Illicit market and informal economy monitoring in refugee settlements
Very-high-resolution optical satellites can track market expansion, structure density and vehicle patterns in large refugee settlements. The method flags spatial anomalies for ground investigation; it cannot distinguish licit from illicit trade.
Sensors
- Maxar WorldView-3: 30 cm panchromatic, 1.24 m multispectral (8 bands including SWIR), revisit roughly 1 day at mid-latitudes with off-nadir tasking. The SWIR bands help distinguish roofing materials and surface types that standard RGB cannot separate.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral (6 bands). Stereo and tri-stereo collection in a single pass allows 3-D structure height estimation, useful for distinguishing permanent stalls from temporary awnings.
- Planet SkySat: 50 cm panchromatic, 1 m multispectral. Can be tasked for video clips of 90 seconds, which reveals vehicle movement patterns that a single still frame misses. Revisit on tasking is typically same-day or next-day.
- Planet PlanetScope SuperDove: 3 m multispectral (8 bands), near-daily global revisit. Too coarse to resolve individual stalls, but useful for monitoring the outer boundary of market areas and detecting expansion into previously unoccupied land over weeks and months.
What the satellite actually sees, and what it does not
A 30 cm image of Bidi Bidi, Cox's Bazar or Kakuma resolves individual vehicles, the footprint of a market stall, and whether a structure has a corrugated-iron roof or a tarpaulin. It does not resolve faces, read signage reliably, or indicate what is being sold. This distinction matters enormously for any programme that might affect the lives of people who are already in a precarious situation.
The honest framing is that imagery provides spatial and temporal anomaly flags. A market area that doubles in footprint between two collection dates, or a cluster of heavy goods vehicles that appears overnight in a location with no prior vehicle congregation, is a signal worth investigating on the ground. It is not evidence of anything specific. Analysts and field teams must treat it accordingly.
Why settlements attract this kind of monitoring in the first place
UNHCR and UNODC field assessments have documented that large, under-resourced settlements can become nodes for smuggled goods, counterfeit pharmaceuticals and, in the most serious cases, trafficked persons. The combination of high population density, limited formal economic infrastructure and constrained law-enforcement access creates conditions that organised criminal networks exploit. None of this is a criticism of refugees themselves; it reflects the structural vulnerabilities of the environment.
Satellite monitoring is one of the few methods that can cover the full spatial extent of a settlement like Cox's Bazar, which at its peak housed over 800,000 people across roughly 26 square kilometres, without requiring continuous physical presence. Ground-based surveillance is expensive, dangerous and often politically contested. Periodic aerial survey is costly and infrequent. Commercial satellite tasking, by contrast, can be scheduled at regular intervals and compared systematically.
Structure density and market-boundary change detection
The core analytic is change detection between two or more high-resolution collections. Automated building and structure footprint extraction, applied to WorldView-3 or Pléiades Neo imagery, can quantify the number of discrete roofed structures within a defined market polygon and track how that count changes over time. Published studies using similar methods in informal urban settings report detection of structures down to roughly 4 square metres at 30 cm resolution, though accuracy degrades under dense canopy or when structures share walls.
Market boundary expansion is measured by classifying the outer edge of contiguous high-density structure clusters and comparing that boundary across dates. PlanetScope's near-daily revisit is useful here even at 3 m resolution, because the boundary shift between collections is typically tens of metres rather than single metres. When the boundary expands rapidly into an area that was previously open land or a buffer zone, that is the anomaly flag. The cause could be a seasonal market surge, a new population influx, or something else entirely. Imagery alone cannot say which.
Vehicle congregation as an economic-intensity proxy
Vehicle counts and congregation patterns are a well-established proxy for economic activity in remote-sensing literature. In a settlement context, the presence of heavy goods vehicles, minibuses or motorcycles in unusual concentrations, at unusual times or in locations without obvious logistical purpose, is a pattern worth flagging. SkySat video clips are particularly useful because they show movement: a vehicle that appears parked in a still image may be part of a rapid loading and departure cycle that only motion reveals.
The limits here are real. Parking patterns in informal markets are chaotic by nature. A cluster of trucks might indicate a legitimate food-aid distribution or a commercial resupply, not contraband. Without time-series context, a single observation is nearly meaningless. Analysts need at least four to six comparison dates before vehicle congregation patterns become interpretable, and even then the interpretation is probabilistic.
Honest limits of the method
Cloud cover is a persistent problem at Cox's Bazar, which sits in a high-rainfall zone. The monsoon season can produce weeks of continuous cloud that makes optical tasking useless. Analysts working in Bangladesh should plan for significant data gaps from June through September and consider whether the analytic programme can tolerate that gap or needs a supplementary approach.
