# Long-term drying of Mars by sequestration of ocean-scale volumes of water in the crust

## Abstract

Geological evidence shows that ancient Mars had large volumes of liquid water. Models of past hydrogen escape to space, calibrated with observations of the current escape rate, cannot explain the present-day D/H isotope ratio. We simulate volcanic degassing, atmospheric escape, and crustal hydration on Mars, incorporating observational constraints from spacecraft, rovers and meteorites. We find ancient water volumes equivalent to a 100- to 1500-meter global layer are simultaneously compatible with the geological evidence, loss rate estimates, and D/H measurements. In our model, the volume of water participating in the hydrological cycle decreased by 40 to 95% over the Noachian period (~3.7 to 4.1 billion years ago), reaching present-day values by ~3.0 billion years ago. Between 30 and 99% of Martian water was sequestered by crustal hydration, demonstrating that irreversible chemical weathering can increase the aridity of terrestrial planets.

There is abundant geomorphological evidence for large volumes of surface liquid water early in Martian history (1), with estimated volumes equivalent to a ~100 to 1500 m deep global equivalent layer (GEL) (14). Liquid water on Mars decreased over geological time; presently most water is stored in the polar ice caps or as subsurface ice. Estimates for the total modern water inventory, in the atmosphere and as ice, total 20-40 m GEL (58). The availability of water to participate in the hydrologic cycles of terrestrial planets is expected to influence their climate and habitability. However, the processes that caused the decline of available water on Mars are poorly constrained.

Previous studies have suggested that Mars experienced substantial water loss due to atmospheric escape, supported by the current atmospheric deuterium to hydrogen isotope ratio (D/H) of 5-10 × SMOW (Standard Mean Ocean Water on Earth; D/H at 1 SMOW is 155.76 × 10−6) (5, 911). The D/H value at ~4 Ga was 2-4 × SMOW, inferred from Martian meteorites (fig. S1) (12, 13). Existing models used these observations, combined with assumed atmospheric escape fractionation factors (αescape) of 0.016-0.32 during loss, to estimate integrated atmospheric escape of at least 10-200 m GEL (fig. S1) (4, 5, 11, 14, 15). These estimates imply an initial 50-240 m GEL water on ancient Mars, consistent only with the lower range of geological estimates (100-1500 m GEL) (14). This has been interpreted as implying a large, unknown reservoir of water on present-day Mars (4).

For present-day Mars, the rate of atmospheric water loss is measured from the H escape flux, because water vapor dissociates and its hydrogen escapes. Spacecraft measurements of the current H escape flux, 1 to 11 × 1026 H atoms s−1, are equivalent to escape of 3-25 m GEL water over 4.5 Ga (16, 17) and cannot explain all the water loss. Another potential water loss mechanism is crustal hydration through irreversible chemical weathering, in which water and/or hydroxyl are incorporated into minerals. Orbital and in situ data show that widespread chemical weathering has produced a substantial reservoir of hydrous minerals on Mars, potentially totaling hundreds of meters GEL in the crust (5, 18). We hypothesize that crustal hydration during the first 1-2 billion years decreased the volume of the hydrologically-available water reservoir, followed by subsequent atmospheric loss that fractionated the Martian atmosphere to its current observed D/H ratio. We simulate water loss through geological time to constrain Mars’ water history and compare the simulations to D/H data from the Curiosity rover (5) and laboratory analyses of Martian meteorites (fig. S1) (12, 13, 19, 20, 21).

## A hydrogen isotope water reservoir model

We developed a water budget and D/H model that integrates water sinks and sources, including crustal hydration, volcanic degassing and atmospheric escape (Fig. 1) (5). Most previous models included only atmospheric escape (4, 11, 14); one model (15) also included volcanic degassing. We treat liquid water, ice, and atmospheric vapor as a single exchangeable reservoir, an isotopic modelling technique originally developed for carbon reservoir models (22). We assume that liquid and solid phases, not vapor, dominate the exchangeable reservoir and fractionation between them is negligible [the fractionation factor is αice-liquid = 1.02 (23)]. Our simulations are constrained such that the exchangeable reservoir can never be negative and must reproduce 20-40 m GEL water today. The initial exchangeable reservoir size (Xex,0), i.e., the ancient hydrologically available water inventory, is a free parameter except during sensitivity analyses. We determined permitted ranges of source and sink fluxes for crustal hydration (Fcrust), volcanic degassing (Fvolcanic), and atmospheric escape (Fesc) during the Noachian (~4.0 – 3.7 Ga), Hesperian (~3.7 – 3.0 Ga), and Amazonian (~3.0 Ga to present) periods of Martian geological history following observational and previous model constraints (Fig. 1 and table S1) (5). Models were evaluated by their ability to reproduce the D/H ratio of the present-day exchangeable reservoir (Rex,end) of 5-10 × SMOW. We also compare our simulation results to a compilation of Curiosity rover Sample Analysis at Mars (SAM) data sets that record a D/H composition range of 3-5 × SMOW for gas released from Hesperian samples during high temperature (>374°C) combustion experiments (5).

