Unlicensed gambling arcade and illicit cash-business cluster detection
Persistent night-time illumination inconsistent with declared land use, combined with high vehicle-turnover signatures in very-high-resolution daytime imagery, can flag candidate unlicensed gambling or cash-front clusters for ground investigation.
Sensors
- VIIRS Day/Night Band (Suomi NPP / NOAA-20): 750 m ground sample distance at nadir; detects low-level artificial radiance down to roughly 2 × 10⁻¹⁰ W cm⁻² sr⁻¹ μm⁻¹. Daily global revisit makes it the primary tool for detecting persistent anomalous illumination over weeks or months. Cannot resolve individual buildings; flags districts or blocks for follow-on tasking.
- Planet SuperDove (PlanetScope): 3 m resolution, 8 spectral bands, daily revisit over most landmasses. Used for vehicle-presence counting and parking-area occupancy analysis across candidate sites identified by VIIRS. Daytime only; cloud is a real constraint in tropical and monsoon climates.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral. Sufficient to count individual parked vehicles, read signage legibility, and detect physical modifications to shopfronts such as blacked-out windows or external CCTV clusters. Tasked on demand; not a persistent sensor.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral, stereo-capable. Comparable to WorldView-3 for site characterisation. Tri-stereo acquisition allows 3-D building-height estimation, useful for distinguishing single-storey commercial units from multi-floor premises where upper floors may host different activities.
What a cash-front looks like from 400 miles up
A legitimate retail unit in a mixed commercial zone follows a predictable illumination pattern: bright during trading hours, dark by ten or eleven at night. An unlicensed gambling arcade, or a cash-intensive front business running a secondary operation after hours, does not. It stays lit. Foot traffic continues. Vehicles arrive and depart in pulses that have no relationship to the declared business type.
This is the core observable. VIIRS Day/Night Band radiance time-series, processed over weeks, can identify blocks where night-time light output is statistically elevated relative to the surrounding neighbourhood and inconsistent with the land-use register. The method is well-established in informal-economy research: published work has applied VIIRS DNB to map electricity access, informal settlement extent, and economic activity in areas with poor administrative data. The same physics applies here, applied to a narrower enforcement question.
The two-layer detection logic
Neither sensor layer is sufficient on its own. VIIRS at 750 m cannot tell you which unit on a block is the problem. WorldView-3 at 31 cm can show you a car park, but a single daytime image tells you nothing about whether that car park is busy at 2 am. The method only becomes useful when the layers are combined in sequence.
Layer one: VIIRS radiance anomaly screening. A rolling 30-night composite is compared against a baseline period. Candidate blocks are those where median radiance in the 22:00–04:00 window exceeds the local baseline by a threshold chosen relative to neighbourhood variance. This is coarse. It produces a list of blocks, not addresses.
Layer two: very-high-resolution daytime vehicle-turnover analysis. PlanetScope daily imagery over the flagged blocks is used to count parked vehicles at multiple times of day across a 30-day window. Sites with high variance in vehicle count (many cars mid-morning, few at noon, many again at midnight, if a night pass is available) are inconsistent with most legitimate retail categories. WorldView-3 or Pléiades Neo is then tasked for a single high-resolution characterisation pass: signage, physical modifications, CCTV density, proximity to other flagged units.
The output is a ranked candidate list, not a conviction. Ground investigation is the necessary next step. Satellite data reduces the area that investigators need to cover.
What the method cannot do
Honest limits matter here, because the enforcement stakes are high and false positives waste investigator time.
VIIRS cannot resolve below 750 m. In dense urban grids, a single VIIRS pixel covers dozens of premises. Elevated radiance in that pixel could come from a legitimate late-night restaurant cluster, a 24-hour pharmacy, or a well-lit petrol station. The anomaly screen is a hypothesis generator, not a finding.
Cloud cover breaks the night-light time-series. In tropical cities with persistent monsoon cloud, a 30-night composite may contain fewer than ten usable observations. Gaps introduce uncertainty in the baseline.
Vehicle counting from PlanetScope at 3 m is reliable for large car parks and arterial roads. On narrow urban streets where vehicles park nose-to-tail, individual vehicle delineation becomes ambiguous. Accuracy figures from published object-detection studies vary widely by scene density and shadow conditions.
The method says nothing about what is happening inside a building. A site that is anomalously bright and busy at night might be a call centre, a bakery, or a legitimate late-licence venue. Satellite data identifies the anomaly; human intelligence and legal process determine its nature.
Clustering matters more than individual sites
Unlicensed gambling operations and cash-front businesses rarely operate in isolation. Enforcement experience in multiple jurisdictions documents the tendency for such premises to cluster within a few hundred metres of each other, sometimes sharing security arrangements or operating under common beneficial ownership. This geographic clustering is itself a signal.
