Main

The Perseverance rover landed on 18 February 2021 at 18.44° N 77.45° E, near the northwest rim of Jezero crater, on the inner northwest slopes of Isidis Planitia1. On board the rover is the most complete environmental station sent to date to another planet: the Mars Environmental Dynamics Analyzer (MEDA) instrument2. It includes new capabilities, compared with previous missions3,4,5,6,7,8, that enable better characterization of the diversity of physical processes driving near-surface environmental changes in Jezero. MEDA acquires data autonomously, on a regular and configurable basis, in sessions that typically cover more than 50% of a sol. Sampling sessions, typically one hour long, alternate every sol between even and odd hours, allowing for a complete characterization of daily and seasonal cycles every other sol (Extended Data Fig. 1 shows the temporal coverage of the measurements made). MEDA also provides context for the investigations that other rover instruments and systems are conducting and supports the planning of Ingenuity flights, as well as landing of a possible future mission to return samples collected by Perseverance.

In this Article, we present results for the first 250 sols of the mission (solar longitudes Ls = 6°–121°), northern hemisphere spring to early summer.

The active atmospheric surface layer

The atmospheric surface layer (ASL) is the lower part of the atmosphere in direct interaction with the surface, having a depth that varies on Mars from a few metres during daytime to tens of metres at night9,10. In the ASL, energy and mass exchanges between surface and atmosphere occur, and its hydrological cycle provides constraints on the photochemistry of surface and near-surface air. Most of the atmospheric dynamics in this layer are driven by radiative processes10. Albedo, net radiative flux and thermal inertia (TI) are key elements of that forcing and result in the radiative surface energy budget (SEB). MEDA’s thermal infrared sensor (TIRS) and radiation and dust sensor (RDS) enable quantification of all SEB terms (see Methods for a detailed description of each term) on the surface of Mars (Fig. 1a), an important step in improving the predictive capabilities of numerical models.

Fig. 1: SEB components measured in situ in Jezero.
figure 1

a, Diurnal variation of the SEB on sol 30 obtained from MEDA (symbols) and simulated with single-column models (solid lines; for further details, see Methods and Extended Data Fig. 2). b, Ground temperature measured by MEDA (black symbols) and obtained by solving the heat conduction equation (solid red line) using MEDA’s net heat flux, G. The best fit is obtained for TI = 230 J m–2 K–1 s–1/2. c, Footprint (green shade) of the ground temperature sensor on sol 30. For horizontal terrains, this footprint covers an area of a few square metres. d, Diurnal evolution of broadband albedo on sols 125 (blue) and 209 (red). The albedo shows a minimum close to noon, with increasing values as the solar zenith angle increases. This non-Lambertian behaviour is similar in other sols, regardless of the type of terrain and geometry of the rover. e,f, As in c, for sols 125 (e) and 209 (f), with MEDA-derived TI values of 605 and 290 J m–2 K–1 s–1/2, respectively. Shown data result from the mean of 300 samples at the beginning and half of each hour, plus or minus the standard deviation. lmst, Local Mean Solar Time; SWd, downwelling solar flux; SWu, solar flux reflected by the surface; LWd, downwelling long-wave atmospheric flux; LWu, upwelling long-wave flux emitted by the surface; Tf, turbulent heat flux. Credit: panels c,e,f from NASA/JPL-Caltech.

When Perseverance is parked every sol, TI is obtained by minimizing the difference between measured and numerically simulated values of the diurnal amplitude of ground temperature (Fig. 1b). The MEDA-derived TI values range from 180 to 605 J m–2 K–1 s–½, as in Gale, ref. 11. The surface albedo (Fig. 1d) is obtained around the clock, using downwelling (0.2−1.2 μm) and reflected (0.3−3.0 μm) solar flux measurements (Fig. 1c,e,f). A radiative transfer model, COMIMART12, is used to convert both fluxes to 0.2−5.0 μm. The SEB is then measured and used as an upper boundary condition to solve the heat conduction equation for homogeneous terrains in models13,14. Figure 1a,b shows the diurnal cycle of retrieved fluxes and surface temperature, respectively.

Figure 1d shows the diurnal evolution of measured albedo on the sols with the highest (sol 125) and lowest (sol 209) values of the studied period. On both sols, the minimum value is reached near noon and increases as solar zenith angle (SZA) approaches 90°. The relative maximum at ~8:00 and ~17:00 occurs when SZA = ~55° and the specular reflection is within TIRS field of view (FoV). This behaviour points to non-Lambertian albedo at the surface, not observable from surface satellites in nadir pointing.

Importantly, modelling of the thermal and radiative environment shows that the effect of the radioisotope thermoelectric generator (RTG) on TIRS FoV ground heating is negligible. This fact has been verified by careful analysis of the ground temperature in consideration of the winds, which indicate that the effects are <0.5 K.