We calculate a permitted range of Fcrust from measurements of H wt.% in Mars surface materials and global remote sensing observations of hydrated minerals. The mass fraction of crustal H2O is based on rover measurements from Gale crater, orbital global infrared and neutron spectrometer data, and measurements of the NWA 7034 Martian meteorite (0.5-3 wt% H2O) (5). The volume of the crustal reservoir is based on orbital measurements of clay exposure depths in the Valles Marineris canyon and craters 5-10 km in depth (5, 18). We adopt permitted ranges of 100-900 m GEL of water in Noachian-aged crust and 10-100 m GEL in Hesperian-aged crust based on this analysis (table S1) (5, 18). Although Fcrust is based on observations of hydrated minerals, we consider crustal water as a single reservoir representing any combination of ice and liquid water, formerly participating in the hydrologic cycle, that now no longer exchange isotopes with the exchangeable reservoir. Fvolcanic is determined using previous thermochemical models of the Martian mantle (24). Different parameterizations of those models (24) predict outgassing of 10-120 m GEL H2O from volcanic processes since 4.1 Ga (5, 24). We consider Noachian and Hesperian Fesc values between 1025 and 1030 H atoms s−1 and adopt the measured current escape rate of 5×1026 H atoms s−1 for the Amazonian (table S1) (5). These escape fluxes are compared to simulations using the 1D photochemical model Kinetics (25, 26) with adopted past solar extreme ultraviolet flux, variable atmospheric pressures, mesospheric and surface temperatures (table S2) (5).

## Controls on D/H and water loss

In our model, step-wise mixing between the exchangeable reservoir and the depleted volcanically outgassed water vapor (0.8-2×SMOW) (fig. S1 and table S1) (5, 19, 27) causes the D/H of the exchangeable reservoir to decrease (5). We do not include fractionation associated with degassing or its redox sensitivity as these are negligible compared to the large range of potential D/H compositions of the volcanic gas inferred from meteorites (5). Atmospheric escape causes D/H of the exchangeable reservoir to fractionate toward heavier values, which we modeled through stepwise Rayleigh distillation, a common isotopic reservoir modelling technique, at each 10-Myr time step with a fractionation factor of atmospheric escape (αescape) of 0.002-0.32 (2830). The fractionation factor between smectite, the most common hydrated mineral found on Mars, and water [

$αsmectite-H2O=0.95$

(5)] is used in the stepwise Rayleigh distillation model as a first order approximation of fractionation by crustal hydration (table S3) (5); we find this fractionation is minor compared to that caused by atmospheric escape.

The D/H of the exchangeable reservoir increases during the Noachian in all our simulations, and through the Hesperian in most of them, due to a combination of crustal hydration and atmospheric escape (Figs. 2 and 3). Higher Fesc,N and Fesc,H increase D/H fractionations of the exchangeable reservoirs (Fig. 2, A and B). We find that the Noachian and Hesperian H escape flux ranges that satisfy the model constraints (fig. S2) have a wide allowable range, ~0.1-1000× the current 5×1026 H atoms s−1 escape flux. Independently, our Kinetics photochemical simulations (5) produce the same range (~1025 to 5×1029 H atoms s−1) (fig. S3). We considered multiple scenarios including (i) a range of standard ancient Mars conditions, (ii) high-altitude water injection (60 ppm at 100 km), and (iii) fixing a surface H2 mixing ratio of 10−3, higher than present-day levels of 10−5 (26). The maximum Kinetics-permitted escape flux (~5×1029 H atoms s−1) and our D/H model maximum permitted flux (4×1029 H atoms s−1) match the diffusion-limited escape of 5×1029 H atoms s−1 we calculate, using equations from (31). The injection of high-altitude water and increased surface H2 concentrations both increase the production of high-altitude H2; one or both would be required for loss fluxes 100-1000 × higher than present (fig. S3).