Spatial autocorrelation analysis on the candidate site list, using standard methods such as Moran's I or kernel density estimation, can identify whether flagged sites are distributed randomly or are significantly clustered. A statistically significant cluster of anomalous-illumination sites within a defined radius is a stronger hypothesis than any single site in isolation. It also suggests a more efficient ground-investigation strategy: a single operation covering a cluster rather than dispersed individual visits.
Archive depth matters for this analysis. VIIRS data extends back to 2012 (Suomi NPP) and is available through NASA EARTHDATA. A multi-year radiance trend can show whether a cluster appeared suddenly (consistent with a new operation opening) or has been present for years (consistent with a long-running enterprise that has evaded previous enforcement).
Practical scope and who commissions this work
The most plausible commissioners are national tax and revenue authorities, financial intelligence units, and municipal licensing enforcement bodies. The use case fits jurisdictions where the density of unlicensed premises is high enough that ground-level surveillance cannot cover the full territory, and where administrative land-use data is sufficiently reliable to make anomaly detection meaningful.
The method scales poorly in cities where land-use registers are themselves unreliable or where informal commercial activity is so prevalent that anomalous illumination is the norm rather than the exception. In those settings, the baseline is contaminated and the anomaly screen loses discriminating power.
Satellize can run this two-layer screening workflow on open VIIRS data combined with commercial tasking on client licence, delivering a ranked candidate-site report with supporting imagery chips and spatial-cluster statistics. The approach is methodologically similar to the informal-economy activity analysis underlying the Tonga crop-estimation programme, applied to a different observable.
A screening run covering a defined urban area of roughly 500 km² is a reasonable starting scope for a pilot. Results are probabilistic. They are intelligence inputs, not evidence, and should be treated accordingly by any legal or enforcement process that follows.
Typical figures
| Night-light spatial resolution (VIIRS DNB) | 750 m at nadir (Suomi NPP / NOAA-20) |
| Daytime optical resolution (vehicle counting) | 3 m (PlanetScope SuperDove); 31 cm panchromatic (WorldView-3 / Pléiades Neo) |
| Night-light revisit | Daily global (VIIRS); cloud limits usable observations in tropical climates |
| Daytime optical revisit | Daily (PlanetScope); on-demand tasking within 1–3 days (WorldView-3, Pléiades Neo) |
| Minimum detectable radiance anomaly (VIIRS DNB) | Approximately 2 × 10⁻¹⁰ W cm⁻² sr⁻¹ μm⁻¹; practical anomaly threshold set relative to local neighbourhood variance |
| Spectral bands used | VIIRS DNB (panchromatic, 0.5–0.9 μm); PlanetScope 8-band (0.44–0.88 μm); WorldView-3 panchromatic + 8-band VNIR |
| VIIRS archive depth | Suomi NPP from 2012; NOAA-20 from 2018 (NASA EARTHDATA) |
| Typical screening area per pilot | Defined urban area of 200–1000 km²; larger areas increase baseline calibration complexity |
| Latency (anomaly report) | 30-night composite baseline requires approximately 6–8 weeks of usable observations before first candidate list is produced |
| Deliverable formats | GeoJSON candidate-site layer, imagery chips (GeoTIFF), PDF ranked-site report with cluster statistics |
Analytics Satellize can run
| Night-light anomaly map | VIIRS DNB radiance time-series compositing; percentile-based anomaly detection relative to neighbourhood baseline; published in informal-economy and electricity-access literature | GeoJSON polygon layer of anomalous-illumination blocks with radiance delta and confidence score; updated monthly |
| Vehicle-turnover signature index | Multi-date PlanetScope vehicle detection using object-detection or parking-area segmentation; variance in daily vehicle count computed across 30-day window | Per-site time-series CSV and GIS point layer with turnover-variance score and flagged time windows |
| High-resolution site characterisation report | Tasked WorldView-3 or Pléiades Neo imagery; manual and semi-automated feature extraction (signage visibility, window occlusion, external infrastructure density) | PDF site sheet with annotated imagery chip, feature checklist, and investigator notes field |
| Spatial cluster analysis | Kernel density estimation and Moran's I spatial autocorrelation on candidate-site point layer; standard GIS methods | Cluster map (GeoTIFF and GeoJSON) with statistical significance scores; identifies priority investigation zones |
| Multi-year trend report | VIIRS archive radiance trend analysis from 2012 baseline; change-point detection to identify when anomalous illumination first appeared at candidate sites | Per-cluster trend chart and narrative summary indicating approximate onset date and radiance trajectory |
| Ranked candidate-site intelligence brief | Composite scoring combining night-light anomaly magnitude, vehicle-turnover variance, cluster membership, and site-characterisation flags; scoring weights set with client enforcement team | Ranked PDF brief listing top candidate sites with supporting evidence summary; designed as an intelligence input for ground-investigation tasking |
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.