Near-surface thermal profile

Another new feature enabled by MEDA is the simultaneous tracking of temperatures at four heights—surface, 0.85 m, 1.45 m and about 40 m—around the clock (Fig. 2a). The variation of these temperatures along a sol reflects the four main regimes of the ASL: (1) daytime convection, (2) an evening transition where the convective boundary layer collapses, (3) a night-time steady regime and (4) a morning transition where the inversion fades and a convective boundary layer grows. Figure 2b–d shows an example of the daily evolution of the thermal gradient. Daytime convection peaks at noon with the maximum of the derivative of the temperature (T, in Kelvin) with the height (z, in meters) (dT/dz)max ≈ –35 K m–1 while night-time stable stratification peaks at 20:00 with (dT/dz)max ≈ +8 K m–1 in the first metre from the surface, reaching values well above the adiabatic gradient g/Cp = 0.0045 K m–1. Figure 2e shows the seasonal evolution of mean temperatures, where the daily average thermal gradient is dominated by the daytime convective period. In most sols, night-time thermal stability weakens as the night progresses, and unstable conditions often develop from 2:00 onward.

Fig. 2: Daily cycle of temperatures under typical values of TI.
figure 2

a, Temperature at the surface (Tsurf; red), at z = 0.85 m (blue), z = 1.45 m (green) and z ≈ 40 m (green yellow). b, Thermal gradients from the surface to the near surface at 0.85 m (blue) and 1.45 m (green). c, Thermal gradient from the surface to 40 m (green yellow) and from 1.45 m to 40 m (grey). The adiabatic thermal gradient, dT/dz = –0.0047 K m–1, is shown in b and c with a horizontal light-blue line (a horizontal line that is partially covered by the curve (TsurfTz=40 m)/Δz, which can be seen quite well in the center of the figure (LTST from 10 to 15 h)). Shown data in ac result from the mean on 12 min intervals (740 samples), plus or minus the standard deviation. d, Seasonal evolution of mean daily air temperatures. TIRS and atmospheric temperature sensor temperatures (solid lines) at the surface (red), z = 0.85 m (blue), z = 1.45 m (green) and z = 40 m (green yellow) are compared with predictions from the Mars Climate Database21 (dashed lines). Yellow line shows the maximum irradiance measured by RDS. e, Statistical analysis of power spectra of temperature fluctuations at 0.85 m (blue), 1.45 m (green) and 40 m (yellow) between 10:00 and 15:00 h. Lines show the fit to the data in each frequency range. Figures reflect the estimated exponential indices. TI = 330 J m–2 K–1 s–1/2. Dots are averages over a 12 min window. ltst, local true solar time.

A similar observation was reported for InSight15 and attributed to the radiative influence of the hardware. In addition, Curiosity’s Rover Environmental Monitoring Station instrument, measuring on the deeper Gale crater, has seen this inversion broken only during the global dust storm16. With MEDA, we observe that the night-time inversion depends on the local terrain properties. Measurements over high TI terrain result in the break-up of the night-time thermal inversion due to warm night-time surfaces (Extended Data Fig. 3). However, air temperatures at different levels are not sensitive to the TI of the specific terrain, and the influence of the terrain on air temperatures decreases progressively from 0.85 to 40.00 m (Fig. 2d and Extended Data Fig. 3). Winds driven by horizontal gradients of surface thermal properties17 may be the origin of the discrepancies we observe from measurements with respect to the predictions from one-dimensional radiative equilibrium models.

Temperature fluctuations are common throughout the sol; these rise after sunrise, peaking near noon (amplitudes ΔTmax ≈ 10 K), and are convective in nature. These subside before sunset and increase again during the break-up of the night-time inversion, suggesting strong nocturnal fluctuations (Extended Data Fig. 4). The oscillations are created by turbulent processes in the atmosphere whose characteristics can be investigated by analysing the spectral power density of temperatures, pressures and winds18,19,20. Figure 2e shows the power spectral densities of temperatures during the convective period, averaged over 250 sols. The results show typical slopes of turbulence in daytime hours, with changes at other times of the sol; these will be analysed in later works. MEDA enables the identification of different dynamical regimes, forced, inertial and dissipative, more clearly than with previous instruments19,20 and at different altitudes.

Pressure fluctuations

Turbulence is also present in pressure and horizontal wind measurements. Figure 3a shows examples of the daily pressure cycle on different sols; this cycle has the contribution of different dynamic phenomena, as discussed in the following. Analysis of the rapid pressure fluctuations shows a power spectrum similar to that of temperature (Extended Data Fig. 5), different from that expected from Kolmogorov turbulence but similar to that measured by InSight15. During the stable night-time period, the pressure fluctuations are at the detector noise level.

Fig. 3: Fluctuations in the daily pressure cycle and wind patterns.
figure 3

a, Daily pressure cycles for sols 101–105, showing deviations from the mean values (in 10 s intervals) caused by different dynamic phenomena. b, Mean values of wind intensity, averaged in intervals of 5 min. Error bars show the standard deviations. c, Wind azimuths in different time slots (dotted lines, ~120° azimuth, show the regional slope direction in the Isidis basin). We observe that upward winds predominate during the day and downward winds during the night, a trend attributable to local flows acting in Jezero and the interaction with other scales (top). Histogram of wind speed at those hours, that is, the frequency distribution of the occurrence of each speed, f(v) (bottom). d, Mean directions from which the winds blow, averaged in intervals of 5 min. Error bars show the standard deviations.

