Urban green-corridor fragmentation and ecological permeability
Satellite-derived canopy and impervious-surface maps, combined with graph-theoretic connectivity models, reveal where urban green corridors have been severed and where least-cost paths for wildlife still exist.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands (B04, B08) at 5-day revisit with two satellites. Sufficient for impervious-surface classification and NDVI-based green-space delineation at city scale, but spectrally unable to separate grass from low tree canopy reliably at this pixel size.
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral (8 bands including red-edge and near-infrared), with a short-wave infrared package. This resolution allows individual tree crowns to be delineated and hard surfaces to be mapped at sub-metre precision. Revisit is roughly 1 day at off-nadir angles, but tasking cost limits it to priority zones rather than whole-city coverage.
- Planet SuperDove: 3 m multispectral (8 bands, including two red-edge channels) with near-daily revisit globally. The red-edge bands improve grass-versus-canopy discrimination compared with classic four-band imagery. Useful for change monitoring across a full urban extent at a cost point between Sentinel-2 and Maxar.
- Copernicus Urban Atlas: A pan-European land-use/land-cover product at minimum mapping unit of 0.25 ha for urban areas and 1 ha for peri-urban areas, derived from very-high-resolution imagery. Provides a validated baseline for impervious-surface extent and green-space classification in European cities, updated approximately every six years.
Why a hedgerow matters more than its area suggests
Ecological connectivity is not simply a function of how much green space a city contains. A city could retain 30 per cent green cover and still be ecologically dead if every patch is surrounded by six lanes of traffic and impermeable hardstanding. What matters is whether patches are close enough, and the intervening matrix permeable enough, for urban-adapted species to move between them. Hedgehogs, common pipistrelles, ground beetles and many songbirds all have documented dispersal limits that make a 200-metre gap across asphalt effectively impassable.
Graph-theoretic connectivity models formalise this intuition. Each green patch becomes a node; edges between nodes are weighted by the cost of crossing the intervening land cover. A road with high traffic volume carries a high cost; a garden boundary hedge carries almost none. Least-cost path analysis then identifies which routes are functionally open, which are marginal and which have already been severed. The output is not a map of green space. It is a map of movement opportunity.
The spectral confusion problem nobody should paper over
Sentinel-2 at 10 m is the workhorse for city-scale green-space mapping, and it earns that role. Five-day revisit, free archive back to 2015, and ten-band multispectral coverage make it the only sensor that can track seasonal change across an entire metropolitan area without budget constraints. The problem is that grass and tree canopy produce nearly identical NDVI values in summer. A manicured football pitch and a mature oak canopy can occupy the same spectral position in bands B04 and B08. For connectivity analysis, this distinction is critical: grass provides movement habitat for some species and none for others, and its permeability to a bat or a woodland beetle is entirely different from that of closed canopy.
Planet SuperDove's red-edge bands (centred near 705 nm and 740 nm) help. Canopy tends to show a steeper red-edge slope than managed grass, and the difference becomes more reliable when imagery is composited across multiple acquisition dates to reduce phenological noise. Even so, at 3 m pixels, a mixed hedge-and-gap boundary will still produce blended spectra. The honest position is that grass-versus-canopy accuracy at medium resolution sits in the range of 75 to 85 per cent in published urban classification studies, depending on city morphology and season. WorldView-3 at 1.24 m, with its full eight-band stack and short-wave infrared channels, pushes that closer to 90 to 95 per cent for crown-scale delineation, but the cost of wall-to-wall coverage of a large city is prohibitive. The practical answer is a stratified approach: Sentinel-2 for extent, SuperDove for seasonal discrimination, WorldView-3 for targeted validation of critical corridor nodes.
Impervious surfaces as the real barrier layer
Mapping what wildlife cannot cross is at least as important as mapping what it can. Impervious-surface fraction, derived by spectral unmixing of Sentinel-2 or by pixel-level classification of very-high-resolution imagery, provides the resistance layer that feeds the cost surface. Concrete and tarmac are assigned high resistance values; permeable surfaces such as gravel gardens, green roofs or unpaved paths carry lower values. Road-network vector data (OpenStreetMap or national datasets) can be overlaid to add traffic-volume penalties where those data exist.
The Copernicus Urban Atlas provides a validated starting point for European cities, with impervious-surface classes already distinguished from green urban areas and sports facilities. Outside Europe, equivalent products are patchier. Global impervious-surface datasets derived from Landsat and Sentinel-2 exist at 10 to 30 m resolution, but their accuracy in rapidly urbanising cities with informal settlements can be lower than in well-mapped European contexts. Any corridor analysis that feeds planning decisions should include an accuracy assessment layer, not just a connectivity map.
From cost surface to corridor priority score
Once the resistance raster and patch layer are assembled, connectivity metrics such as Probability of Connectivity (PC) and the delta-PC importance index can be computed for each patch and each potential stepping stone. PC is a graph-theoretic metric published by Saura and Pascual-Hortal that quantifies the probability that two randomly chosen points in the landscape fall in connected habitat. Delta-PC measures how much overall connectivity drops if a given patch is removed. High delta-PC patches are the ones a planner must protect; low delta-PC patches may be candidates for enhancement rather than strict preservation.
