Artisanal mining deforestation frontier advance rate mapping
Artisanal mining expands through forest clearance that precedes bare-earth excavation by days to weeks. SAR coherence loss and optical NDVI change, read at weekly cadence, map the advancing frontier independently of the pit itself.
Sensors
- Sentinel-1 SAR (C-band, ESA): 10 m ground range resolution in IW mode; 6-day repeat at the equator with both satellites active. C-band backscatter drops sharply when forest canopy is removed, and interferometric coherence collapses within one pass after clearance, giving a cloud-independent alert within roughly 12 days of the event.
- PlanetScope (Planet Labs): 3 m native resolution, daily revisit globally in clear conditions. Four-band (blue, green, red, NIR) imagery supports NDVI differencing at fine spatial detail, resolving clearings as small as a few hundred square metres. Cloud cover remains the limiting factor; daily revisit compensates by increasing the probability of a clear observation within any given week.
- Sentinel-2 MSI (ESA): 10 m in visible and NIR bands, 5-day revisit with both satellites. Thirteen spectral bands allow NDVI, SWIR-based burn scar discrimination, and red-edge indices that distinguish stressed but standing vegetation from fully cleared ground. Free and open archive to 2015.
- RADARSAT Constellation Mission (CSA): Three satellites providing 4-day exact repeat in compact polarimetry mode, 5 m to 100 m selectable resolution. Compact polarimetry preserves more scattering-mechanism information than single-pol SAR, improving discrimination between forest, low regrowth, and bare earth in humid tropical conditions.
Why the frontier matters more than the pit
Open-pit detection from satellite is well understood: bare earth reflects differently from forest, and the geometry of an excavation is hard to hide. The problem is timing. By the time a pit is visible, the decision to mine has already been acted upon, the forest is gone, and any enforcement response is chasing history.
The deforestation frontier is the earlier signal. Artisanal and small-scale mining (ASM) operations in forested terrain typically clear a working area before breaking ground, felling trees to establish access tracks, processing zones, and buffer space around the intended excavation. This clearing phase can precede the bare-earth signature by days to several weeks. Mapping the rate and direction of that frontier gives regulators and land managers a lead indicator rather than a lagging one.
What SAR coherence loss actually measures
Synthetic aperture radar coherence is a measure of how similar the phase of two SAR acquisitions is over the same ground. Intact forest canopy produces low coherence at C-band because wind moves leaves and branches between passes. Bare earth, by contrast, produces high coherence: the surface is stable. The transition from low to high coherence therefore marks the moment of clearance, not the moment of observation.
Sentinel-1's 6-day repeat (12-day with a single satellite) means a coherence change can be detected within one to two passes of the clearance event. The practical alert latency, including processing time, is typically under two weeks. Importantly, this works through cloud. The Amazon and Congo basins, where a large share of forested ASM activity occurs, can sustain weeks of continuous cloud cover that renders optical sensors blind. SAR is unaffected.
The limit worth stating plainly: C-band coherence is sensitive to surface moisture as well as vegetation change. Heavy rainfall on bare soil can temporarily depress coherence, producing false positives. Cross-validation with optical data, when cloud permits, reduces this ambiguity materially.
Optical NDVI change at 3 m: what it adds and what it cannot do
PlanetScope's 3 m resolution resolves individual clearings that a 10 m product would undercount. A 30 by 30 m clearing, common in early-stage ASM, occupies only nine Sentinel-2 pixels; at PlanetScope resolution it contains 900 pixels, enough to characterise shape, edge regularity, and the presence of tracks. These geometric cues help separate ASM clearance from smallholder agriculture, which tends to produce more regular field boundaries.
NDVI differencing between a baseline composite and a current composite flags pixels where green biomass has been removed. The GLAD forest alert system, published by the University of Maryland and operating on Landsat-8 and -9 at 30 m, demonstrates that weekly compositing with per-pixel confidence scoring can push alert latency below eight days in regions with adequate clear-sky frequency. PlanetScope's daily revisit can achieve comparable or better latency at finer resolution, though the processing pipeline to handle the data volume is non-trivial.
The honest constraint: cloud cover in equatorial forest regions regularly exceeds 80 percent on any given day. A weekly cloud-free observation is not guaranteed everywhere. SAR and optical should be treated as complementary, not interchangeable.
Measuring advance rate and direction
A single alert locates a clearing. A time series of alerts, aggregated over weeks to months, reveals something more useful: the vector of expansion. By fitting a convex hull or directional kernel to successive frontier positions, it is possible to estimate the rate of advance in metres per week and the predominant bearing of movement. This matters operationally because it allows a prediction of where the frontier will be in 30 or 60 days if the current trajectory continues.
Published studies using Landsat time series in the Peruvian Amazon have demonstrated that ASM-driven deforestation can advance at rates of several hundred metres per year along river corridors, with acceleration phases tied to commodity price cycles. SAR-plus-optical fusion at weekly cadence improves temporal resolution by roughly an order of magnitude relative to Landsat-only approaches, which is the relevant comparison for near-real-time monitoring.
Frontier advance rate is also a useful normalisation for comparing sites. Two sites with equal total cleared area may have very different risk profiles if one has been static for six months and the other is advancing at 50 m per week.
