Informal settlement mapping for network densification prioritisation
Standard census data can lag informal urban growth by a decade or more. Multi-temporal VHR optical and SAR imagery reveals building-footprint density at sub-metre precision, giving network planners a current proxy for subscriber demand where official maps go blank.
Sensors
- Planet SuperDove: 3 m optical resolution, daily revisit over most land areas, 8 spectral bands including red-edge. Sufficient to resolve individual building footprints in dense informal settlements and detect new construction between monthly composites.
- Maxar WorldView-3: 30 cm panchromatic, 1.24 m multispectral. Resolves roof material, structure size and lane geometry at a level that supports individual-building delineation. Revisit is tasked rather than systematic, typically 1 to 4.5 days depending on latitude and cloud.
- Sentinel-1 SAR (C-band): 20 m resolution in Interferometric Wide Swath mode, 6-day repeat at the equator (12-day for a single satellite). Cloud-independent backscatter change detects new built-up area even under persistent cloud cover, which is critical in humid tropical cities.
- Sentinel-2 MSI: 10 m resolution in visible and NIR bands, 5-day revisit with both satellites. Useful for settlement boundary delineation and change detection at city scale, though individual structures below roughly 50 m² are not reliably resolved.
Why the census map is already wrong
National censuses in most low- and middle-income countries run on 10-year cycles. In fast-growing cities, informal settlements can double their built footprint in three to five years. A network planner relying on 2015 enumeration data to dimension a 2025 small-cell rollout is, in effect, planning for a city that no longer exists.
Administrative boundaries compound the problem. Informal settlements frequently straddle ward or district lines, or occupy peri-urban land that sits outside any enumerated zone entirely. The result is a systematic blank spot in demand models precisely where population density is highest and existing infrastructure is thinnest.
What a dense roofscape gives away
Building-footprint density, measured as the fraction of ground area covered by structures within a given grid cell, is a well-established proxy for population density in informal settlements. Studies using WorldView and Pleiades imagery have demonstrated strong correlation between rooftop count per hectare and household density derived from field surveys, though the relationship varies by settlement morphology and roof-sharing patterns.
At 30 cm resolution, WorldView-3 resolves individual corrugated-iron or concrete-block structures down to roughly 10 to 15 m² in footprint. Automated building-extraction models, typically convolutional neural networks trained on labelled VHR datasets, achieve intersection-over-union scores in the 0.65 to 0.80 range on informal settlement test sets, which is adequate for density mapping at 100 m grid resolution even if individual polygon boundaries are imprecise.
Planet SuperDove at 3 m is coarser but offers daily coverage. It is better suited to detecting where new construction is occurring month-to-month than to precise footprint delineation. The combination of Planet for change detection and WorldView for baseline characterisation is more cost-effective than tasking WorldView at high cadence across an entire city.
SAR backscatter fills the cloud gap
Optical imagery is blind under cloud. In cities like Lagos, Dhaka or Kinshasa, cloud cover exceeds 70 percent of days during the wet season. Sentinel-1 C-band SAR is unaffected. Urban areas return high backscatter values due to double-bounce scattering off vertical walls and horizontal ground surfaces. New construction raises local backscatter by several decibels over bare soil or vegetation baselines, making it detectable in coherence-change or intensity-change analyses even at 20 m resolution.
SAR backscatter is not a substitute for optical footprint mapping. It cannot reliably distinguish a new single-storey dwelling from a small commercial shed, and it is sensitive to soil moisture changes that can mimic construction signals after heavy rain. The practical workflow is to use Sentinel-1 change maps to flag areas of probable new construction, then task optical imagery to those areas for confirmation and characterisation.
The ambiguity that cannot be resolved from orbit
High building-footprint density does not equal high occupancy. A dense informal settlement may have many small single-room dwellings each housing one or two people, or it may have similarly sized structures housing extended families of eight to twelve. Satellite imagery cannot distinguish these cases. Occupancy rates, which directly determine subscriber headcount, require ground truth: field surveys, mobile network operator passive data, or mobile money transaction density as a proxy.
Roof material is partially informative. Corrugated iron is common in lower-income settlements; concrete or tiled roofs correlate with higher income and, in some markets, higher smartphone penetration. WorldView-3 SWIR bands can distinguish metal from concrete roofing with reasonable accuracy. But this is a weak signal and should be treated as one input to a demand model, not a standalone estimate.
