Forest clearance detection and heritage site exposure risk in tropical regions

Tropical forest cover is the primary physical defence of countless unexcavated heritage sites. Sentinel-1 SAR and optical NDVI time-series together detect clearance events even through persistent wet-season cloud, flagging exposure risk before looters or machinery cause irreversible damage.

Sensors

Sentinel-1 SAR (C-band, ESA): 5.4 GHz C-band synthetic aperture radar; 10 m ground range resolution in Interferometric Wide swath mode; 6-day repeat at the equator with both satellites active. Cloud-penetrating and illumination-independent, making it the primary sensor for wet-season monitoring when optical coverage fails for weeks at a time. Backscatter change between forest and bare soil is typically 3–6 dB, well above the noise floor.
Landsat 8/9 OLI (USGS/NASA): 30 m multispectral resolution; 16-day single-satellite repeat, 8-day combined. NDVI and NBR derived from OLI bands 4 and 5 provide the primary optical deforestation signal. Archive from 1972 (MSS) through present allows decadal baselines. Useless under dense cloud, which is why SAR pairing is essential in the tropics.
Planet NICFI tropical forest data: 4.77 m resolution monthly and biannual basemaps covering tropical forest nations under the Norway NICFI programme. Sufficient resolution to distinguish individual tree-fall gaps from machinery clearance. Available from 2015 onwards for qualifying users. Still cloud-limited; revisit advantage is in the monthly composite rather than daily tasking.
PRODES / DETER (INPE, Brazil): Brazil's National Institute for Space Research publishes near-real-time deforestation alerts (DETER) and annual mapping (PRODES) derived from Landsat and MODIS. Minimum mapping unit for DETER is approximately 25 ha, which is useful for large clearances but misses the smaller targeted cuts that often precede heritage looting.
Sentinel-2 MSI (ESA): 10 m visible and near-infrared bands; 5-day revisit at the equator. Produces higher-resolution NDVI than Landsat and can resolve clearance patches down to roughly 0.1 ha. Cloud contamination rates in humid tropics frequently exceed 70% of acquisitions in wet season, so time-series compositing is mandatory rather than optional.

What the canopy was actually doing

Dense tropical forest does more than obscure archaeological sites from aerial observation. The root systems bind and stabilise earthworks. The canopy intercepts rainfall, reducing direct erosion of exposed stone and compacted soil surfaces. The humidity discourages casual access. Clearance removes all three protections simultaneously, and it does so fast: mechanised land-clearing in the Brazilian Amazon or Cambodian lowlands can strip several hectares in a single day.

The problem for heritage managers is that the same cloud cover that makes tropical sites difficult to photograph also makes satellite monitoring unreliable if you rely on optical sensors alone. During the Cambodian wet season, Sentinel-2 and Landsat acquisitions can be cloud-contaminated for six to ten consecutive weeks. A clearance event that begins and ends within that window leaves no optical trace. This is not a minor operational inconvenience; it is the structural weakness that SAR-optical fusion is designed to address.

Why SAR backscatter change is the right first signal

C-band radar at 5.4 GHz interacts primarily with the upper canopy volume and, once that is removed, with the soil surface beneath. The transition from closed-canopy forest to bare or disturbed ground produces a backscatter decrease of roughly 3 to 6 dB in VV polarisation and a different but also detectable shift in VH, depending on soil moisture and residual debris. That magnitude of change is large relative to the sensor noise floor and consistent enough to be detected with straightforward bi-temporal differencing or a change-point algorithm applied to a Sentinel-1 time-series stack.

Published work on Cambodian heritage landscapes around Angkor and the Cardamom Mountains has used exactly this approach to flag clearance inside and adjacent to protected zones during wet seasons when no optical data were available. The method's honest limit is spatial: at 10 m resolution, a clearance of less than roughly 0.05 ha may not produce a statistically clean signal above background variation caused by wind or soil-moisture change. Targeted clearance of a single small structure can therefore be missed. The response is to lower the detection threshold and accept more false positives, then triage with optical data when cloud breaks.

NDVI time-series as the longer memory

Optical NDVI from Landsat or Sentinel-2 serves a different function from SAR in this workflow. Rather than detecting individual events quickly, NDVI time-series built from multi-year archives establish what normal seasonal variation looks like at each pixel. A clearance event then appears as a step-change drop that exceeds the seasonal envelope, persisting across subsequent acquisitions rather than reverting. This persistence distinguishes genuine clearance from cloud shadows, burn scars that regrow, or agricultural cycles in adjacent land.

The Guatemalan Petén, which contains hundreds of mapped and thousands of suspected Maya sites, is a well-documented application area. Landsat-derived forest-loss products, including the University of Maryland Global Forest Change dataset, show annual clearance rates in the Petén biosphere reserve that correlate with documented site looting incidents. The correlation is not proof of causation in individual cases, but it is operationally useful: areas of recent clearance adjacent to known site polygons become the priority list for ground verification.

Planet NICFI data adds a resolution tier between Sentinel-2 and commercial tasking. At roughly 5 m, individual structure outlines can sometimes be distinguished in fresh clearances, which helps assess whether a clearance is agricultural in intent or targeted at a specific feature.

The exposure cascade: what happens after the trees come down

Clearance is not the end of the damage sequence; it is the beginning. Once canopy cover is removed, the site faces several concurrent threats that satellite monitoring can track separately. Erosion of earthworks and middens becomes measurable through DEM differencing if high-resolution stereo data are available. Surface artefact scatter becomes visible to high-resolution optical sensors that would previously have seen only canopy. And access improves dramatically: roads cut for timber or agricultural machinery become the same roads used by looters.