Data limits that an honest programme must account for
Spatial resolution sets a floor on detectable clearing size. At 10 m (Sentinel-1 IW), clearings smaller than roughly 0.01 hectares are below reliable detection. PlanetScope lowers that floor considerably, but its commercial archive is not free and coverage is not universal. Sentinel-2 sits between the two in both resolution and cost.
Temporal gaps matter. A site that clears rapidly during a period of persistent cloud, then stabilises before the next clear optical observation, may be underestimated in total area cleared. Fusion of SAR and optical records is the standard mitigation, but it introduces its own classification complexity at the boundary between sensor types.
Attribution is a separate problem from detection. Satellite data can confirm that a clearing occurred and estimate when, but distinguishing ASM from legal logging, subsistence agriculture, or infrastructure clearance requires contextual layers: concession boundaries, road networks, proximity to known mineralised zones, and, where available, ground truth. Satellize integrates these contextual layers into frontier alert products for clients operating in high-risk jurisdictions, drawing on the same open-constellation pipeline used in the Tonga crop-estimation programme.
From alert to actionable output
The practical deliverable for a government or enterprise client is not a raster of coherence change. It is a weekly polygon layer showing new clearings above a minimum area threshold, attributed with estimated clearance date, area, frontier advance vector, and a confidence score derived from the number of independent sensor observations supporting the alert. Delivered as a GeoJSON or shapefile feed into an existing GIS, this integrates directly with ranger dispatch systems or compliance reporting workflows.
The advance rate metric, computed monthly from the weekly polygon series, supports a different audience: policy analysts and supply-chain due-diligence teams who need a trend indicator rather than a point alert. A chart showing frontier advance rate in metres per week over a 12-month period, overlaid on commodity price data, is a different kind of evidence from a map, and often more persuasive in a regulatory submission.
If you are scoping a frontier monitoring programme for a specific concession boundary or national park buffer zone, the right starting point is a historical baseline run over the past 24 months of Sentinel-1 and Sentinel-2 archive, which is free and open, to establish pre-programme deforestation rates before any operational alerting begins.
Typical figures
| Spatial resolution (SAR) | 10 m (Sentinel-1 IW mode); 5–25 m selectable (RADARSAT Constellation Mission) |
| Spatial resolution (optical) | 3 m (PlanetScope); 10 m (Sentinel-2 visible/NIR); 30 m (Landsat-8/9) |
| Revisit cadence | 6 days (Sentinel-1, dual satellite); daily (PlanetScope, cloud-dependent); 5 days (Sentinel-2); 4 days (RADARSAT Constellation) |
| Alert latency (SAR coherence) | Typically 6–14 days from clearance event to processed alert |
| Alert latency (optical NDVI) | 2–8 days from clearance event under clear-sky conditions; indeterminate under persistent cloud |
| Minimum detectable clearing (Sentinel-1) | Approximately 0.01 ha (one to four pixels); smaller clearings below reliable detection threshold |
| Spectral bands used | C-band VV/VH (SAR); NIR and red for NDVI; SWIR for burn/moisture discrimination (Sentinel-2 bands 11, 12) |
| Cloud penetration | Full (SAR); none (optical — mitigated by high revisit and compositing) |
| Archive depth (open data) | Sentinel-1 from 2014; Sentinel-2 from 2015; Landsat from 1972 (30 m) |
| Delivery formats | GeoJSON polygon feed, GeoTIFF change raster, CSV advance-rate time series, PDF compliance report |
Analytics Satellize can run
| Weekly new-clearing alert layer | SAR coherence change detection (Sentinel-1 6-day interferometric pair) fused with PlanetScope NDVI differencing against rolling 30-day baseline composite | GeoJSON polygon feed with clearance date estimate, area, and per-polygon confidence score; updated weekly |
| Monthly frontier advance rate metric | Directional kernel density estimation applied to cumulative weekly clearing centroids; advance vector fitted by least-squares to successive frontier convex hulls | CSV time series and chart: advance rate in m/week and bearing, per defined area of interest |
| 24-month historical baseline deforestation rate | Sentinel-1 coherence stack and Sentinel-2 NDVI time series processed over open archive; pre-programme rate established before operational alerting begins | GeoTIFF cumulative cleared-area map and summary statistics report, delivered once at programme initiation |
| Frontier trajectory 30/60-day projection | Linear extrapolation of fitted advance vector from most recent 8-week window; uncertainty cone derived from directional variance in historical series | GIS polygon layer showing projected frontier position with confidence envelope; updated monthly |
| Sensor-fusion confidence scoring | Per-pixel agreement count across SAR coherence, optical NDVI, and SWIR bare-earth indices; alerts classified as high (3 sensors agree), medium (2), or low (1) confidence | Attributed polygon layer; low-confidence alerts flagged separately for manual review |
| Contextual attribution overlay | Spatial join of alert polygons against concession boundaries, protected area limits, known mineralised zone buffers, and road network proximity; rule-based ASM likelihood scoring | Attributed GeoJSON with ASM likelihood class and nearest concession or protected area reference; included in weekly feed |
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.