Crustal hydration during early Mars history also increases D/H fractionation of the exchangeable reservoirs, with the permitted range of Fcrust,N depending on the assumed Fcrust,H (Fig. 2C). This is primarily because higher Fcrust,N decreases the exchangeable reservoir size, not because of the fractionation [

$αsmectite-H2O=0.95$

(5)] associated with clay formation. As the exchangeable reservoir is reduced through crustal hydration, less atmospheric escape is needed to produce the modern D/H of the atmosphere. During the Noachian, decreasing exchangeable reservoir size and increasing D/H are a feature of all our simulations. Changes to the assumed timing of the boundary between the Noachian and Hesperian (tN-H) and balance of Fcrust,N to Fcrust,H only slightly affect the Noachian D/H fractionation (Figs. 2C and 3C). During the Amazonian, the exchangeable reservoir size is low, and its D/H increases slightly in all our simulations due to the lack of crustal hydration, low H escape flux (assumed equal to the present rate), and a low volcanic degassing flux (Figs. 2 and 3). In contrast, the D/H evolution during the Hesperian is less well constrained because models with low total volcanic outgassing (10-20 m GEL) result in D/H increases while models with high outgassing (60-120 m GEL) result in D/H decreasing or staying approximately constant (Fig. 3, A and B). The amount of volcanic degassing controls the required sizes of Fcrust and Fesc for different Xex,0 to reproduce the present-day D/H ratio (Rex,end) (figs. S4 to S6). Evolution of Hesperian D/H is also sensitive to the absolute timing of the debated (5) boundary between the Hesperian and Amazonian periods (tH-A) because in our model that boundary sets the hydration and volcanic flux magnitudes (Fig. 3C).

## Crustal hydration as a water sink

Considering the simulations over our whole parameter space, we find that the amounts of water lost through crustal hydration and atmospheric escape vary in ratios ranging from 3:8 to 99:1 (Fig. 4 and figs. S4 to S6), equivalent to ~30-99% of initial water being lost through crustal hydration (5). The maximum proportional contribution of atmospheric escape occurs when the volume of the crustal water reservoir is minimum and vice versa. Any larger proportional escape would produce D/H heavier than the present-day observed value (> 10 × SMOW). However, the absolute allowed volumes of integrated crustal hydration and atmospheric escape are dependent on the size of the initial exchangeable reservoir (figs. S4 to S6). For some of our model solutions, no difference in the average atmospheric escape flux relative to the present-day flux is required to account for the observed increase in D/H and decrease in the exchangeable water reservoir (Fig. 4 and figs. S3 and S4). Both the maximum and minimum escape-to-space cases (Fig. 4 and figs. S4 to S6) occur with intermediate assumed initial exchangeable reservoir volumes (~500 m GEL).

Accounting for water loss by both crustal hydration and atmospheric escape (figs. S4 to S6) resolves the apparent contradiction between the estimates of integrated H escape, the D/H ratio of present-day Mars, and geological estimates of a large, ancient exchangeable reservoir (1, 4). These can be reconciled because the amount of atmospheric escape needed for the atmosphere to reach the present-day D/H ratio is reduced by the removal of large initial water volumes via crustal hydration. Our models require larger Noachian exchangeable reservoirs (100-1500 m GEL) than previous work (50-240 m GEL) because we include crustal hydration (Fig. 4F). The whole parameter space allows for initial exchangeable water reservoirs of 100-1500 m GEL at 4.1 Ga, 20-300 m GEL at the Noachian/Hesperian boundary, and a near-constant 20-40 m GEL throughout the Amazonian (Fig. 4F). We chose a preferred solution based on observational constraints on the parameter space (Table 1 and Fig. 4F). In this preferred simulation, the Noachian and Hesperian H escape fluxes are twice that of today, i.e., Fesc,N= Fesc,H ~ 1027 H atoms s−1. The Kinetics simulations indicate that the most probable long-term H escape flux was similar to today, though there may have been shorter duration enhancements e.g., during dust storms or surface fluxes of H2 from geologic processes (5) (figs. S2 and S3). In the preferred model, crustal hydration removes 500 m GEL and 50 m GEL during the Noachian and Hesperian, respectively, corresponding to roughly 3 wt% H2O in Noachian crust of 5 km thickness and 1 wt% H2O in Hesperian crust of 1 km thickness (18). This is compatible with the range of present-day water contents and crustal reservoir depths measured from orbit and rovers (5). Fvolcanic is assumed based on volcanic degassing simulations (24) which themselves assumed fmantle = 100 ppm based on meteorite measurements (5). This is compatible with observational constraints on crustal production rates and water contents of Martian meteorites (5). Our preferred simulation is therefore similar to the minimum escape case shown in Fig. 4C. These simulations adopt Rex,0 = 4 × SMOW based on meteorite measurements (5) and produce a present-day D/H ratio of ~5.3 × SMOW.