Convection generates transient events detectable through several MEDA sensors, especially in temperature and pressure data (Extended Data Fig. 6). Some events are dust devils (DDs), also detected as slight drops (0.4–26.0%) in radiation sensor readings or imaged with Perseverance cameras22. Jezero exhibits the highest abundance of DDs so far detected by a mission on the surface of Mars23,24. The pressure drops detected in this period range from 0.3 to 6.5 Pa in intensity and last from 1 to 200 s. DDs where simultaneous MEDA wind data are available have estimated diameters from 5 to 135 m (ref. 23), with vortices having rotational speeds of ~4–24 m s–1. A small number of these produced measurable albedo changes on the surface as they removed dust, observed through variations in the upward/downward radiation ratio measured with TIRS and RDS sensors (Extended Data Fig. 7)23.

Wind speeds show a daily cycle, with maximum values of 7 m s–1 in the afternoon and near null between 4:00 and 6:00 (Fig. 3b). Strong gusts are detected with maximum speeds of 25 m s–1 at midday. Turbulence creates fluctuating winds of about 2–4 m s–1 at night, where they are probably responses to horizontal shear flows25, and 5–7 m s–1 during the convective hours (Fig. 3c). Pre-landing atmospheric modelling predicted that daytime upflows and night-time downflows on the Isidis basin slopes would dominate the overall wind pattern, with Jezero local topography causing a relatively small but measurable effect13,14. Wind data support the dominance of daytime upslope currents from roughly southeast, with a reversal of winds at night14 (Fig. 3c,d). These diurnal wind patterns drive aeolian erosion at Jezero23 and show a variety of behaviours resulting from a complex interaction among regional circulation, slope winds and interaction with the general circulation and the large-scale Hadley cell flow.

Atmospheric dust properties

SkyCam is tracking the regular morning–evening opacity cycle in the 600–800 nm wavelength range as a function of time (Fig. 4a)2. During the clear season covered by this study, persistently higher opacity is observed in the morning (optical depth (OD) 0.5) than in the afternoon (OD 0.4) (refs. 26, 27). Analysis of RDS data at different wavelengths and observation geometries allows determination of the optical properties of the dust, and derivation of the OD variation at high temporal resolution, similar to the more sparsely sampled Viking record28 (Fig. 4b). The dust optical properties and OD are estimated by comparing the temporal variation of the measured sky spectral intensity with radiative transfer simulations (Fig. 4c shows an example). Most of the particle-size information is obtained when the Sun trajectory is near one of the RDS lateral sensor’s FoV. From these observations, we found particle sizes ranging from 1.2 to 1.4 μm, consistent with previous studies29,30,31, with corrections suggested by refs. 32,33,34. To estimate the non-sphericity of those particles, the T-matrix approach was used to compute the phase function, single-scattering albedo and extension cross section.

Fig. 4: Daily monitoring of atmospheric opacity and characterization of dust particles and clouds.
figure 4

a, OD derived from SkyCam images to follow the day–night cycle in the visual range (550 nm). Values are observed to be higher in the morning than in the afternoon, consistent with measurements made occasionally with MastCam-Z22. b, RDS observations also allow the derivation of the OD at a temporal resolution of 1 s, enabling the study of short-duration events, such as DD. An example of a temporal increase in opacity due to a nearby DD occurred on sol 21 around 15:11. c, An example of dust particle radius and OD estimation using RDS observations at different wavelengths and radiative transfer simulations. The best fit is obtained for an effective radius reff = 1.4 μm. d, Variation of the colour index (CI), defined as the ratio between RDS observations at zeniths at 450 and 950 nm (ref. 41) to the SZA measured on sol 296. The SZA of maximum CI indicates that this cloud layer is above 45 km. e, Aerosol OD (AOD) at 9 µm retrieved from TIRS observations for sols 30 and 200. TIRS observations enable AOD to be retrieved at all local times. The large diurnal variation for sol 200 is probably caused by water-ice clouds.

These first sols of the mission fell within the aphelion cloud belt season and near the peak latitude for water-ice clouds35. It is therefore likely that some of the afternoon opacity in Fig. 4a, including the increase around sol 70 and most of the morning–afternoon difference, is due to water-ice hazes36,37. Clouds were observed around sols 70 and 180, but discrete clouds were not typically present during daytime22. During this season, a low dust OD (0.3–0.6), with low variability, is typical of other sites38,39. We have also found cloud signatures during daytime and twilight. In the latter cases (Fig. 4d)40, we could constrain the cloud altitudes using radiative transfer simulations41. In most cases, we found cloud altitudes around or above 40 km and particle sizes larger than 1 μm (indicative of water-ice particles).

The daily OD evolution was also retrieved from TIRS infrared measurements, strengthening the case for night-time clouds. Figure 4e shows the diurnal variation of thermal infrared aerosol OD, contributed by both dust and water-ice clouds, as a function of local time for two representative sols. Because TIRS observes thermal infrared radiation, this retrieval is possible for all local times, including during the night, a capability not available on previous rovers except by Mini-TES on board Mars Exploration Rover, which could make only limited and occasional night-time observations42. The OD observed by TIRS during sol 30 (Ls = 20°) shows a moderate variation with greater opacity at night than during the day. By sol 200, the aphelion season cloud belt43 was near its peak annual amplitude, and data reveal a notable diurnal variation in OD with maximum clouds shortly after dawn. This ability to track OD throughout each sol is a powerful tool that leads to new insights about how dust and water-ice clouds interact with the surface and the rest of the atmosphere.

