Orchard tree count and individual canopy area mapping
Very-high-resolution optical imagery and canopy height models let analysts count individual orchard trees and measure each crown. The results feed subsidy verification, insurance underwriting and productive-capacity assessments for permanent crops including olives, almonds and citrus.
Sensors
- Maxar WorldView-3: Panchromatic band at 0.31 m native resolution (licensed to 0.5 m for commercial delivery); eight VNIR multispectral bands at 1.24 m. The pan-sharpened product at 0.5 m resolves individual crowns in orchards with inter-tree spacing as tight as 3 m, and the SWIR bands assist in separating canopy from bare soil.
- Airbus Pléiades Neo: 0.3 m panchromatic, 1.2 m multispectral (four bands plus a red-edge channel). Stereo and tri-stereo acquisition modes produce photogrammetric canopy height models with vertical accuracy of roughly 0.5 m RMSE over flat terrain, enabling height-filtered crown delineation without a separate lidar mission.
- Planet SuperDove: 3 m multispectral resolution across eight bands including red-edge and NIR. Insufficient for individual-crown delineation in most orchards but useful for dense time-series monitoring of canopy greenness and seasonal phenology between higher-resolution acquisition campaigns.
- Airborne or spaceborne lidar (GEDI, mission-specific): NASA GEDI produces canopy height profiles at 25 m footprint spacing, too coarse for individual-tree work but useful for regional calibration. Airborne discrete-return lidar at point densities above 10 pts/m² resolves crown volume directly; it remains the reference standard against which photogrammetric CHMs are validated.
What a floating roof gives away
A mature olive tree occupies roughly 15 to 35 square metres of canopy depending on variety and pruning regime. At 0.5 m resolution, that crown is represented by 60 to 140 pixels, enough for an object-based image analysis algorithm to trace its boundary, compute its area, and flag it as distinct from the soil between rows. The geometry is unambiguous in a way that spectral indices alone cannot provide.
This matters because orchard subsidies in many jurisdictions, including the EU Common Agricultural Policy, are tied to declared tree counts and planted area. A satellite-derived count that disagrees with a farmer's declaration by more than a few percent triggers an audit. The satellite does not need to be perfectly right; it needs to be consistent and independently verifiable, which is a different and more achievable standard.
From pixels to crowns: the detection chain
Object-based image analysis (OBIA) segments a very-high-resolution image into homogeneous objects rather than individual pixels. A typical workflow applies a multiresolution segmentation algorithm, then classifies objects by spectral signature, shape compactness and the normalised difference vegetation index. Crowns in summer imagery of olive or almond orchards typically produce NDVI values above 0.4 against bare-soil NDVI values below 0.15, a contrast wide enough to survive moderate atmospheric haze.
Shadow is both a problem and a clue. At low sun angles, each tree casts a shadow whose length is a function of height. Analysts who account for solar geometry at acquisition time can extract a rough height estimate even from a single-pass image, which helps separate mature trees from young plantings. Photogrammetric canopy height models from stereo pairs formalise this: Pléiades Neo tri-stereo acquisitions routinely produce CHMs that distinguish a 4 m tree from a 2 m tree with sub-metre vertical uncertainty over flat ground.
The hard limit is canopy closure. When inter-tree spacing falls below roughly one crown diameter, adjacent crowns merge into a continuous mass in the image. Algorithms that work well at 6 m spacing fail at 3 m. This is not a solvable problem at current commercial resolutions; it is a known constraint that should be declared in any deliverable.
Canopy area as a proxy for productive capacity
Crown projected area correlates with leaf area index and, through that, with photosynthetic capacity and potential yield. For olive and almond, published field studies have found that trunk cross-sectional area and crown area together explain a substantial share of inter-tree yield variance, though the relationship is variety-specific and weakens in heavily pruned or hedgerow-trained systems.
For insurance underwriting, canopy area measured before a frost or hail event provides the pre-loss baseline against which post-event imagery is compared. A hailstorm that strips leaves reduces NDVI within days; a frost that kills branches changes crown geometry over weeks. Neither signal requires ground inspection to detect at orchard scale, which is the practical value for an insurer writing policies across thousands of hectares.
Subsidy verification is the more immediate commercial application. A government agency tasking a single WorldView-3 collect over a declared orchard can receive a tree-count layer within days of image delivery. Discrepancies above a set threshold, say 10 percent of declared trees, are flagged for field follow-up. The satellite pass costs a fraction of a physical inspection programme at regional scale.
Where the method fails, and what to do about it
Young orchards present the most consistent difficulty. Trees planted within the previous two to three years may have crowns smaller than 1 m in diameter, which places them at or below the resolution floor of even WorldView-3 pan-sharpened imagery. A 0.3 m Pléiades Neo acquisition improves detection, but very small crowns are still confused with isolated weeds or soil irregularities. In these cases, a canopy height model that confirms objects are above 0.5 m provides a useful discriminator.
