Mine and quarry rehabilitation progress monitoring for land reuse
Multispectral change detection and vegetation indices applied to Sentinel-2, Landsat and commercial imagery give regulators and developers an independent, time-stamped account of whether a former extractive site is actually recovering, or just appears to be.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at mid-latitudes with both satellites. Bands 4, 8 and 8A support NDVI and red-edge indices sensitive to sparse pioneer vegetation. Free archive from 2015.
- Landsat archive (Landsat 1–9): 30 m multispectral resolution; Landsat 1 data from 1972 provides the deepest publicly available baseline for any site. Landsat 8 and 9 offer 16-day revisit with 30 m OLI bands and 100 m thermal. USGS Collection 2 surface reflectance products are analysis-ready.
- Planet SuperDove: 3 m resolution, daily revisit globally, eight bands including red-edge. Useful for detecting small revegetation patches, individual spoil-heap movements and subtle surface texture changes that 10 m imagery misses. Commercial licence required.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral. Suitable for void-edge mapping, berm geometry verification and surface crack detection at a level of detail relevant to geotechnical assessments. Tasked on demand; not an archive product.
What the spectral record actually captures on a recovering mine site
Bare spoil, compacted overburden and residual mineral dust reflect very differently from living vegetation. The Normalised Difference Vegetation Index, computed from red and near-infrared reflectance, rises from near zero on bare ground toward 0.6–0.8 on dense grassland or scrub. That gradient is measurable at Sentinel-2's 10 m resolution. More usefully, the red-edge bands (Sentinel-2 bands 5 and 6, centred near 705 nm and 740 nm) are sensitive to chlorophyll content at low canopy densities, which matters on sites where pioneer species are sparse and patchy. A site that looks green in a photograph may still show suppressed red-edge response if the vegetation is stressed or thin.
Void infill is a different problem. Where backfilling has occurred, the surface texture and reflectance change, but the most reliable signal is topographic: repeat digital surface models derived from stereo Pléiades Neo imagery or from SAR coherence methods can detect elevation changes of less than a metre over areas of a few hundred square metres. Combining spectral and topographic signals gives a more complete picture than either alone.
Building a baseline that predates the planning application
Landsat's archive is the longest continuous land-surface record in existence. Landsat 1 acquired its first imagery in July 1972; Landsat 5 operated until 2013, giving five decades of 30 m multispectral data over most of the Earth's land surface. For a quarry that opened in 1985 and closed in 2010, it is straightforward to reconstruct the pre-extraction spectral state, the progressive disturbance footprint and the post-closure trajectory, all from publicly available USGS Collection 2 data.
That historical depth matters legally. Rehabilitation bonds are often structured around a target condition defined at closure. If a developer or regulator disputes what 'recovered' looks like, the Landsat record provides an independent, court-admissible chronology. Sentinel-2 takes over from 2015 with finer spatial resolution and more spectral bands, so the transition from Landsat to Sentinel-2 is a step up in analytical capability, not a gap.
Honest limits: what satellite data cannot resolve
Cloud cover is the most persistent constraint. Temperate and tropical sites can lose weeks of usable imagery in winter or monsoon seasons. Persistent cloud can delay detection of a revegetation failure by a full growing season if the analyst relies on a single sensor. The practical mitigation is sensor fusion: Sentinel-2, Landsat 8/9 and Planet SuperDove have different orbital repeat times, so the joint probability of all three being cloud-obscured on the same day is lower than for any one alone. Even so, cloud is a real limitation and any honest monitoring programme should state the expected number of cloud-free acquisitions per year for the specific site latitude.
Spatial resolution sets a floor on detectability. At 10 m, a revegetation patch smaller than roughly 20–30 m across is likely to be mixed with bare soil in the same pixel, suppressing the NDVI signal. At 3 m (Planet SuperDove), patches of a few metres become detectable, but at additional cost and with less archive depth. Pléiades Neo at 30 cm can resolve individual plants, but that level of detail is expensive to acquire repeatedly and is best reserved for targeted verification rather than routine monitoring. Subsurface conditions, groundwater contamination and soil chemistry are entirely outside what optical or SAR satellite data can assess.
Change detection methods and how they are applied
The standard approach is bi-temporal differencing of a vegetation index, typically NDVI or the Soil-Adjusted Vegetation Index (SAVI, which applies a correction factor for exposed soil background and is more appropriate on partially revegetated sites than bare NDVI). A pixel-level change map is produced by subtracting the index value at a reference date from the value at the monitoring date, then thresholding the result to classify pixels as significantly improved, unchanged or degraded. The threshold is calibrated against the noise floor of the sensor and the natural seasonal variation at the site.
More sophisticated approaches use time-series methods such as LandTrendr (developed at Oregon State University and applied to the Landsat archive) to fit piecewise linear trajectories to each pixel's spectral history. This separates genuine long-term recovery trends from seasonal fluctuation and short-term disturbance events like topsoil stripping or reseeding. The output is a per-pixel characterisation of when recovery began, how fast it is progressing and whether it has stalled. That kind of trajectory analysis is directly relevant to bond-release assessments, where regulators need to judge not just current state but rate of progress toward a target.
