Illegal amber extraction pit and landscape scarring detection
Illicit amber mining in Ukraine's Rivne and Zhytomyr oblasts leaves dense pit fields and pump-discharge channels that are legible in multispectral satellite imagery. Object-based analysis of Sentinel-2 and PlanetScope data can map active zones, track seasonal spread and quantify disturbed area for enforcement agencies.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at mid-latitudes with two satellites. Bare-soil and moisture indices (NDVI, BSI, NDWI) computed from 13 spectral bands distinguish freshly excavated sandy spoil from surrounding forest. Free archive from 2015.
- PlanetScope SuperDove: 3 m resolution, daily revisit globally. Eight spectral bands including red-edge allow finer delineation of individual pit clusters and detection of smaller disturbance patches below Sentinel-2's practical mapping threshold. Commercial licence required.
- Maxar WorldView-2: 46 cm panchromatic, 1.85 m multispectral. Eight spectral bands including coastal blue and near-infrared-2. Used for validation of pit boundaries and morphological characterisation of spoil mounds and pump-channel geometry. Tasked on demand; archive coverage of affected oblasts is patchy.
- Airbus Pléiades: 50 cm panchromatic, 2 m multispectral. Tri-stereo capability allows digital surface model generation to estimate spoil-mound volume and confirm canopy removal depth. Useful for court-ready documentation of individual sites.
What a pit field looks like from 786 km up
Illegal amber extraction in Ukraine's Polesia region follows a recognisable pattern. Miners use high-pressure water pumps to liquefy the sandy substrate, then sluice amber fragments to the surface. The result is a field of circular to irregular pits, typically 2 to 15 metres across, surrounded by pale spoil mounds of displaced sand. Pump-discharge channels carry turbid water outward, leaving linear bright signatures that persist for weeks after extraction stops.
In Sentinel-2 imagery, freshly disturbed sandy soil produces a strong positive signal in the bare-soil index (BSI), calculated from shortwave-infrared band 11 and red band 4, against the surrounding dark forest. NDVI drops sharply to near zero over active pits. The combination is distinctive: dense clusters of low-NDVI, high-BSI pixels inside forested terrain, with no agricultural or infrastructure explanation. At 10 m resolution, individual pits are sub-pixel, but pit clusters of a few hundred square metres are detectable. PlanetScope at 3 m begins to resolve individual pit outlines, and WorldView-2 at 1.85 m multispectral makes them unambiguous.
Why cloud and canopy are the real adversaries
Polesia receives significant cloud cover, particularly in autumn and winter. A single Sentinel-2 pass is often unusable. Practical mapping relies on multi-temporal compositing: stacking cloud-masked acquisitions over a 30 to 60-day window and selecting the least-clouded pixel per location, a method well-established in the published literature on tropical deforestation monitoring and directly applicable here.
Forest canopy is the harder problem. Mature pine canopy in the Rivne and Zhytomyr forests partially conceals early-stage pit fields. Extraction that begins under closed canopy may not produce a detectable bare-soil signal until canopy removal is substantial. This is an honest limit: satellite optical methods detect the surface consequence of mining, not mining itself. Sites operating under intact canopy for their first weeks are likely to be missed. SAR coherence change detection using Sentinel-1 C-band can flag canopy disturbance before bare soil is fully exposed, and is a useful complement, though it is covered separately in the sibling page on illegal logging road detection.
Seasonal timing matters. Spring and early summer, when deciduous understorey is leafing out but extraction is active, can produce ambiguous signals. Late summer and early autumn, after months of activity have widened pits and removed canopy, give the clearest optical signatures.
Object-based analysis: why pixels alone are not enough
Per-pixel classification of bare soil flags agricultural fields, sand bars and construction sites alongside mining pits. Object-based image analysis (OBIA) resolves much of this ambiguity. OBIA segments the image into spectrally and spatially coherent objects, then classifies objects using shape, texture and context rules alongside spectral indices.
Amber pit clusters have a characteristic morphology: irregular aggregations of small bright objects inside forested land cover, with no road network or building footprint nearby. Agricultural bare soil forms large, geometrically regular objects adjacent to field boundaries. Construction sites sit within or adjacent to settlements. These contextual rules, applied in published OBIA frameworks such as eCognition-style multiresolution segmentation, substantially reduce false positives. Published remote-sensing studies of artisanal mining in other contexts report producer accuracies above 85 percent for pit detection using OBIA on sub-5 m imagery, though figures for amber specifically depend on pit density and canopy cover at each site.
Tracking expansion: change detection across seasons and years
A single map of current disturbance is useful. A time series is what enforcement agencies actually need. Quarterly composites from Sentinel-2 going back to 2015, when the archive begins, allow reconstruction of how pit fields have grown, contracted during enforcement operations, and re-expanded. The Sentinel-2 archive is free and open, which means historical analysis has no incremental data cost.
PlanetScope's daily cadence, available from around 2016 for most of Ukraine, allows finer temporal resolution where the commercial licence is in place. Change between consecutive monthly composites can be expressed as a disturbed-area delta in hectares, giving enforcement planners a quantified measure of activity intensity. Periods of rapid expansion correlate with high amber-market prices and reduced law-enforcement presence, patterns that have been documented in investigative journalism covering the Rivne region.
