The present epoch may not be representative in determining the history of water on Mars

Significance

There is broad interest in the possible existence of life on Mars. Understanding the history of water is important for determining Mars' habitability. We examine the data on the history of water, with special attention to the uncertainties in the analyses and in determining whether data pertaining to different time periods are consistent with each other. The results tell us that Mars has lost a lot of water to space as well as to other “sinks” and that the amount of water that Mars once had could have supported life.

Abstract

Understanding the history of water on Mars is important for understanding both its geological and potential biological history. The abundance and physical state of water has evolved through time, from the surface having an early warmer and wetter environment to the present-day colder and drier one. Although multiple lines of evidence support this change, attempts to determine the abundance of water on the planet, the history of water at the surface, and the sequestration into both permanent and exchangeable sinks have yielded a wide range of results. I explore the uncertainties in the processes and interpretation, to understand our ability to quantitatively determine the water inventory and its changes through time. Results indicate that the present state of models and of the data constraining them preclude determining a unique history for water. This uncertainty does not affect the conclusion that significant amounts of water have been lost to space and to other sinks and that these losses are consistent with the changes in climate and surface environment.

Water is important on Mars both for its geological and geochemical effects and for its ability to support potential biological activity. The behavior of water on Mars throughout its history can be inferred from two very different approaches. First, we can observe the current distribution of water over the planet, determine the polar and atmospheric processes responsible for the observed seasonal cycle of water and for the annual loss of the H and O atoms to space, extrapolate to how those processes might have changed through time, and add up the amounts of water involved. Second, we can observe present-day indicators that tell us about the longer-timescale history of water and infer what the state of water has been and what processes must have operated through time. With this second approach, we see morphological features on ancient Mars that are understood to have resulted from the geological actions of liquid water, we have identified minerals on the surface that required the presence of liquid water at the surface or in the crust early in Martian history in order to form, and we can measure the ratios of stable isotopes that are indicative of the time-integrated and globally integrated loss of a significant amount of gas to space.

Are these approaches consistent with each other? Are the processes that we infer to be operating today, acting over long times, capable of explaining the transition seen in the geological observations from an ancient Mars that had stable and seemingly abundant liquid water to the present cold and dry planet? Or do we need to postulate the occurrence of different processes or of different boundary conditions that affected those processes as they would have operated in the past? In addition, can we use this information to infer the abundance and distribution of water and how it has changed through time? Although we now know a lot about the behavior of water on all timescales, from daily up through the billion-year timescales of geological time, these are still open questions.

In this paper, I will examine the evidence for whether the key processes affecting Mars water have operated continually through time, whether there have been different processes operating in the past or whether the same processes operating today operated with different efficacy in the past. Throughout, the emphasis will be on the temporal variability of the processes on timescales ranging from 1 y to 4 billion years and on how the uncertainties in determining the processes translate into an inability to determine a unique history for water. Our focus will be largely on the rate of loss to space and its impact on the ratio of D/H in water in the atmosphere and in the reservoirs with which water can exchange. We will start with the seasonal water cycle and work toward longer timescales.

1. The Seasonal Water Cycle and Interannual Variability

The seasonal cycle of atmospheric water vapor is important in that the H₂O supplies the H and O that can escape to space and result in a net loss of water from the planet through time. The seasonal water cycle on Mars was first observed in the 1960s using spectroscopic observations made telescopically from Earth (e.g., ref. 1). However, the full spatially resolved seasonal cycle of atmospheric water has been observed only starting in 1976 to 1978 with data from the Viking mission (2). Measurements from several different spacecraft extend this record up through the present, spanning more than twenty Mars years (e.g., refs. 3 and 4). The combined interpretation over these years is that the northern-hemisphere polar ice cap loses its seasonal CO₂-ice covering during summer, exposing a water-ice cap that lies underneath it (Fig. 1); the water ice heats up, and water vapor sublimates (goes from solid directly to vapor) into the atmosphere (e.g., ref. 5). This atmospheric water then can be distributed somewhat globally by the atmospheric circulation. Some of the water will exchange diffusively on diurnal and seasonal timescales with the regolith, where it can adsorb onto individual regolith grains. The regolith can hold much more water than the atmosphere, even within just the top centimeter; as a result, adsorption and exchange can occur both in response to the seasonal variations in atmospheric water driven by the polar-cap and as a driver of those variations (6–8).

Fig. 1.

Mars north-polar residual cap, showing the ice deposits, the dark lanes, and the surrounding layered terrain. (NASA mosaic from Viking images.)

