Protected-area boundary encroachment detection
Agricultural clearing, road-cutting and settlement expansion inside gazetted protected areas can be detected weeks earlier than ground patrols can confirm them. Multispectral and SAR change-detection applied to Sentinel time series flags legal-boundary crossings automatically, separating permanent conversion from seasonal burn or flood.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at the equator with both satellites. NDVI, NBR and NDWI time series expose vegetation loss, burn scars and soil exposure. Cloud cover is the main operational limit in humid tropics.
- Sentinel-1 SAR (C-band): 20 m GRD resolution in IW mode, 6-to-12-day revisit depending on latitude. C-band backscatter drops sharply when closed-canopy forest is cleared, independent of cloud or smoke. Useful for confirming structural change flagged spectrally, and for detecting change during prolonged overcast seasons.
- Planet SuperDove: 3 m resolution, near-daily revisit in eight spectral bands including red-edge. Confirms sub-hectare incursions that Sentinel-2 resolves ambiguously. Requires commercial tasking licence; archive from 2021 for SuperDove generation.
- Landsat 8/9 OLI: 30 m resolution, 16-day revisit per satellite (8-day combined), free archive to 1972 for earlier sensors. Provides the long baseline needed to establish pre-disturbance spectral norms and to date the onset of gradual encroachment over years.
Why boundaries fail without continuous observation
A gazetted boundary is a legal line. On the ground it is often unmarked, contested and far from any ranger post. Agricultural expansion in forest-margin communities rarely announces itself: a family clears half a hectare, crops it for a season, then expands. By the time a patrol reaches the site the clearing is established, the crop is in the ground and the political calculus of eviction has shifted.
Satellite change-detection changes that calculus by compressing the detection lag from months to days. The critical metric is not spatial resolution alone but the combination of revisit frequency and the sensitivity of the chosen index to the specific land-cover transition. A 10 m NDVI drop detectable in a single Sentinel-2 pass can represent a clearing of roughly 0.1 hectares, small enough to catch incursions before they become entrenched.
What a floating roof gives away, and what bare soil does not
Agricultural clearing and settlement expansion produce different spectral signatures. Clearing for annual crops exposes bare soil with high shortwave-infrared reflectance and low NIR, driving NDVI sharply negative relative to the pre-disturbance baseline. Settlement expansion adds roofing material (metal, tile, plastic) with characteristic thermal and shortwave responses. Road incursion shows as a linear feature with persistently low NDVI and compacted-soil spectral properties.
Distinguishing these from ephemeral disturbance is the harder problem. Seasonal flooding suppresses NDVI in wetland margins. Controlled burns inside some protected areas are legal management tools. A fire scar from an escaped burn looks spectrally similar to a cleared plot for several weeks. The standard approach is to combine the Normalised Burn Ratio (NBR) with temporal trajectory analysis: a burn scar recovers spectrally within one to three growing seasons; a conversion to cropland does not. SAR coherence provides a second line of evidence because cleared and tilled soil has markedly different coherence properties from recovering post-fire vegetation.
The honest limit here is that C-band SAR struggles to separate dense secondary regrowth from intact primary forest once regrowth exceeds roughly three to four metres in height. Detecting selective logging or low-intensity encroachment under a partial canopy requires either L-band SAR (ALOS-2 PALSAR-2, though revisit is limited) or very high-resolution optical imagery on clear days.
Anchoring the algorithm to the legal boundary
Change detection without a boundary layer produces noise. The operational workflow ingests the protected-area polygon, typically sourced from the WDPA (World Database on Protected Areas), clips the analysis to a configurable buffer zone inside and outside the boundary, and runs the change algorithm only within that spatial domain. This reduces false-alert volume dramatically and focuses ranger attention on legally actionable events.
Buffer width matters. A 500 m interior buffer catches incursions early; a 2 km exterior buffer monitors the pressure zone where new settlements and farms are most likely to expand inward. Both zones can be monitored simultaneously, with the interior zone triggering higher-priority alerts. WDPA boundary data carries its own uncertainties: some polygons are generalised to 1:250,000 scale, which means a 250 m positional error is plausible. Any alert within that margin of the boundary line should be flagged as ambiguous rather than confirmed.
Tropical cloud cover: the operational reality
In the Congo Basin, the Amazon and much of Southeast Asia, Sentinel-2 usable-observation frequency drops to fewer than four clear scenes per month during wet seasons. A change that occurs in week one of a cloudy period may not be spectrally confirmed for six weeks. That is long enough for a clearing to be planted and for the window of low-cost intervention to close.
The mitigation is sensor fusion. Sentinel-1 SAR operates through cloud and smoke. Running a parallel SAR-based change detector, calibrated against the optical baseline, allows provisional alerts to be issued during overcast periods and confirmed or retracted when the next clear optical scene arrives. This two-stage alert architecture reduces both false negatives (missed clearings) and false positives (cloud shadows misread as soil exposure). It does add processing complexity and requires careful cross-sensor calibration, particularly at forest edges where SAR layover and foreshortening can mimic structural change.