Regulatory compliance and independent verification
Rehabilitation bonds in most jurisdictions require the site operator to demonstrate that vegetation cover has reached a specified percentage, that voids have been filled to a specified level and that surface stability has been achieved before the bond is released. Those conditions are typically assessed by on-the-ground survey, which is expensive, infrequent and subject to access disputes. Satellite-derived evidence does not replace a ground survey, but it provides an independent, continuous record that can flag non-compliance between survey visits and support or challenge the operator's own reporting.
Planning conditions attached to development approvals for former extractive sites often include staged rehabilitation milestones. A satellite monitoring programme can be structured to produce a compliance report at each milestone date, with a consistent methodology and a documented chain of custody for the imagery. Satellize has applied this kind of structured, archive-backed monitoring to agricultural land assessment in the Kingdom of Tonga crop-estimation programme; the same analytical framework transfers directly to rehabilitation compliance, where the question is similarly one of measuring a biological signal against a documented baseline over time.
For developers acquiring former mine or quarry sites, the satellite record also provides due diligence value. A site that shows stalled or reversed vegetation recovery in the Sentinel-2 time series is a signal worth investigating before exchange of contracts, not after.
Structuring a monitoring programme for a specific site
A practical programme for a mid-sized quarry of, say, 50–200 hectares would typically combine a Landsat historical baseline going back to the site's operational start, Sentinel-2 for annual and seasonal monitoring from 2015 onward, and Planet SuperDove for high-frequency checks during critical revegetation windows such as the first two growing seasons after seeding. Pléiades Neo tasking would be reserved for specific verification events: a bond-release inspection, a planning milestone, or investigation of an anomaly flagged by the lower-resolution time series.
The deliverable at each monitoring interval is a change map with per-class area statistics, a vegetation index trend chart for the site as a whole and for defined sub-zones, and a written assessment against the stated rehabilitation targets. That assessment should state explicitly what the satellite data can and cannot confirm, so that the regulator or developer knows what additional ground-truthing is needed. Satellite evidence is most powerful when it is honest about its own resolution floor.
Typical figures
| Spatial resolution (routine monitoring) | 10 m (Sentinel-2 MSI); 30 m (Landsat 8/9 OLI) |
| Spatial resolution (verification tasking) | 3 m (Planet SuperDove); 1.2 m multispectral / 30 cm pan (Pléiades Neo) |
| Revisit frequency | 5 days (Sentinel-2, mid-latitudes); 16 days (Landsat 8/9); daily (Planet SuperDove); on-demand (Pléiades Neo) |
| Archive depth | 1972 to present via Landsat 1–9; 2015 to present via Sentinel-2 |
| Key spectral bands | Red (665 nm), NIR (842 nm), red-edge (705 nm, 740 nm) for vegetation indices; SWIR (1610 nm, 2190 nm) for soil and moisture discrimination |
| Minimum detectable revegetation patch | Approximately 20–30 m diameter at 10 m resolution; approximately 5–10 m at 3 m resolution |
| Cloud cover constraint | Site-dependent; temperate sites typically achieve 15–30 usable Sentinel-2 acquisitions per year; tropical sites fewer |
| Topographic change detection | Sub-metre elevation change detectable via stereo Pléiades Neo DSM differencing over areas of several hundred square metres |
| Delivery formats | GeoTIFF change maps, GeoPackage vector outputs, PDF compliance reports, time-series CSV statistics |
Analytics Satellize can run
| NDVI and SAVI trend maps | Per-pixel vegetation index computation and bi-temporal differencing on Sentinel-2 and Landsat surface reflectance products | GeoTIFF change layer with per-class area statistics; PDF summary chart for each monitoring interval |
| Long-run spectral trajectory analysis | LandTrendr-class piecewise linear time-series fitting on Landsat archive pixels | Per-pixel recovery onset date and rate-of-change layer; GIS polygon output of stalled or regressing zones |
| Revegetation coverage percentage by sub-zone | Supervised classification of multispectral imagery into bare soil, sparse vegetation and dense vegetation classes, validated against site survey points | Zonal statistics table keyed to rehabilitation bond sub-areas; compliance status flag per zone |
| Void and spoil-heap surface change detection | Digital surface model differencing from stereo Pléiades Neo acquisitions; SAR coherence change detection as a lower-cost alternative | Elevation difference GeoTIFF; volume-change estimate for defined void polygons |
| Rehabilitation milestone compliance report | Structured comparison of satellite-derived metrics against documented planning conditions and bond targets at specified milestone dates | Dated PDF report with methodology statement, uncertainty bounds and explicit list of conditions requiring ground-truth verification |
| Pre-acquisition due diligence summary | Historical Landsat and Sentinel-2 archive review; anomaly flagging against expected recovery trajectory | Single-document briefing identifying spectral anomalies, stalled recovery zones and data gaps, with annotated imagery |
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.