Abandoned shallow mine void and crown-hole collapse risk mapping
Shallow historic mine workings produce millimetre-scale precursor subsidence before crown-hole collapse. Persistent-scatterer and small-baseline InSAR can detect that signal weeks to months in advance, if you choose the right wavelength for the terrain.
Sensors
- Sentinel-1 (C-band SAR, 5.6 cm wavelength): Interferometric Wide swath mode delivers 5 x 20 m ground resolution with 6-day repeat at mid-latitudes using both satellites. C-band is coherent over hard urban and peri-urban scatterers but loses coherence rapidly over vegetated farmland between acquisitions, limiting PS-InSAR coverage density in rural mining districts.
- ALOS-2 PALSAR-2 (L-band SAR, 23.6 cm wavelength): L-band penetrates vegetation canopy and maintains coherence over fields and woodland where C-band fails. Stripmap mode gives 3 x 3 m resolution; standard repeat is 14 days. The longer wavelength tolerates greater temporal decorrelation, making it the preferred sensor for rural room-and-pillar workings under crops.
- COSMO-SkyMed (X-band SAR, 3.1 cm wavelength): Spotlight mode achieves better than 1 m resolution, useful for resolving small pillar-compression bowls in urban fringes and infrastructure corridors. X-band is highly sensitive to surface change and loses coherence even faster than C-band over bare agricultural soil between rain events. Best applied where dense surface scatterers exist.
- Copernicus DEM GLO-30: A globally consistent 30 m posting digital elevation model derived from TanDEM-X. Used to remove topographic phase from interferograms and to provide baseline terrain context for subsidence bowl geometry. Vertical accuracy is nominally better than 4 m LE90 over most terrain types, sufficient for phase-to-displacement conversion but not for detecting the bowls themselves.
What a pillar does before it fails
Room-and-pillar coal mines leave a grid of coal pillars supporting a rock roof, sometimes at depths of only 10 to 50 metres. Over decades, pillars compress under overburden load and groundwater weakens the floor strata. The surface above deflects downward in a broad, shallow bowl, typically a few centimetres of settlement spread across tens of metres. That bowl is the signal InSAR is looking for.
Solution-mined evaporite cavities behave differently. Dissolution of salt or gypsum creates a cavity whose roof thins progressively until it fails suddenly, often with little or no precursor surface deformation. The distinction matters for risk assessment: pillar compression gives you time, roof failure over an evaporite void may not. InSAR can characterise the gradual case well. For the sudden case, it provides spatial context for cavity locations rather than reliable advance warning.
Why wavelength decides whether you see anything at all
Interferometric coherence, the quality of the phase measurement between two SAR acquisitions, degrades when the surface changes between passes. Vegetation growth, tillage and rain all scatter C-band energy differently from one pass to the next. Over a wheat field in a 12-day window, Sentinel-1 C-band coherence can drop below 0.3, which is effectively noise. Persistent-scatterer (PS) methods partially recover the signal by identifying individual stable reflectors, but rural mining districts often lack the density of hard scatterers needed: typically fewer than five PS per square kilometre over open farmland, compared with hundreds per square kilometre in a town.
ALOS-2 L-band at 23.6 cm wavelength penetrates the canopy and interacts with the soil and woody stems beneath it. Temporal coherence over agricultural land at 14-day intervals is measurably higher than C-band, enabling small-baseline subset (SBAS) processing across vegetated areas where PS-InSAR produces only sparse results. The trade-off is sensitivity: a full wavelength of phase change corresponds to 11.8 cm of line-of-sight displacement for L-band versus 2.8 cm for C-band, so L-band is less sensitive to very small motions but more likely to produce a usable measurement at all.
X-band systems such as COSMO-SkyMed are best reserved for infrastructure corridors, road verges and building stock above known workings, where dense scatterers allow millimetre-class PS detection. Combining all three wavelengths across a study area is not redundant; it is the only way to avoid systematic blind spots.
The archive is the asset
Sentinel-1 data extends back to April 2014 for Europe, with some areas covered from late 2014. ALOS-2 began systematic acquisition in 2014; its predecessor ALOS PALSAR ran from 2006 to 2011. For many British and European coalfields, this means a decade or more of archived SAR imagery exists over ground that was already subsiding before anyone commissioned a study.
Processing the archive with SBAS or PS-InSAR produces a time-series of surface velocity at each coherent point, typically expressed in millimetres per year in the satellite line-of-sight direction. Velocities exceeding 5 to 10 mm per year over a spatially coherent bowl of 20 to 100 m diameter, in an area with documented shallow workings, constitute a credible precursor signal. Velocities below 2 to 3 mm per year are within the noise floor of most processing chains and should not be interpreted as motion. Honest uncertainty bounds on each measurement point are not optional; they are the difference between a defensible risk assessment and a liability.
What the method cannot do
InSAR measures displacement in the satellite line-of-sight direction, which for Sentinel-1 ascending geometry is roughly 23 degrees from vertical. Purely vertical subsidence is recovered well. Horizontal motion, which accompanies some pillar failures near old roadways, is only partially captured and requires combining ascending and descending passes to decompose.
Crown-hole collapse itself is almost never captured in an interferogram. The event is instantaneous on SAR timescales, the surface rupture destroys coherence, and the resulting void is typically 2 to 15 m across, smaller than the resolution cell of most SAR modes used for wide-area mapping. What InSAR delivers is a precursor velocity map and a ranked list of locations showing anomalous motion. It does not replace ground investigation; it tells a geotechnical engineer where to look first.