Superimposed on this, some water is being desublimated (going from vapor to solid) onto the south-polar-cap due to the year-round presence of cold CO₂ ice, and some of the atmospheric water vapor is returned to the north-polar cap during northern winter and spring (2, 9). Some water may be sublimating from the south cap during southern summer, in regions that lose their CO₂-ice cover in summer, and may contribute to the summertime increase in southern-hemisphere atmospheric water vapor (10–12). The amounts of water moving onto and off of the south cap are uncertain, as are the amounts of water returned to the north cap in winter (9). Thus, it’s unclear whether there is a net annual transfer of a significant amount of water between the two polar-caps or even in which direction any net transfer might go.

In the north, summertime sublimation certainly occurs from the exposed water ice that is identified by its relatively high albedo, its water-ice composition as determined from its reflectance spectrum, and measured temperatures that support significant sublimation (13). However, the dark lanes within the cap (Fig. 1) are thought to consist of alternating layers of dust and water ice, or of dust and ice mixed in different proportions. As it takes very little intermixed dust to make water ice appear dark (14, 15) and the bulk of the volume of the dark lanes is thought to be water ice based on its radar properties (e.g., ref. 16), the surface of these dark lanes could have a significant component of water-ice right at the surface.

At the present epoch, the peak summertime temperature of the “white” polar material is about 205 K (17, 18). The total local net sublimation, summed over the entire summer season, would result in the removal of ~100 microns of surface ice (13). At the same time, the peak summertime temperature of the dark lanes is about 236 K (17). At these higher temperatures, if the dark lanes have a significant component of water ice at the surface with only a small amount of dust contamination, the integrated summertime sublimation from a single location could be as much as 500 times higher (see ref. 19). The dark lanes comprise several percent of the surface area of the polar region; if they do consist predominantly of water ice at their surface, sublimation from them could dominate the summertime supply of water to the atmosphere. Of course, even a thin surface layer of dust will “protect” any buried water ice and reduce sublimation (20). We don’t know the surface configuration of the dark lanes in detail, however, so we don’t know the relative importance of supply of water from the white stuff versus supply from the “dark” stuff. The source of water is important in extrapolating sublimation to other epochs, as the dark lanes and bright ice will respond differently to changes in insolation.

There is significant interannual variability of the water cycle. The major year-to-year differences in the water cycle occur during southern-hemisphere summer, when different amounts of water vapor appear in the southern atmosphere (21). Fig. 2 shows the amounts of water vapor measured in the southern-hemisphere atmosphere during the southern summer season for a number of Mars years. There is maybe a factor-of-two variation between most years, consistent with interannual variability of the atmospheric circulation or small variations in the amount of water-ice exposed on the south residual cap each summer; there does not appear to be a consistent pattern of increasing or decreasing water vapor through time.

Fig. 2.

Global-average atmospheric water content as measured in different years. Solid dots are averages of measurements in the range L_s = 330 to 360° for each year, and open circles are in the range L_s = 250 to 300°. Vertical lines represent approximate two-sigma variations, between separate measurements for whole-disk observations and along latitude for spatially resolved observations. Data taken from refs. 2, 3, and 21–24.

However, Earth-based telescopic observations showed a very different behavior in 1969. During that year, the amount of water in the atmosphere during southern summer was greater than that has been observed any time since, and it was a factor of 4 to 6 greater than what was seen a few years later from Viking (Fig. 2) (1, 21). The Earth-based measurements represent essentially a global-average. A re-examination of the original photographic plates from the observations confirmed that the atmospheric water abundance was this high (21). One potential explanation for this behavior is that the south-polar-cap may have lost its CO₂-ice cover to a much greater extent either that year or in preceding years, exposing an underlying water-ice cap that then would heat up and sublimate large amounts of water into the atmosphere (see also ref. 12).

This anomalous 1969 behavior appears consistent with the geological observations of ice covering the south cap. These observations show significant variations in the cover of CO₂ ice, producing what has been termed the “swiss cheese” terrain (Fig. 3); the “holes” in the CO₂ covering expand each year, exposing more of the underlying water ice (Fig. 4). The behavior of this ice over several years suggested that an event involving a major resurfacing by CO₂ ice occurred within the last several tens of Mars years (10, 11, 25). Could the south cap have had no CO₂-ice cover leading up to the 1969 summer, with a major resurfacing event causing it to jump to a state in which it has been largely covered by CO₂ ice throughout the year? If it did, was the resurfacing event a one-time jump in the nature of the cap, or could it represent a portion of a longer cycle in which the south cap switches from covered to uncovered and then back again? And, if it is the latter, how often is the south cap predominantly covered during southern summer versus predominantly uncovered?

