Urban expansion and boundary mapping

Satellite imagery from Sentinel-2, Landsat and Sentinel-1 SAR can track the outward spread of built-up areas with annual or sub-monthly cadence, exposing where urban boundaries on official maps have fallen behind reality.

Sensors

Sentinel-2 MSI: 13 spectral bands at 10–60 m resolution; the 10 m SWIR-composite enables NDBI and NDVI differencing for impervious-surface detection. Five-day revisit at the equator (2–3 days at mid-latitudes with both satellites). Free archive from 2015.
Landsat 8/9 OLI: 30 m multispectral resolution, 16-day revisit per satellite (8-day combined). The continuous archive from 1972 via earlier Landsat generations makes it the only freely available record long enough to reconstruct decades of urban growth. SWIR bands support NDBI calculation.
Sentinel-1 C-band SAR: Interferometric Wide Swath mode delivers 10 m range × 10 m azimuth imagery regardless of cloud or darkness. Coherence loss between repeat passes (6-day or 12-day) flags newly cleared or paved surfaces. Particularly useful in persistently cloudy tropical cities where optical revisit is effectively much lower than nominal.
Planet SuperDove: 3 m resolution, daily revisit, eight spectral bands including red-edge. Useful for resolving peri-urban fringe detail that sits below Sentinel-2's 10 m floor, though at commercial cost and without a long free archive. Best used to sharpen boundaries already identified in Sentinel or Landsat change maps.

What the 10-metre floor actually means for city edges

The 10 m pixel of Sentinel-2 is often cited as a strength. At the urban fringe it is also a genuine limit. A single-storey house on a 100 m² plot is smaller than one pixel. A row of such houses becomes detectable only when enough of them cluster to shift the mixed-pixel reflectance signature measurably toward the built-up end of the spectrum. In practice, Sentinel-2 reliably detects new residential blocks once they cover roughly 0.5 ha or more of contiguous impervious surface. Below that, the signal is ambiguous.

Landsat 8/9 at 30 m is coarser still, but its value is chronological rather than spatial. No other freely available archive lets an analyst compare today's city extent with its 1990 or 2000 footprint at consistent radiometric quality. For long-run growth accounting, that temporal depth outweighs the resolution penalty. Where sub-10 m detail is genuinely needed at the fringe, commercial imagery from Planet or similar providers can be layered in, but the cost and archive gaps should be planned for honestly.

How spectral indices expose new construction

The Normalised Difference Built-up Index (NDBI) contrasts SWIR reflectance against near-infrared. Bare concrete, brick and asphalt reflect more strongly in SWIR than vegetation does, so NDBI rises as a plot is cleared and built upon. Differencing NDBI between two dates isolates pixels that have shifted toward built-up character. Paired with NDVI differencing (which falls as vegetation is removed), the two indices cross-validate each other: a pixel that simultaneously gains NDBI and loses NDVI is a strong change candidate. A pixel that gains NDBI without losing NDVI is more likely a soil exposure or seasonal artefact.

Seasonal timing matters considerably. Comparing a wet-season image with a dry-season image will produce spurious change signals because bare soil and stressed vegetation mimic built-up spectral signatures. Analysts should either match acquisition seasons across years or use time-series methods (harmonic fitting, for instance) that model seasonal variation and isolate the residual trend. This is not an exotic requirement; it is basic practice, and any workflow that ignores it will overcount urban growth.

SAR coherence as a complementary signal

Synthetic Aperture Radar measures the phase relationship between microwave pulses scattered back from the surface. Vegetation scatters incoherently between passes because leaves and branches move. Bare soil and built surfaces scatter more coherently. When a vegetated plot is cleared and paved, coherence between a pre-event and post-event Sentinel-1 pass rises sharply. When a building is erected on previously bare ground, coherence can drop temporarily during construction (disturbed soil, scaffolding) and then rise again once the structure stabilises.

The practical advantage of SAR coherence is cloud independence. West African coastal cities, much of South and Southeast Asia, and tropical Latin America experience cloud cover that renders optical revisit effectively seasonal rather than monthly. Sentinel-1's 6-day or 12-day repeat cycle (depending on acquisition mode and latitude) is unaffected. The limitation is interpretive: coherence change is a geometric and dielectric signal, not a spectral one. It tells you the surface has changed character; it does not by itself distinguish a new road from a new warehouse from a flood-deposited sediment layer. Fusion with optical data resolves most of that ambiguity.

Boundary mapping versus change detection: a practical distinction

Change detection answers the question: what was not built-up before and is now? Boundary mapping answers a different question: where exactly does the built-up area end today? The two analyses share methods but serve different administrative purposes. A planning department enforcing a growth boundary needs a current perimeter they can overlay on cadastral parcels. A national statistics office estimating urban population needs a time-series of area extent. A land-administration agency updating official maps needs both, reconciled to a common datum.