A complex humidity cycle

Measuring the relative humidity (RH) in the ASL on diurnal and seasonal timescales is a key element in understanding hydrological processes in the Martian atmosphere44. MEDA’s humidity sensor (HS) often finds a nocturnal hydrological cycle more complex than anticipated in numerical predictions13. Figure 5a shows the daily and seasonal behaviour of RH, while Fig. 5b shows the daily maxima recorded in the period studied. Likewise, Fig. 5c shows the seasonal variations of night-time water-vapour volume mixing ratio (VMR). Within the diurnal cycle, the maximum, typically in the range 15–30% in RH (referring to HS temperature), occurs in the early morning, with maximum VMR being reached around midnight. Nocturnal water-vapour amounts at Jezero during the seasons are lower than those at Gale and model predictions13.

Fig. 5: Daily and seasonal cycles of RH and VMR observed at Jezero.
figure 5

a, Daily and seasonal evolution of RH for sols 80–250, Ls 44–122°. The uncertainty in the estimation of this magnitude, which is temperature dependent, is <3.5% in the temperature range recorded by HS. b, Seasonal evolution of the maximum RH values observed, reaching an absolute maximum of 29% RH, referring to the temperature recorded by HS itself, for sols 80–250, Ls 44–122°. c, Seasonal behaviour of the night-time VRM derived from RH for sols 64–250, averaging seconds 2–5 from the beginning of each acquisition session. The uncertainty in the estimation of this magnitude, for the temperatures and the RH range shown in a, is <20%. Maintenance regeneration heatings of the sensor heads are marked as blue bars (these data should not be taken into consideration).

MEDA measured a seasonal minimum in night-time VMR near Ls = 70°, with higher abundance and greater variability at the end of the season (Fig. 5c). In addition, a large increase in VMR was observed on the evening of sol 104 (see Extended Data Fig. 8 for details of the evolution of VMR and RH on those sols), accompanied by cooling atmospheric temperatures. The increase in RH slowly returned to typical values, while the temperature continued dropping. This behaviour may be due to a single dry air mass advected over the rover bringing cold, dry air, which then remains in the area. An alternative explanation is that it is related to the local surface and possible exchange processes. In the early morning of sol 105, frost conditions were possible as seen by comparing the calculated frost point (at 1.45 m) with the TIRS ground temperature. A resulting hypothesis is that, if subsurface exchange is occurring, the actual frost point at the atmosphere–surface interface may be lower owing to less vapour present in the atmosphere.

Non-local dynamical phenomena

The daily pressure cycles at Jezero showed a rich variability, reflecting the action of different dynamical mechanisms in the atmosphere under a variety of spatial and temporal scales (Fig. 3a). On the seasonal scale, as the northern polar cap sublimated, the daily mean pressure increased from 735 Pa on sols 15–20 (Ls = 13°–16°) to a maximum of 761 Pa on sols 99–110 (Ls = 52°–57°). It then gradually decreased to 650 Pa in sol 250 (Ls = 125°) as the southern polar cap grew (Fig. 6a).

Fig. 6: Atmospheric pressure variability as a consequence of various dynamic mechanisms at various scales.
figure 6

a, Seasonal evolution of daily mean pressure values and their daily ranges. b, Detrended standard deviation of pressure, as a function of sol and LTST, after subtracting least squares fit of the observed pressure. Removing the tidal components shows that the dominant contribution to pressure variability occurs during the hours with strong convection and after a calm period. c, The first three components of the Fourier analysis of the pressure cycle: diurnal (24 h period; black), semidiurnal (12 h period; blue) and terdiurnal (6 h period; red) components. d, Equivalent Fourier analysis of the temperature data at z = 1.45 m: diurnal (blue) and semidiurnal (red) components.

Extended Data Fig. 9a shows the deviations of the daily pressure averages of Fig. 6a from a polynomial fit of degree 5, observing oscillations with periods in the range of 3–5 sols and amplitudes varying between 1 and 3 Pa, a signature of what could be high-frequency travelling waves arising from baroclinic instabilities also reported elsewhere13. In addition, Extended Data Fig. 9b shows the regular oscillations of the residuals resulting from a fit to the approximately 1 h measurement series that, on average, have a peak-to-peak amplitude of 0.2–0.4 Pa and periods between 12 and 20 min. The properties of these oscillations are similar to those reported in Gale, observed by the Rover Environmental Monitoring Station instrument on the Curiosity rover45 and in Elysium Planitia by the InSight mission15, which have been interpreted as produced by the passage of gravity waves. Figure 6b shows the seasonal and diurnal pressure variability where the daily patterns of pressure changes are observed.

Thermal tides cause the large modulation of the daily pressure (P) cycle46. A Fourier analysis of that cycle shows that up to six components are present, with maximum amplitudes ranging from 0.2 to 10.0 Pa. Figure 6c shows the wide variability of the normalized amplitude of diurnal and semidiurnal tides. The maximum relative change observed in the studied period is (δP/〈P〉)max ≈ 0.013. Smaller changes also occurred in tidal components 3 and 4. The semidiurnal component showed a very strong drop between sols 20 and 50; although still under investigation, it may suggest a relation to the dust loading present on those sols or the development of disturbances at the polar cap edge46. Tides are also detected in the Fourier analysis of the temperature data with half amplitudes of 26 K (diurnal), 2–6 K (semidiurnal) and 2–4 K (terdiurnal), as shown in Fig. 6d. The tidal variability is related mainly to changes in atmospheric opacity produced by clouds and dust loading at different altitudes22. The period studied is the non-dusty season on Mars, which makes the variability relatively low.

