) (d) as predicted by the GCM. In c,d, blue and red horizontal dashed lines depict CO230 and H2O frost point14, respectively. On both volcano calderas at the time of CaSSIS image acquisition, CO2 frost point is not reached. This indicates favourable conditions for H2O ice. The simulations were conducted at geographical coordinates 18.75° N, −133.75° E for Olympus Mons and −8.75° N, −121.25 °E for Arsia Mons.

Fig. 5: Microclimatic conditions simulated over Olympus Mons.
figure 5

ad, Utilizing MarsWRF high-resolution mesoscale modelling (at the time of CaSSIS observation Fig. 1d), this figure presents the influence of Olympus Mons’s topography on its local climate, as shown by elevation gradients (a), surface atmospheric pressure (b), near-surface horizontal wind patterns (c) and the deviation in temperature between the Martian surface and the local CO2 frost point (d). The topography of the Olympus Mons caldera is demonstrated to cause noticeable variations in local pressure, wind velocities and temperature gradients. The black outline across all panels highlights the boundary of the Olympus Mons caldera, while the black dashed rectangle marks the area observed by CaSSIS, as referenced in Fig. 1d. The CO2 frost point in the area of CaSSIS observation is exceeded by about 10 K. By contrast, the CO2 frost point in Arsia Mons is exceeded by around 60 K (Extended Data Fig. 7d).

Table 1 CaSSIS frost detection times and predicted surface temperatures by the GCM

Microclimate and water ice amount

Our high-resolution mesoscale simulations reveal the distinct microclimatic conditions induced by the topography of the Tharsis volcanoes, as shown in Fig. 5 and Extended Data Fig. 7. Specifically, within the calderas of Olympus Mons and Arsia Mons, we observe a substantial reduction in surface atmospheric pressure and near-surface horizontal wind speeds compared with the surrounding areas. For example, within the caldera of Olympus Mons (Fig. 5b), the atmospheric pressure is estimated at only 110 Pa, compared with 160 Pa at the mountain’s base. Similarly, in the area of Arsia Mons (Extended Data Fig. 7b), the pressure is about 100 Pa, notably lower than the over 200 Pa found in the adjacent plains. Moreover, the near-surface horizontal wind speeds within Olympus Mons (Fig. 5c) are estimated at less than 10 m s–1, in stark contrast to the approximately 30 m s–1 observed along the volcano’s flanks. In the case of Arsia Mons (Extended Data Fig. 7c), the wind speeds are below 5 m s–1 within the caldera, compared with roughly 20 m s–1 on the flanks, highlighting the profound impact of volcanic topography on localized weather patterns.

Furthermore, our GCM simulations suggest that the thickness of water frost deposits is on the order of 1 µm (Methods). However, this estimate carries considerable uncertainty due to the unknown quantities of water-vapour-column abundances. To refine this estimate, we reference radiative transfer calculations50,51, which suggest a minimum thickness of 100 µm, while laboratory experiments52 imply a thickness of about 10 µm (Methods). By adopting the median thickness of 10 µm for the water frost, and considering that the frost deposits are confined to the calderas of Olympus, Arsia, Ascraeus Montes and Ceraunius Tholus, we estimate that there is a transfer of approximately 1.5 × 108 kg of water ice between the surface and the atmosphere (Methods).

Possible sources of water vapour

The seasonal trends as shown by the set (n = 13) of CaSSIS observations suggests an atmospheric phenomenon driven by water transport due to large-scale seasonal changes, such as sublimation of the seasonal ice cap in the opposite hemisphere and transportation of humid air into the volcano calderas by upslope winds. Seasonal processes have been observed at a wide range of Martian latitudes53 and may also apply to the Tharsis region. For example, the activity of the Aphelion Cloud Belt peaks at Ls 40°–140° (ref. 54), and in general little cloud activity is observed at Ls ~245°–320°10. Similarly, afternoon orographic clouds have been detected by Mars Reconnaissance Orbiter’s Mars Color Imager55 over the Tharsis volcanoes10. The seasonal observation of water-vapour enrichment over Tharsis17 shows increased abundances around northern spring equinox (Ls 0°), consistent with the CaSSIS detections of frost close to this season in Olympus Mons. Therefore, we hypothesize that this water-vapour enrichment17 may be the source of the frost deposits detected in our study. The transport of water vapour from high latitudes to the Tharsis highlands could be facilitated by large-scale atmospheric eddies56. This process could be further augmented by strong upslope winds, driven by a combination of thermal effects and mountain gravity waves57, facilitating the movement of moisture over the volcano calderas. The local topography-induced circulation57 and microclimatic conditions within the caldera (shown in Fig. 5 and Extended Data Fig. 7) may create favourable conditions for water frost condensation during the cold Martian nights. Within these calderas, ~150,000 tons of water ice is exchanged daily between the regolith and the atmosphere during the cold Martian seasons. Although this amount is relatively a small fraction of the seasonal inventory of water vapour in the Martian atmosphere (~1012 kg) (ref. 14), it is important in the context of localized Martian environmental processes. Understanding these micro-environments is crucial for a comprehensive understanding of Mars’s hydrological cycle.

