Informal settlement mapping and growth monitoring
Spectral texture, roof-material signatures and morphological analysis of very-high-resolution imagery can detect and delineate informal settlements, track their lateral growth over time, and flag where planning records diverge from reality.
Sensors
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral (8 bands including SWIR at 3.7 m GSD), 1.1 m SWIR. SWIR bands 6–8 are the primary discriminator for corrugated metal versus concrete roofing. Tasked revisit typically 1–4 days depending on latitude and cloud.
- Planet SuperDove: 3 m multispectral, 8 bands including red-edge and NIR. Daily revisit over most urban areas. Resolution is insufficient for individual roof classification but well suited to detecting settlement boundary expansion at monthly intervals.
- Sentinel-2 MSI: 10 m (visible/NIR), 20 m (SWIR, red-edge). Five-day revisit at the equator. Free and open; the 20 m SWIR bands retain some discriminating power for dense metal-roofed clusters. Best used for multi-year growth trend analysis rather than individual structure mapping.
- EROS-C (ImageSat International): 0.5 m panchromatic. Operated by ImageSat International; provides an alternative very-high-resolution archive with coverage across Africa, the Middle East and South Asia where informal settlement growth is most rapid. Panchromatic only, so spectral classification requires fusion with multispectral sources.
What a corrugated-iron roof gives away
Informal settlement mapping rests on a physical fact: the materials used to roof unplanned housing are spectrally distinct from those used in formal construction. Corrugated galvanised steel has a high and relatively flat reflectance across the SWIR region (roughly 1.4–2.5 µm). Painted concrete tiles and asphalt shingles do not. WorldView-3's eight SWIR bands, available at 3.7 m GSD, were specifically designed for mineral mapping in mining applications, but they turn out to be equally useful for separating iron-sheet roofing from concrete in dense urban scenes. Blue and grey plastic sheeting, common in transitional or post-disaster informal construction, has a different SWIR signature again, with stronger absorption near 1.9 µm due to polymer absorption features.
Thermal infrared adds another dimension. Metal roofs heat rapidly during the day and cool rapidly at night; concrete and clay tile roofs have higher thermal mass and show a flatter diurnal temperature curve. ASTER (90 m) and Landsat 8/9 TIRS (100 m resampled to 30 m) are too coarse for individual roof attribution in most settlements, but they can confirm the aggregate thermal character of a district and help separate informal clusters from low-density formal housing at the neighbourhood scale.
Texture and morphology: the structural fingerprint
Spectral signature alone is not enough. Low-cost warehouse roofing and some agricultural storage buildings share the corrugated-metal spectrum. The distinguishing factor is morphology. Informal settlements have characteristic spatial statistics: small, irregular plot sizes, high building density, narrow or absent road setbacks, and a texture signature in the grey-level co-occurrence matrix (GLCM) that differs measurably from formal low-rise residential. At WorldView-3 resolution, GLCM contrast and homogeneity features calculated on a 5×5 pixel window have been used in published literature to achieve overall accuracy above 85 % in distinguishing informal from formal fabric, though results degrade in transitional zones where the two types mix.
Object-based image analysis (OBIA) is the standard processing framework. Segments are grown from the panchromatic or pan-sharpened image, then attributed with spectral, textural and shape features before classification. The shape features (compactness, elongation, area distribution) carry as much discriminating information as the spectral bands in many scenes. This matters because it means the method is not purely dependent on having SWIR coverage; a well-resolved panchromatic image from EROS-C, fused with Sentinel-2 spectral data, can still produce a useful classification.
Measuring growth: what revisit rate actually buys you
A single classification tells you where informal settlement exists. A time series tells you how fast it is growing and in which direction. Planet's SuperDove constellation, with daily revisit and a consistent 3 m resolution, is the most practical source for monthly change detection at city scale. The method is straightforward in principle: co-register two images, run the same OBIA classification, and compute the difference in the delineated boundary polygon. In practice, the main sources of error are seasonal vegetation change (which alters texture), shadow angle variation between acquisition dates, and atmospheric inconsistency.
For longer historical baselines, Sentinel-2 and Landsat 8/9 provide free archives extending back to 2015 and 1972 respectively. Decadal growth rates derived from Landsat are well established in the academic literature for cities in sub-Saharan Africa, South Asia and Latin America. Quantifying annual lateral expansion to within a few hundred metres is achievable at Landsat resolution; quantifying it to within tens of metres requires VHR. A reasonable workflow pairs Landsat for the 10-to-30-year trend with Planet or WorldView for the current boundary and recent growth rate.
Where the method gets confused
Honest disclosure is warranted here. The main source of false positives is low-density formal development that uses corrugated-metal roofing, common in rural peri-urban fringes across many lower-income markets. A newly built but legally permitted warehouse estate can score identically to an informal settlement on spectral and density metrics. Cadastral data, where it exists and is current, resolves this ambiguity. Where it does not exist or is out of date, the satellite analysis itself cannot reliably distinguish the two.