From pixel flag to ranger dispatch
A detection algorithm produces a raster of change probability. Converting that into an actionable alert requires several post-processing steps: minimum mapping unit filtering (typically 0.5 to 1 hectare to suppress noise), boundary intersection to confirm the change polygon overlaps the protected area, and severity classification based on the magnitude and persistence of the spectral shift.
Alerts can be delivered as GIS layers (GeoJSON or Shapefile) pushed to protected-area management platforms, as email or SMS notifications to ranger coordinators, or as a feed into a national monitoring dashboard. Latency from satellite overpass to alert delivery depends on processing pipeline design; with Sentinel data available on Copernicus Dataspace within hours of acquisition, same-day alert delivery is achievable for high-priority areas. Satellize applies this boundary-anchored change-detection approach across open constellations, with the same analytic architecture it uses in the Tonga crop-estimation programme adapted for conservation boundary contexts.
The irreducible limit is verification. A satellite alert is evidence of a probable land-cover change at a specific location. It is not a legal determination of encroachment. Ground verification, or at minimum very-high-resolution commercial imagery, is required before enforcement action. The alert's value is in directing that verification effort efficiently rather than replacing it.
Archive depth and the question of when it started
Establishing when encroachment began is often as important as detecting that it has occurred. Landsat's continuous archive from 1984 (with usable OLI data from 2013) allows analysts to reconstruct the spectral history of any site and date the onset of clearing to within one 16-day repeat cycle. For legal proceedings or compensation claims this historical record is frequently decisive.
Sentinel-2 archive depth runs from mid-2015 for Sentinel-2A and from 2017 for the combined constellation. For events after 2017 the 5-day revisit provides a much finer temporal resolution than Landsat alone. Combining both archives gives a 40-year spectral record at 30 m and a 7-plus-year record at 10 m, which is sufficient to characterise both long-run pressure trends and recent acute events.
Typical figures
| Optical spatial resolution | 10 m (Sentinel-2 visible/NIR), 30 m (Landsat OLI), 3 m (Planet SuperDove) |
| SAR spatial resolution | 20 m GRD (Sentinel-1 IW mode) |
| Revisit frequency | 5 days (Sentinel-2 combined), 6–12 days (Sentinel-1 by latitude), 16 days (Landsat combined), near-daily (Planet SuperDove) |
| Spectral bands used | Blue, Green, Red, Red-edge, NIR, SWIR1, SWIR2 (Sentinel-2 B2–B12); C-band VV/VH (Sentinel-1) |
| Minimum detectable clearing | ~0.1 ha at 10 m resolution under clear sky; ~0.5 ha practical minimum after noise filtering |
| Alert latency | Same-day feasible for Sentinel data; 12–24 h typical with automated pipeline |
| Boundary positional uncertainty | 250 m or more for WDPA polygons generalised to 1:250,000; alerts within this margin flagged ambiguous |
| Archive depth | Sentinel-2 from 2015; Landsat from 1984 (OLI from 2013); Planet SuperDove from 2021 |
| Cloud-cover limitation | Optical unusable under cloud; SAR fusion mitigates but does not eliminate wet-season detection gaps |
| Delivery formats | GeoJSON, Shapefile, GeoTIFF change-probability raster, alert feed (email/API) |
Analytics Satellize can run
| Boundary-crossing change alert | Bi-temporal NDVI differencing with minimum-mapping-unit filter and WDPA polygon intersection | GeoJSON alert layer with change magnitude, date of detection and boundary-overlap flag; same-day email notification |
| Permanent-versus-ephemeral conversion classification | NBR trajectory analysis over rolling 12-month time series to separate burn-scar recovery from sustained soil or crop exposure | Classified raster (confirmed conversion / probable burn / recovering / no change) updated each Sentinel-2 pass |
| SAR-based cloud-season provisional alert | Sentinel-1 C-band backscatter change detection (VV/VH ratio) calibrated against optical baseline; two-stage alert with optical confirmation step | Provisional alert GeoJSON issued within 24 h of SAR pass; status updated to confirmed or retracted on next clear optical scene |
| Encroachment onset dating | Landsat OLI + Sentinel-2 combined time-series breakpoint detection (BFAST or equivalent published method) applied to full archive | Site-level PDF report with spectral trajectory chart, estimated onset date and confidence interval; suitable for legal documentation |
| Pressure-zone expansion monitoring | Land-cover change mapping in configurable exterior buffer zone (500 m to 2 km outside boundary) to track settlement and agricultural expansion vectors | Quarterly GIS layer showing expansion polygons, direction of pressure and rate of change in hectares per month |
| Multi-year encroachment trend report | Annual land-cover classification using Landsat and Sentinel-2 time series; area statistics inside protected-area polygon by class | Annual summary report with area tables, change maps and year-on-year comparison; GeoTIFF classification layers included |
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.