It is conceivable that dormant volcanoes can emit CO2, water vapour and minor amounts of SO2 (ref. 58) via diffuse outgassing from the regolith59,60. If the observed water frost deposits are of volcanic origin, their distribution may constrain models for present-day outgassing from the interior. However, on Mars, SO2 has not been detected61 and no thermal hotspots have been found62. A volcanic source for the condensate cannot completely be ruled out, but further tests for trace species (CO2, H2S and SO2) would be useful to explore the likelihood of this potential mechanism. Consequently, we conclude here that the newly detected frosts on Tharsis volcano calderas are probably of atmospheric origin.


CaSSIS frost observations

We surveyed ~4,200 CaSSIS images (acquired up to 2022 February 05) with illumination geometries of 50–90° incidence within dusty, low thermal inertia (<100 TIU) regions (60° N—30° S). Only images that include the latest CaSSIS radiometric and absolute calibration were used in this study43,63,64.

The images used in this study consisted of early (6:00–9:00 lst) and late (15:00–18:00 lst) times. Analysis and comparison in these two local time regimes may help the distinction between early-morning and late-afternoon phenomena. During the survey, it was noticed that most CaSSIS images acquired at extremely high solar incidence angles of 85–90° contain colour and calibration artefacts due to the decrease in SNR and/or an increase in aerosol contribution from the atmosphere63. Consequently, the images with colour artefacts were labelled as ambiguous and were not used for further analysis.

Frost detections relied on the use of CaSSIS NPB (near infrared (NIR) = 940, panchromatic (PAN) = 670, blue (BLU) = 497 nm) and synthetic RGB (red–green–blue; PAN and BLU only) products. These filter configurations allow a convenient separation between frosty and frost-free terrains. In CaSSIS colour products, frosty areas appear bluish, and/or whitish, and sometimes are bright only in the BLU filter (relative to frost-free areas; also see Supplementary Figs. 9 and 10). In support, we observe bluish frost deposits in HRSC colour images shown in Fig. 1b and Extended Data Fig. 1 (composites of blue (440 nm), green (530 nm) and red (750 nm) channels).

As shown by previous studies21,65 deposits are usually correlated with topography (prefer poleward-sloping terrains). Therefore, if both conditions were met (colour and topographic correlation), it was considered a strong indication of surface frost. As a final procedure, each of these candidate detections was then analysed using a spectral profile tool in the Environment for Visualizing Images software. This procedure extracts the pixel irradiance over flux (I/F) values between two manually selected points crossing the potentially frosted region in each filter. The profiles were then normalized by a mean I/F of a nearby frost-free, relatively flat region of interest (ROI), a well-established method to cancel out some of the atmospheric and topographic effects49,66,67,68,69,70. If the frost deposits were brighter in the BLU filter than the surrounding frost-free terrains by at least 3% (within CaSSIS absolute uncertainty43), then such images were flagged as potential frost detections. This survey yielded many frosty sites (not shown here) at latitudes ~40° N and ~30° S. However, because these latitude bands are dominated by known seasonal frost deposits21,23,65 and we do not have a robust method to distinguish between seasonal and diurnal frost, we further narrowed our filtering criteria. The final frost detections analysed here were restricted to equatorial ~20° N to ~10° S latitudes (outside of the seasonal mid-latitude regions). In this work, only equatorial sites that included visible evidence of frost are considered.

The spectral profile shown in Fig. 2e was computed by dividing each pixel along the profile by an average pixel value extracted from an ROI in Extended Data Fig. 5d. The ROI (>100 pixels in size) was selected on a frost-free and relatively flat terrain as suggested by the low slope values in the CaSSIS digital elevation model of this site. CaSSIS digital elevation models were produced by a pipeline developed at the Astronomical Observatory of Padova, National Institute for Astrophysics71,72.

