Subsurface coal seam fire extent and progression mapping
Underground coal seam fires betray themselves at the surface through thermal anomalies and millimetre-scale ground subsidence. Combining thermal infrared imagery with Sentinel-1 InSAR makes it possible to map fire extent, track the burn front, and estimate void growth without entering the ground.
Sensors
- ASTER TIR (Terra): Five thermal infrared bands (8–12 µm) at 90 m spatial resolution. Daytime and night-time acquisitions available; night-time passes suppress solar-reflected contamination and reveal residual subsurface heat more clearly. Revisit is approximately 16 days at the equator but can be tasked more frequently at mid-latitudes. Archive extends to 1999.
- Landsat-9 TIRS: Two thermal bands (Band 10 at 10.9 µm, Band 11 at 12.0 µm) at 100 m resolution, resampled to 30 m in data products. Split-window or single-channel algorithms yield land surface temperature to roughly ±1–2 K under clear sky. 16-day revisit; combined with Landsat-8 the effective revisit halves to 8 days.
- Sentinel-1 C-band SAR (InSAR): C-band (5.405 GHz) synthetic aperture radar in Interferometric Wide Swath mode at 5 × 20 m resolution. Repeat-pass interferometry detects line-of-sight surface displacement to sub-centimetre precision over 6- or 12-day intervals. Coherence loss above active burn fronts is itself a diagnostic signal. Free and open archive from 2014.
- ECOSTRESS (ISS): Five TIR channels (8–12.5 µm) at approximately 70 × 70 m pixel size, mounted on the International Space Station. Non-sun-synchronous orbit gives variable overpass times, including night passes, and sub-daily revisit at some latitudes. Useful for capturing diurnal thermal cycles that help distinguish subsurface heat from surface heating.
- VIIRS Day/Night Band and M-band TIR (Suomi-NPP / NOAA-20): M-band thermal channels at 375 m resolution with twice-daily global coverage. Coarse for spatial mapping but valuable for continuous temporal monitoring and for flagging new anomalies that then trigger tasking of higher-resolution sensors.
What a burning seam leaves on the surface
Underground coal seam fires heat the overlying strata by conduction and by convective transport through fractures. At the surface this produces a thermal anomaly that, in well-documented cases such as the Jharia coalfield in Jharkhand, India, can be tens of degrees Celsius above background in the most active zones. The anomaly is not uniform. It follows the geometry of the seam, the depth of overburden, and the network of fractures that carry hot gases upward. That spatial structure is itself diagnostic.
As combustion consumes coal, voids form. The overburden above them deflects downward, sometimes catastrophically, but more often as a slow, measurable subsidence that InSAR captures before any surface crack is visible to a field observer. The combination of a thermal high and a subsidence bowl at the same location is a strong discriminator against other heat sources such as industrial plant or sun-baked bare rock.
Thermal detection: what the sensors can and cannot see
ASTER and Landsat-9 TIRS both measure emitted thermal radiation in the 8–12 µm atmospheric window. Land surface temperature products derived from these instruments can resolve anomalies of 1–3 K above background under clear-sky conditions, which is sufficient to detect active fire zones where overburden is thin or fractured. Night-time acquisitions are preferred: solar heating of bare ground during the day can produce surface temperatures comparable to or exceeding the anomaly from a deep fire, masking the signal entirely.
Cloud is the hard constraint. Both sensors are passive optical in the thermal, which means a single monsoon season can produce months of unusable imagery over South Asian coalfields. The Jharia field, for instance, sits in a region where June through September cloud cover routinely exceeds 80 percent of days. ECOSTRESS partially compensates through its variable overpass timing, but it does not eliminate the problem. Analysts must plan around seasonal windows and use multi-year compositing to build reliable extent maps.
Spatial resolution sets a floor on what can be resolved. A fire zone narrower than roughly 200 m will be detectable as a temperature elevation but not geometrically resolved at 90–100 m pixel sizes. Depth matters too: fires deeper than approximately 50 m of competent, unfractured overburden may produce surface anomalies below detection thresholds in cool seasons.
InSAR as the second witness
Sentinel-1 interferograms measure phase difference between repeat passes to infer surface displacement in the satellite line-of-sight direction. Over an active burn front, the subsidence signal is typically a few millimetres to a few centimetres per month in the early stages, accelerating as the void grows. Time-series methods such as Persistent Scatterer InSAR or Small Baseline Subset (SBAS) processing aggregate many interferograms to separate the deformation signal from atmospheric noise, reaching displacement sensitivities of a few millimetres per year under favourable coherence conditions.
The Jharia coalfield has been studied extensively using this approach. Published work using ERS, Envisat and later Sentinel-1 data has mapped subsidence bowls that correlate spatially with known fire zones and thermally anomalous areas identified from ASTER. The InSAR signal continues through cloud cover and monsoon seasons where thermal sensors go blind, making it the more temporally continuous of the two data streams.
One honest limitation: C-band InSAR loses coherence over vegetated or disturbed surfaces. Active subsidence zones with fresh cracking, loose rubble or dense scrub return low-coherence interferograms that are difficult to interpret. L-band SAR (ALOS-2 PALSAR-2) penetrates vegetation better and maintains coherence over longer baselines, but it is not freely available at Sentinel-1 cadence or cost.
