Urban tree canopy carbon inventory for municipal offset programmes
Municipalities issuing urban-forestry carbon credits need verified crown inventories, not self-reported tree lists. Sub-metre multispectral imagery and lidar enable individual crown delineation, species-group classification, and above-ground carbon estimation via published urban allometrics.
Sensors
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral across 8 VNIR bands plus 8 SWIR bands. SWIR aids impervious-surface separation and some species discrimination. Revisit roughly 1 day at off-nadir angles up to 45 degrees, though cloud and tasking demand affect actual delivery.
- Airbus Pléiades Neo: 0.30 m panchromatic, 1.2 m native multispectral (4 standard bands plus red-edge and deep-blue on Neo). Red-edge band is particularly useful for canopy vigour and early stress detection in street trees. Constellation of two satellites gives a revisit of around 12 hours at mid-latitudes under ideal conditions.
- Planet SkySat: 0.50 m collect, 0.72 m ortho product, 4-band plus panchromatic. Weaker spectral depth than WorldView-3 or Pléiades Neo, but tasking flexibility and a larger constellation make it practical for rapid change detection between annual inventories. Not ideal as the primary classification sensor.
- Leica CityMapper-2 (airborne lidar): Dual-channel waveform lidar delivering point densities above 20 pts/m² in typical city survey configurations. Provides canopy height models at 0.25 m resolution and enables direct crown volume estimation, which substantially tightens allometric uncertainty. Coverage is city-commissioned and not on-demand from orbit; acquisition cost and weather windows are the binding constraints.
Why a spreadsheet inventory fails a carbon auditor
Most municipal tree registers are built from ground surveys conducted over years or even decades, with species, diameter at breast height (DBH), and condition recorded tree by tree. The data are patchy, aging, and almost never spatially complete. A street-tree planted in 2018 may appear in the register; a park tree felled in 2021 may still be earning credits. For a carbon programme, that is not a minor administrative problem. It is a permanence failure waiting to be audited.
Satellite and airborne remote sensing does not replace the ground survey for DBH measurement, but it provides an independent, spatially exhaustive, dated snapshot of canopy extent. That snapshot can be repeated annually. The combination of a verified crown layer with allometric equations published for urban trees gives an auditor something a spreadsheet cannot: a reproducible, time-stamped carbon estimate with a quantified uncertainty range.
What a floating roof gives away, and what it hides
Urban canopy mapping is harder than forest mapping for a simple geometric reason. Street trees grow beside buildings, and buildings cast shadows that are spectrally indistinguishable from dark asphalt or bare soil in standard multispectral imagery. A WorldView-3 collect at 0.31 m panchromatic resolution resolves individual crowns clearly, but a crown partially occluded by a five-storey building will have its area underestimated unless the analyst corrects for shadow geometry using the solar azimuth and elevation at acquisition time.
Lidar sidesteps much of this problem. Because the sensor measures the first and last returns from a pulse, it can distinguish a tree crown at 12 m above ground from the building facade 2 m behind it. Where CityMapper-2 data are available, the canopy height model is the primary delineation input, with multispectral imagery used for species-group classification rather than crown boundary detection. Where lidar is absent, which is the common case for cities that have not commissioned an airborne survey, the workflow relies on multispectral imagery alone, and shadow-correction becomes a mandatory pre-processing step rather than an optional refinement.
From pixel to tonne: the allometric chain
Crown delineation at sub-metre resolution, typically using a watershed segmentation or region-growing algorithm applied to a canopy height model or a normalised difference vegetation index surface, produces a polygon for each detected tree. The polygon area gives crown projected area. From crown area, published urban allometric equations, such as those compiled in the i-Tree model framework developed by the USDA Forest Service, relate crown dimensions to DBH and thence to above-ground biomass. The equations carry species-specific coefficients, which is why species-group classification from multispectral imagery matters.
Classification accuracy is the honest weak point. WorldView-3 and Pléiades Neo can separate broadleaf deciduous from conifer and, with training data, can distinguish a small number of dominant genera such as Quercus, Tilia, and Platanus in temperate cities. Separating species within a genus is generally not achievable from passive optical imagery alone. The practical approach is to classify into allometrically meaningful groups, perhaps four to eight, rather than species, and to propagate the within-group allometric variance into the final carbon uncertainty estimate. A well-executed workflow should be able to report above-ground carbon per tree with a coefficient of variation in the range of 20 to 40 percent, depending on species diversity and canopy closure. That uncertainty is large relative to plantation forestry but is still sufficient for registry-grade reporting if disclosed correctly.
Revisit cadence and the permanence question
Carbon credits issued against urban trees carry permanence risk that is qualitatively different from forest projects. A single storm can remove hundreds of street trees overnight. Disease, particularly ash dieback caused by Hymenoscyphus fraxineus, has removed millions of trees from European urban canopies since 2012. Construction projects remove trees legally and illegally. A carbon programme that does not monitor annually is not monitoring permanence at all.