NOMAD-LNO spectral processing

The NOMAD instrument is a suite of three high-resolution spectrometers also on board TGO, offering nadir infrared observations through its LNO channel37,73. This channel covers the 2.2–3.8 µm spectral range where several spectral features of ice are distributed over different wavelengths. Nevertheless, the NOMAD-LNO spectrometer has the particularity of not observing the entire spectral range at once. The data are acquired through small spectral windows, representing specific diffraction orders of the diffraction grating. Each LNO observation can select a maximum number of 6 diffraction orders every 15 seconds to ensure the best possible SNR74,75,76. The LNO footprint (instantaneous field of view) is 17.5 km × 0.5 km (ref. 75), which provides enough spatial scale to resolve the caldera of Olympus Mons. In this work, we use spectrally and radiometrically calibrated LNO data converted into a reflectance factor. The 2.7 µm ice band is the strongest in the LNO spectral range, resulting from both CO2 and H2O ice absorption. Although the use of this band is not suitable for quantifying the amount of ice (easily saturated), it is effective for detecting homogeneous deposits (both CO2 and H2O ice), as demonstrated with the ice index value77. This spectral parameter uses two diffraction orders. It is based on the combination of high reflectivity at continuum wavelengths with a more pronounced absorption in the 2.7 μm band. Initially defined as the spectral ratio between the reflectance factors of order 190 (continuum part, 2.32–2.34 µm) and 169 (short wavelengths shoulder of the 2.7 µm band, 2.61–2.63 µm) (ref. 77), we adjust the ice index by considering the available orders of the joint CaSSIS–NOMAD observations, that is, orders 190 and 168 (2.64–2.65 µm).

In nadir mode, the variability in the reflectance factors is caused mainly by the surface albedo variations resulting from the different absorption of the Martian surface mineralogy78,79,80. To remove spatial albedo variations over the explored Martian surface, we normalize the LNO reflectance factors to the Martian albedo. The adjusted ice index (II) can thus be defined as:


where Ri is the LNO reflectance factor value averaged around the central wavelength i of the LNO spectrum, fitted by a third-degree polynomial to mitigate the spectral oscillations resulting from the instrumental characteristics of the LNO channel, which become significant on the edges of each order (Supplementary Figs. 1 and 2). OMEGAi is the OMEGA albedo map80 based on reflectance spectra in the near infrared as NOMAD-LNO Ri. Two OMEGA albedo maps are used in this work: one defined at 2.32 µm for order 190 and the other defined at 2.62 µm for order 168. Studies have shown that this spectral parameter identifies spatially extensive and abundant ice deposits when the index values are three sigma higher than their average value over ice-free mid-latitude terrain77,81.

Mars GCM modelling

We perform the Martian GCM simulation for the entire MY 36 using the MarsWRF model, which is the Mars adaptation of the general-purpose planetary atmosphere model, planetWRF47. Here the GCM set-up is based on a previous study46 examining the Martian planetary boundary and dust–turbulence interaction over a decade, from MY 24 to MY 34, which hosted three global dust storms. The reference model set-up46 was validated against NASA’s Mars Climate Sounder (MCS) observations on board the Mars Reconnaissance Orbiter, radio occultation observations from ESA’s Mars Express orbiter, as well as the in situ observations from NASA’s Mars Science Laboratory Curiosity rover. This model set-up consists of a semi-interactive two-moment dust transport model46 within the MarsWRF framework, in a way that the dust is lifted, mixed by model winds and sedimented, as guided by observed maps of column-integrated dust optical thickness82,83. Via this method, model processes govern the vertical dust distribution and related dust radiative heating, yet the horizontal dust distribution is guided to match the orbiter observations. In this model, the horizontal dust distribution is constrained to follow observations. In this model, the two-stream correlated k-distribution scheme is used for the short-wave and long-wave radiative transfer84. We use a Mars-specific boundary-layer turbulence parameterization scheme, which allows us to obtain the surface–atmosphere exchange coefficients85 Surface properties of the MarsWRF model, such as the topography, albedo, emissivity and thermal inertia, are acquired from the datasets of the Mars Orbiter Laser Altimeter2 and Thermal Emission Spectrometer (TES78) observations, where the details are presented in another study47. Here we increased the horizontal model grid spacing of the GCM from 5° × 5° to 2.5° × 2.5° (ref. 46), enabling better spatial coverage to provide more realistic boundary and initial conditions to our mesoscale simulations. We used 52 vertical sigma layers extending up to the model top of 100 km. The predicted surface temperatures are shown in Table 1.

Our modelling methodology is based on a previous study by MarsWRF85,86. Mesoscale simulations for Fig. 5 and Extended Data Fig. 7 were forced with initial and boundary conditions acquired by GCM simulations corresponding to the same seasonal conditions of CaSSIS observations shown in Figs. 1 and 2. The plots we present in terms of winds, pressure and temperature correspond to the local hours of observations. We nested three mesoscale domains in our GCM domain (see Supplementary Fig. 12 for details). Mesoscale domains use prescribed boundary conditions, derived either from GCM predictions (as in the case of d2) or from another mesoscale domain (d3 and d4). The GCM grid has a horizontal resolution of approximately 150 km. We progressively increased the horizontal resolution with a factor of three for our nested mesoscale domains. Our innermost domain, d4, has a horizontal resolution of 5.47 km. To assess the accuracy of our mesoscale predictions, we compared MarsWRF surface temperature predictions with the surface temperature observations by MCS and TES available for Olympus Mons and Arsia Mons regions at around 3:00 lst (Supplementary Fig. 13). We considered a sufficient Ls range of MCS and TES observations (Ls 310–360 for Olympus Mons and Ls 75–100 for Arsia Mons) to provide a sufficient set of observations to acquire a temperature map to be compared with MarsWRF simulations. These observations range from 1:00 lst to 4:00 lst, and MarsWRF estimations at the corresponding local times are compared for validation. The modelled surface temperatures for Olympus Mons caldera are within 10 K of the observations and within a few degrees Kelvin for Arsia Mons. It is important to note that these predictions carry uncertainties, particularly in regions with complex topography such as the Tharsis volcanoes.

