Martian Water & Ice

Where the water is

Martian Water Inventory: Caps, Buried Glaciers and Hydrated Soil

Four reservoirs, in descending order of size and ascending order of usefulness.

The north polar residual cap is nearly pure water ice, roughly 1.6 million km³ of it — enough to cover the planet to about 1.5 m if spread evenly. It is also at 85°N, in the dark for half the year, and no crewed architecture has ever seriously proposed going there. The reservoirs that matter for a program are the smaller, dirtier ones closer to the equator.

The last row deserves a note because it keeps reappearing in proposals. Atmospheric water capture needs no excavation, no prospecting and no planetary-protection argument — and it processes an enormous volume of very thin air for a small yield. It is a credible trickle-charge source for a small crew and not a propellant plant.

The prospecting gap

Nobody Has Drilled: Why the Maps Are Inference

The deepest anyone has sampled Martian subsurface ice is a trench a few centimeters deep, dug in 2008.

NASA’s Phoenix lander scraped white material at 68°N in 2008, watched it sublimate over four sols, and confirmed water ice within roughly 5–7 cm of the surface at that site. That remains the only direct observation of buried Martian ice. Everything else is remote sensing: Mars Odyssey’s gamma-ray and neutron spectrometers measuring hydrogen in the top meter, MRO’s SHARAD and Mars Express’s MARSIS radars imaging deeper layering, CRISM and THEMIS mapping hydrated minerals and thermal inertia.

The SWIM project — Subsurface Water Ice Mapping, run for NASA by the Planetary Science Institute — consolidates those datasets into consistency maps showing where ice is likely within the top few meters across the northern mid-latitudes. It is the best product available and it is explicitly a consistency map, not a resource assessment. An architecture that quotes a production rate is quoting a grade nobody has measured, which is the same failure mode the lunar program has with polar ice.

The dedicated fix has been proposed and shelved. Mars Ice Mapper, a NASA concept with the Canadian, Japanese and Italian agencies carrying a subsurface-sounding radar specifically to find crew-accessible ice, was dropped in the 2022 budget cycle and has not been replaced. China’s Zhurong rover carried the RoPeR ground-penetrating radar at Utopia Planitia and found layered deposits but no detectable ice in the upper tens of meters at its equatorial site — a useful negative result, and the only subsurface radar traverse ever driven on Mars. Tianwen-3’s site selection for the 2028 and 2031 windows uses water proxies among its criteria.

Demand

How Much Water a Mars Program Actually Needs

The number swings sevenfold on a vehicle decision made before anyone sets foot on the planet.

Water demand is dominated by propellant, and propellant demand is set by the architecture. Under direct-to-surface, where Starship lands and refuels for Earth return, the surface must produce 683 t of methalox per window, which implies roughly 570 t of water. Under the orbit-staged ferry, ascent-only propellant drops the requirement to about 107 t of methalox and roughly 75 t of water. Crew consumables are a rounding error beside either.

The net mining rate is lower than the propellant mass suggests, because the Sabatier reaction returns roughly half the water it consumes through the electrolysis loop. For a 103 t Phobos-node cycle the chain runs at 132 kg/day of propellant, but only about 65 kg/day of that is newly mined water.

The consequence for site selection is the whole argument. At 570 t/window a site needs high-grade ice and that means going north, to 39°N and the seasonal power penalty that comes with it. At 75 t/window ordinary equatorial regolith at 3–8 wt% suffices, no clay deposit or special deposit required, and the site can be chosen on landing safety and insolation instead. Isidis Planitia ranks first under orbit staging and Arcadia Planitia under direct-to-surface, with the break-even at an ice weighting of about 12 out of 100. See Mars Site Selection.

Two feedstocks

Massive Ice or Hydrated Soil: Two Mines, Two Machines

One is a mining problem with a prospecting risk. The other is a heating problem with no prospecting risk at all.

Ice-cemented regolith and buried glaciers give high grade — tens of percent water by mass, up to nearly pure ice — and demand excavation in material that behaves like frozen concrete, or a Rodriguez well, which melts a cavity in place and pumps the water out. Rodwells are proven at Antarctic stations and work only in massive ice with no interstitial debris, so their applicability here is exactly the thing nobody has measured. The site must be poleward of about 30°, which costs seasonal insolation and adds a plane-change penalty to the orbital node.

Hydrated regolith gives 3–8 wt% anywhere, including the equator, bound into minerals and released by heating to a few hundred degrees. No prospecting, no latitude penalty, no deposit to find — and 1,500 to 11,400 t of material to move and heat per window depending on architecture. It trades a geological unknown for a straightforward energy and throughput problem, which is the kind of trade an engineering program should usually take.