Fig. 3.

Mars “swiss cheese terrain” on the residual south-polar-cap. Elevated areas are covered with CO₂ ice, depressions have a water-ice floor. (NASA image from Mars Global Surveyor.)

Fig. 4.

Map of Mars south-polar residual cap, showing depth of water-ice feature in spectral reflectance. Colors go from red (ice free) to blue (deepest water-ice absorption feature). (From ref. 10.)

One possible answer comes from observations related to the presence and distribution of near-surface ground ice in the middle and high latitudes. At these latitudes, the regolith is cold enough that water vapor will diffuse in from the atmosphere and condense as ice within the top meter of regolith (26). Measurements of the thermophysical properties of the soil allow surface and near-surface temperatures to be calculated throughout the year and predictions to be made of where ice should be stable (e.g., ref. 27). Comparison can be made of the predicted ice distribution and that obtained from neutron measurements and from exposures of water ice by recent impacts. The best fit of the modeled distribution of ice to the observations occurs if the water content of the atmosphere were 2 to 3 times the present observed value (e.g., refs. 27–30). As the abundance of ground ice responds to the atmospheric conditions on the ~100 to 1000 y timescale, this could suggest that the long-term average atmospheric water content has been larger than the present value.

An alternative explanation for the ice distribution could be that the water vapor had been confined to a near-surface layer, increasing the number density at the surface and widening the region of stability. This behavior could occur if the atmosphere were unusually dust-free, resulting in colder temperatures and a confinement of the water into a near-surface atmospheric layer. In either case, such an increased average could result if Mars had roughly equal numbers of years of the two types of behavior.

Combined, there is a significant uncertainty in the average or representative atmospheric water content and in the underlying processes at the present epoch. It’s not even clear that we’ve observed the extreme types of behavior. This uncertainty will carry over into difficulty in determining the rate of loss of water to space and in extrapolating the behavior of the seasonal water cycle to other epochs, as discussed below.

2. Seasonal Dust Cycle Effects and Interannual Variability

Atmospheric dust has the potential to affect the seasonal behavior of water vapor and the loss of H and O to space. Airborne dust can affect the seasonal water cycle in several ways: i) Dust in the atmosphere absorbs sunlight, emits infrared energy, and changes the thermal behavior of the atmosphere that drives the atmospheric circulation; as a result, transport of water, energy, and dust will be affected (e.g., refs. 31 and 32). ii) Dust transported through the atmosphere can be deposited on either polar-cap, changing the albedo and affecting the polar energy balance (e.g., ref. 15). iii) Water adsorbed onto airborne dust can be transported along with the dust, redistributing water in a way that may not be observable directly and that could possibly affect the net seasonal cycle. The year-to-year variability of the atmospheric dust cycle will affect the efficacy of all of these processes.

Dust grains can hold significant quantities of adsorbed water if the dust is a heavily altered palagonite-like material (see ref. 33). Adsorbed water can constitute 10% or more of the mass of the dust if the relative humidity is 0.4 or greater, and as much as 8% for relative humidity of 0.1 (34, 35). Fig. 5 shows example atmospheric temperature and saturation–temperature profiles under clear and dusty conditions. Humidity this high can occur at altitudes above 10 km if the dust loading is low and at all altitudes throughout the lowermost 25 km at night during both clear and dusty periods, especially near the surface [(36); Fig. 5]. Dust raised into the atmosphere at the onset of dust-storms has the potential to carry with it significant amounts of water relative to the extant water vapor column. For example, atmospheric dust having a vertical opacity of 10 during the peak of the dust-storm can carry as much as several precipitable micrometers (pr μm) of water into the atmosphere, and release it during the day when temperatures are high and humidity is low; this compares to a nominal water-vapor content of the atmosphere of ~5 to 10 pr μm during southern-hemisphere summer when the dust-storms occur (2, 3). Similarly, when the dust-storm abates, the dust can carry a significant fraction of the water out of the atmosphere and back onto the surface, especially if the dust settles out at night.

Fig. 5.

Comparison of atmospheric temperature profiles with water-vapor saturation curves. Saturation curves are labeled with an integrated column of water vapor (W) in precipitable micrometers (pr μm) and are calculated assuming that the water is uniformly mixed with altitude. Models show temperature profiles assuming a dust-free atmosphere (solid lines) and a dusty atmosphere (dashed lines). Horizontal bars and plus marks show the range of temperatures measured by Mariner 9 and Viking during dust-free periods and dusty periods. (After ref. 37.)

As with water vapor, the behavior of dust as observed by spacecraft may not be representative of the present epoch. There are two observations that suggest significant variability.