A rich and dynamic near-surface environment

Mars 2020 mission includes a payload to monitor an environment that exhibits a rich diversity of behaviours. Many of the measurements made by MEDA so far are the first time they have been obtained on Mars, revealing interesting surprises in Jezero’s atmosphere.

The SEB is measured for the first time in situ. The design of future engineering systems, the understanding and modelling of photochemical reactions at the surface and the interpretation of satellite measurements benefit from these results. An example is the characterization of the non-Lambertian reflection of the surface that must be considered in the interpretation of orbital observations of variations in albedo when trying to understand changes in the physical properties of the surface.

Globally, the measured daily temperature cycle agrees with model predictions (although with some deviations in the vertical temperature gradient), expected magnitude of thermal oscillations and the seasonal evolution. In addition, the observed vortex convective activity matches the predictions of large eddy simulation using the MarsWRF model23. However, when analysing vertical temperature profiles, we find a diversity of nocturnal responses that raise intriguing questions about what is happening at the different locations traversed by the rover.

Several independent radiation sensors and methods have measured the occurrence and even development of night-time clouds long before the peak of the cloud season. The ability to track OD throughout each sol is a powerful tool that has shown the prevalence of clouds near dawn and will lead to new insights about how dust and water-ice clouds interact with the surface and the rest of the atmosphere.

The observed nocturnal hydrologic cycle is more complex than anticipated by the models and is also observed in Gale. This unpredicted behaviour can be due to a variety of causes yet to be explored in detail.

Thermal tides show smaller pressure amplitudes at Jezero when compared with Curiosity’s observations47 or Viking48. While the general behaviour at Jezero was predicted by the models13,14, there are differences in its amplitude and timing probably due to the interaction of local topography with the air mass exchanges between the interior and exterior of the basin. Another interesting result is the existence of multisol waves at a time of the season when baroclinic waves have been observed to have very limited activity46.

Overall, MEDA observations show a dynamic environment rich in atmospheric phenomena that is different from other locations on Mars studied by previous missions. The characterization of Jezero’s atmosphere plays an important role in the development of the Mars Sample Return mission and in the exposure of the samples being collected by Perseverance.

Methods

MEDA operational strategy

MEDA can operate continuously and independently of the rover’s battery charge cycles, 24 hours a day and subject to a measurement programme sent from Earth (Extended Data Fig. 1). The observation sessions are subject to power availability and, eventually, to incompatibility with other activities to be performed by the rover.

Due to these restrictions, the usual sequence of measurements that MEDA is carrying out consists of the acquisition of all the magnitudes that the instrument records every other hour. On the following sol, the measurement hours are reversed so that every two sols, a total coverage of all daily magnitudes is made.

TIRS measurements and analysis

The TIRS is an infrared radiometer with five channels that measure downward radiation (IR1), air temperature (IR2), reflected short-wave radiation (IR3), upward long-wave radiation (IR4) and ground temperature (IR5)2. We use the ratio between TIRS IR3 and the ‘total light’ Top7 detector of the RDS (description follows) measurements as a proxy for surface albedo. Since TIRS and RDS measure in different bands, a correction is made using the COMIMART radiative transfer model12.

TI can be straightforwardly derived across Perseverance’s traverse by using MEDA values of the net heat flux into the ground as the upper boundary condition to solve the heat conduction equation for homogeneous terrains. This quantity governs the thermal amplitude in the shallow subsurface from diurnal to seasonal timescales, and therefore accurate estimations of TI can be useful to constrain the thermal environment of the samples collected by the Mars 2020 mission. For this estimation, numerical models need to simulate the SEB (see paragraphs that follow in this section). MEDA measures the SEB, allowing for a more in situ-based estimation of the TI.

Concerning TIRS infrared fluxes, the two upward-viewing sensors of TIRS (IR1 and IR2) enable the total column optical depth of aerosol above the rover to be retrieved. The observed signal in TIRS IR1 is sensitive to a combination of atmospheric temperatures and total aerosol optical depth (dust plus water-ice cloud). The atmospheric temperature profile is taken from concurrent observations by the interferometric thermal infrared spectrometer (EMIRS) instrument on board the Emirates Mars Mission49 with temperatures near the surface modified to match the observed TIRS IR2 signal. A radiative transfer model including aerosol scattering is then used to find the aerosol optical depth that would produce the observed TIRS IR1 signal. The estimated uncertainty in these retrievals is ±0.03.

RDS measurements and analysis

The RDS comprises two sets of eight photodiodes (RDS-DP) and a camera (SkyCam)2,50. One set of photodiodes is pointed upward, with each one covering a different wavelength range between 190 and 1,200 nm. The other set is pointed sideways, 20° above the horizon, and they are spaced 45° apart in azimuth to sample all directions at a single wavelength.

