Are there active hot springs on Mars? THEMIS looked everywhere at night. There are none it could see.

Mars Odyssey's THEMIS camera was built, in part, to catch the pre-dawn glow of hydrothermal vents. This project re-ran that search over every nighttime image the instrument has taken in 23 years, worked out exactly how small a hot spring it could have missed, and turned the empty result into a number.

Fewer than two active, vigorous hot springs exist on the whole of Mars (95% credible), and the warm water that feeds such springs lies, on average, five to seven kilometres down.

136,564 nighttime infrared images covering 99.8% of the planet, each spot seen a median of ten times. Not one compact warm feature is hotter than sunlit rock can be at night. Injecting synthetic vents into real images shows the survey would have caught any spring exposing more than a few thousand square metres of warm water, and a Poisson-zero posterior turns that into the bound above.

136,564

nighttime images searched

Every analysable THEMIS band-9 nighttime image, 2002–2025, ±87° latitude. 7.88 million warm clusters catalogued.

< 2

active vigorous vents on Mars

95% upper limit (Jeffreys prior) on a vent showing a full 100 m pixel of ≥300 K water. Marginalized over water temperature: < 2.0.

17 m²

of boiling water THEMIS could see

Enough to raise one 100 m pixel by the instrument's 1 K noise. The real limit is confusion with warm rock, at a few thousand m².

5–7 km

median depth to liquid water

Pure water 7.2 km; brines 5.4 km. Probability of pure liquid water within 1 m of the surface: zero in six million draws.

Why look

When THEMIS was designed, objective number two on its list was to "search for pre-dawn thermal anomalies associated with active sub-surface hydrothermal systems." The logic is simple. During the Martian night the surface cools by radiation and its temperature is set by how much heat it stored during the day. A hot spring adds heat of its own, and by the pre-dawn hours it should stand out as a warm spot that has nothing to do with sunlight. The THEMIS team reported no such spots in their early results, and the instrument went on to map minerals and thermal inertia instead. The null result is cited widely but was never turned into a quantitative limit. That is what this project does.

What a hot spring would look like

A THEMIS nighttime pixel is 100 m across, ten thousand square metres. A spring does not need to fill it. The infrared radiance of a pixel is the area-weighted mix of everything inside it, and at 12.6 µm a patch of boiling water at 373 K outshines cold regolith at 190 K by a factor of several hundred. Working the Planck function through the band, a pool of just 17 m², a puddle four or five metres across, raises the whole pixel by the 1 K noise level. Warm 300 K water needs 39 m². Fresh lava needs about one square metre. THEMIS was not insensitive to hot springs; it was exquisitely sensitive to them.

Minimum detectable hot-spring area versus vent temperature — Minimum area of a vent inside a 100 m nighttime pixel that raises the pixel by 1, 3 or 5 K, versus vent temperature. Background regolith at 190 K.

The catch: rock is warm at night too

Search the night images and warm spots are everywhere. A first pass over 230 images of the young volcanic regions the THEMIS team named as priorities found 33,365 of them. Bedrock, boulder fields and fresh lava hold the day's heat far longer than dust, so at night they glow 5 to 30 K warmer than their surroundings. About half the detections are linear scan artifacts, a third are diffuse geologic units, and the compact remainder cluster along terrain boundaries in the rocky Cerberus plains. None is a point source, and none is hot in absolute terms: they sit at 170 to 200 K, the temperature of cold stone, not of water.

Cerberus nighttime image with flagged warm anomalies — A nighttime image over the Cerberus plains: brightness temperature (left), background-removed residual (middle), flagged warm clusters (right). The warm spots follow the terrain in bands, the signature of thermal inertia, not of vents.

So warmth alone is not a signature. What separates a spring from a rock is that the spring is hot in an absolute sense, hotter than any sunlit surface can be at 4 a.m., and that it is warm by day as well, because its heat does not come from the sun.