First are observations of global-scale dust events. Regional-scale dust events occur in a pattern of three sequential events during the annual dust-storm season of L_s~200–345^o (38). About once every several Mars years, one of these regional events will expand to global scale. However, the 1956 global dust event, which is the largest ever observed, was more extensive and more intense than any event that had been observed during the preceding century by telescopic observers. While the observations are limited due to the variable Earth–Mars geometry, the apparent complete absence of such global-scale events prior to 1956 is striking (39).

Second, the nature of the background (non-dust-storm) level of dust loading may have changed significantly. There is a modest background level of dust in the atmosphere throughout the year, with dust opacities increasing during the dust-storm season. This pattern of increasing dust during southern spring and summer has been observed by spacecraft and is repeatable. The background dust loading is thought to come from dust raised by a combination of local and regional dust-storms that occur annually and dust devils that are now thought to be quite common [see (32, 40)]. However, this behavior may have changed during the last century. The background level of dust loading over the last century has been inferred from Earth-based telescopic observations of the Martian limb. These observations come from measurements of the size of Mars as measured telescopically, as was done from 1879 through 1958 (41). When dust is present in the atmosphere, the radius of Mars as observed from Earth represents the altitude at which the line-of-sight atmospheric opacity reaches 1. Converting this altitude to a vertical opacity, the resulting dust opacity during those years must have been ~5× lower than during the year observed by Viking and in subsequent years.

While each type of observation has its limitations and cannot be deemed definitive, both suggest that there are significant variations in the dust cycle on the decades-to-centuries timescale. Again, it isn’t clear whether this change would represent a one-time jump in behavior or whether it reflects our having observed only part of a longer cycle, and it isn’t clear what type of behavior is more typical of the present epoch. It also isn’t clear whether the abrupt changes in the water and dust behavior occurring at around the same time (1950s/1960s), if real, is coincidence or whether they represent processes that somehow couple together.

3. Escape of H to Space and Interannual Variability

The seasonal behaviors of water and dust can affect the loss of H₂O to space. Water vapor in the atmosphere can be photodissociated into the constituent H and O atoms by extreme ultraviolet photons from the Sun. After diffusing to the top of the atmosphere, the H atoms are light enough that some of them can escape to space via Jeans’ (thermal) escape. Escape of H to space is thought to have been a significant process due to the observed substantial enrichment in the ratio of D/H in water remaining behind—the heavier D escapes less readily, so that when H and D are reincorporated back into H₂O, the D is enriched relative to the H and D/H increases (42). The current atmospheric water has D/H ~5 to 6× the inferred initial Martian value (43, 44), suggesting loss of a significant fraction of the water to space. Unfortunately, the history of D/H cannot be determined easily from in situ analyses and meteoritic samples, due in part to the uncertain provenance of the water contained within them and to the ability of water to exchange with the surrounding environment and change the D/H (45; see also ref. 46).

Yung et al. (42) calculated the relative escape rates of D and H under a set of assumed nominal atmospheric conditions. Their escape rate was ~10⁸ cm⁻²-s⁻¹ which, if constant over 4.5 billion years, would result in loss of hydrogen from an amount of water equivalent to a global layer of water ~3.6 m thick (i.e., a global equivalent layer, GEL, of 3.6 m). The relative loss rates of H and D are such that loss of this amount of water loss could produce the observed D/H if ~95% of the water was lost; this would leave behind a residual of ~20 cm H₂O.

However, the situation is more complicated than this nominal atmospheric behavior would suggest, for a number of reasons:

(i) Exchange of atmospheric water with a non-atmospheric reservoir of water will serve to buffer the D/H, mitigating the enrichment resulting from escape. The largest exchangeable non-atmospheric reservoir is the north-polar water-ice cap and mid- and high-latitude ground ice, containing ~20 to 30 m GEL H₂O (47). This water is thought to exchange with the atmosphere on timescales of ~10⁵ and 10⁶ years (48), and the cap itself may not have been present prior to ~4.5 Mya (49). Thus, if exchange does occur and if the ice in the cap has D/H similar to that of the atmosphere, then the estimate of loss of 95% of the water is not consistent with the estimated loss rate and would have required a substantially greater loss rate earlier in history (50). For instance, if the cap represents the 5% residual after loss of 95% of the water, then as much as 500 m GEL H₂O must have been lost to space. Clearly, this result depends on the currently unknown D/H and exchange rate of the polar ice. In addition, while the water vapor sublimated into the north-polar summertime atmosphere does have D/H similar to the atmosphere (51), the D/H throughout the full volume of polar ice is unknown.

