Forest and trees

Trees and vegetation Default model

Tree and shrub cover mask over a residential area

Example result: tree and shrub cover mask.

Cost

8 credits per km² + imagery

Example: 10 km² on Mapbox zoom 18 ≈ 110 credits.

Segments every area covered by trees and shrubs, including sparse forest and shrubland. It works in most environments, cities included. Height classes and individual tree crowns are options you switch on when you start a run.

What you get

A polygon layer of tree and shrub cover.

Works best with

0.6–0.15 m imagery (zoom 18–20) taken in the growing season. Leafless or snow-covered trees are not detected.

Usage examples

Measure canopy cover per district or parcel
Compare two dates to find cleared forest
Build a green-space inventory for a city

Version: 2026-07-03
Geo domain: Global
Method: Segmentation
GSD / zoom: 0.6–0.15 m / z18–20

The model is trained on 0.6–0.3 m imagery from different regions and climate zones. Its resolution is enough to find small groups of trees and narrow tree lines. Use imagery from the growing season: leafless trees and snow-covered vegetation are not the target class.

Scenarios with this model

Each scenario is the Forest and trees model with one option switched on. Choose options in the run form of Mapflow Web or the QGIS plugin.

Map forest and tree cover

Default run

All tree and shrub cover, down to small groups and tree lines.

8 cr/km² + imagery · zoom 18–20

Map forest and tree cover

Vegetation height along power lines

+ Height classes

Vegetation split into shrubs, 4–10 m forest and forest over 10 m.

18 cr/km² + imagery · zoom 18–19

Vegetation height along power lines

A point for every tree crown on a golf course

Count individual trees

+ Tree crowns

A point for every tree crown, free-standing trees included.

13 cr/km² + imagery · zoom 19

Count individual trees

Vegetation height along power lines

Forest and trees + Height classes

Cost

18 credits per km² + imagery

Example: 10 km² on Mapbox zoom 18 ≈ 210 credits.

Runs the forest model with height estimation. Vegetation is classed as shrubs under 4 m, forest from 4 to 10 m and forest over 10 m, which is what vegetation managers need near power lines.

What you get

Vegetation polygons with a height class.

Works best with

0.6–0.3 m imagery (zoom 18–19) taken in the growing season.

Usage examples

Prioritise clearance work in power line corridors
Plan right-of-way maintenance by height class
Screen vegetation along roads and railways

Tip. Buffer your line layer in QGIS (Vector › Geoprocessing › Buffer) and use the buffer as the area, so you pay only for the corridor.

Note

Height classes:

shrubs lower than 4 m;
forest from 4 to 10 m;
forest higher than 10 m.

The classes support vegetation management in power line corridors and similar zones, see the professional solutions by Geoalert. The thresholds can be customized for your requirements.

Count individual trees

Forest and trees + Tree crowns

Cost

13 credits per km² + imagery

Example: 10 km² on Mapbox zoom 19 ≈ 220 credits.

Extracts tree crowns from forest and from single trees, as points with a crown radius or as crown polygons.

What you get

Tree crown points with a radius, or crown polygons.

Works best with

0.3 m imagery (zoom 19) to separate individual trees.

Usage examples

Street-tree and park inventories
Count trees in orchards and plantations
Track tree loss on a construction site

Important

Use 0.3 m imagery (about zoom 19) with the Tree crowns options when you need individual trees.

All options

Height estimation: Classifies the forest mask by height: shrubs under 4 m, forest 4–10 m, forest over 10 m.
Tree crown polygons: Extracts tree crowns from forest and from free-standing trees, as polygons.
Tree crown points: Extracts tree crowns from forest and from free-standing trees, as points with a crown radius.

Sample results

Forest model result: solid forest mask — Solid **forest** mask

Forest model result: tree crowns as points — **Tree crowns**, points

Forest model result: forest with height classes — **Forest with heights** (raster output)

📊 Benchmarks

Forest and trees v.2026-09-10 (Global, segmentation) was evaluated on 8 areas of interest against manually annotated ground truth. Metrics are area-based: IoU is the intersection-over-union of the predicted and ground-truth vegetation masks, and F1, precision and recall are computed on the overlapping mask area. The v.2026-09-10 update refines tree-crown extraction results.

