Illegal mining disturbance impact on forest carbon project areas
Artisanal and small-scale mining expands into forest carbon project areas faster than ground teams can track. Bare-soil signatures, turbid river plumes, and pit geometries visible from orbit make the disturbance legible, and datable, independent of what the project developer reports.
Sensors
- Planet SuperDove: 3 m resolution, 8-band multispectral including red-edge and near-infrared. Daily revisit over most tropical latitudes. Sufficient to resolve individual pit perimeters and track week-on-week areal expansion. Cloud cover remains a practical constraint in wet-season Amazonia, where clear acquisitions can be sparse for days at a stretch.
- Sentinel-2 MSI: 10 m in visible and NIR bands, 20 m in shortwave infrared. Five-day revisit at the equator with both satellites operating. SWIR bands (1610 nm and 2190 nm) are particularly sensitive to bare-soil mineralogy and moisture, aiding discrimination of active mining spoil from older clearings. Free and openly archived since 2015.
- Sentinel-1 SAR: C-band synthetic aperture radar at 10 m (IW mode). Penetrates cloud cover entirely, which matters enormously in tropical forest regions. Detects structural change, loss of canopy backscatter, and standing water in pits, but cannot identify spectral mineralogy. Best used as an all-weather change trigger that cues optical follow-up.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral, plus 16 SWIR bands at 3.7 m. Tasked commercially. At this resolution, individual sluice boxes, access tracks, and spoil heaps are distinguishable. Revisit is irregular and costly; the practical use is targeted confirmation of anomalies flagged by coarser daily sensors, not routine monitoring.
Why mining disturbance is a carbon accounting problem, not just an environmental one
Forest carbon projects sell credits on the premise that a defined area of forest is being protected. When artisanal or small-scale mining, known in the literature as ASM, expands inside that boundary, the carbon stored in cleared biomass is released. If the developer does not report it, the credits sold against that area are unsupported by real sequestration. That is not a technicality. It is the core claim of the project, invalidated.
The problem is structural. ASM operations move fast, often operating at night or during overcast periods, and project ground teams are rarely resourced to patrol boundaries continuously. In the Amazon basin, gold-rush expansions have been documented advancing several kilometres in a matter of weeks. Voluntary carbon registries rely heavily on developer self-reporting, with periodic third-party audits that may be years apart. The gap between ground reality and reported status is where fraudulent or simply negligent credit issuance occurs.
What a mining scar looks like from orbit
Active ASM leaves a distinctive spectral and geometric signature. Bare laterite and alluvial spoil have high reflectance in the red and SWIR bands and low reflectance in NIR, producing normalised difference vegetation index values close to zero or negative. This contrasts sharply with surrounding forest canopy, which has NDVI values typically above 0.7. Sentinel-2's SWIR bands add mineralogical discrimination: iron-rich soils common to gold-bearing laterites show characteristic absorption features that differ from agricultural clearings or logging gaps.
Geometry matters too. ASM pits tend to be small, irregular, and clustered along stream channels or ridge lines following ore deposits. Logging clearings follow road networks. Agricultural conversion produces rectilinear or arc-shaped boundaries. At 3 m (Planet SuperDove) or 10 m (Sentinel-2), these geometrical differences are resolvable, and machine-learning classifiers trained on labelled examples can separate ASM from other disturbance types with reasonable accuracy, though confusion with small-scale agriculture on similar soils remains a known source of error.
Water turbidity is a secondary but valuable indicator. Mercury and sediment from sluicing operations turn river reaches downstream of active sites visibly brown-orange. Sentinel-2's red and green bands capture this turbidity plume. Tracing the plume upstream can locate active sites even when direct canopy views are obscured by cloud, because rivers are often clearer in imagery than the surrounding forest.
SAR fills the cloud gap that optical sensors cannot
Tropical forest regions, particularly in the Congo Basin and western Amazon, can be cloud-covered for weeks continuously during wet season. An optical-only monitoring approach will miss disturbance events that begin and end between clear acquisitions. Sentinel-1 C-band SAR provides reliable acquisitions regardless of cloud, with a six-day repeat at the equator (twelve days for a single satellite).
Forest canopy produces a characteristic volume-scattering SAR return. When canopy is removed, the return drops sharply and the surface-scattering signal from bare soil or standing water dominates. This change is detectable within one or two SAR passes of the clearing event. The limitation is that SAR alone cannot confirm the cause of the change: a mining pit, a windthrow gap, and a small agricultural clearing can produce similar backscatter signatures. The practical workflow fuses SAR alerts with optical confirmation, using SAR as the trigger and optical imagery, when available, for classification and area measurement.
Turning detections into carbon accounting evidence
Detection of a mining scar is only the first step. Carbon auditors need area estimates, dates of disturbance onset, and ideally a trajectory showing whether the disturbance is active or stabilising. Time-series analysis of the Sentinel-2 archive, which extends to 2015, allows retrospective dating of clearing events to within the revisit interval. Planet's archive, available commercially from around 2016, can refine that to weekly precision in many areas.
Area measurement carries uncertainty that should be stated honestly. At 10 m resolution, the minimum reliably detectable clearing is roughly 0.1 hectares, though this depends on the contrast between the clearing and its surroundings. Sub-pixel mixing at clearing edges introduces area errors that compound when aggregating many small pits. At 3 m, the minimum detectable area drops to around 0.01 hectares, but Planet tasking is not continuous everywhere and archive gaps exist.