HS measurements and analysis

The HS provides directly the local RH and local sensor temperature. Combined with the pressure data provided by the MEDA pressure sensor (PS), water-vapour VMR can be calculated, too. The HS has two measurement modes: continuous measurement and high-resolution interval mode (HRIM). In HRIM, the HS is powered on for only 10 s and then powered off to avoid self-heating. HRIM provides the measurements with the best accuracy, but continuous measurements are beneficial for monitoring changes in RH during short periods.

PS measurements and analysis

The PS is actually a set of two capacitive transducers that provide the hydrostatic pressure as a function of the local temperature, for which the sensor also provides its own temperature2.

They are the Barocaps RSP2M and NGM (due to the internal operation of the sensor, only one or the other can work, not both simultaneously), which can be operated at 0.5 or 1 Hz; the RSP2M has a worse resolution and worse stability than the NGM, but the RSP2M’s warm-up time is shorter.

Atmospheric temperature sensor measurements and analysis

Atmospheric temperature sensors (ATSs) are thin thermocouple sensors with three sensors distributed azimuthally around the remote-sensing mast (RSM) at an altitude of 1.45 m and two sensors on the front sides of the rover at an altitude of 0.85 m, which can provide local temperature measurements at a configurable rate. The location of three sensors around the RSM ensures that at least one of them is located downwind during most of the time, producing a clean measurement of air temperature. The two sensors at 0.85 m are more shielded from the environment.

A systematic comparison of ATS data and wind measurements, including the rover orientation on each individual sol, guided us to use the following rules to select the appropriate ATS to characterize the unperturbed atmosphere. During daylight hours, the sensor at z = 1.45 m measuring the lowest temperature is generally the one located downwind. In certain wind conditions, two sensors can be located downwind, and they show very similar temperatures (within 0.1 K) and equivalent oscillations. At z = 0.85 m during daylight hours, we select the sensor with the lowest temperature. To take into account changes in wind direction that modify the selected ATS, we consider slow transitions from one sensor to another at each level when the sensor measuring the lowest temperature changes. During night-time, the smaller values of the winds and the sheltered location of the two ATSs at z = 0.85 m typically result in one sensor generally much warmer than the other. Comparison with environment winds indicates that the lowest temperature sensor is always the most exposed to environment winds with the lowest thermal perturbations from the rover. In the RSM, all three ATSs can experience radiative cooling effects from the rover deck. Thus, when two ATSs have highly correlated values and thermal oscillations, we use the average of them instead of the ATS with the lowest temperature. Thermal perturbations from the radioisotope thermal generator are easily observed at night at z = 1.45 m and identified from the wind data and rover orientation and do not have any noticeable effect on the results here presented. The radioisotope thermal generator, located at the back of the rover, does not generally cause detections in the sheltered detectors in the front of the rover at z = 0.85 m.

Wind measurements and analysis

The wind sensor consists of two horizontal booms placed on the RSM 1.5 m above the rover base and rotated in azimuth 120° with respect to each other.

This placement allows at least one of them to be out of thermal disturbance and out of the rover geometry for any wind direction. Therefore, the level of confidence and the analysis of the perturbing effect that the rover causes on the measurements made will properly depend on the wind direction and speed.

Thus, the data retrieval procedure implemented on the ground consists of the weighted combination of the local wind-speed retrievals from each individual sensor, thereby obtaining the free-flow wind estimate. The weighting of each contribution is established on the basis of the results of the computational fluid dynamic models developed to evaluate how the free wind flow is affected by the rover hardware, thus allowing interpretation of the local wind measurements at each boom location. More details of this process are provided in ref. 2.

SkyCam image analysis

The SkyCam imager is integrated inside the RDS sensor (described in the preceding) and permanently pointed at the Martian sky. The orientation of the camera is not motorized, and its optics are fixed.

SkyCam optical depths were measured via extinction determined through direct solar imaging2. The image field of view includes an annular ND-5 coating: when the Sun is within that coated region twice each sol, it appears comparable in brightness to the sky outside the annulus. Flux from the Sun was integrated after removal of background signal (mostly dark current). From the flux and atmospheric path at the time of each image, optical depth is calculated as τ = ln(F/F0)/η, where τ is normal optical depth, F is observed solar flux, F0 is a calibration parameter representing solar flux in the absence of an atmosphere and η is air mass, which is the ratio of atmospheric column mass on the observed ray to normal column mass. Because SkyCam cannot be calibrated by observing a wide range of air masses or changing the camera versus atmosphere geometry40, it was calibrated by comparison with MastCam-Z-derived optical depths22.

The current estimates of SkyCam opacity uncertainties are 0.07, while those for AM–PM differences are 0.08, and neither varies much among points. The variation among AM–PM differences is not substantial for sols 50–250. However, the population average is 0.10 ± 0.01 for sols 50–250 and –0.01 ± 0.04 before sol 50.