In the Brazilian Amazon, FUNAI and IPHAN have both documented cases where clearance for cattle pasture or soy production has exposed previously unknown or poorly mapped archaeological sites, sometimes destroying them before any formal assessment could occur. The satellite record in these cases becomes the primary evidence of what existed and when it was lost. That forensic function is distinct from the preventive one, but it matters for legal proceedings and for heritage loss quantification.

Combining sensors: what the fusion actually looks like in practice

A practical monitoring system for a defined heritage landscape runs three parallel streams. First, a Sentinel-1 change-detection alert triggers within 6 to 12 days of a clearance event regardless of cloud. Second, an optical compositing pipeline checks the same location against the next cloud-free Sentinel-2 or Landsat acquisition, which may arrive days or weeks later, to confirm and characterise the clearance. Third, a PRODES or equivalent national deforestation product provides independent validation for large events and feeds into reporting to national heritage authorities.

The alert geometry matters. Heritage site polygons from national registers or UNESCO World Heritage boundary files define the primary zone of concern, but a buffer of 500 m to 2 km around each polygon is equally important: clearance that approaches a site boundary is a warning sign even before it crosses it. Road-network proximity adds a second spatial filter, since isolated clearance far from any track is less immediately threatening than clearance that connects to existing access.

Satellize runs this kind of multi-stream fusion on open constellations for clients who need sovereign, non-commercially-intermediated access to the alert pipeline. The Tonga crop-estimation programme uses a structurally similar NDVI time-series approach, applied to agricultural rather than heritage objectives, which gives a sense of how the underlying analytics transfer across land-cover monitoring problems.

Honest limits and what they mean for programme design

No satellite system currently provides same-day detection of small targeted clearances under persistent cloud. Sentinel-1's 6-day revisit is the best freely available cadence for cloud-penetrating observation, and it will miss events that begin and end within a single revisit window. Commercial SAR constellations such as ICEYE or Capella can shorten that window to hours, but at cost and with smaller swath widths that require pre-positioning over known risk areas.

The minimum detectable clearance at Sentinel-1 resolution is roughly 0.05 to 0.1 ha for a confident detection, which is large enough to miss the kind of small, targeted excavation that characterises professional looting. That gap is real and should be stated plainly to any heritage authority designing a monitoring programme. Satellite monitoring is most effective as a landscape-scale early-warning system, not as a substitute for ground patrols or drone surveys at individual sites.

Archive depth is a genuine asset. Sentinel-1 data runs from 2014, Landsat from 1972. For any site where clearance is suspected to have occurred before a monitoring programme was established, retrospective analysis of the archive can reconstruct the timeline of forest loss, which has direct value for legal and insurance purposes.

Typical figures

SAR spatial resolution (Sentinel-1 IW mode)	5 × 20 m (range × azimuth), posted to 10 m GRD product
SAR revisit at equator	6 days (Sentinel-1A + 1B combined); single satellite 12 days
Optical resolution (Sentinel-2 MSI)	10 m (visible/NIR bands); 20 m (red-edge, SWIR)
Optical resolution (Landsat 8/9 OLI)	30 m multispectral; 15 m panchromatic
Minimum detectable clearance (SAR change)	~0.05–0.1 ha for confident detection; smaller events ambiguous
Cloud penetration	Sentinel-1 SAR: full; optical sensors: zero penetration of dense cloud
Archive depth	Sentinel-1 from 2014; Landsat from 1972; Planet NICFI from 2015
Alert latency (SAR-based)	6–12 days from clearance event to processed alert under cloud
Typical coverage	Sentinel-1 IW swath 250 km; national-scale coverage achievable within weeks
Delivery formats	GeoTIFF change layers, GeoJSON alert polygons, PDF site-risk reports

Analytics Satellize can run

SAR backscatter change alert	Bi-temporal or time-series change-point detection on Sentinel-1 VV/VH stack; threshold calibrated to local forest type	GeoJSON polygon alert layer with date, magnitude and confidence score; delivered within 12 days of acquisition
NDVI step-change detection	Seasonal decomposition of Landsat or Sentinel-2 NDVI time-series; anomaly flagged when value falls outside multi-year seasonal envelope	Per-pixel change date raster and summary CSV of affected area by heritage site buffer zone
Heritage site exposure risk score	Spatial intersection of clearance alerts with site polygon registry and road-network proximity; weighted composite risk index	Monthly ranked site-risk report in PDF and GIS layer format for heritage authority review
Retrospective clearance timeline	Annual forest-loss stack from Landsat archive (1972 to present) combined with PRODES/Global Forest Change product for Brazilian sites	Site-specific timeline chart and GeoTIFF stack showing year-of-loss for each cleared pixel within defined heritage buffer
Wet-season monitoring continuity report	Cloud-cover fraction calculation per acquisition; SAR gap-fill flagging; optical confirmation rate tracking	Quarterly data-quality audit report showing proportion of monitoring period covered by SAR versus optical, with coverage gaps identified
Clearance approach vector mapping	Sequential clearance polygon centroids fitted to road-network graph; directional trend analysis toward heritage site boundaries	Vector layer showing clearance progression trajectory and estimated time-to-boundary under current rate

Sources

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.