For leakage accounting, the relevant question is whether ASM activity displaced by the project boundary has shifted to adjacent areas. This requires monitoring a buffer zone outside the project boundary, not just the project area itself. The same spectral methods apply; the analysis simply needs to be spatially extended.
Limits that an honest analyst will name upfront
Cloud cover is the most significant operational constraint. In the wettest tropical regions, optical revisit gaps of two to four weeks are common even with Planet and Sentinel-2 combined. A fast-moving ASM front can clear several hectares in that window without leaving a clear optical record of the onset date.
Spectral confusion with other bare-soil disturbance types, particularly recent road construction and small-scale agriculture on laterite soils, means that automated classifiers produce false positives that require human review. The mineralogical SWIR bands on Sentinel-2 and WorldView-3 reduce but do not eliminate this ambiguity.
Very small or highly shaded pits, such as those in steep terrain or under partial canopy, may fall below detection thresholds at 10 m. WorldView-3 tasking can address this for specific suspect locations, but it is not economically viable as a routine monitoring tool across large project areas. SAR is insensitive to the mineralogical and colour differences that distinguish ASM from other disturbance; it detects structural change only.
Satellize applies this detection stack, combining Sentinel-1 change alerts, Sentinel-2 spectral classification, and targeted commercial tasking, to carbon project monitoring engagements. The methodology is the same class used in the organisation's Tonga crop-estimation programme: open-constellation baselines with commercial sensor uplift where precision demands it.
What the output package looks like for a registry auditor
A credible monitoring package for a carbon registry audit contains several components. A georeferenced change map showing all detected disturbance polygons within and adjacent to the project boundary, with dates of first detection and estimated area. A classification layer distinguishing ASM-probable from other disturbance types, with a confidence score and a note on which signatures drove the classification. A time-series chart for each significant polygon showing the NDVI or SAR backscatter trajectory before, during, and after the disturbance event. And a turbidity-plume record for any river systems that pass through or border the project area.
This evidence package can be delivered as GIS layers (GeoJSON or GeoTIFF), a PDF report with annotated imagery, or a structured data feed that integrates with registry audit workflows. The archive depth available from Sentinel-2 and Landsat (the latter extending to the 1970s for coarser historical context) means that baseline conditions prior to project registration can be reconstructed, which is essential for additionality and permanence assessments.
Typical figures
| Spatial resolution (routine monitoring) | 10 m (Sentinel-2 MSI); 3 m (Planet SuperDove) |
| Spatial resolution (targeted confirmation) | 31 cm pan / 1.24 m MS (WorldView-3, tasked on demand) |
| SAR resolution | 10 m (Sentinel-1 IW mode, cloud-independent) |
| Revisit (optical) | Daily (Planet SuperDove, subject to cloud); 5 days at equator (Sentinel-2 A+B) |
| Revisit (SAR) | 6 days at equator (Sentinel-1 A+B combined) |
| Minimum detectable clearing area | ~0.1 ha at 10 m; ~0.01 ha at 3 m (contrast-dependent) |
| Key spectral bands for ASM detection | Red, NIR, SWIR-1 (1610 nm), SWIR-2 (2190 nm); red-edge for transition-zone mapping |
| Archive depth | Sentinel-2 from 2015; Landsat for coarser historical context from 1972; Planet commercially from ~2016 |
| Alert latency | 24–72 hours from acquisition to flagged change, depending on cloud and processing pipeline |
| Delivery formats | GeoTIFF, GeoJSON, PDF audit report, structured data feed |
Analytics Satellize can run
| ASM disturbance polygon map | NDVI and SWIR spectral change detection on Sentinel-2 time series; geometric classifier to separate ASM from agricultural and logging signatures | GeoTIFF and GeoJSON layer with polygon boundaries, first-detection dates, and estimated area per polygon |
| All-weather change alert | Sentinel-1 SAR backscatter change detection (log-ratio thresholding on IW GRDH product) | Automated alert feed with coordinates and acquisition date, triggering optical follow-up tasking |
| River turbidity plume trace | Suspended sediment index derived from Sentinel-2 red and green bands; upstream source attribution by channel network tracing | Georeferenced plume polygons with upstream origin estimate and date series |
| Disturbance classification confidence layer | Random forest or gradient-boosted classifier trained on labelled ASM, agriculture, and logging samples using Sentinel-2 multispectral and Sentinel-1 dual-polarisation features | Raster confidence layer (ASM-probable / other disturbance / stable forest) with per-class probability scores |
| Leakage buffer monitoring report | Same spectral and SAR change detection applied to a defined buffer zone outside the project boundary; comparison of disturbance rates inside and outside boundary | Quarterly PDF report with disturbance area tables, trend charts, and annotated imagery for registry submission |
| Retrospective baseline reconstruction | Annual composite analysis of Sentinel-2 and Landsat archive to establish pre-project forest condition and any prior disturbance history | Time-series GeoTIFF stack with annual disturbance area statistics from 2015 (Sentinel-2) or earlier (Landsat) |
| High-resolution confirmation imagery | Targeted WorldView-3 tasking over anomalous polygons; manual and semi-automated feature extraction of pit structures, access tracks, and equipment | Annotated 30 cm imagery tiles with feature inventory (pit count, estimated active area, infrastructure visible) |
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.