Nocturnal artificial-light mapping as a poaching-pressure proxy in protected areas
Persistent or anomalous artificial light inside protected-area boundaries can betray illegal camps, bush-meat processing, and night vehicle movements. VIIRS Day/Night Band time series, backed by DMSP-OLS archive depth, make those signals legible, with honest caveats about what a 375-metre pixel cannot see.
Sensors
- VIIRS Day/Night Band (Suomi NPP / NOAA-20): Panchromatic low-light channel, 375 m native pixel at nadir, sensitive to radiances from roughly 3 × 10⁻⁹ W/cm²/sr to about 2 × 10⁻² W/cm²/sr. Daily global coverage at roughly 01:30 local overpass. The primary workhorse for anomalous-light detection in near-real time.
- DMSP-OLS archive: Coarser 2.7 km pixel, but the archive runs from 1992 to 2013, providing the only multi-decade baseline against which VIIRS trends can be contextualised. Useful for establishing pre-protection or pre-intervention light-level baselines.
- Sentinel-1 SAR (C-band, IW mode): 10 m resolution, 6-day revisit at mid-latitudes. Used here not for light detection but for land-cover masking: distinguishing agricultural clearings, active burn scars, and settlement edges from dark interior pixels, which sharpens the specificity of anomalous-light flags.
- VIIRS VNIR active-fire product (375 m): The VIIRS I-band active-fire product runs alongside the Day/Night Band and is essential for separating thermal emission from burning vegetation, a common false-positive source, from persistent anthropogenic light sources at the same pixel location.
What the physics actually allows
The VIIRS Day/Night Band detects upwelling radiance in a broad panchromatic window centred near 700 nm, with a noise-equivalent delta-radiance of roughly 2 × 10⁻¹⁰ W/cm²/sr under ideal conditions. A single vehicle headlamp at 375-metre resolution produces a radiance signal well below that threshold once averaged across the pixel. A cluster of vehicles, a generator-powered processing site, or a sustained camp fire is a different matter: multiple independent light sources within one pixel sum coherently, and persistent signals over several consecutive overpasses cross the detection floor with high confidence.
That 375-metre footprint is the central honest limit of this method. A single poacher with a torch, a hand-held spotlight used briefly, or a small cooking fire that extinguishes before the 01:30 overpass leaves no trace in VIIRS. The method is a camp-scale and vehicle-convoy indicator, not a single-person tracker. Buyers who expect single-individual detection will be disappointed; buyers who want to prioritise ranger patrols toward sectors showing persistent anomalous illumination will find it useful.
Separating poaching signals from agricultural burning and urban spill
The two dominant false-positive sources are agricultural burning and urban-edge radiance spill. Burning vegetation produces a strong VIIRS DNB signal, but the VIIRS 375 m active-fire product flags thermal anomalies simultaneously. Any DNB-bright pixel that co-registers with an active-fire flag on the same overpass is reclassified as a fire event and excluded from the anomalous-light layer. This step removes the majority of seasonal burning false positives, though smouldering fires below the active-fire detection threshold can still contaminate the record.
Urban spill is handled through a static exclusion mask derived from the DMSP-OLS stable-lights composite and updated with Sentinel-1 settlement backscatter. Pixels within a defined radiance gradient of known settlements are down-weighted in the anomaly score. Protected-area interior pixels that are consistently dark across the DMSP archive and the early VIIRS record form the baseline population; a z-score or median-absolute-deviation threshold applied to that population flags statistically unusual brightening events. The threshold is tunable: a conservative setting reduces false alarms at the cost of missing weaker signals; a sensitive setting catches more events but demands more ranger-time for ground-truthing.
Reading the archive: what a thirty-year baseline reveals
The DMSP-OLS archive, despite its 2.7 km pixel and analogue gain instability, is the only satellite record that reaches back to the early 1990s. Intercalibrated DMSP-OLS composites, produced by groups including NOAA's National Centers for Environmental Information, allow decade-scale trend analysis. A protected area that shows steadily increasing interior light from the mid-2000s onward, even at low absolute radiance, is telling a different story from one that has remained dark throughout.
Transitioning from DMSP to VIIRS requires a cross-calibration step because the two sensors have different spectral responses and gain settings. Published intercalibration coefficients exist in the literature, and the VIIRS monthly composites produced by the Earth Observation Group at the Colorado School of Mines provide a consistent product series from 2012 onward. The overlap period, 2012 to 2013, is the join point. Gaps and inconsistencies in that join are a known limitation and should be declared in any trend report.
Correlating light anomalies with ranger incident records
Satellite-derived light anomalies gain operational credibility when they can be tested against independent ground truth. Ranger incident logs, where they exist and are shared, provide that test. A retrospective correlation study compares the spatial and temporal distribution of DNB anomaly events against logged poaching incidents, snare-removal locations, and illegal camp discoveries. Where the correlation is strong, the light-anomaly layer can be presented to park management as a leading indicator rather than a lagging one.