Annual re-tasking of WorldView-3 or Pléiades Neo over the project area, compared against a baseline crown polygon layer using change-detection algorithms, can flag crown loss down to individual tree level. The detection threshold is not zero: a tree that loses 30 percent of its crown to disease but retains its structure may not trigger a polygon-level alert. Canopy health indices derived from the red-edge band of Pléiades Neo or the SWIR bands of WorldView-3 can catch physiological stress before structural loss occurs, giving a municipality time to intervene rather than simply record the credit reversal.
Honest limits before you commission a survey
Cloud cover is the obvious operational constraint. A single cloud-free collect per year is achievable in most temperate cities but is not guaranteed. Leaf-off acquisitions in winter simplify building-shadow correction but reduce spectral separability between species groups. The optimal acquisition window for a temperate city is typically late spring to early summer, when canopy is at full leaf and solar elevation is high enough to minimise shadow extent.
Minimum detectable tree size depends on spatial resolution and crown shape. A young street tree with a crown diameter below 1.5 m will often be missed or merged with adjacent vegetation in a 1.24 m multispectral product, even if the panchromatic band resolves it. This matters for additionality claims on recently planted trees. The honest answer is that trees planted within the last three to five years are likely undercounted in a purely optical workflow, and a ground-truth sample is necessary to estimate the omission rate.
Satellize has applied canopy analytics in agricultural contexts, including the Kingdom of Tonga crop-estimation programme, and brings the same uncertainty-quantification discipline to urban carbon workflows. The method stack is not proprietary; what differs is the rigour of the uncertainty propagation and the format of the evidence package delivered to the registry.
Typical figures
| Primary optical resolution | 0.30 to 0.50 m panchromatic (WorldView-3, Pléiades Neo, SkySat); 1.2 to 1.24 m native multispectral |
| Lidar point density (where available) | Greater than 20 pts/m² typical for Leica CityMapper-2 city surveys; canopy height model at 0.25 m |
| Revisit (tasked optical) | Around 1 day (WorldView-3 off-nadir); around 12 hours (Pléiades Neo constellation); subject to cloud and tasking demand |
| Spectral bands used | VNIR 8-band (WorldView-3), red-edge and deep-blue (Pléiades Neo), SWIR 8-band (WorldView-3) for impervious-surface masking |
| Minimum detectable crown diameter | Approximately 1.5 m in optical-only workflow; smaller crowns detectable with lidar CHM input |
| Above-ground carbon uncertainty (per tree) | Coefficient of variation 20 to 40 percent, depending on species diversity and canopy closure; must be disclosed in registry submissions |
| Optimal acquisition window (temperate) | Late spring to early summer, full leaf, solar elevation above 45 degrees to minimise building shadow extent |
| Archive depth (commercial sensors) | WorldView-3 archive from 2014; Pléiades Neo from 2021; SkySat from approximately 2017 |
| Delivery formats | GeoPackage or Shapefile crown polygon layer, GeoTIFF canopy height model, CSV per-tree carbon table with uncertainty, PDF evidence report |
Analytics Satellize can run
| Individual crown delineation layer | Watershed segmentation or region-growing on normalised DSM or NDVI surface; shadow-correction pre-processing using solar geometry at acquisition time | GIS polygon layer with crown area, centroid coordinates, and detection confidence flag |
| Species-group classification | Random forest or support vector machine classifier trained on multispectral reflectance signatures; four to eight allometrically meaningful groups; accuracy assessed by stratified ground-truth sample | Attributed crown layer with species-group label and per-class classification confidence |
| Per-tree above-ground carbon estimate | Crown-area-to-DBH allometric conversion using i-Tree or equivalent published urban equations; species-group-specific coefficients; Monte Carlo uncertainty propagation | CSV table of per-tree carbon (kg C) with 90 percent confidence interval; aggregated municipal total with combined uncertainty |
| Annual canopy change detection | Polygon-to-polygon comparison between baseline and update crown layers; area-change threshold flagging; canopy health index from red-edge or SWIR bands for stress detection | Change GIS layer flagging crown loss, gain, and stress; summary table of credit reversals requiring registry notification |
| Shadow-adjusted canopy cover map | Solar-geometry-corrected binary canopy mask separating tree crowns from building shadows; impervious-surface exclusion using SWIR-based index | GeoTIFF canopy cover raster at 0.5 m resolution; per-land-use-zone canopy cover statistics |
| Registry-grade evidence package | Compiled acquisition metadata, processing logs, accuracy assessment tables, uncertainty quantification, and chain-of-custody documentation formatted to Verra or Gold Standard MRV requirements | PDF and structured data archive suitable for third-party verification submission |
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.