(ii) The estimated escape rate does not account for the presence of a “hot” component of H in the Martian exosphere. Hot H (having energy greater than thermal energies) could result from the impact of energetic solar wind or exospheric atoms or from photochemical reactions. Evidence for a hot component comes from Mars Express, HST, and MAVEN IUVS observations of the extended corona of H surrounding the planet (52–55). Having an energy greater than thermal means that the escape rate for H would be greater than predicted just from the average temperature. Actual escape rates could be as much as twice the calculated thermal rate and, importantly, the efficiency of D loss relative to H loss would be larger (56). As a result, getting to the observed D/H would require loss of a greater amount of water. It is not obvious what the importance of escape of hot H would be at other epochs, so that extrapolations of loss and of the effects on D/H would be uncertain.

(iii) The abundance of H in the exospheric corona that supplies escaping H is not a constant. It varies by an order of magnitude and likely results in an order-of-magnitude variation in the escape rate (53, 57–59). As observed now through multiple annual cycles, the trend appears to be seasonal, with the greatest abundances occurring during the southern-hemisphere spring and summer (e.g., refs. 55, 60, and 61).

(iv) The seasonal cycle of airborne dust can affect the atmosphere. The background level of atmospheric dust increases substantially during southern spring and summer, and regional dust-storms that can expand into global events occur preferentially during this same season. The increase in atmospheric temperature resulting from the presence of dust allows water vapor to be carried to higher altitudes. There, it can be photodissociated more easily, plus it is closer to the exobase altitude from which escape occurs. These effects result in an increase in the abundance of H in the corona surrounding the planet above the collisional atmosphere (62). The coronal abundance increases by a factor of ten during the southern-summer season and increases by another factor of several during regional dust events (63, 64). In addition to the increased temperatures in the lower atmosphere, increased upper-atmosphere temperatures, as occur during parts of the solar cycle, affect the loss rate (65).

(v) The H escape rate also appears to be connected to the rate of loss of O that is also released from H₂O (e.g., ref. 66). Oxygen escape to space is driven by nonthermal processes, involving processes and reactions that are separate from those driving H escape. If only H escaped, O would build up in the atmosphere and, through a series of reactions, would catalyze the reformation of H₂O and thereby reduce the H escape rate. The net effect would be loss of H and O in the ratio of 2:1, so that it drives an overall loss of H₂O (67). The timescale for H and O to reach equilibrium is 10³ to 10⁵ years, so that variations within a single year or between years can occur without the H and O reaching equilibrium. As an example, a sudden increase in the lower-atmosphere H₂O abundance initially would result in a significant increase in the H escape rate; on 10³- to 10⁵-year timescales, however, the H escape rate would decline again to match the O escape rate. Observations of the H and O escape suggest that they are near the 2:1 ratio that would effectively remove water (68).

(vi) Finally, atmospheric O can come from photodissociation of either H₂O or CO₂, such that the history of O is also tied to the history of C (69); the rate of loss of C to space at the present epoch is not yet well understood (e.g., ref. 70). In addition to being lost to space, oxygen also can react with surface minerals and oxidize the surface (e.g., ref. 71), and carbon can react with the surface to form carbonate minerals (72). Thus, the history of O and C will be intimately connected to the timing of the oxidation of the surface and of carbonate formation. Clearly, understanding the current H escape rate requires understanding the mechanisms for removing O and even C from the atmosphere, both to space and to the surface and subsurface.

The combined uncertainties in the seasonal water cycle and dust cycle preclude stating with any certainty whether H and O are escaping in the ratio of 2:1 or even what the average or representative escape rate is at the present epoch. In particular, the observations of both the water and dust cycles made by spacecraft may not have shown the full range of behaviors or even a representative or average behavior for the present epoch. In addition, the dependence of the H escape rate on the water and dust behavior, both of which can have significant interannual variability, means that we do not know the average escape rate at the present epoch. A single best-estimate value could be off by as much as an order of magnitude.

4. Orbital Element Variations

The Martian climate is thought to respond to temporal changes in the orbital elements, in particular its axial obliquity, on timescales longer than 10⁴ years. The axial obliquity of Mars currently is 25.2°; this is the angle between the polar axis and the normal to the plane of Mars’ orbit around the Sun. (For comparison, the Earth’s obliquity is about 23.5°.) The Martian obliquity varies due to the gravitational pull primarily of Jupiter (e.g., refs. 73 and 74). The obliquity variations have natural frequencies of about 10⁵ and 10⁶ years, and the value can oscillate about its average by as much as 15°. On timescales longer than 10⁷ years, however, the “guiding center” of the obliquity varies chaotically and the obliquity cannot be predicted (73, 75). Values can range to as low as 0° or as high as about 70°. Importantly, the most likely value statistically is about 42°, significantly larger than the present 25.2^o (74).