Searching the whole planet

The full nighttime archive is 136,567 images. Rather than pull 45 MB radiance cubes for each, the search used the 1–2 MB brightness-temperature products, and it began with a free step: the archive index already lists the warmest pixel in every image. A sub-pixel spring must make its image's warmest pixel an outlier. Normalizing every image's warmest pixel against what bedrock reaches at that latitude and season flags 833 images out of 137,952 index rows. Every flagged image was then downloaded and analysed pixel by pixel. That step found that 194 of them are empty byte-fill products with no data, a fact that the error check only caught later, and 36 more are whole-image calibration artifacts reading 300 K everywhere. Of the 630 genuine images, 106 contain a compact warm cluster. The hottest of those compact features is 256.5 K, in a perihelion equatorial scene where bare rock reaches exactly that. Nothing compact sits above the 260 K ceiling that sunlit bedrock can hold through the night.

Absolute warmest temperature versus localized excess for every candidate-bearing image — Every genuine candidate-bearing image on Mars, plotted by the absolute temperature of its warmest pixel and the local excess of its warmest cluster. A hot spring would be compact and hotter than the bedrock ceiling. That region is empty.

In parallel, every one of the 136,564 images was streamed through the full anomaly detector, producing 7.88 million warm clusters and a global atlas. The persistent ones, 21,385 cells warm in two or more overpasses, all lie in the equatorial belt of rocky young lava and crater ejecta. None is an isolated hot point.

Global nighttime warm-anomaly atlas — The global nighttime anomaly atlas: density of all warm anomalies, density of focused hot-spring-like ones, peak excess, and cross-overpass persistence. Persistent focused anomalies trace the rocky equatorial terrains, above all the Cerberus and Elysium plains.

Turning nothing into a number

A non-detection is only as good as the survey behind it. Three ingredients turn "we saw nothing" into a limit. First, coverage: rasterizing every image footprint shows 99.8% of Mars was imaged at night, with a median of ten independent looks per location. Second, detection efficiency: synthetic sub-pixel vents of known temperature and area were injected into 310 real images and recovered through the same pipeline. The result is the most important methodological finding of the project. The barrier is not the 1 K instrument noise; it is confusion with warm rock. A vent must out-shine the natural bedrock ceiling, about 239 K, to be flagged, so the true 50%-recovery area for a single look is 3,000 to 8,000 m² of 273–350 K water, not the naive 20 m². Ten looks per site then push the completeness for anything above that size to essentially 100%.

Third, the statistics of zero. With no detections and an effective searched area that covers 99.8% of the planet, the Poisson posterior with a Jeffreys prior puts the number of active vigorous vents on Mars below 1.9 at 95% credibility, below 3.0 with a flat prior, and below 4.6 at 99%. Marginalizing over liquid-water temperatures from 273 to 373 K gives 2.0. The limit relaxes for weaker sources, to 4.4 vents at 800 m² and to about 86 at 80 m², where the sub-pixel confusion floor takes over.

Bayesian upper limit on the number of active vents versus vent area — Left: 95% upper limit on the number of active hot springs on Mars as a function of vent area within a 100 m pixel, for several vent temperatures. Right: survey completeness. The floor of 1.92 is the Jeffreys-prior Poisson limit at full coverage.

What could still hide

THEMIS is blind to heat that arrives diffusely. A steady extra flux q re-radiates as a surface warming of q divided by the radiative stiffness of the surface, about 2 W per square metre per kelvin at Martian night temperatures. So any heat flow below about 2 W/m², eighty times the planet's background, produces less than 1 K and is invisible. Over a 100 km square that is 20 gigawatts hiding in plain sight. Warm water under an insulating dust blanket could sit less than a metre down and show nothing; under conductive ice-cemented ground it must be 60 m or more deep. But an unreplenished warm body freezes within centuries, and replenishment means circulation, and circulation that reaches the surface makes exactly the focused vents the survey rules out. The activity that robustly hides is heat delivered by conduction alone, with no surface outlet.

Minimum invisible depth of warm water versus overburden conductivity — How shallow warm water can be and still produce less than 1 K at the surface, as a function of the conductivity of the overburden: about 1 m under air-fall dust, tens of metres under duricrust, 60–170 m under rock or ice-cemented ground.