AOI (location)	Predicted features	IoU	F1	Precision	Recall
Italy — Crotone	254	0.862	0.926	0.944	0.909
Spain — Madrid	368	0.813	0.897	0.880	0.915
Philippines — Balanga	220	0.806	0.892	0.864	0.923
Spain — Velilla de San Antonio	1070	0.784	0.879	0.894	0.864
Spain — Cuenca	409	0.763	0.865	0.859	0.872
Uzbekistan — Tashkent	156	0.730	0.844	0.905	0.791
Spain — Rus	249	0.617	0.763	0.699	0.841
Argentina — La Banda	93	0.576	0.731	0.653	0.829
Global (mean of 8 AOIs)	2819	0.744	0.850	0.837	0.868

Area-based IoU / F1 / Precision / Recall measured against ground-truth vegetation masks; segmentation evaluation run 2026-07-03 (unchanged in v.2026-09-10). Compared with the previous version v.2025-06-14 (mean F1 0.513, IoU 0.396).

Benchmarks - tree crowns

Latest update — 🌲 Forest and trees v.2026-09-10 (tree crown polygons). The crown-polygon output was evaluated on 7 areas of interest (AOI) against manually annotated ground truth. Metrics are area-based on the crown masks: IoU is the intersection-over-union of the predicted and ground-truth crown masks, and F1 / Precision / Recall are computed on the overlapping mask area.

AOI (location)	Predicted crowns	IoU	F1	Precision	Recall
Italy — Crotone	254	0.842	0.914	0.930	0.898
Spain — Madrid	398	0.823	0.903	0.907	0.899
Philippines — Balanga	309	0.780	0.876	0.849	0.904
Spain — Cuenca	433	0.732	0.846	0.875	0.818
Spain — Velilla de San Antonio	1134	0.694	0.819	0.855	0.786
Spain — Rus	239	0.689	0.816	0.759	0.883
Argentina — La Banda	111	0.624	0.769	0.700	0.852
Global (mean of 7 AOIs)	2878	0.741	0.849	0.839	0.863

Area-based IoU / F1 / Precision / Recall measured on tree-crown polygon masks; evaluation run 2026-10-01. Compared with the previous version v.2026-07-03 (mean F1 0.824, IoU 0.706).

Tree count accuracy. For tree-counting use cases we also track how closely the number of detected crown polygons matches the ground-truth crown count. Relative count error is |detected − ground truth| / ground truth.

AOI (location)	GT crowns	v.2026-09-10 detected (count err)	v.2026-07-03 detected (count err)
Italy — Crotone	327	254 (22.3%)	254 (22.3%)
Spain — Madrid	478	398 (16.7%)	368 (23.0%)
Philippines — Balanga	193	309 (60.1%)	220 (14.0%)
Spain — Cuenca	321	433 (34.9%)	409 (27.4%)
Spain — Velilla de San Antonio	1212	1134 (6.4%)	1070 (11.7%)
Spain — Rus	258	239 (7.4%)	249 (3.5%)
Argentina — La Banda	125	111 (11.2%)	83 (33.6%)
Mean count error (7 AOIs)		22.7%	19.4%

Ground-truth crown counts from the updated report (2026-10-01). v.2026-09-10 is substantially closer to the true count where the previous model under-detected (La Banda, Madrid, Velilla de San Antonio), but it over-segments dense canopy at Balanga and Cuenca — producing more crowns than ground truth — so its mean relative count error (22.7%) is slightly above v.2026-07-03 (19.4%). ⚠️ The false positives may also indicate the lack of details in the manual labels as the mask-overlap accuracy (IoU / F1) still favours v.2026-09-10 in those areas.

Benchmarks - canopy height (CHM)

The Height estimation output (Canopy Height Model, CHM) is benchmarked against lidar-derived ground truth on 5 m hexagons (max aggregation). The current model CHM v.2026-08-31 reduces the mean absolute height error from 4.70 m to 3.24 m (−31%) versus CHM v.2026-07-03 across 38 areas of interest, with the largest gains in tall, structurally complex forest.