Refugee and displaced-person settlement mapping for emergency connectivity demand assessment
Rapid informal settlement growth can outpace any connectivity plan. Very-high-resolution optical change detection and SAR coherence loss let planners count shelters, estimate spatial extent, and size emergency network deployments before the first mast goes in.
Sensors
- Planet SuperDove: 3 m ground sample distance, 8-band multispectral (coastal blue through NIR), daily global revisit. Ideal for tracking the day-by-day footprint expansion of a settlement and detecting new shelter clusters as bare earth is cleared.
- Maxar WorldView Legion: 30 cm panchromatic, 1.2 m multispectral, up to 15 revisits per day over a target area. Provides the spatial resolution needed to count individual shelter units and distinguish tarpaulin from permanent roofing material.
- Sentinel-1 (ESA): C-band SAR, 10 m resolution in Interferometric Wide Swath mode, 6-day repeat at the equator (12-day for a single satellite). Coherence loss between pre- and post-displacement acquisitions reveals ground disturbance and new construction even through cloud cover, which is critical during monsoon conditions.
- Pleiades Neo (Airbus): 30 cm panchromatic, 1.2 m multispectral, stereo and tri-stereo capable. Useful for generating a surface model of the camp to estimate built volume and identify elevated water or fuel storage that indicates semi-permanent infrastructure.
Why shelter count is the right starting metric
A network planner sizing emergency connectivity needs one number above all others: how many simultaneous users will the system carry at peak load? Population registers do not exist for informal settlements. Household surveys take weeks. Shelter count from satellite imagery is not a perfect substitute, but it is the fastest defensible proxy available within 24 to 72 hours of a displacement event.
The logic is straightforward. Each visible shelter unit represents one household. Published UNHCR emergency standards assume 3.5 to 5 persons per household in acute displacement settings, though real occupancy in crisis conditions frequently exceeds that range. A count of 20,000 shelter units therefore implies a population somewhere between 70,000 and 120,000 people, with the honest caveat that ground-truthing is essential before committing capital to infrastructure. That uncertainty range is wide enough to matter for spectrum licensing and backhaul dimensioning, but it is far narrower than no data at all.
What a tarpaulin roof gives away, and what it does not
At 30 cm resolution, a WorldView or Pleiades Neo image resolves individual shelter ridgelines, doorway orientations, and the gap spacing between units. Automated object detection using convolutional neural networks trained on labelled humanitarian imagery (the xView and SpaceNet datasets are the most cited public benchmarks) can achieve shelter detection rates above 80 percent on high-contrast tarpaulin against bare earth. Accuracy drops sharply when shelters are constructed from local materials that spectrally resemble the surrounding ground, or when tree canopy partially occludes the settlement.
Planet SuperDove at 3 m cannot resolve individual shelters in a dense camp, but it can delineate the settlement boundary with reasonable accuracy by detecting the spectral transition from natural land cover to disturbed bare earth and synthetic roofing. That boundary polygon, updated daily, is what drives the spatial extent estimate used in radio propagation planning. The two sensors are complementary: SuperDove provides the temporal cadence, WorldView provides the count density.
SAR coherence loss when optical sensors go dark
Optical imagery fails in cloud. Sentinel-1 does not. When a displacement event occurs during a rainy season, SAR coherence change detection becomes the primary detection method. The principle is that C-band radar backscatter from undisturbed vegetation or bare soil is highly coherent between repeat passes. Construction activity, ground clearing, and the installation of tarpaulin or metal-sheet shelters disrupt that coherence. The decorrelation signal is detectable at Sentinel-1's 10 m resolution even when individual structures are sub-pixel.
The limitation is that coherence loss is not shelter-specific. It flags any ground disturbance, including agricultural activity, flooding, or road construction. In practice, analysts combine coherence change maps with optical data from cloud-free windows and any available pre-event land-cover classification to isolate settlement-related disturbance. This multi-source approach reduces false positives but adds processing time. Expect 24 to 48 hours of analyst time to produce a validated coherence-change product for a 500 km² area of interest.
Translating spatial extent into a network architecture brief
The output of the remote-sensing analysis is a geospatial dataset: a settlement boundary polygon, a shelter-density heat map at 10 to 50 m grid resolution, and a population estimate with an explicit confidence interval. That dataset feeds directly into the network planning workflow.
Connectivity planners use the shelter-density grid to identify candidate macro-cell locations that maximise coverage over the highest-density zones. The spatial extent of the camp determines whether a single high-power site is sufficient or whether a distributed small-cell architecture is needed. The population estimate drives the traffic model used to dimension backhaul, typically satellite VSAT or microwave to the nearest fibre point of presence. Humanitarian funding applications to bodies such as GSMA's Disaster Response programme or the UN Emergency Telecommunications Cluster require documented evidence of the affected population size; a satellite-derived shelter count with a stated methodology satisfies that requirement in a way that field estimates often cannot.