Cloud cover is the other persistent constraint. Tropical cities, where informal settlement growth is fastest, frequently have cloud cover exceeding 70 % on any given day. Planet's daily revisit improves the odds of obtaining a usable acquisition within a given month, but multi-month gaps in usable imagery are still common in equatorial regions during the wet season. SAR (Sentinel-1 C-band, 10 m) can detect built-up area extent through cloud, but its ability to classify roof material is limited; it contributes primarily as a cloud-independent change detector rather than a material classifier.
Resolution also imposes a minimum detectable settlement size. At Sentinel-2's 10 m, a cluster of fewer than roughly 20–30 structures is unlikely to produce a reliable classification. At WorldView-3's 0.31 m pan resolution, individual structures of 10–15 m² are detectable, but the cost of tasking large informal settlement areas at that resolution is significant, and cloud-free tasking cannot be guaranteed.
From classification to a usable planning product
The analytic outputs that actually get used in planning and investment decisions are boundary polygons with attributed growth vectors, not raw classified rasters. A polygon layer showing settlement extent at three or four dates, with arrows indicating the direction and rate of expansion in metres per year, is immediately legible to a municipal planner or a development-finance institution assessing resettlement risk. Overlaying that with flood-zone or road-access data (covered in sibling pages in this library) converts a mapping exercise into a risk-prioritisation tool.
Satellize runs informal-settlement classification and growth-monitoring workflows on open constellations and commercial tasking depending on the resolution requirement and budget of the client programme. The analytics pipeline draws on the same OBIA and spectral-classification methods described here, applied to whichever sensor combination the coverage and cloud situation makes viable. As with the Tonga crop-estimation programme, the deliverable is a structured analytic product rather than a raw image dump: boundary GIS layers, growth-rate tables, and a written interpretation that places the numbers in planning context.
Typical figures
| Best spatial resolution (roof classification) | 0.31 m pan / 1.24 m MS (WorldView-3); 0.5 m pan (EROS-C) |
| Practical resolution for boundary mapping | 3 m (Planet SuperDove); 10 m (Sentinel-2 visible/NIR) |
| Revisit for change detection | Daily (Planet); 1–4 days tasked (WorldView-3); 5 days (Sentinel-2 at equator) |
| Key spectral bands | SWIR 1.4–2.5 µm for roof-material discrimination; panchromatic + NIR for texture/OBIA |
| Thermal complement | Landsat 8/9 TIRS at 100 m (resampled 30 m); ASTER 90 m for aggregate district characterisation |
| Minimum detectable settlement cluster | ~20–30 structures at Sentinel-2 10 m; individual structures >10–15 m² at WorldView-3 |
| Historical archive depth | Landsat from 1972; Sentinel-2 from 2015; Planet from 2016 (selected cities) |
| Cloud limitation | Optical blocked by cloud; SAR (Sentinel-1, 10 m) provides cloud-independent built-up extent but not material classification |
| Typical classification accuracy (published literature) | >85 % overall accuracy in homogeneous scenes; lower in transitional peri-urban zones |
| Deliverable formats | GeoJSON / Shapefile polygon layers, GeoTIFF classified rasters, tabular growth-rate summaries |
Analytics Satellize can run
| Informal settlement boundary delineation | Object-based image analysis (OBIA) with GLCM texture and SWIR spectral features on VHR imagery | Polygon GIS layer with material-class attributes (metal roof, plastic sheet, concrete), compatible with QGIS / ArcGIS |
| Multi-date growth vector mapping | Co-registered boundary differencing across Planet or Sentinel-2 time series; lateral expansion measured in metres per year per boundary segment | Annotated GIS layer with directional growth arrows and tabular expansion rates by settlement zone |
| Roof-material classification | Supervised classification using WorldView-3 SWIR bands 6–8 combined with pan-sharpened visible; training labels from field or VHR visual interpretation | Classified raster with per-pixel material probability scores; summary statistics by administrative unit |
| Decadal trend analysis | Landsat 8/9 and Sentinel-2 archive processing; normalised difference built-up index (NDBI) and modified normalised difference water index (MNDWI) time series to isolate impervious surface expansion | Annual impervious-surface extent table with confidence intervals; PDF report with trend charts |
| Cloud-gap-filled change alert | Sentinel-1 SAR backscatter change detection to flag new construction events through cloud cover; optical confirmation when next clear acquisition is available | Monthly alert feed with flagged change polygons and acquisition-date metadata |
| Confusion-risk flagging (informal vs. formal low-cost) | Cadastral overlay and road-network proximity scoring to identify areas where spectral classification is ambiguous; manual review queue generated for high-ambiguity zones | Confidence-scored polygon layer; flagged ambiguous zones highlighted in review report |
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.