Mapping the burn front: method and honest uncertainty
The operational workflow fuses the two signals. Thermal anomaly maps from multi-date ASTER or Landsat-9 composites define candidate fire zones. InSAR displacement maps, typically processed as monthly or quarterly velocity fields, are overlaid to identify where subsidence and thermal elevation co-occur. The intersection of the two defines a high-confidence active fire zone; areas showing only one signal are flagged as probable but uncertain.
Progression mapping requires consistent time-series analysis. The burn front advances in the direction of available oxygen, which in practice means along fracture networks and toward surface outcrops. Sequential seasonal maps can estimate advance rates. Published studies of Jharia have reported advance rates of several metres to tens of metres per year in different sectors, though these figures vary considerably by seam depth, coal quality and fire age.
The method cannot determine fire temperature at depth, total combustion volume, or the precise geometry of the void. It maps the surface expression of a subsurface process. That distinction matters when the output is used for risk assessment: a subsidence bowl visible in InSAR indicates that void formation has already occurred, not that it is imminent.
Operational context and what the data is actually used for
Governments and mining authorities use fire-extent maps for three distinct purposes. First, safety zoning: identifying areas of surface where subsidence risk is high enough to prohibit habitation or vehicle movement. Second, fire-fighting prioritisation: directing water injection, sand-stowing or surface excavation to the most active sectors. Third, liability and compliance documentation: demonstrating to regulators that fire extent is being monitored and, ideally, contained.
Each purpose places different demands on the data. Safety zoning needs the highest spatial accuracy and short latency. Fire-fighting prioritisation needs temporal frequency to track response. Compliance documentation needs archival consistency and audit-ready provenance. A single sensor or processing run rarely serves all three simultaneously.
Satellize runs thermal anomaly and InSAR time-series analytics on open constellations for clients requiring periodic fire-progression reports or near-real-time alert feeds. The Jharia coalfield remains the best-documented public reference for method validation in this application. Clients commissioning this analysis for other coalfields should expect a calibration period of two to three seasonal cycles before progression rates can be stated with confidence.
Typical figures
| Thermal spatial resolution | 90 m (ASTER TIR), 100 m native / 30 m resampled (Landsat-9 TIRS), ~70 m (ECOSTRESS) |
| InSAR spatial resolution | 5 × 20 m (Sentinel-1 IW mode); displacement maps typically posted at 20–40 m after multi-looking |
| Thermal revisit | 8 days combined Landsat-8/9; ~16 days ASTER; sub-daily VIIRS for coarse flagging; variable ECOSTRESS |
| InSAR revisit | 6 days (Sentinel-1 ascending + descending combined); 12 days single geometry |
| Minimum detectable temperature anomaly | ~1–3 K above background under clear sky (passive TIR sensors); dependent on overburden depth and season |
| InSAR displacement sensitivity | Sub-centimetre per epoch; ~mm/year in time-series stacking under good coherence conditions |
| Cloud limitation | Thermal sensors fully blocked by cloud; InSAR unaffected by cloud but loses coherence over disturbed or vegetated ground |
| Archive depth | Landsat thermal from 1982 (TM); ASTER from 1999; Sentinel-1 InSAR from 2014 |
| Spectral bands used | 8–12 µm TIR window (thermal); C-band 5.405 GHz (InSAR) |
| Delivery formats | GeoTIFF temperature anomaly maps, displacement velocity rasters, GeoPackage / Shapefile fire-zone polygons, PDF progression reports |
Analytics Satellize can run
| Thermal anomaly extent map | Multi-date land surface temperature retrieval (single-channel or split-window algorithm) from ASTER / Landsat-9 TIRS; night-time compositing to suppress solar noise | Seasonal GeoTIFF and polygon layer showing temperature-above-background zones, with confidence class by anomaly magnitude |
| InSAR subsidence velocity map | SBAS or Persistent Scatterer time-series processing of Sentinel-1 interferogram stack; atmospheric correction using ERA5 or GACOS tropospheric models | Annual or quarterly displacement velocity raster (mm/year line-of-sight) and subsidence bowl polygons with peak displacement values |
| Fused fire-zone confidence map | Spatial intersection of thermal anomaly and InSAR subsidence layers; high-confidence zones require both signals; single-signal zones flagged as probable | GIS layer with three-class confidence rating (confirmed, probable, watch area), updated per seasonal analysis cycle |
| Burn-front progression report | Sequential seasonal fire-zone polygons differenced to estimate lateral advance rate and direction; trend fitted over minimum three-year archive | Annual PDF report with advance-rate estimates by sector, uncertainty bounds, and annotated map series |
| Near-real-time thermal alert | VIIRS 375 m M-band anomaly flagging (twice-daily) triggers tasking of ECOSTRESS or next available Landsat pass for confirmation | Email or API alert with coordinates, anomaly magnitude, and sensor confirmation status within 48 hours of VIIRS flag |
| Coherence-loss change detection | InSAR coherence map time-series; new coherence-loss patches over previously stable ground indicate fresh subsidence or surface disruption above advancing burn front | Monthly change-detection layer showing newly incoherent areas, delivered as GeoPackage with date-of-change attribute |
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.