One practical note: camp boundaries change faster than network deployment cycles. A settlement that covers 4 km² on the day of the initial assessment may cover 7 km² six weeks later. Building in a revisit cadence, at minimum weekly using SuperDove, is not optional if the network design is to remain valid through the deployment period.
Honest limits of the method
Several failure modes deserve explicit acknowledgement. Dense tree canopy over a settlement, common in forested displacement zones in Central Africa or Southeast Asia, can suppress shelter detection rates below 50 percent even at 30 cm resolution. SAR coherence methods partially compensate but are not reliable in tropical forests where baseline coherence is already low due to vegetation motion.
Occupancy rate uncertainty is the single largest source of error in population estimates. The 3.5 to 5 persons per household figure is a planning assumption, not a measurement. In acute crisis phases, multi-family shelter sharing is common and can push real occupancy above 8 persons per unit. Conversely, in protracted displacement settings, household sizes normalise toward lower values. No satellite sensor can measure occupancy directly. Ground surveys, mobile network signalling data where available, and food distribution records are the only reliable correctives.
Archive depth is an asset here. Sentinel-1 data is freely available from 2014, and Landsat imagery extends to 1972. For settlements with a longer history, change detection over multi-year timescales can reveal growth trajectories that inform long-term infrastructure planning rather than emergency response alone. Satellize has run similar multi-temporal change analysis for agricultural monitoring, including the Kingdom of Tonga crop-estimation programme, and the same methodological pipeline applies to settlement mapping.
From pixels to a fundable connectivity proposal
The final product a telecoms operator or humanitarian organisation needs is not a satellite image. It is a document that answers three questions: how many people, where exactly, and how fast is the situation changing. Remote sensing answers all three with sufficient rigour for capital allocation decisions, provided the uncertainty bounds are stated honestly.
A well-structured analysis package includes the settlement boundary as a GeoJSON or shapefile, the shelter-density grid as a GeoTIFF, a population estimate table with low and high scenarios, a time series of settlement extent from first detection to the analysis date, and a written methodology note that a funder's technical reviewer can audit. That package takes roughly five to ten working days to produce for a new site, assuming commercial tasking is available and cloud cover is not severe. Subsequent updates on a weekly cadence are faster once the baseline is established.
Typical figures
| Optical spatial resolution (shelter counting) | 30 cm (WorldView Legion, Pleiades Neo); 3 m (Planet SuperDove boundary delineation) |
| SAR spatial resolution | 10 m (Sentinel-1 Interferometric Wide Swath); 3 m (Sentinel-1 Stripmap, limited coverage) |
| Temporal revisit | Daily (Planet SuperDove global); up to 15 passes/day over target (WorldView Legion); 6-day (Sentinel-1 dual-satellite) |
| Spectral bands used | Panchromatic, Red, NIR, Coastal Blue (optical); C-band 5.4 GHz VV/VH (SAR coherence) |
| Minimum detectable shelter unit | Single tarpaulin shelter (~3 × 4 m) detectable at 30 cm resolution; clusters of 5–10 shelters detectable at 3 m |
| Settlement extent accuracy | Boundary delineation typically within 10–30 m at 3 m resolution; validated against ground truth in published humanitarian RS studies |
| Population estimate uncertainty | ±30–60% depending on occupancy rate assumptions; ground-truthing required to narrow range |
| Archive depth | Sentinel-1 from 2014 (free); Landsat from 1972 (free); commercial archives from approximately 2008 |
| Cloud penetration | SAR (Sentinel-1) fully cloud-independent; optical sensors blocked by cloud cover above ~70% scene fraction |
| Delivery formats | GeoJSON, Shapefile, GeoTIFF, PDF analytical report, CSV population estimate table |
Analytics Satellize can run
| Settlement boundary polygon | Supervised land-cover classification and spectral change detection on Planet SuperDove time series; bare-earth and synthetic-material spectral signatures | GeoJSON / Shapefile updated on requested cadence (daily to weekly) |
| Shelter density heat map | Object-based image analysis and CNN-based shelter detection on WorldView or Pleiades Neo imagery; density gridded at 10–50 m resolution | GeoTIFF raster layer with shelter count per grid cell |
| Population estimate with confidence interval | Shelter count multiplied by published UNHCR occupancy rate distributions; Monte Carlo uncertainty propagation over occupancy range | CSV table with low, central, and high population scenarios; methodology note for funder submission |
| SAR coherence change map | Interferometric coherence differencing between pre- and post-event Sentinel-1 acquisitions; decorrelation thresholding to isolate construction disturbance | GeoTIFF coherence-loss layer; useful during cloud-obscured optical windows |
| Settlement growth time series | Multi-date boundary polygon stack from Planet SuperDove daily archive; area and shelter-count trend extracted per epoch | Time-series chart (PNG/PDF) and tabular data for inclusion in connectivity funding applications |
| Candidate macro-cell siting overlay | Viewshed and coverage radius modelling over shelter-density grid; identifies locations maximising population coverage per site | GIS layer of ranked candidate tower locations with estimated coverage footprints |
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.