Geothermal surface heat flux mapping for resource prospecting and plant monitoring
Active geothermal fields produce land-surface temperature anomalies detectable from orbit, but separating crustal heat from solar heating requires careful emissivity correction and multi-temporal analysis. This page explains the physics, the sensors, and the honest limits of what a 90-metre pixel can and cannot resolve.
Sensors
- Landsat 9 TIRS: Two thermal bands centred at 10.9 µm and 12.0 µm, 100 m native resolution resampled to 30 m in products. 16-day repeat at the equator, improving to higher frequency at mid-latitudes. The split-window pair allows atmospheric correction and emissivity separation. Archive continuity with Landsat 8 TIRS stretches back to 2013, giving a decade-plus baseline for trend detection.
- ASTER TIR: Five thermal bands from 8.1 to 11.7 µm at 90 m spatial resolution, with on-board emissivity retrieval via the Temperature-Emissivity Separation (TES) algorithm. ASTER is not acquired continuously; scenes must be requested or found in archive. Its multi-band TIR capability makes it the reference instrument for emissivity mapping of volcanic and hydrothermal surfaces, where lithology varies sharply.
- ECOSTRESS: ISS-mounted radiometer covering 8.0 to 12.0 µm at approximately 70 m resolution. Its irregular ISS orbit produces variable overpass times, which is actually useful for geothermal work: comparing day and night acquisitions from the same week isolates features that remain warm overnight, a strong indicator of subsurface heat rather than solar loading. Revisit is non-deterministic, typically 1 to 5 days at mid-latitudes.
- Sentinel-3 SLSTR: Dual-view thermal radiometer at 1 km resolution in TIR channels, with daily global coverage. Too coarse to resolve individual fumarolic fields, but useful for regional-scale heat-flux anomaly surveys across large volcanic provinces and for near-real-time change alerts when a major surface expression event occurs.
What a warm pixel actually means
A satellite thermal sensor measures upwelling radiance from the land surface. That radiance is a function of two things: the true kinetic temperature of the surface and its emissivity, the efficiency with which a material radiates relative to a perfect blackbody. Bare silica sinter, the white mineral crust common around geothermal springs, has emissivity near 0.85 in the 10 µm window. Wet clay sits closer to 0.97. If you treat the whole scene as a uniform emissivity of 0.97 and then encounter a patch of sinter, you will underestimate its temperature. The anomaly you are hunting may be partly masked, or partly invented, by the emissivity field itself.
This is why ASTER's five-band TIR coverage matters for prospecting. The TES algorithm uses spectral contrast across those bands to retrieve emissivity and temperature simultaneously, without needing a ground truth measurement. Landsat TIRS has only two bands, so it relies on an assumed emissivity map, typically derived from a vegetation index and a land-cover classification. In a geothermal field where the surface is mineralogically complex and rapidly changing, that assumption introduces errors of 1 to 3 K, which is the same order of magnitude as the anomaly you are trying to detect at the margins of a field.
Solar heating looks almost identical. Here is how to separate them.
On a cloudless afternoon, a dark basalt outcrop at low albedo can reach surface temperatures 15 to 20 K above a neighbouring grassy meadow. That differential is larger than most geothermal anomalies. Slope aspect compounds the problem: south-facing slopes in the northern hemisphere receive more direct insolation and read warmer in any single daytime acquisition, regardless of what is happening underground.
The solution is temporal stacking. Geothermal anomalies are persistent. Solar heating is diurnal and seasonal. A pixel that is anomalously warm at 02:00 local time, when insolation is zero, is not being heated by the sun. ECOSTRESS's variable overpass times make it particularly useful here: a stack of night-time passes over a suspected field, compared against a stack of day-time passes from the same period, isolates the thermally persistent component. Across a Landsat archive, the same logic applies seasonally. A feature that ranks in the top decile of surface temperature in winter, when solar input is minimal, and also in summer is almost certainly not a solar artefact. A feature that appears only in summer is probably a slope-aspect or albedo effect.
Atmospheric water vapour is a third confound. The 10 µm and 12 µm bands of Landsat TIRS attenuate differently through a humid atmosphere. The split-window correction uses the difference between the two bands to estimate and remove atmospheric contribution. Errors in this correction, particularly over humid volcanic terrain where water vapour is locally elevated by fumarolic emission, can introduce biases of 0.5 to 2 K. Honest prospecting reports should carry that uncertainty range explicitly.
The 90-metre floor and what it costs you
ASTER's 90 m resolution and Landsat's 100 m native TIRS resolution set a hard floor on what is detectable from orbit. A fumarole vent a few metres across, even at 200°C, contributes to the mixed pixel radiance in proportion to its fractional area. At 1% of a 90 m pixel, its thermal contribution is swamped by the background. In practice, geothermal features detectable from orbit are either large hydrothermal pools and mud pools (tens of metres across), extensive sinter aprons, or diffuse zones of elevated ground temperature across hundreds of metres.
This is not a reason to abandon orbital thermal sensing for prospecting. It is a reason to use it correctly. Orbital data is excellent for narrowing a search area from a regional scale down to a few square kilometres, for monitoring whether a known field is expanding or contracting over years, and for detecting new surface expression events. Ground-based surveys and airborne thermal cameras, which can achieve sub-metre resolution, then take over for detailed characterisation. The two approaches are complementary, not competing.
Multi-temporal stacks as a reservoir monitoring tool
Once a geothermal plant is operating, the surface thermal expression of the field changes. Injection wells, production wells, and pressure drawdown all alter the subsurface fluid pathways that feed surface features. Hot springs can migrate, intensify, or go cold. Mud pools can dry out. New fumarolic activity can appear along fault traces.