The decorrelation argument favors having both. Two settlements that both depend on mid-latitude shallow ice share one failure mode; a pair where one uses equatorial bulk hydration and the other uses shallow ice do not. Water form, not distance, is what makes a second site independent.

What it costs

The Energy Cost of Martian Water Is Set by Grade, Not by Chemistry

1.1 kWh/kg at good grade, 4.1 at poor. The spread is all in how much rock gets heated.

Water extraction sits low on the closure ladder — below oxygen at about 7 kWh/kg realized, far below metals at 20–40 — but the range within the rung is wide, and the variable is ore grade. Melting ice costs the latent heat and little else. Baking 5 wt% hydrated soil means heating 20 kg of rock for every kilogram of water, and the rock’s heat capacity dominates the bill. Heat recovery between batches is the single largest lever on the number.

At the orbit-staged requirement of 100 t of water per year, the plant draws roughly 57 kW at good grade and 114 kW at 1 wt%. The oxygen rung alongside it draws about 80 kW. Against a 150 kWe bus, the two do not both fit at low grade — which makes closure a dispatch problem rather than a capability problem, and makes the propellant plant the site’s designated load-follower, curtailed first because its product is storable.

Other jobs

Buffer, Shield and Life Support: One Substance, Four Ledgers

The same tonnage counts three times, which is the strongest argument for holding a large reserve.

As a buffer, water decouples the stages of the ISRU chain. Extraction, electrolysis, Sabatier, liquefaction and storage form a serial chain where duplicating one stage buys nothing if another is single-string; a buffer sized as production rate times mean time to repair fixes that instead. At 65 kg/day, 0.91 t covers 14 days and 1.94 t covers 30. Two tonnes of water decouples every stage in the chain for a month.

As shielding, water is about 2 t/m² for the areal density that regolith delivers at 3–5 t/m², and it is hydrogen-rich, which matters against solar particle events specifically. The arithmetic splits by area: over a hull’s 450 m² footprint water shielding would need 900 t, which is impossible against any plausible production rate, but over a 20 m² storm shelter it is 40 t — a credible reserve doing double duty. Regolith over the hull, water over the shelter inside it.

As life support, water closure is what makes consumables affordable. Imported consumables run 5.18 t per resident per window open-loop, 1.72 t with water closed to 90% and oxygen from CO₂ electrolysis, and 0.47 t once food is grown locally. Water closure is the first and cheapest of those three steps.

Brines

Liquid Water on Mars: Brines, the Triple Point, and Why It Does Not Help

Perchlorates depress the freezing point far enough for transient liquid brine. That is a science result, not a resource.

Water’s triple point is 611.657 Pa; mean Martian surface pressure is 610 Pa. Low-lying sites sit above the line — Arcadia Planitia runs about 867 Pa — so liquid is thermodynamically permitted, though temperature rarely cooperates. Calcium and magnesium perchlorates in the soil depress the freezing point of a saturated brine to roughly 204 K, well inside the daily temperature range at many sites, and deliquescent salts can pull vapor directly out of the atmosphere.

The recurring slope lineae once read as seasonal brine flows are now better explained as dry granular flows, and the case for present-day surface liquid has weakened over the past decade rather than strengthened. Even if it exists, transient films of perchlorate brine are not an extractable resource: the volumes are tiny, the chemistry is hostile to equipment, and any process to concentrate it costs more than baking ordinary hydrated soil.

Brines matter for two other reasons. They are the strongest candidate for a present-day habitable niche, which is what drives planetary protection policy below. And the triple-point pressure has a hardware consequence with nothing to do with resources: sublimation cooling, the thermal rejection method behind every spacewalk since Apollo, requires ambient pressure below the triple point, so it does not work at a low-elevation Martian site.

Planetary protection

Special Regions: The Cheapest Constraint to Change, and Nobody Has Started

Subsurface ice is exactly where the policy is most restrictive, and crewed missions have no compliant answer yet.

COSPAR planetary protection policy designates Special Regions — places where terrestrial organisms could plausibly propagate, defined by temperature and water activity thresholds. Subsurface ice and brine-bearing ground are the archetype, and a crewed mission is by definition a large, warm, leaking source of terrestrial microbes sitting on top of the resource it came to mine. Every crewed architecture assumes access to ice; no crewed architecture has an approved planetary-protection plan for touching it.

This belongs in the same category as the radiation-worker reclassification question: a regulatory gate that costs nothing in hardware, moves the architecture significantly, and is currently nobody’s active program. Waste disposal, drill contamination, and the handling of returned subsurface samples are all unresolved, and the resolution will arrive as policy rather than as engineering.

Record

Provenance and Data Sources