Derivation and importance of SEB

MEDA enables in situ quantification of the SEB on Mars. To this end, conservation of energy at the surface–atmosphere interface of Mars requires that

$$G = ({\mathrm{SW}}_{\rm{d}}-{\mathrm{SW}}_{\rm{u}}) + ({\mathrm{LW}}_{\rm{d}}-{\mathrm{LW}}_{\rm{u}}) + T_{\mathrm{f}}-L_{\mathrm{f}}$$
(1)

where G represents the net heat flux into the ground, SWd is the downwelling solar flux, SWu is the solar flux reflected by the surface, LWd is the downwelling long-wave atmospheric flux, LWu is the upwelling long-wave flux emitted by the surface, Tf is the turbulent heat flux and Lf is the latent heat flux. By convention, radiative fluxes directed towards the surface (warming) and nonradiative fluxes (Tf, Lf, and G) directed away from the surface (cooling) are taken as positive in equation (1). Moreover, the radiative fluxes are plugged into equation (1) as positive values, whereas nonradiative fluxes can be plugged in as positive or negative depending on whether they are directed away from or towards the surface.

RDS measures SWd between 0.2 and 1.2 µm, while TIRS measures SWu between 0.3 and 3 µm and LWd and LWu between 6.5 and 30 µm2,52,53. As required in quantifications of the SEB, measured radiative fluxes must be extended to the entire short-wave (0.2–5 µm) and long-wave (5–80 µm) ranges. To extend SWd and SWu, we use the radiative transfer model COMIMART12 along with measured values of aerosol optical depth from the MastCam-Z instrument22. We note that the incoming radiation between 0.2 and 1.2 µm accounts for around 78% of the entire short-wave flux and that the uncertainty in this extension is small because COMIMART includes wavelength-dependent dust radiative properties, accounting for the variations (smaller than 3% for the majority of conditions) in the conversion factor as a function of dust opacity and solar zenith angle. Similarly, we assume a surface emissivity, ϵ, of 0.99 and the Stefan–Boltzmann law to extend LWu. This value of ϵ minimizes the difference between the ground temperature measured by TIRS (8–14 µm) and the ground temperature derived from LWu. To extend LWd, we use the University of Helsinki/Finnish Meteorological Institute adsorptive subsurface–atmosphere single column model (SCM) along with measured values of aerosol optical depth51. Figure 1a shows the diurnal evolution of each term of the SEB terms on a particular sol, both obtained from MEDA (symbols) and simulated by SCM (solid lines). Extended Data Fig. 2 shows the evolution of these magnitudes recorded by MEDA over the first 250 sols. The excellent agreement demonstrates that MEDA’s measurements are robust.

The turbulent heat flux is defined as \(T_{\mathrm{f}} = \rho _{\mathrm{a}}c_{\mathrm{p}}\overline {w^\prime T^\prime }\), where ρa is the air density, cp = 736 J Kg–1 K–1 is the specific heat of CO2 gas at constant pressure and \(\overline {w^\prime T^\prime }\) is the covariance between turbulent departures of the vertical wind speed, w’, and temperature, T’. These departures are typically calculated over periods of about a few minutes16. As MEDA measurements of w are not yet available, we use the drag transfer method54 to indirectly calculate the turbulent heat flux as:

$$T_{\mathrm{f}} = k^2U_{\mathrm{a}}\rho _{\mathrm{a}}c_{\mathrm{p}}f(R_{\mathrm{B}})\frac{{(T_{\mathrm{g}} - T_{\mathrm{a}})}}{{{\mathrm{ln}}^2(z_{\mathrm{a}}/z_0)}}$$
(2)

where k = 0.4 is the von Karman constant, za = 1.45 m is the height at which the air temperature and horizontal wind speed (Ua) are measured, z0 is the surface roughness (set to 1 cm (ref. 55)) and f(RB) is a function of the bulk Richardson number that accounts for the thermal stability in the near surface of Mars56.

We note that Lf has been neglected in the SEB because formation or sublimation of surface ice has not been detected at Jezero to date, with maximum near-surface relative humidity values below 25% for the first 250 sols.

Derivation of thermal inertia using MEDA measurements

We use MEDA measurements of the SEB as the upper boundary condition to solve the heat conduction equation in the soil for homogeneous terrains:

$$\frac{{\partial T\left( {z,t} \right)}}{{\partial t}} = \left( {\frac{{{\mathrm{TI}}}}{{\rho c_{\mathrm{p}}}}} \right)^2\frac{{\partial ^2T\left( {z,t} \right)}}{{\partial z^2}}$$
(3)
$$- \frac{{{\mathrm{TI}}}}{{\rho c_{\mathrm{p}}}}\frac{{\partial T\left( {z = 0,t} \right)}}{{\partial z}} = G = \left( {{\mathrm{SW}}_{\mathrm{d}}-{\mathrm{SW}}_{\mathrm{u}}} \right) + \left( {{\mathrm{LW}}_{\rm{d}}-{\mathrm{LW}}_{\rm{u}}} \right) + T_{\mathrm{f}}-L_{\mathrm{f}}$$
(4)
$$T\left( {z = z_{\mathrm{d}},t} \right) = T_{\mathrm{d}},$$
(5)

where TI is the thermal inertia, ρ is the soil density, c is the soil specific heat and zd is the depth at which the subsurface temperature is constant and equal to Td. Here we assume that ρc = 1.2 × 106 J m–3 K–1 and zd = 3 × L, where \(L = \left( {\frac{{{\mathrm{TI}}}}{{\rho _c}}} \right)\sqrt {\frac{2}{\omega }}\) is the diurnal e-folding depth and ω = 7.0774 × 10−5 s–1 is the angular speed of Mars’s rotation.