A multi-temporal stack of Landsat TIRS scenes, processed consistently with the same atmospheric correction and emissivity assumptions, can detect these changes. The method is straightforward: compute land surface temperature for each cloud-free acquisition, register all scenes to a common grid, then calculate the per-pixel trend over time and the variance. Pixels with a statistically significant cooling trend may indicate reduced upwelling. Pixels with increasing variance may indicate episodic activity linked to pressure fluctuations. The Landsat archive, continuous since 1982 with thermal capability from Landsat 4 onwards, provides a baseline against which modern plant-era changes can be assessed. The TIRS-era archive from 2013 onwards is the most radiometrically consistent portion for quantitative work.
One honest caveat: a cooling trend in the thermal image does not unambiguously mean reservoir depletion. It could reflect a change in surface water availability, a shift in precipitation patterns, or vegetation recovery altering the emissivity field. Interpretation requires integrating the thermal time series with production data, well-head pressure records, and precipitation climatology. The satellite layer is evidence, not a verdict.
Putting it into an analytics workflow
A practical prospecting workflow begins with Sentinel-3 SLSTR to identify candidate regions across a volcanic province, flagging areas where the 1 km thermal signal persistently exceeds the regional background by more than a defined threshold across multiple seasons. Those candidate zones are then examined in the Landsat and ASTER archives at 90 to 100 m resolution, applying the temporal filtering described above to reject solar artefacts. Confirmed anomalies are ranked by spatial extent, thermal excess above background, and persistence score.
For operating plants, the workflow shifts to a monitoring cadence. Every cloud-free Landsat overpass over the field is ingested, processed to land surface temperature, and compared against the historical baseline. Anomalous departures trigger a review. ECOSTRESS night-time passes, when available, provide an independent check. Satellize runs this kind of multi-source thermal stack analysis on open constellations as a standard analytics product, with the Tonga crop-estimation programme being an example of how persistent time-series methods transfer across very different physical phenomena.
Cloud cover is the operational constraint that is easiest to underestimate. Many of the world's most productive geothermal regions sit in humid volcanic terrain with persistent cloud. Iceland, Indonesia, the Philippines, and the East African Rift all present significant cloud fractions. In such environments, the effective revisit for a usable thermal acquisition may be monthly rather than fortnightly, and multi-year stacks are needed to build statistically reliable anomaly maps. Buyers should request cloud-fraction statistics for their specific area of interest before commissioning a monitoring programme.
Typical figures
| Spatial resolution (TIR) | 90 m (ASTER TIR), 100 m native / 30 m resampled (Landsat 9 TIRS), ~70 m (ECOSTRESS), 1 km (Sentinel-3 SLSTR) |
| Revisit period | 16 days (Landsat 9 at equator), non-deterministic 1–5 days (ECOSTRESS/ISS), daily (Sentinel-3 SLSTR); ASTER on-demand |
| Spectral bands | 10.9 µm and 12.0 µm (Landsat TIRS); 8.1–11.7 µm across 5 bands (ASTER TIR); 8.0–12.0 µm (ECOSTRESS); 10.85 µm and 12.0 µm (SLSTR) |
| Land surface temperature accuracy | ±1–2 K under good atmospheric conditions (Landsat split-window); ±1.5 K typical for ASTER TES; higher uncertainty over humid volcanic terrain |
| Minimum detectable anomaly (area) | Features occupying at least several percent of a pixel are detectable; isolated vents below ~30–50 m diameter are not resolvable from orbit |
| Archive depth | Landsat TIRS from 2013 (L8) and 2021 (L9); ASTER from 2000; ECOSTRESS from 2018; Sentinel-3 SLSTR from 2016 |
| Latency (open data) | Landsat: typically <12 hours to USGS distribution; Sentinel-3: same-day via Copernicus Data Space; ECOSTRESS: days to weeks |
| Cloud sensitivity | All TIR bands are opaque to cloud; effective revisit in humid volcanic regions may be monthly or less |
| Delivery formats | GeoTIFF land surface temperature grids, anomaly shapefiles, time-series CSV per zone of interest, PDF prospecting reports |
Analytics Satellize can run
| Geothermal anomaly map | Multi-temporal median compositing of Landsat TIRS LST with solar-artefact rejection via night/day ratio and seasonal persistence filter | GeoTIFF anomaly layer with ranked candidate zones; PDF summary with methodology notes and uncertainty bounds |
| Emissivity-corrected land surface temperature | Split-window algorithm (Landsat) or Temperature-Emissivity Separation (ASTER TES); atmospheric correction using NCEP reanalysis water vapour fields | Per-scene GeoTIFF LST grids with associated emissivity layer and per-pixel uncertainty estimate |
| Persistent thermal anomaly score per pixel | Time-series percentile ranking across multi-year Landsat archive; pixels scoring above 90th percentile in both winter and summer seasons flagged as geothermally persistent | Ranked anomaly shapefile with persistence score and seasonal breakdown, suitable for GIS import |
| Surface expression change detection for operating fields | Per-pixel linear trend and variance analysis across Landsat TIRS time series; change points identified using sequential Mann-Kendall or similar non-parametric test | Monthly monitoring report with trend maps, flagged change-point dates, and comparison against plant production calendar where provided |
| Night-time thermal persistence check | ECOSTRESS night-time acquisitions composited and differenced against daytime stack to isolate non-solar heat sources | Day/night differential GeoTIFF and annotated report identifying features confirmed as non-solar |
| Regional volcanic province screening | Sentinel-3 SLSTR 1 km thermal anomaly detection across large areas using background temperature modelling and seasonal normalisation | Regional screening GeoTIFF and ranked candidate list for follow-on high-resolution analysis |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.