Baltic amber zones in Poland and Lithuania present the same spectral and morphological signatures. Enforcement agencies in those countries face analogous challenges: forested terrain, seasonal cloud, and extraction that moves quickly once a site is detected. The analytical pipeline transfers directly.
From map to evidence: what the data can and cannot prove
Satellite-derived disturbance maps establish the spatial extent and temporal progression of landscape scarring. They do not identify individual perpetrators, establish ownership of the land, or prove that the extracted material is amber rather than sand. For legal proceedings, satellite evidence typically functions as corroboration for ground inspections and financial investigations, not as standalone proof.
Accuracy statements matter in court. Any delivered product should carry an honest uncertainty estimate: minimum detectable patch size (practically around 0.05 ha for Sentinel-2-based detection, smaller for PlanetScope), commission and omission error rates from validation, and clear documentation of the cloud-masking and compositing method used. Pléiades tri-stereo imagery, which can generate digital surface models accurate to around 1 m vertically, adds a volumetric dimension: spoil-mound volume estimates give a rough proxy for the quantity of material excavated, which has evidentiary value in establishing scale of operation.
Satellize runs this kind of multi-sensor disturbance analysis on open and commercial constellations. The same OBIA and change-detection pipeline used in the Tonga crop-estimation programme adapts to extraction-pit mapping with adjustments to the training class definitions and index thresholds.
Practical limits a procurement officer should know
Minimum detectable disturbance at Sentinel-2 resolution is approximately 0.05 to 0.1 ha for a coherent pit cluster, assuming reasonable cloud-free compositing. Isolated single pits of 2 to 5 m diameter are below detection without commercial very-high-resolution tasking. Revisit is effectively 5 days for Sentinel-2 in clear conditions, but cloud-free composites over Polesia in winter may require 60-day windows, meaning rapid-response alerting is seasonal.
PlanetScope daily revisit improves temporal resolution substantially but the 3 m pixel means pit boundaries are approximate rather than precise. For legal-grade boundary mapping, Pléiades or WorldView-2 tasking is needed, at commercial cost and with lead times of 24 to 72 hours depending on cloud probability. Archive coverage of specific sites before 2016 is sparse across all commercial sensors.
Spectral confusion with legal sand extraction and forestry clear-cuts is real. Contextual rules and ancillary land-use data reduce but do not eliminate false positives. Any operational system should include a human review step before alerts are passed to enforcement.
Typical figures
| Spatial resolution (mapping) | 10 m (Sentinel-2), 3 m (PlanetScope), 1.85 m multispectral / 46 cm pan (WorldView-2) |
| Revisit cadence | 5 days (Sentinel-2 twin satellites), daily (PlanetScope), on-demand tasking (WorldView-2, Pléiades) |
| Minimum detectable disturbance | ~0.05–0.1 ha pit cluster (Sentinel-2 OBIA); individual pits from ~3 m diameter with PlanetScope |
| Key spectral bands | Red (B4), NIR (B8), SWIR (B11, B12) for BSI and NDWI; red-edge (B5–B7) for vegetation stress; coastal blue for turbidity in discharge channels |
| Archive depth | Sentinel-2 from 2015 (free); PlanetScope from ~2016 (commercial licence); Maxar archive variable by location |
| Cloud-masking approach | Multi-temporal compositing over 30–60 day windows using SCL cloud mask (Sentinel-2) or UDM2 (PlanetScope) |
| Vertical accuracy (Pléiades stereo DSM) | ~1 m RMSE for spoil-mound volume estimation |
| Latency (change alert) | 24–72 hours after cloud-free acquisition; seasonal degradation in winter over Polesia |
| Coverage | Global; Rivne and Zhytomyr oblasts fully covered by Sentinel-2 and PlanetScope |
| Delivery formats | GeoTIFF disturbance raster, GeoJSON pit-cluster polygons, PDF enforcement report, change-delta time-series CSV |
Analytics Satellize can run
| Baseline disturbance map | OBIA multiresolution segmentation on Sentinel-2 BSI and NDVI composites; contextual land-use filtering | GeoJSON polygon layer of all detected pit clusters with area in hectares and first-detection date |
| Quarterly change-detection report | Bitemporal comparison of cloud-masked Sentinel-2 composites; disturbed-area delta calculation | PDF report with maps and hectare-change table; GeoTIFF showing gain and loss by quarter |
| High-resolution site confirmation | PlanetScope or Pléiades OBIA for individual pit boundary delineation and spoil-mound identification | Site-level GeoJSON with pit count, total disturbed area and morphological classification |
| Spoil-mound volume estimate | Pléiades tri-stereo digital surface model differenced against pre-disturbance SRTM baseline | Per-site volume estimate in cubic metres with stated vertical uncertainty; court-ready metadata |
| Active-extraction alert feed | Near-daily PlanetScope change detection flagging new bare-soil objects within monitored area boundary | GeoJSON alert feed with timestamp, location and estimated new disturbed area; email or API notification |
| Multi-year expansion time series | Annual Sentinel-2 composite stack from 2015 to present; cumulative disturbed-area curve per administrative boundary | Time-series CSV and chart; animated GIF of annual disturbance maps for briefing use |
| False-positive audit layer | Cross-reference of flagged objects against cadastral land-use data and OSM road/settlement buffers | Reviewed GeoJSON with confidence classification (high / medium / requires ground check) per polygon |
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.