As the tilt varies, it changes the amount of solar energy incident on the polar-caps, affecting both the seasonal CO₂ cycle and the summertime supply of water vapor to the atmosphere (e.g., ref. 19). We expect summertime polar-cap temperatures to be higher during periods of high obliquity when the Sun will be higher above the horizon, resulting in increases in the amount of water sublimated through the summer season. When the obliquity is 42°, for example, as much as 100x as much water can sublimate into the atmosphere; the integrated summertime sublimation can be another factor of 10 higher at 60^o (19). The amount of water sublimated is difficult to predict accurately, however, as we don’t know what happens to the albedo of the ice, the composition (CO₂ or H₂O ice and intermixed dust) of the residual polar-caps, or the relative contributions of sublimation from the exposed water ice or from possible ice in the dark lanes at high obliquity.

We also don’t understand the mechanisms driving regional and global-scale dust events at the present epoch—we don’t know what triggers them, why we only get planet-encircling events in some years, or whether there really is a decades-to-centuries-long cycle driving dust events (32). At higher obliquity, the massive deposits of CO₂ ice locked up below the surface in the south-polar ice cap would presumably sublimate into the atmosphere, doubling the atmospheric pressure (76). It’s not clear what this will do to the occurrence rates of dust-storms—whether the rate or the intensity will stay the same, increase, or even decrease.

Given the uncertain water content and dust behavior of the atmosphere at higher obliquity, combined with the role that these seasonal cycles play in the escape to space of H from H₂O, it’s not possible at this time to determine what the rate of loss to space will be at higher obliquities. We expect that the loss of H will be greater at higher obliquity due to the higher atmospheric H₂O abundance, but the dependence of the H loss rate on the O loss rate, and the lack of an identified mechanism for the O loss rate to vary with obliquity, makes it difficult to predict what the loss will be.

Making it even more difficult, the chaotic nature of the variations in obliquity means that we cannot know the history of the obliquity. Even if we did know precisely how the loss rate varied with obliquity, we would not be able to determine the integrated loss through time. Intuitively, however, we expect that the average obliquity has been higher than the current value and that this has driven the loss of H to space at a rate greater than the current loss rate.

5. Longer-Timescale Geological Influences

On billion-year timescales, geological influences can play a significant role in the evolution of surface and near-surface water. Of particular relevance are outgassing from the mantle associated with volcanic activity, hydration of minerals within the crust, the percolation or diffusion of water into the crust and segregation there, and the multiple geological processes involving water that occurred on early Mars. The impact of these processes on D/H depends on the timing and abundance of permanent loss of H₂O from the surface/atmosphere environment and on the ability of water to exchange with non-atmospheric reservoirs of H₂O (42, 45, 50, 77, 78).

Outgassing of juvenile water will increase the amount of water at the surface and partially reset the D/H of the water toward mantle values (47). The water content of the mantle has been inferred from measurements of H₂O and Cl in the Martian meteorites; estimates go up to about 250 ppm (e.g., re.f 79), with mantle D/H being close to terrestrial (80, 81). As the upper-mantle rock partially melts prior to the eruption of magma to the surface, water can diffuse into the melt from surrounding rock, increasing the abundance in the magma relative to that in the mantle. Once emplaced into the crust or onto the surface, some fraction of the water in the magma/lava will be released into the atmosphere.

During the last 3 b.y., volcanic outgassing and escape to space are likely to have been the dominant processes affecting water on Mars (82). Based on the global D/H value inferred for Mars at about 3 b.y. ago, these processes appear to have controlled the evolution of D/H at the surface.

Integrated over time starting from the time of onset of the Martian geologic record in the late Noachian epoch, however, volcanic outgassing does not appear to be able to explain the amounts of water seen at the surface. Based on the volume of volcanic material identified at the surface and on the derived abundance of water in the mantle, outgassing can have supplied no more than ~40 m H₂O GEL over time (83). This compares to estimates of the surface water inventory that range between ~400 and 2,000 m GEL (45, 77). Instead, the bulk of the water must have been released to the surface earlier in Martian history, probably as a result of planetary formation, core formation, and early crustal formation. Subsequent evolution then would have been dominated by a drawdown through interaction with the crust and loss to space. Of course, these estimates are uncertain due to the wide variation of derived water abundances in the mantle, whether they reflect uncertainty in the analysis techniques or variable composition within the mantle, and uncertainties in the transition from mantle water to surface water via volcanism (83).