Table 1 Summary of parameters assumed or calculated in our preferred scenario.

We list the assumed parameter values for our preferred simulation (Fig. 4F) and our reasoning for each choice. This preferred simulation reproduces a D/H composition ~5.3 × SMOW for the present-day atmosphere and an initial exchangeable reservoir size of ~570 m GEL.

## Consequences for Mars evolution

If the planet accreted with 0.1-0.2 wt% water (32), the large Noachian exchangeable reservoirs predicted by the model are consistent with Mars primordial water volumes. A Martian primordial volume of >1100 m GEL (potentially thousands of meters GEL) could have been produced by catastrophic outgassing of the mantle (~500-6000 m GEL) (33, 34), delivery of water through impacts (600-2700 m GEL) (35), and/or capture of gasses from the protoplanetary disc (36). However, the high hydrogen loss rates indicated by the D/H ratio at 4.1 Ga recorded within meteorites (4, 11) and possible evidence for hydrodynamic escape in Xe isotopes (37) suggest that a large part of the primordial atmosphere and water were lost during the pre-Noachian period. Our proposed volumes of 100-1500 m GEL during the early Noachian are within the lower end of these predicted primordial volumes and would therefore be compatible with the loss of a large part of the primordial atmosphere. Following loss of the primordial atmosphere, isotope measurements of C and Ar suggest that loss of a large fraction of these elements from the remaining Martian atmosphere and the reservoirs that exchange with the atmosphere would have occurred after 4.1 Ga (19, 3739). This matches our proposed trajectory of water loss within the exchangeable reservoir, which is reduced by 80-99% after 4.1 Ga within our model simulations.

Our modeled initial reservoirs are also consistent with geological estimates of Noachian and Hesperian surface water volumes. A 100-150 m GEL ocean during the Hesperian (1, 40) has been suggested from geomorphological observations and is compatible with our preferred simulation. A larger 550 m GEL ocean that has been suggested at the Noachian/Hesperian boundary (3) is possible in simulations where Fcrust and Fesc are both maximized in the Noachian and Hesperian, requiring the initial exchangeable water reservoir at 4.1 Ga to be ~1500 m GEL (Fig. 4F). Even larger oceans of 1000-1500 m GEL have been proposed based on geomorphology (1, 2); these would be permitted only in certain simulation scenarios during the early Noachian and not later epochs (Fig. 4F).

Our models are compatible with the major observed trajectories of Martian climate. A high volume Noachian exchangeable reservoir is consistent with geomorphological evidence for large volumes of Noachian surface waters and observed widespread hydrated mineral formation. Aqueous alteration of the crust could have produced periods of warmer and wetter climates (supplementary text) (4143) through accumulation of H2 in the atmosphere (figs. S4 to S6). In cases where atmospheric escape dominates water loss over the crustal hydration sink, H loss could be balanced by atmospheric oxygen escape (18-58 m GEL) and crustal oxidation (~30-380 m GEL) (supplementary text). However in cases where crustal hydration dominates water loss, short-term accumulation of H2 could have occurred (supplementary text). In our Kinetics simulations, the accumulation of H2 in the atmosphere results in increased H escape flux (fig. S3) (5).

The permitted parameter space of our D/H model allows either (i) a Hesperian exchangeable reservoir that was initially large but smaller than the Noachian reservoir (≤300 m GEL) and decreased or (ii) a Hesperian reservoir that was similar to present-day levels of 20-40 m GEL (Fig. 4F). In case (i), the Hesperian may have had sustained periods of warm and wet climate, which could have caused chemical weathering on a global scale and potentially formed an ocean (1, 40). In case (ii), the Hesperian climate was likely similar to the Amazonian climate with the exception of few local and short-lived instances of surface liquid water reservoirs (44). During the Amazonian period, the low H escape flux and low volcanic degassing flux counter each other, producing low model water availability within the exchangeable reservoir consistent with geomorphological and mineralogical evidence of an arid climate (Fig. 4F) (31, 45).

Crustal hydration would produce a buried water reservoir with composition reflecting the Noachian exchangeable reservoir of ~2-4 × SMOW. Martian meteorites with ages of 1.6 to 0.1 Ga have D/H values of ~2-3 × SMOW (20, 21). Previously proposed explanations include a distinct subsurface fluid reservoir, mixing between low D/H igneous and high D/H present-day atmospheric material, or terrestrial contamination (20, 21). We suggest that exchange between younger igneous rocks and fluids derived from hydrated Noachian (~2-4 × SMOW) crust could account for the intermediate D/H in these meteorites.