Surface frost thickness and mass estimations

The MarsWRF GCM incorporates the phase transition and transport mechanisms of water vapour and ice, facilitating a parameterization of the Martian hydrological cycle that aligns with the methodologies outlined by previous studies87. This parameterization enables the model to approximate the surface frost layer thickness to about 1 μm at the locations in our study. However, it is important to acknowledge the inherent uncertainties associated with such estimations, particularly due to the limitations of physical parameterizations within Martian atmospheric models. These uncertainties are most pronounced in the prediction of atmospheric variables in regions lacking empirical observational data, such as the deposition rates of atmospheric volatiles.

In a recent experimental investigation, one study52 systematically evaluated the interaction between water frost deposition and the optical properties of a Martian soil simulant, specifically Mars Global Simulant (MGS-188). The experimental design involved the controlled deposition of water frost on the surface of the simulant, followed by precise measurements of both the spectral reflectance and the thickness of the frost layer. The findings indicate that a frost layer thickness ranging from approximately 10 to 20 μm is required to significantly attenuate the characteristic red slope of the spectral reflectance, aligning with the observed morning frost brightening in the blue wavelengths by approximately 10–20% as detected by the CaSSIS instrument. Furthermore, the study demonstrates that a relatively thin frost layer of about 100 μm is sufficient to flatten the visible spectrum, effectively neutralizing the spectral features.

Radiative transfer models50,51 can provide an additional constraint on the frost thickness estimation via the minimum optical depth (τ) necessary for frost visibility at CaSSIS visible and LNO near-infrared wavelengths. For example, with a τ of 10–2, we anticipate a minimal impact on albedo, less than 0.1 at CaSSIS visible wavelengths and negligible at LNO near-infrared wavelengths, given the single-scattering co-albedo is around 10−6 for visible light and less than 10−1 at 2.6 μm. However, LNO observations indicate a discernible albedo reduction at near-infrared wavelengths, suggesting a higher optical depth than 10−2. This implies that the frost’s grain radius and/or thickness must exceed 5 μm and 1 μm, respectively. If the grain radius is about 1 μm, then the frost layer’s thickness could be significantly greater, approximately 100 μm.

To conduct a preliminary quantification of the frost mass, we assumed a uniform frost layer thickness across all identified frost-covered regions, as observed by CaSSIS. The geographical extent of the frost coverage was approximated to the combined surface areas of the calderas of Martian volcanoes such as Arsia Mons, Olympus Mons, Ascraeus Mons and Ceraunius Tholus. By integrating the uniform frost thickness with the delineated area and adopting the density value for pure ice, we derived an initial estimate of the total frost mass. This approach provides a rudimentary yet insightful approximation of the frost mass, acknowledging the broad-scale estimative nature of this calculation.

Boulder size measurements

To investigate a potential effect of the diurnal frost cycle on the overall geomorphology and landscape evolution, we studied the shape of mass-wasted boulders across six sites of interest. Here we compare the sizes of boulders on volcanoes with frost as determined by CaSSIS (two sites in Olympus Mons and one in Arsia Mons) and on volcanoes where frost has not been detected (Tharsis Mons, Jovis Tholus and Ulysses Tholus). Because frost accumulates preferentially on poleward-facing slopes on Mars29, here we focused only on north-facing and south-facing slopes. This might reveal whether there are considerable differences in boulder sizes due to frost weathering89.

We used eight map-projected High Resolution Imaging Science Experiment (HiRISE)90 images in Geographic Information System (QGIS) to determine the three principal dimensions of each identified boulder. The first dimension is defined as the longest distance between two points on the boulder as visible from orbit. Similarly, the second dimension is defined as the diameter of the boulder orthogonal to the first dimension. Last, the third dimension is defined as the height of the boulder as estimated using shadow length and solar incidence angle. In total, we identified and measured 63 boulders across the six sites. All derived measurements were plotted on ternary diagrams91 using the Tri-Plot software92. These diagrams relate the three principal dimensions of each boulder, visualizing its overall shape as well as similarities and differences within and across the studied sites.