Resolution floors matter too. At 30 cm, a standard market stall is a few pixels. Individual goods are invisible. The method cannot distinguish a medicine stall from a food stall, let alone identify counterfeit packaging. It also cannot detect activity inside structures. A covered market, a warehouse or even a large tarpaulin defeats optical observation entirely. These are not edge cases; they are routine features of informal markets in all three settlement contexts.
Finally, there is the question of who conducts the analysis and under what governance framework. Satellite monitoring of refugee populations raises serious ethical and legal questions about consent, data retention and the potential for misuse. Any programme of this kind should be designed with explicit human-rights safeguards and clear protocols for how anomaly flags are handed to investigators.
Putting the analytics to work
A practical programme would combine quarterly WorldView-3 or Pléiades Neo collections for structure-density analysis with monthly PlanetScope time-series for boundary monitoring, supplemented by SkySat video tasking triggered by specific anomaly flags from the lower-resolution layer. This layered approach manages cost while concentrating high-resolution collection on locations and periods that the coarser data has already flagged as unusual.
Satellize runs this kind of layered change-detection workflow on open and commercial constellations, with output delivered as GIS layers and periodic written assessments rather than raw imagery. The Tonga crop-estimation programme uses a comparable multi-resolution logic, though the subject matter is obviously different. Organisations considering a settlement-monitoring programme should begin with a scoping collection over a single defined market area, establish a baseline, and run two or three comparison dates before committing to a standing tasking arrangement. That sequence will reveal what the imagery can and cannot resolve in the specific settlement before any significant investment is made.
Typical figures
| Best available spatial resolution | 30 cm (WorldView-3 panchromatic, Pléiades Neo panchromatic) |
| Multispectral resolution | 1.2–1.24 m (WorldView-3, Pléiades Neo); 3 m (PlanetScope SuperDove) |
| Revisit rate (tasked VHR) | 1 day or less at mid-latitudes with off-nadir collection (WorldView-3, Pléiades Neo, SkySat) |
| Revisit rate (monitoring layer) | Near-daily (PlanetScope SuperDove, weather permitting) |
| Spectral bands available | Panchromatic, 8-band multispectral including NIR and SWIR (WorldView-3); 6-band multispectral (Pléiades Neo); 8-band multispectral (SuperDove) |
| Minimum detectable structure footprint | Approximately 4 m² at 30 cm resolution under clear sky, open-roof conditions; degrades significantly under canopy or shared-wall construction |
| Cloud-cover limitation | Optical sensors fully blocked; Cox's Bazar monsoon season (June–September) typically produces multi-week data gaps |
| Archive depth | WorldView-3 archive from 2014; Pléiades Neo from 2021; PlanetScope daily archive from approximately 2016 |
| Typical collection latency | 12–48 hours from tasking to analyst-ready imagery for commercial VHR sensors |
| Delivery formats | GeoTIFF, COG, GIS vector layers (GeoJSON, Shapefile), structured change-detection reports |
Analytics Satellize can run
| Structure density map and change index | Automated building footprint extraction via object-based image analysis (OBIA) applied to VHR optical imagery; bi-temporal differencing | GIS polygon layer with structure count per grid cell and percentage change between collection dates; quarterly summary report |
| Market boundary expansion layer | Supervised classification of high-density structure clusters in PlanetScope time-series; outer-boundary vectorisation and area differencing | Monthly GIS boundary layer with expansion vectors and flagged zones exceeding a defined area-change threshold |
| Vehicle congregation anomaly flag | Object detection for vehicle-class features in VHR imagery; count and spatial clustering compared against baseline distribution | Point-layer alert file identifying grid cells with vehicle counts exceeding two standard deviations from baseline, with collection date and coordinates |
| Roofing-material classification | Spectral unmixing using WorldView-3 SWIR bands to distinguish corrugated metal, tarpaulin and concrete; used as a proxy for structure permanence | Classified raster layer with legend; updated per collection cycle |
| 3-D structure height estimate | Stereo or tri-stereo photogrammetry from Pléiades Neo same-pass collection; digital surface model differencing against bare-earth baseline | Digital surface model (GeoTIFF) and derived height-change layer for structures exceeding 2 m, flagging potential multi-storey or covered-market construction |
| Anomaly narrative assessment | Analyst synthesis of change-detection outputs, vehicle flags and boundary data against known settlement layout and reported events | Written assessment (PDF and structured data) identifying priority areas for ground-truth investigation, with explicit confidence ratings and stated assumptions |
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.