The correlation is rarely clean. Ranger patrols are not spatially uniform: heavily patrolled sectors generate more incident records simply because rangers are present, creating a detection-effort bias. Areas with weak patrol coverage may show light anomalies that never appear in incident logs because no ranger visited to confirm them. Any honest analysis must account for patrol-effort distribution before drawing conclusions about where poaching pressure is highest. This is a statistical design problem, not a satellite problem, but it falls to the analyst to flag it.
Operational limits worth stating plainly
Cloud cover attenuates the DNB signal. Persistent cloud, common in tropical protected areas during the wet season, can produce data gaps of days to weeks. Monthly composites mitigate this by taking the maximum or median radiance across all cloud-free overpasses in the period, but a genuine poaching event that occurs entirely within a cloudy window will not appear in the composite. Wet-season gaps should be declared in any product specification.
Saturation is a problem near urban edges. The DNB has a high-gain and a low-gain mode; bright urban cores can saturate the high-gain mode, producing blooming artefacts that extend several pixels into adjacent dark areas. Protected areas adjacent to lit towns or industrial sites require a wider exclusion buffer, which reduces the effective monitored area. There is no correction for saturation artefacts that recovers the underlying signal; the affected pixels must simply be masked.
Satellize runs DNB anomaly pipelines on open VIIRS data as part of its analytics offering. The Tonga crop-estimation programme is the company's named public engagement, but the underlying time-series methods, z-score anomaly detection on pixel radiance stacks, transfer directly to protected-area light monitoring.
Turning a radiance map into a patrol-priority layer
The end product that park managers actually use is not a radiance map. It is a ranked list of grid cells, updated monthly or after each significant anomaly event, indicating which sectors warrant increased patrol attention. Each cell carries a confidence tier based on the number of anomalous overpasses, the magnitude of the radiance excess above baseline, and whether co-located SAR data shows any new clearings or track signatures.
A tiered alert system avoids the operational problem of alert fatigue. Tier-one events, multiple consecutive anomalous overpasses with high radiance excess and no active-fire flag, trigger a direct ranger notification within 48 hours of the composite being processed. Tier-two events, single anomalous overpasses or marginal exceedances, accumulate in the monthly summary report. The latency between satellite overpass and a processed alert is typically 24 to 72 hours for near-real-time DNB products, depending on data-downlink scheduling and processing pipeline configuration. That is fast enough to be tactically useful for mobile ranger units, provided the protected area has communication infrastructure to receive and act on the alert.
Typical figures
| Primary sensor spatial resolution | 375 m (VIIRS DNB at nadir) |
| Archive sensor spatial resolution | 2.7 km (DMSP-OLS stable-lights composites) |
| Contextual land-cover sensor resolution | 10 m (Sentinel-1 IW mode) |
| Revisit frequency | Daily (VIIRS, ~01:30 local overpass); 6-day (Sentinel-1 at mid-latitudes) |
| Archive depth | DMSP-OLS from 1992; VIIRS from 2012 |
| Minimum detectable signal | Cluster-scale sources (multiple vehicles or a generator-powered camp); single torch below detection floor |
| Alert latency (near-real-time pipeline) | 24 to 72 hours post-overpass, subject to downlink and cloud cover |
| Cloud-cover limitation | DNB signal attenuated by cloud; wet-season monthly composites may carry multi-week gaps |
| Saturation risk | High-gain DNB saturates near urban cores; affected pixels must be masked, not corrected |
| Delivery formats | GeoTIFF anomaly rasters, GeoJSON alert polygons, monthly PDF summary report, ranked patrol-priority CSV |
Analytics Satellize can run
| Monthly DNB anomaly composite | Median-radiance stack across cloud-free VIIRS overpasses; z-score or MAD thresholding against per-pixel historical baseline | GeoTIFF raster layer with anomaly-score values, clipped to protected-area boundary |
| Near-real-time light-event alert | Single-overpass radiance excess detection with active-fire co-registration filter to exclude burning events | GeoJSON point or polygon alert with radiance excess value, timestamp, and confidence tier, delivered within 72 hours |
| Decade-scale interior-light trend report | Intercalibrated DMSP-OLS to VIIRS time series using published Earth Observation Group coefficients; linear trend fitting per pixel | PDF report with trend maps and sector-level summary statistics, covering 1992 to present where archive allows |
| False-positive fire-exclusion layer | Spatial join of VIIRS 375 m active-fire product against DNB anomaly pixels; co-registered events reclassified and removed | Cleaned GeoTIFF anomaly layer with fire-excluded pixels flagged separately for audit |
| Patrol-priority ranking grid | Multi-criteria scoring combining anomaly frequency, radiance excess magnitude, and Sentinel-1 land-cover change flags; tiered by confidence | CSV ranked sector list and corresponding GIS polygon layer, updated monthly |
| Ranger-incident correlation analysis | Retrospective spatial-temporal correlation between DNB anomaly events and supplied incident-log records, with patrol-effort bias correction | Statistical report with correlation coefficients, bias-adjusted hotspot map, and recommendations for threshold calibration |
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.