## Comparative planetary evolution

We conclude that the increasing aridity of Mars over its history was caused by the sink of chemical weathering of the crust (Fig. 4), recorded in the widespread Noachian hydrated minerals on the planet’s surface (18). On Earth, crustal hydration also occurs but plate tectonics enables recycling of crustal water that is eventually outgassed to the atmosphere through volcanism (46). This has facilitated sustained participation of water in the hydrologic cycle throughout geological history on Earth (46). The ancient age of most hydrated minerals (43) indicates that any such recycling did not persist on Mars. Irreversible chemical weathering therefore plays a role in regulating the habitability of terrestrial planets, by controlling the timescales of sustained participation of water in the hydrologic cycle.

Our model makes testable predictions for D/H measurements of the rock and ice record (Figs. 2 and 3): a substantial long-term secular increase in D/H over the Noachian and potentially Hesperian with little change over the Amazonian. Under a variable climate, our model also indicates the geological record might contain evidence of short-term D/H cyclicity: transient warm periods with greater atmospheric H2O (42) would periodically increase crustal hydration and escape flux, rapidly increasing D/H, whereas during cold periods, the D/H would decrease or increase slowly, depending on the balance between volcanic degassing and atmospheric escape.

## References and Notes

1. Materials and methods are available as supplementary materials.
2. J. F. Mustard, Sequestration of volatiles in the Martian crust through hydrated minerals: A significant planetary reservoir of water, in Volatiles in the Martian Crust (Elsevier, ed. 2, 2019), pp. 247–264.

3. L. Wernicke, B. Jakosky Quantifying the water stored in hydrated minerals on Mars. AGU Fall Meeting, #P41C-3462 (2019).

4. Y. L. Yung, W. B. DeMore, Photochemistry of Planetary Atmospheres (Oxford University Press, Oxford, UK, 1998).

5. R. M. Haberle, Solar system/syn, atmospheres, evolution of atmospheres | Planetary Atmospheres: Mars, in Encyclopedia of Atmospheric Sciences (Elsevier, ed. 2, 2015), pp. 168–177.

Acknowledgments: We thank A. Hoffmann, P. Mahaffey, C. Webster, H. Franz, J. Stern, D. Breuer, J. Dickson, J. Eiler, J. Grotzinger, Y. Liu, E. Stolper, and the Sample Analysis at Mars (SAM) Science Team for discussion. We thank B. Jakosky, two anonymous referees, and editor K. Smith for suggestions that strengthened the manuscript. Funding: R.H., B.L.E. and Y.L.Y were supported by a NASA Habitable Worlds grant (#NNN13D466T, later changed to #80NM0018F0612). Part of this work was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration (grant 80NM0018D0004). E.L.S. was supported by a NASA Earth and Space Science Fellowship (NESSF) (grant #80NSSC18K1255). D.J.A. was supported by a Future Investigator in NASA Earth and Space Science and Technology (FINESST) fellowship (grant #80NSSC19K1548). Author contributions: E.L.S. drafted the manuscript, developed the code and performed the simulations for the D/H model. B.L.E. developed the concept of the water budget and D/H model. E.L.S., B.L.E., and R.H. developed the D/H model concept. E.L.S., B.L.E., R.H., D.J.A., and Y.L.Y adapted the Kinetics model input parameters for this study. D.J.A. implemented the Kinetics adaption and D.J.A and Y.L.Y. analyzed the Kinetics output. All authors participated in the writing and editing of the manuscript. Competing interests: We declare no competing interests. Data and materials available: The equations used for the D/H model, and our adopted parameter ranges are given in the supplementary material. The Mars D/H model code, input and output files, the Kinetics data files used for fig. S3, and visualization scripts are all available in the CaltechDATA repository at DOI: 10.22002/D1.1879. The Kinetics software was developed by a combination of authors (D.J.A. and Y.L.Y.) and a large number of non-authors (25, 26), so we do not have permission to distribute the source code. An executable version with adjustable input parameters, to reproduce all simulation scenarios in this paper, is available at the same DOI. The SAM data were taken from the Planetary Data System at https://pds-geosciences.wustl.edu/msl/msl-m-sam-2-rdr-l0-v1/mslsam_1xxx/data/ ; we used level 2 data for samples eid25094, eid25123, eid25173, eid25413, eid25484, eid25515, and eid25538 (5).