How deep is the water in fact? A cryosphere model with a linear geotherm, Monte-Carlo sampled over the planet with literature priors on heat flow, crustal conductivity and the freezing point of Martian brines, puts the median depth to pure liquid water at 7.2 km and to any brine at 5.4 km. In six million draws pure water never comes within a metre of the surface, and comes within a kilometre in eight of them. A two-dimensional version using the Odyssey gamma-ray heat-flow map and MOLA topography sharpens this to a latitude-dominated atlas: 4–6 km at the equator, 14 km at the poles, with the giant volcanoes deepening the isotherm and Hellas shallowing it.

Distribution of depth to liquid water on Mars — Depth to liquid water over the surface of Mars, area-weighted. The only draws shallower than 300 m are the 2.5% of the surface where a 210 K eutectic brine is liquid at the ambient annual-mean temperature; that is cold brine, not warm water.

GRS heat flow and 2-D depth-to-water atlas — Surface heat flow from Odyssey GRS thorium and potassium (top) and the resulting median depth to liquid water with MOLA elevation (bottom).

Cross-checks with daytime data

Rock and vents differ by day. High-inertia rock is warm at night but cool by day; a heat source is warm in both. The 21,385 persistent warm-at-night cells were therefore re-examined in daytime THEMIS images. The first version of this test, it turned out, sampled the wrong pixels: it read the day and night images at the mean position of each cell, a median of three pixels from the detected warm pixels, and so measured background against background. The corrected analysis reads the detected pixels themselves. At those pixels the night excess reproduces to 0.1 K, and the daytime excess is uncorrelated with it (r = −0.04). The strongest night features have no more daytime excess than the weakest. There is no warm-both branch. A one-dimensional thermal model fit to each day–night pair finds that 94–98% of candidates need zero internal heat, and the remainder sit on sunlit slopes the flat model cannot represent; the honest per-candidate ceiling on internal warming is a few kelvin, set by terrain texture and model assumptions rather than by the data.

Day versus night behaviour of persistent candidates at the detected pixels — Night excess at the detected pixel versus daytime excess at the same place, for 21,223 persistent cells. A heat source would put its night excess back as a proportional day excess along the dotted 1:1 line. Nothing does.

Finally, every 3 km cell of the planet was checked. All 335,425 day and night images were binned into destriped global mosaics, giving a clean global thermal-inertia map as a by-product. Above the mosaic's 14 K floor, the largest residual patches are polar frost; the ten largest patches at temperate latitudes are all cooler than their surroundings by day, the fingerprint of thermal inertia. No broad region of Mars is warm both day and night.

Global destriped day and night thermal mosaic — Global THEMIS mosaic from every cached image: nighttime thermal-inertia anomaly (top; dust provinces dark, rock bright), night excess over the day-and-albedo manifold (middle), and cells above the 15 K screen (bottom): polar frost and orbit-track striping, nothing compact and warm-both.

What this does and does not say

It says no hot spring exposing more than a few thousand square metres of warm water has been active anywhere THEMIS imaged at night, at 95% credibility fewer than two on the planet, and the persistent warm spots that fill the archive are rock.
It does not say anything about small, cool seeps below the bedrock confusion floor, roughly a hundred square metres of near-freezing water per pixel, where the bound weakens to tens of vents. Nor does it constrain heat delivered diffusely below 2 W/m², or warm water more than a few metres under insulating dust.
The thermal model used for the per-candidate and mosaic tests is one-dimensional, flat and has no atmosphere; adding a realistic sky flux changes nothing material, but slopes are not represented, which is why the per-candidate ceiling is a few kelvin rather than the noise level.
An error check on 2026-09-07 found and fixed three analysis bugs, a bookkeeping error in the outlier census and a dozen smaller slips. None changed the top-line result; one supporting claim was withdrawn and replaced. The full log is on the Error check page.

Methods and caveats: the twelve-step chain in detail, with every number and its provenance.
All figures with captions, including the superseded originals.
Data and code: the report PDF, scripts, and catalogues.

Casey Handmer. Analysis June–July 2026; reconstruction, error check and this site 2026-09-07. THEMIS data courtesy of NASA/JPL/Arizona State University via the PDS. Suggested citation: Handmer, C. (2026), Did THEMIS find Martian hot springs? A global reproduction, completion and Bayesian bound, themis.caseyhandmer.com.