Minerals have been identified at the surface that incorporate water into their structure, as H₂O or OH (e.g., ref. 71). These minerals include Fe, Mg, and Al phyllosilicates or hydrated silica, for example. The minerals are typically present in the subsurface, having been exposed at the surface by later erosional processes. Typically, they formed early in Martian history, although some formed as late as the late Hesperian, ~3.0 b.y.a., and even into the Amazonian (71, 84).

Enough is now known about the distribution of these minerals that an estimate can be made of their total abundance based on the geological context and the inference of their abundance in places where they cannot be identified directly (85, 86). In turn, this allows an estimate to be made of how much water they contain (87, 88). Although a global abundance as high as 500 m H₂O GEL could not be categorically ruled out, the best estimate was in the range of 130 to 260 m H₂O (87). When spectroscopic mapping can be done at higher spatial resolution, it is likely that additional deposits of hydrated minerals will be identified (86); thus, this estimate probably constitutes a lower limit on the abundance.

Finally, we consider the possible presence of either free liquid water or solid ice residing in the pore space within the crust and the megaregolith. The geological argument for the presence of water in the crust goes back to Carr (89). He calculated the amount of water required to carry the debris released to form the outflow channels as being a minimum of 35 m H₂O GEL. He argued that the flooding appeared to have drained water from roughly a tenth of the Martian crust and that the remaining 9/10ths of crust also should contain comparable amounts of water. This gave a minimum of an additional 315 m H₂O that would still be present in the crust. An upper limit comes from the estimated pore volume of the crust down to the self-compaction depth and is ~1,000 m H₂O (90). Analysis of the subsurface temperatures expected for the likely geothermal gradient suggests that any water would be in the form of ice near the surface, but could be liquid at depths below a few kilometers due to geothermal heating (90). Interestingly, at least one test of the global groundwater hypothesis has failed to turn up evidence for liquid water—Ejecta from the 219-km-diameter impact crater Lyot, which easily should have penetrated to a depth that would access groundwater, failed to show any evidence for water having been present (91). Other evidence comes from the detection of near-surface H in the top meter of the mid- and high-latitude regolith, inferred to be present largely as water ice (e.g., ref. 92). The near-surface ice is in diffusive equilibrium with the atmosphere. Unfortunately, it is not known how deep this ice extends—whether it is only a near-surface layer or extends to much greater depth.

There are no observations of the deep crust that tell us directly whether any water or ice actually is present there. For the time being, we have to take this as a potential reservoir containing an unknown amount of water.

Finally, consider the geomorphological analyses that suggest that abundant surface water was present during the Noachian epoch and into the Hesperian. Features indicative of long-lived liquid water include valley networks that are thought to have been formed by water runoff, large-scale erosion of ancient surfaces that degraded features such as impact craters smaller than about 15 km scale, and degradation of larger impact craters that show spur-and-gully topography indicative of erosion by surface water (e.g., ref. 93, and references therein). While the processes involved in producing the geomorphological features are clear, the amounts of water involved are uncertain due to the lack of knowledge about the early climate and the efficacy and robustness of recycling of water via an early water cycle.

The outflow channels debouched into the northern lowlands, and it is almost inescapable that the water would have ponded there and formed at least a temporary large lake or ocean (e.g., refs. 47, 94, and 95); and references therein). The amounts of water involved could have been as large as 100 m GEL. The water most likely evaporated or sublimated into the atmosphere and became part of the inventory of water involved in polar and atmospheric processes (94, 95). Importantly, the large amounts of water involved, and the evidence for abundant liquid water in the Noachian and Hesperian epochs, point toward a different type of climate behavior than what we see at the present epoch. The climate may have been more analogous to the present-day terrestrial climate than to the present-day Martian climate. The implications of such a climate for the evolution of water have not been explored in detail.

These sources, sinks, and reservoirs for water are, in the end, difficult to quantify uniquely, and the ramifications for the evolution of D/H and the amounts of water lost are somewhat model dependent. Clearly, though, large amounts of water have been involved, much more than is present today near the surface and in the polar ice caps. One approach is to identify how much water has been lost to each sink and exchangeable reservoir, consistent with the observational constraints. This approach has been taken in the modeling by Scheller et al. (77). Their best estimate is that Mars had between 100 and 1,500 m GEL H₂O and that, of this, as much as 530 m GEL has been lost to space with the bulk of the rest having gone into crustal hydration. Jakosky and Hallis (45) took a different approach of quantifying how much water can be identified today in each known sink, based to the maximum extent possible on the observations. They estimated that Mars initially had 400 to 2,000 m GEL H₂O and that about 100 to 500 m has been lost to space.