The honest position on accuracy is that Sentinel-2 and Landsat change maps typically achieve overall accuracies of 85–95 % in published validation studies, but those figures are averages across a scene. Accuracy is lower at the fringe, where mixed pixels dominate, and lower still in informal settlement areas where roofing materials (corrugated iron, tarpaulin, thatch) have reflectance signatures that overlap with both soil and vegetation. Sub-metre commercial imagery is the appropriate tool when parcel-level boundary precision is required. For strategic planning and policy monitoring at city or regional scale, 10 m is generally adequate.

Putting it into practice: archive depth, delivery and honest timelines

A credible urban expansion analysis starts with a baseline. For most cities in the developing world, the Landsat archive provides usable imagery back to the early 1980s, though cloud-free coverage in any given year can be sparse in humid climates. Sentinel-2 provides a cleaner, higher-resolution baseline from 2015. Between those two anchors, an analyst can construct a growth chronology with decadal steps (Landsat) refined to annual or semi-annual steps (Sentinel-2) and then to sub-monthly monitoring for the current period.

Satellize runs this kind of multi-sensor change stack on open constellations as a standard analytics workflow, with outputs delivered as GIS layers, change-area statistics by administrative zone, and optional alert feeds when new construction exceeds a user-defined threshold. The Tonga crop-estimation programme demonstrated that the same open-archive approach scales to small island contexts with irregular cloud cover, which is directly relevant to cities in similar climatic settings. For clients who need finer spatial resolution than Sentinel or Landsat provides, commercial tasking can be added on client licence. The lead time for a first baseline analysis of a single city is typically a matter of weeks, not months, because the archive already exists.

Typical figures

Best available spatial resolution (free data)	10 m (Sentinel-2 MSI visible/NIR/SWIR bands)
Best available spatial resolution (commercial)	3 m (Planet SuperDove); sub-metre from other providers on tasking
Revisit cadence (optical, free)	2–5 days (Sentinel-2, latitude-dependent); 8 days combined Landsat 8+9
Revisit cadence (SAR, free)	6–12 days (Sentinel-1 Interferometric Wide Swath, orbit-dependent)
Archive depth	1972 to present (Landsat); 2015 to present (Sentinel-2); 2014 to present (Sentinel-1)
Minimum detectable built-up patch	Approximately 0.5 ha at 10 m resolution; smaller patches require commercial imagery
Key spectral bands	SWIR (1.57–1.65 µm), NIR (0.84–0.87 µm) for NDBI/NDVI; C-band 5.4 GHz for SAR coherence
Typical change-detection accuracy	85–95 % overall (published validation range); lower at mixed-pixel fringe zones
Delivery formats	GeoTIFF change layers, vector polygons (GeoPackage/Shapefile), zonal statistics (CSV), WMS/WMTS tile feeds
Latency from acquisition to analysis	Sentinel-2 and Landsat data typically available within 24 hours of acquisition; processed change layers within 1–3 working days depending on pipeline

Analytics Satellize can run

Urban extent baseline map	NDBI/NDVI threshold classification on Sentinel-2 or Landsat composite; manual QA on fringe zones	Vector polygon of current built-up perimeter, attributed by density class; GeoPackage or Shapefile
Multi-decade growth chronology	Bi-temporal NDBI differencing across Landsat archive epochs (e.g. 1990, 2000, 2010, 2020, current); area statistics per epoch	Time-series chart and GIS layer stack showing urban footprint at each epoch; CSV area table by administrative zone
Annual change alert layer	Automated Sentinel-2 NDBI differencing on rolling 12-month window; change candidates filtered by minimum patch size and cross-validated against Sentinel-1 coherence	Annual GeoTIFF change raster and vector summary of new built-up patches exceeding user-defined area threshold
SAR coherence change map	Sentinel-1 repeat-pass interferometric coherence differencing (6- or 12-day pairs); coherence gain mapped as proxy for surface hardening	Cloud-independent change layer for periods of persistent optical cloud cover; GeoTIFF with coherence-difference values
Peri-urban fringe classification	Spectral mixture analysis or supervised classification on Planet SuperDove 3 m imagery to resolve sub-10 m built-up patches; trained on Sentinel-2-derived labels	High-resolution fringe classification raster; parcel-level summary table for planning enforcement input
Official boundary gap report	Overlay of current satellite-derived urban extent against client-supplied official administrative or planning boundary polygons; gap area calculated per boundary segment	PDF report with maps and statistics quantifying where built-up area has expanded beyond official limits; GIS layer of exceedance zones

Sources

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.