Urban green-space audit and canopy mapping

Satellite multispectral and LiDAR data can quantify urban green space with far more rigour than ground surveys alone, but separating canopy layers and correcting for seasonal bias demands careful method choices. This page explains what the sensors can and cannot resolve.

Sensors

Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at the equator with twin satellites. Thirteen spectral bands including red-edge (705 nm, 740 nm, 783 nm) that are particularly sensitive to chlorophyll content and canopy stress. Free archive from 2015 onwards. Cloud cover remains the primary operational constraint.
Planet SuperDove: 3 m resolution, eight spectral bands including two red-edge channels. Near-daily revisit in most latitudes. Useful for detecting small street-tree patches that fall below Sentinel-2's pixel floor, though the commercial licence cost is per-area and adds up quickly over large cities.
Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral VNIR (eight bands), 3.7 m SWIR (eight bands). The SWIR bands allow lignin and moisture discrimination that aids species-level grouping. Revisit is roughly 1 to 4.5 days depending on latitude and tasking priority, but cost limits it to targeted validation areas rather than city-wide continuous monitoring.
GEDI (Global Ecosystem Dynamics Investigation): Spaceborne full-waveform LiDAR operating from the International Space Station. Provides canopy height and vertical structure estimates at approximately 25 m footprint spacing along orbital tracks. Not a wall-to-wall raster; coverage depends on ISS inclination (51.6° latitude limit) and track spacing. Published canopy height products serve as an independent validation reference for spectral canopy-area estimates.

What a 10-metre pixel actually sees

A single Sentinel-2 pixel at 10 m resolution covers 100 square metres. A mature urban oak with a canopy spread of 8 m fits inside one pixel with room to spare, which means the pixel's reflectance is a mixture of leaf, branch, shadow, and whatever surface lies beneath. This spectral mixing is the central problem in urban green-space mapping, and it does not disappear by switching to a vegetation index.

NDVI (Normalised Difference Vegetation Index) and EVI (Enhanced Vegetation Index) both exploit the contrast between red absorption and near-infrared reflectance that healthy chlorophyll produces. They work well at identifying where vegetation exists. They are less reliable at telling you whether a high-NDVI pixel is a 15-metre tree canopy or a well-watered lawn. EVI partially corrects for canopy background and atmospheric effects that inflate NDVI in dense vegetation, but neither index resolves the vertical structure question without an independent height reference.

Separating canopy from grass: where spectral data needs help

Grass and tree canopy can produce nearly identical NDVI values in peak summer. The red-edge bands on Sentinel-2 (particularly the 705 nm and 740 nm channels) add discrimination because the red-edge inflection point shifts with leaf area index and canopy depth, but the separation is probabilistic, not definitive. At 10 m, a mixed pixel of 60% grass and 40% tree shadow can mimic a lightly stressed shrub.

The practical solution is to combine spectral indices with a canopy height model. GEDI provides height estimates along its orbital tracks; airborne LiDAR, where a city has commissioned it, provides wall-to-wall height rasters at 0.5 to 1 m resolution. Overlaying a height threshold (commonly 2 m or 3 m is used to define 'tree canopy' in urban forestry standards) onto the spectral classification sharply reduces confusion between grass and woody vegetation. The honest caveat: GEDI's track spacing means large gaps exist between passes, and interpolating between tracks introduces uncertainty that should be reported in any delivered canopy-area figure.

WorldView-3's eight VNIR bands and eight SWIR bands push further toward species discrimination. Differences in leaf water content, lignin concentration, and wax coatings produce distinguishable SWIR signatures for broad species groups. This matters for urban heat island studies (different species provide different shade and evapotranspiration) but requires careful atmospheric correction and is not yet operationally reliable for individual species identification at city scale.

The phenology trap in year-on-year comparison

Deciduous canopy in a temperate city can swing from near-zero NDVI in February to peak greenness in July. If an audit uses imagery from June one year and August the next, the apparent change in green-space area is largely an artefact of phenological timing rather than actual tree loss or gain. This is a common error in municipal green-space reporting.

The correction is straightforward in principle: compare images acquired at equivalent phenological stages, not equivalent calendar dates. Sentinel-2's five-day revisit and multi-year archive make it possible to select cloud-free scenes within a narrow phenological window defined by growing-degree-day models or by the NDVI trajectory of a reference site within the same city. Planet's near-daily cadence makes this window selection easier but does not eliminate the need to define the window explicitly. Any year-on-year comparison delivered to a client should state the acquisition dates and the phenological stage assumption; without that, the number is not auditable.

Canopy health signals the sensors can and cannot carry

Chlorophyll stress shows up in near-infrared and red-edge reflectance before it is visible to the human eye. This is the genuine early-warning capability of multispectral monitoring: a tree suffering drought stress or pest damage will show a declining red-edge slope weeks before its leaves visibly yellow. Sentinel-2's red-edge bands make this detectable at city scale, though the signal is subtle and requires a multi-date baseline to distinguish stress from normal phenological variation.