Least-cost path corridors between high-priority patches can then be mapped at the street level, identifying specific road crossings, wall gaps or garden boundaries where a small intervention, a wildlife crossing, a gap in a wall, a planted median strip, would restore a severed link. This is where satellite data stops being remote sensing and starts being actionable urban planning intelligence. The resolution of the input imagery determines the precision of the output: a corridor identified from Sentinel-2 alone might be 50 metres wide in the map but only 3 metres wide on the ground.
Change detection and the slow loss that planning committees miss
Single-epoch connectivity analysis tells you the current state. Time-series analysis tells you the trajectory. Sentinel-2's archive from 2015 onwards allows annual or seasonal green-space maps to be produced for most cities, revealing incremental losses that no single planning application would flag as significant. A 0.1 ha garden lost to a driveway extension is invisible in isolation. Across a borough over eight years, those losses can sever a corridor entirely.
Satellize's analytics stack runs this kind of multi-temporal classification on open constellations, adding commercial tasking at client licence where sub-metre validation is needed. The Tonga crop-estimation programme uses a similar compositing and classification pipeline adapted for agricultural parcels; the urban green-corridor workflow draws on the same architecture with different class definitions and a connectivity post-processing step. Change alerts can be configured to flag patches that cross a delta-PC threshold, giving conservation officers an early warning before a loss becomes irreversible.
What this analysis cannot do
Satellite-derived connectivity maps are models of physical permeability, not observed animal movement. A corridor that looks open in the imagery may be avoided by target species for reasons no sensor can detect: noise, light pollution, predator presence, microclimate. Ground-truthing with camera traps or acoustic monitoring remains necessary to validate that a mapped corridor is actually used.
Cloud cover is a persistent constraint in temperate and tropical cities. A single-date acquisition in a cloudy climate will have gaps; multi-temporal compositing reduces but does not eliminate this. Tall buildings cast shadows that suppress vegetation signal in nadir-looking sensors, creating false bare-ground classifications at the base of towers. Canopy height cannot be recovered from passive multispectral imagery alone; lidar or stereo photogrammetry would be needed to distinguish a 2 m shrub from a 20 m oak, and neither is routinely available at city scale from current open constellations. These are not reasons to avoid the analysis. They are reasons to state the uncertainty budget clearly in any deliverable.
Typical figures
| Spatial resolution (green-space mapping) | 10 m (Sentinel-2), 3 m (Planet SuperDove), 1.24 m multispectral (WorldView-3) |
| Revisit frequency | 5 days (Sentinel-2, two-satellite); near-daily (Planet SuperDove); ~1 day off-nadir (WorldView-3) |
| Spectral bands used | Red, NIR, red-edge (705 nm, 740 nm), SWIR; panchromatic sharpening for sub-metre products |
| Minimum mappable patch | ~0.25 ha at Sentinel-2 resolution; ~0.01 ha at WorldView-3 resolution |
| Grass-versus-canopy classification accuracy | 75–85% at 10 m (Sentinel-2); 85–92% at 3 m with red-edge (SuperDove); up to ~95% at 1.24 m (WorldView-3) |
| Impervious-surface mapping accuracy | Typically 85–92% overall accuracy at 10 m in published urban studies; lower in informal-settlement contexts |
| Archive depth | Sentinel-2 from 2015; Landsat back to 1972 (30 m); Copernicus Urban Atlas epochs from ~2006 |
| Connectivity metric | Probability of Connectivity (PC) and delta-PC importance index; least-cost path resistance rasters |
| Delivery formats | GeoTIFF classification rasters, GeoPackage vector patch and corridor layers, PDF priority-corridor report |
| Processing latency | Baseline city-scale classification: 2–5 days from imagery acquisition; change-alert mode: configurable threshold triggers |
Analytics Satellize can run
| Impervious-surface fraction map | Spectral unmixing or supervised pixel classification of Sentinel-2 and/or WorldView-3 imagery; validated against Copernicus Urban Atlas where available | GeoTIFF raster of impervious fraction (0–1 per pixel), city-wide coverage, annual update |
| Green-space and canopy-type classification | Random forest or support vector machine classifier trained on red, NIR and red-edge bands; multi-date compositing to reduce phenological ambiguity; separate grass, tree canopy, shrub and bare-soil classes | GeoPackage vector layer of classified green patches with area, perimeter and canopy-type attributes |
| Resistance (cost) surface | Land-cover class resistance values assigned per published urban ecology literature; road-network vector overlay for traffic-volume penalty; composite raster used as input to least-cost path solver | GeoTIFF resistance raster with documented class-cost table |
| Patch connectivity priority scores (delta-PC) | Graph-theoretic Probability of Connectivity and delta-PC index computed using Conefor software or equivalent; dispersal-distance parameters set per target species group (e.g. 200 m for hedgehog, 1 km for common pipistrelle) | Patch vector layer with delta-PC score per feature; ranked priority list in PDF report |
| Least-cost corridor map | Least-cost path and corridor-width analysis between high-priority patch pairs; paths clipped to road-network barrier intersections to identify specific crossing points | GeoPackage corridor polylines and polygons; crossing-point point layer for field-survey targeting |
| Multi-temporal green-space change detection | Annual Sentinel-2 composites classified with consistent class scheme; change matrix computed between epochs; delta-PC recalculated to flag connectivity-significant losses | Change raster (gain/loss/stable) per epoch pair; automated alert when patch delta-PC exceeds configured threshold |
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.