Bare peat, saturated clay and freshly ploughed soil all produce decorrelation that mimics the absence of signal. A negative result over farmland is not evidence of stability. It is evidence of insufficient scatterers, and should be flagged as a data gap rather than a clean bill of health.
Turning a velocity map into a ranked risk layer
The analytic workflow begins with a mining void cadastre, the spatial record of known workings from national mine abandonment plans or historical maps. In Britain, the Coal Authority publishes development high-risk areas derived from its mine-entry database, which provides a useful spatial prior. The InSAR velocity field is then intersected with this prior: locations where measured subsidence bowls spatially coincide with documented pillar grids at the correct depth-to-width ratio receive the highest risk rank.
Bowl geometry carries diagnostic information. A subsidence trough with a width-to-depth ratio consistent with the known seam depth and a Gaussian or near-Gaussian profile is characteristic of distributed pillar compression. A sharper, smaller anomaly may indicate a localised pillar failure or a previously unknown void. Anomalies that do not correspond to any mapped working are flagged separately for historical archive investigation.
Satellize applies this workflow using open Sentinel-1 and ALOS-2 archives, with commercial COSMO-SkyMed tasking added where infrastructure is at risk. Output is a GIS layer of ranked anomalies with velocity time-series at each point, formatted for direct ingestion into a client's risk management system. The same analytic pipeline that underpins the Tonga crop-estimation programme handles the time-series extraction; the physics is different but the data engineering is not.
What to commission and in what order
A sensible starting point is a retrospective SBAS analysis of the full Sentinel-1 archive over the area of interest, processed separately for ascending and descending geometries. This costs nothing in data acquisition and typically returns a velocity map within two to three weeks of processing time. It will immediately identify whether the area has sufficient PS density to support C-band analysis or whether an L-band campaign is needed.
If C-band coverage is sparse, a six-month ALOS-2 acquisition campaign covering four to six passes provides enough imagery for a preliminary SBAS solution. Twelve months gives a more reliable velocity estimate and begins to distinguish seasonal soil-moisture signals from genuine structural motion. Twelve months is the minimum credible monitoring period for any site where a planning or insurance decision rests on the result.
If you have a specific site with documented workings and a planning application pending, request a quote for a targeted PS-InSAR analysis of that polygon with uncertainty-bounded velocity outputs and a written geohazard interpretation. That is the deliverable that satisfies a planning authority or insurer, not a raw displacement raster.
Typical figures
| Spatial resolution (Sentinel-1 IW mode) | 5 x 20 m ground range (multi-looked to ~14 m for interferometry) |
| Spatial resolution (ALOS-2 Stripmap) | 3 x 3 m |
| Spatial resolution (COSMO-SkyMed Spotlight) | < 1 m |
| Revisit period | 6 days (Sentinel-1, two satellites); 14 days (ALOS-2); 1 to 4 days (COSMO-SkyMed, tasked) |
| Minimum detectable velocity (PS-InSAR, urban) | 1 to 2 mm/year line-of-sight (C-band, high PS density) |
| Minimum detectable velocity (SBAS, rural L-band) | 3 to 5 mm/year line-of-sight; poorer where coherence is marginal |
| Archive depth | Sentinel-1: from April 2014 (Europe); ALOS-2: from 2014; ALOS PALSAR: 2006 to 2011 |
| DEM vertical accuracy (GLO-30) | < 4 m LE90 over most terrain; used for phase flattening, not subsidence measurement |
| Delivery formats | GeoTIFF velocity rasters, Shapefile or GeoPackage anomaly layers, CSV time-series per PS point, PDF geohazard interpretation report |
| Processing latency (retrospective archive analysis) | 2 to 4 weeks from commission to delivery, depending on archive volume and area extent |
Analytics Satellize can run
| Retrospective surface velocity map | SBAS-InSAR or PS-InSAR on Sentinel-1 and/or ALOS-2 archive; standard published processing chains (e.g. SNAP, StaMPS, MintPy) | GeoTIFF line-of-sight velocity raster with per-pixel uncertainty, ascending and descending geometries separately |
| Vertical and horizontal displacement decomposition | Combination of ascending and descending velocity fields using published vector decomposition geometry | GeoTIFF vertical velocity layer and east-west horizontal velocity layer, with propagated uncertainty |
| Subsidence bowl detection and characterisation | Spatial clustering of anomalous PS or SBAS points; Gaussian profile fitting to identify bowl width, depth and centre; comparison with expected depth-to-width ratios from published void mechanics | GeoPackage layer of detected bowls with geometry attributes, ranked by velocity magnitude and spatial coherence |
| Risk-ranked anomaly register | Spatial intersection of InSAR anomalies with mine abandonment cadastre (e.g. Coal Authority development high-risk areas) and depth-consistency screening | Tabular register of anomalies with risk rank, mining context, velocity statistics and recommended investigation priority |
| Displacement time-series at nominated points | PS or SBAS time-series extraction at user-specified coordinates; seasonal signal separation using published harmonic decomposition | CSV and chart of cumulative displacement versus date at each point, with seasonal component identified separately from trend |
| Ongoing monitoring alert feed | Incremental PS-InSAR update on each new Sentinel-1 acquisition; threshold exceedance detection against baseline velocity | Automated alert (email or API) when velocity at a monitored point exceeds a client-defined threshold, with supporting interferogram |
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.