What the sensors cannot do: they cannot identify the cause of stress. A declining NDVI trend in a street-tree row could indicate root compaction, salt damage from winter gritting, fungal infection, or drought. Satellite data narrows the list of hypotheses and prioritises which trees warrant ground inspection. It does not replace the arborist.

Building a defensible canopy-cover baseline

Municipal clients typically need a number they can defend in a planning inquiry or a climate-adaptation report: total canopy cover as a percentage of city area, broken down by administrative ward or planning zone. Getting to a defensible number requires four steps. First, a cloud-free or cloud-composited multispectral mosaic for the target season. Second, a height filter applied from GEDI or available airborne LiDAR to separate woody canopy from grass. Third, a spectral classification validated against a stratified sample of ground-truth points or very-high-resolution imagery. Fourth, an explicit uncertainty estimate on the final area figure, which for Sentinel-2-based methods at city scale is typically reported in the literature as plus or minus 5 to 15 percent depending on tree density and urban morphology.

Satellize applies this workflow operationally; the Tonga crop-estimation programme established the same compositing and phenological-correction pipeline that transfers directly to urban canopy contexts. For cities that want to track change over time rather than produce a single baseline, the Sentinel-2 archive from 2015 gives roughly a decade of comparable data, which is enough to detect meaningful trends in canopy cover linked to development pressure or planting programmes.

What to ask before commissioning an audit

Three questions determine which sensor combination is worth paying for. What is the minimum canopy patch size that matters for your reporting standard? If your city's green-space policy counts any patch above 500 square metres, Sentinel-2 is sufficient. If individual street trees matter, you need Planet or WorldView-3 for at least a sample of the city. Second, do you need height-based canopy-tree separation, or is a spectral-only classification acceptable for your purposes? Height separation is more defensible but requires either GEDI integration or a separate LiDAR dataset. Third, is this a one-time baseline or an ongoing monitoring commitment? The answer changes the cost structure entirely, because Sentinel-2's open archive makes annual updates cheap once the baseline workflow is established.

Cloud cover is the constraint that no sensor escapes in humid or monsoon climates. A single cloud-free scene in the right phenological window may require waiting through an entire season. Multi-date compositing reduces this risk but introduces its own assumptions. Be sceptical of any audit that does not disclose its cloud-masking method and the dates of the contributing scenes.

Typical figures

Spatial resolution (spectral)	10 m (Sentinel-2 VNIR); 3 m (Planet SuperDove); 1.24 m VNIR / 3.7 m SWIR (WorldView-3)
Revisit frequency	5 days (Sentinel-2, twin satellites); near-daily (Planet); 1–4.5 days (WorldView-3, tasked)
Canopy height reference	GEDI: ~25 m footprint, non-wall-to-wall, ISS coverage to ±51.6° latitude
Key spectral bands for canopy	Red-edge 705 nm, 740 nm, 783 nm (Sentinel-2); NIR; SWIR 1 and 2 (WorldView-3 for species grouping)
Minimum detectable canopy patch	~100 m² (Sentinel-2, reliable); ~10–25 m² (Planet); individual large trees (WorldView-3)
Archive depth	Sentinel-2: 2015 to present (open); Landsat: 1972 to present (open); Planet: 2016 to present (commercial)
Typical canopy-area uncertainty	±5–15% at city scale for spectral-only methods; reduced with height-model integration
Delivery formats	GeoTIFF canopy-cover raster, GeoPackage or Shapefile polygon layer, ward-level summary CSV, PDF audit report
Cloud cover constraint	Single-date acquisitions may be unusable; multi-date compositing required in humid climates

Analytics Satellize can run

City-wide canopy-cover baseline	Multispectral classification (NDVI/EVI/red-edge) with GEDI height-filter to separate woody canopy from grass; validated against stratified ground-truth sample	GeoTIFF raster and ward-level summary table with uncertainty range stated
Annual canopy-change detection	Phenologically matched Sentinel-2 composites from consecutive years; change pixels flagged by threshold on NDVI delta and confirmed by spectral classification	Change polygon GIS layer with gain/loss attribution and acquisition-date metadata
Canopy stress early-warning	Multi-date red-edge reflectance trend analysis; declining red-edge slope flagged against a rolling baseline for each tree-canopy segment	Ranked alert list of stressed canopy patches with coordinates, for arborist prioritisation
Species-group discrimination (targeted areas)	WorldView-3 eight-band VNIR plus SWIR spectral unmixing; broad species-group classification using published spectral libraries	Species-group polygon layer for designated survey zones; accuracy matrix included
Green-space equity analysis by ward	Canopy-cover raster intersected with administrative boundary and population-density layers; per-capita canopy-area calculated by zone	Ward-level equity report with ranked deficit zones, suitable for planning submissions
Phenological calendar for repeat audits	NDVI time-series analysis of reference sites within the city to define the optimal acquisition window for year-on-year comparability	Recommended acquisition window document and automated scene-selection criteria for future monitoring cycles

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.