Under these assumptions, Td and TI are the only unknowns, which can be solved by best fitting the solution to equations (3)–(5) to measured values of the daily minimum ground temperature and diurnal amplitude in ground temperature, respectively. As analysed in ref. 54, the solution to equations (3)–(5) depends primarily on TI, with considerably smaller variations as a function of Td, zd and ρc.

Derivation of albedo using MEDA measurements

We calculate the broadband albedo in the 0.3–3.0 µm range as

$$\alpha = {\mathrm{SW}}_{\rm{u}}^{0.3 - 3\upmu {\mathrm{m}}}/{\mathrm{SW}}_{\rm{d}}^{0.3 - 3\upmu {\mathrm{m}}}$$
(6)

Here, \({\mathrm{SW}}_{\rm{u}}^{0.3 - 3\upmu {\mathrm{m}}}\) is the reflected solar flux measured directly by TIRS, while \({\mathrm{SW}}_{\rm{d}}^{0.3 - 3\upmu {\mathrm{m}}}\) is the downwelling solar flux calculated with COMIMART by extending RDS Top7 measurements from 0.19–1.20 to 0.3–3.0 µm. On the basis of uncertainties in measured solar fluxes, the relative error in albedo is <10% in the vicinity of noon and <20% towards sunset and sunrise.

Description of COMIMART and UH/FMI SCM

COMIMART includes wavelength-dependent radiative properties of the Martian atmospheric constituents. Dust radiative properties are calculated from the refractive indices derived from satellite observations32,57. We have assumed a dust particle effective radius of 1.5 µm and an effective variance of 0.3. The surface albedo is also assumed to depend on wavelength, with low values in the ultraviolet, increasing towards the near-infrared spectral region58. However, this value barely impacts simulations of downwelling solar fluxes. Solar radiative fluxes are calculated using the delta-Eddington approximation59. The model has been validated for different solar elevations, dust abundances and scattering regimes12, concluding that it is accurate for the majority of conditions that can be found at Jezero crater.

Unlike COMIMART, the University of Helsinki/Finnish Meteorological Institute single-column model (UH/FMI SCM) allows the simulation of long-wave fluxes. SCM handles solar (short-wave) radiation with a fast broadband delta-two-stream approach for CO2 and dust, and thermal (long-wave) radiation with a fast broadband emissivity approach for CO2, H2O and dust. These two schemes have been validated through comparisons from a multiple-scattering resolving model60. As in COMIMART, dust optical parameters were obtained using refractive indices in refs. 32, 57. This confirmed the dust SW single-scattering albedo to 0.90 and the asymmetry parameter to 0.70, which are used in our reference model.

When initialized with aerosol opacities retrieved from MastCam-Z at 880 nm, COMIMART and SCM simulate nearly identical solar fluxes, with relative departures <5% at noon.

Retrieval of dust scattering properties

Dust optical properties are estimated by simulating the RDS measurements2 throughout the sol, using radiative transfer simulations. The irradiance (E) measured by a given RDS channel is computed as:

$$E = C \times [T\left( {\mu _0,\varphi _0} \right) \times e^{\frac{{ - \tau }}{{\mu _0}}} + \mathop {\smallint }\limits_0^{2\pi } \mathop {\smallint }\limits_{ - 1}^0 I\left( {\tau ,\mu _0,\varphi _0,\mu ,\varphi } \right) \times T\left( {\mu ,\varphi } \right) \times R \times \mu \times {\mathrm{d}}\mu {\mathrm{d}}\varphi ]$$
(7)

where I is the scattered intensity simulated with a radiative transfer model, μ0 and φ0 are the cosine of the solar zenith angle and the solar azimuth angle, μ and φ are the cosine of the zenith angle and the azimuth angle of the observation, T and R are the sensor angular response and responsivity, C is a constant and τ is the total vertical opacity. Extended Data Fig. 10a shows RDS Top5 (750 nm) signals simulated for different dust opacities and an effective radius of reff = 1.2 μm.

Here we see that, by varying the dust opacity, the simulated signals change in value and shape. Similar results are found when we vary the effective radius of the particles, keeping constant the dust number density. By changing reff, we vary the dust opacity through the particle cross section, the particle phase function and the single-scattering albedo. Estimation of the free parameters (for example dust number density or dust particle radius) is performed by finding the optimal value of the parameter that minimizes the differences between observations and simulations, as shown in Extended Data Fig. 10b, using the Levenberg–Marquardt procedure.

The dust particle radius is estimated using (1) the observations made by the RDS lateral photodiodes2, whose FoV is narrow (±5°), when the Sun is near the FoV of one of those detectors (this occurs when the Sun is at low elevation); at that time, RDS lateral detectors provide observations at a wide range of scattering angles (sensitive to the particle phase function); (2) the observations made by the top photodiodes at different wavelengths (for example, 450, 650, 750, 950 nm). In this case, the opacity at each channel wavelength depends on the particle radius, so by fitting the observations made at different wavelengths simultaneously, we can derive the size of the dust particles. To better constrain the particle radius, the observations from RDS and SkyCam can be combined: the opacity derived from SkyCam can be used when fitting the observations made by RDS, being in this case the particle radius is the only free parameter in the retrieval.

Once the particle radius is estimated, we can provide the dust opacity at different time intervals, such as the daily or the hourly average, by assuming the particle radius does not change much throughout the sol.