The honest position is that satellite-derived building density gives you a defensible rank-ordering of candidate densification zones. It does not give you a subscriber count. Operators should treat the satellite output as a prioritisation layer that directs where to invest in ground-level demand validation, not as a replacement for it.
Turning density maps into network decisions
The practical output is a gridded demand-potential surface, typically at 100 m or 250 m resolution, combining building-footprint density, change rate over the past 12 to 36 months, and any available socioeconomic proxies. This surface is overlaid against the operator's existing coverage footprint to identify cells where building density exceeds a threshold but signal strength is below a minimum service level.
Candidate small-cell or macro-site locations can then be ranked by the ratio of unserved building density to nearest existing site capacity. This is a straightforward spatial join that any GIS platform can execute once the density layer exists. The satellite work is in producing that layer at sufficient currency and spatial resolution to be trusted.
Satellize has applied analogous density-from-imagery methods in its analytics work, including the Tonga crop-estimation programme where building and land-use classification underpinned area estimates. The same pipeline, adapted for urban morphology rather than agricultural parcels, is the basis of this use case.
Honest limits of the method
Cloud cover is the most operationally significant constraint. In persistently cloudy regions, a usable cloud-free WorldView acquisition may require a tasking window of weeks rather than days. Planet's daily revisit improves the odds but does not eliminate the problem at 3 m resolution.
Vertical structure is largely invisible. A two-storey building and a single-storey building of identical footprint look identical from nadir. Building height matters for propagation modelling and for estimating total floor area as a population proxy, but it requires stereo or LiDAR data, which is covered separately in the building-height extraction page.
Archive depth for commercial VHR imagery varies by location. Well-covered cities may have WorldView archive going back to 2009. Secondary cities in low-income countries may have sparse coverage. Sentinel-1 archive is consistent from late 2014 and is freely accessible, making it the default for multi-year change baselines.
Typical figures
| Best optical resolution (building delineation) | 30 cm panchromatic (WorldView-3); 3 m multispectral (Planet SuperDove) |
| SAR resolution (change detection) | 20 m (Sentinel-1 IW mode); 3 m (Sentinel-1 Stripmap, limited swath) |
| Optical revisit (Planet SuperDove) | Daily global, cloud-permitting |
| SAR revisit (Sentinel-1 constellation) | 6 days at equator; 12 days single satellite |
| Minimum detectable structure (VHR optical) | Approximately 10 to 15 m² footprint at 30 cm resolution |
| Spectral bands used | Visible, NIR, red-edge, SWIR (WorldView-3 SWIR for roof material); C-band SAR (5.4 GHz) for change |
| Archive depth | Sentinel-1 from late 2014 (free); WorldView from 2009 (licensed, coverage varies by city) |
| Typical grid output resolution | 100 m to 250 m demand-potential surface |
| Latency (change detection to delivery) | 3 to 10 days from satellite acquisition, depending on cloud and tasking queue |
| Delivery formats | GeoTIFF density raster, GeoPackage vector footprints, CSV ranked site candidates |
Analytics Satellize can run
| Building-footprint density grid | CNN-based building extraction on VHR optical imagery (WorldView-3 or Planet), aggregated to 100 m grid cells | GeoTIFF raster layer, updated quarterly or on tasking |
| Settlement growth rate map | Multi-temporal change detection on Planet monthly composites and Sentinel-1 backscatter intensity time series | GeoPackage polygon layer showing areas of significant new construction over a user-specified period (12, 24 or 36 months) |
| Roof-material classification | Supervised spectral classification using WorldView-3 SWIR bands to distinguish metal, concrete and tile roofing as a socioeconomic proxy | Per-building attribute in footprint GeoPackage; summary statistics by grid cell |
| Unserved high-density zone ranking | Spatial join of building-density surface against operator-supplied coverage footprint; ranking by density-to-coverage ratio | Ranked CSV of candidate densification zones with grid-cell coordinates, density score and coverage deficit |
| Settlement boundary delineation | Object-based image analysis on Sentinel-2 or Planet imagery to extract informal settlement extents distinct from formal urban fabric | GeoPackage polygon layer of settlement boundaries with area and perimeter attributes |
| Multi-year change animation | Time-stacked Planet or Sentinel-2 composites rendered as annual snapshots | Georeferenced image stack (GeoTIFF series) suitable for internal stakeholder presentation |
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.