Species confusion is rare in monoculture orchards but real in mixed plantings. Olive, carob and wild scrub produce similar spectral signatures in summer multispectral imagery. Adding a red-edge band, available on both SuperDove and Pléiades Neo, reduces but does not eliminate this ambiguity. Where species identity matters for the analysis, a ground-truth sample of at least 50 to 100 manually verified trees per species is needed to train and validate the classifier.
Cloud cover is the operational constraint that does not appear in accuracy tables. A single tasked acquisition has a non-trivial probability of cloud contamination over Mediterranean and subtropical orchard regions in winter and spring. Contracting for multiple acquisition attempts over a season, or pairing a commercial task with a Sentinel-2 time series to confirm cloud-free windows, is standard practice.
Putting a number on the uncertainty
Published validation studies on OBIA-based tree detection in olive and citrus orchards report detection rates of 85 to 95 percent and commission errors (false positives) of 5 to 15 percent under favourable conditions: mature trees, spacing above 4 m, summer imagery with NDVI contrast above 0.3. Performance degrades toward the lower end of those ranges in dense plantings or where the orchard has been recently pruned. Any deliverable should include a per-block accuracy estimate derived from a held-out validation sample, not just a global figure.
Satellize applies these methods on commercial tasking arranged on client licence, pairing WorldView-3 or Pléiades Neo imagery with open Sentinel-2 time series for seasonal context. The approach is structurally similar to what underpins the Tonga crop-estimation programme, scaled to permanent-crop inventories. The output is a GIS layer with per-tree crown polygons, area attributes and a summary count table suitable for direct import into a subsidy-management or insurance platform.
Typical figures
| Spatial resolution (primary sensor) | 0.3 m pan (Pléiades Neo) to 0.5 m pan-sharpened (WorldView-3); multispectral 1.2 to 1.24 m |
| Minimum detectable crown diameter | Approximately 1.5 m under ideal conditions at 0.5 m resolution; 1.0 m at 0.3 m resolution. Smaller crowns are below reliable detection. |
| Canopy height model vertical accuracy | 0.5 m RMSE typical for Pléiades Neo tri-stereo over flat terrain; degrades on slopes above 10 degrees |
| Revisit / tasking cadence | WorldView-3: 1 to 4.5 days revisit depending on latitude and off-nadir tolerance. Pléiades Neo: 1 to 3 days. Planet SuperDove: daily at most latitudes. |
| Spectral bands used | Panchromatic, Blue, Green, Red, Red-edge, NIR, SWIR (WorldView-3 adds eight SWIR bands at 3.7 m) |
| Typical scene area per task | WorldView-3: up to 225 km² per collect. Pléiades Neo: up to 400 km² per strip. |
| Archive depth | WorldView-3 archive from 2014; Pléiades Neo from 2021. Archival imagery may not meet required off-nadir or sun-angle criteria for CHM generation. |
| Delivery format | GeoPackage or Shapefile of crown polygons with area and NDVI attributes; summary CSV of tree counts per cadastral block; optional raster CHM in GeoTIFF |
| Detection accuracy (published range) | 85 to 95 percent recall, 5 to 15 percent commission error, for mature orchards with inter-tree spacing above 4 m |
Analytics Satellize can run
| Individual tree crown polygons | Multiresolution object-based image analysis (OBIA) segmentation and NDVI-based classification on pan-sharpened VHR imagery | GIS polygon layer with per-crown area, centroid coordinates and NDVI value; one layer per acquisition date |
| Orchard-level tree count and density map | Automated crown counting with manual QA on a stratified validation sample; count aggregated to cadastral or farm boundary | Summary table (CSV) of tree counts and mean crown area per parcel, flagging parcels where satellite count differs from declared count by more than a configurable threshold |
| Canopy height model | Dense image matching (photogrammetric stereo) from Pléiades Neo stereo or tri-stereo pairs; normalised against a bare-earth DTM | GeoTIFF raster at 0.5 m pixel spacing; per-tree height attribute appended to crown polygon layer |
| Pre- and post-event canopy change layer | Bi-temporal OBIA comparison: crown area and NDVI differenced between a pre-event and post-event acquisition to detect defoliation, structural damage or tree loss | Change polygon layer with damage classification (loss, partial defoliation, no change); formatted for insurance loss-adjustment workflow |
| Seasonal canopy greenness time series | Red-edge NDVI extracted from Planet SuperDove daily composites, aggregated to crown polygons from the VHR delineation layer | Per-tree or per-block time-series CSV covering the growing season; anomaly flags where greenness falls below a species-specific baseline |
| Subsidy-verification report | Comparison of satellite-derived tree count against declared cadastral register; discrepancy scoring per parcel using a configurable tolerance band | PDF and GIS package formatted to a client-specified subsidy-authority template, with accuracy statement and validation sample metadata |
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.