These estimates differ from other recent estimates of the Martian water inventory, such as those by Carr and Head (47, 94). Part of the difference is semantic, with the current estimates including water that hydrated minerals during the Noachian while the earlier estimates explicitly excluded it. Part is real, based on having an updated understanding of the loss of H to space and the fractionation of D/H, the amount of water potentially released through volcanic outgassing, and the role played by early catastrophic outgassing of water. In the end, it’s not possible to better quantify the amounts of water involved given the currently available data.

6. The History of Water— What We Don’t Know and What We Do Know

Evidence suggests that a significant amount of volatiles has been lost to space and that this loss has contributed to the evolution of water on Mars. Measurements of the ratio of D/H in Martian water provide a very powerful way to show that loss to space has been significant. However, uncertainties in how the various processes affect water loss today and did so in the past preclude obtaining a unique determination of the amount lost. Specific issues are

• •The amount of water vapor present in the atmosphere and averaged over, say, the last thousand years is not known to within a factor of perhaps several. As this water is the source for H that escapes, the escape rate at the present epoch (as distinct from “this year”) may be equally uncertain.
• •The behavior of atmospheric dust on timescales longer than the last few decades is uncertain, in terms of both the behavior of global-scale dust events and the background loading of dust during southern-hemisphere summer. As the presence of dust can increase the H escape rate, the average H loss rate and its effect on D/H at the present epoch are uncertain.
• •The H escape rate is tied at least in part to the O loss rate. We don’t understand how the coupling between them would vary as the H₂O abundance varies—whether the loss rate today is in the ratio of 2:1, whether it would be in the ratio of 2:1 under different boundary conditions, or how all of the processes involved in the chain leading from H₂O to escape change as the H₂O or dust abundances vary with time.
• •Although the water content of the atmosphere should increase significantly during periods with higher obliquity, calculations of how much water will sublimate from the polar ice caps during summer are very uncertain. The additional H and O in the atmosphere should enhance loss to space, but the relevant processes have not been explored.
• •The behavior of atmospheric dust at higher obliquity is uncertain as well. As we don’t understand its behavior at the present epoch, it’s hard to extrapolate to periods when the atmospheric pressure would double and the solar forcing of the atmosphere as a function of latitude would differ. Would dust-storms occur more often, would they be more intense? As the H loss rate responds to the radiative effects of airborne dust, predicting the loss rate at high obliquity is difficult.
• •Uncertainties in the timing and amount of water sequestered into the crust and regolith, and the detailed location of any water, make it difficult to determine the size of the exchangeable non-atmospheric reservoir for water through time; in turn, this translates into uncertainties in the interpretation of D/H.

Well, then, what do we know and what can we say with some certainty about the history of water?

• •Significant quantities of atmospheric gas have been lost to space through time. The relative enrichment of the heavier isotope in all of the light stable isotopic pairs of atmospheric gas (D/H, ¹⁵N/¹⁴N, ¹³C/¹²C, ¹⁸O/¹⁶O, ³⁸Ar/³⁶Ar) can best be explained by significant loss to space.
• •The observed ratio of D/H requires that a large fraction (>85%) of the “exchangeable” water must have been lost to space; the exact fraction depends on details of the loss process, but could be 95% or more. Translating this fraction into an absolute amount of water lost requires assumptions on i) the D/H of non-atmospheric reservoirs of water such as the polar-caps, ii) the timing and amounts of exchange with non-atmospheric water, and iii) the amounts of water lost permanently and the timing of permanent loss to non-atmospheric reservoirs. Reasonable assumptions require loss of a minimum of ~100 m H₂O and possibly as much as 500 m.
• •Adding in water that has been lost to other sinks (polar-caps, ground ice, hydrated minerals, free water or ice in the crust), Mars could have had between a few hundred and nearly 2,000 m GEL H₂O. This amount of water appears to be consistent with the amounts required to have formed the geological features on ancient surfaces that have been attributed to water and ice.

The history of water inferred from the present-day water cycle and behavior of water and from the geological evidence on long-term behavior appear to be consistent with each other. However, there are uncertainties in both approaches, in terms of the absolute amounts of water involved and the exchange between different reservoirs and sinks. These preclude determining a single unique history for water or for the amounts of water that have been at the surface at this time. This uncertainty should not keep us from concluding that Mars had a warmer and wetter surface environment early in its history, that loss to the multiple sinks can account for where the water has gone, or that Mars had an extremely complex behavior of water throughout its history.

Data, Materials, and Software Availability

All study data are included in the main text.