🌲 Canopy Height Model (CHM) — per-location benchmark (v.2026-08-31)

This page details the Canopy Height Model (CHM) produced by the 🌲 Forest and trees Height estimation option. CHM v.2026-08-31 is compared against CHM v.2026-07-03 on the most-improved sample per country, using lidar-derived ground truth aggregated to 5 m hexagons (max height).

The metric is the mean absolute error (MAE) of canopy height, in metres, against the lidar ground truth. Across 38 areas of interest, CHM v.2026-08-31 lowers the mean MAE from 4.70 m to 3.24 m (−31%) versus CHM v.2026-07-03, with the largest gains in tall, structurally complex forest.

How to read these maps. The Ground truth panel shows the lidar-derived canopy height as a green ramp (darker green = taller) over the aerial image. Each model panel is an error map: the colour ramp shows the absolute difference between the model's predicted canopy height and the lidar ground truth, per hexagon — 0 = exact match, rising to 15 m. Emptier, cooler maps mean predictions closer to lidar.

← → to browse · ESC to close

🇷🇺 Russia — RU_4

Model	MAE (m) ↓	vs lidar GT
CHM v.2026-08-31	2.73	best
CHM v.2026-07-03	8.06	−5.33 m worse

Ground truthlidar canopy height (darker = taller) on aerial

CHM v.2026-07-03error vs lidar · MAE 8.06 m

CHM v.2026-08-31error vs lidar · MAE 2.73 m

20,980 ground-truth hexagons. Mean absolute height error reduced by 5.33 m (−66%).

🇨🇭 Switzerland — SWISS_3

Model	MAE (m) ↓	vs lidar GT
CHM v.2026-08-31	3.36	best
CHM v.2026-07-03	7.78	−4.42 m worse

Ground truthlidar canopy height (darker = taller) on aerial

CHM v.2026-07-03error vs lidar · MAE 7.78 m

CHM v.2026-08-31error vs lidar · MAE 3.36 m

1,740 ground-truth hexagons. Mean absolute height error reduced by 4.42 m (−57%).

🇪🇪 Estonia — EST_473628

Model	MAE (m) ↓	vs lidar GT
CHM v.2026-08-31	2.75	best
CHM v.2026-07-03	5.42	−2.67 m worse

Ground truthlidar canopy height (darker = taller) on aerial

CHM v.2026-07-03error vs lidar · MAE 5.42 m

CHM v.2026-08-31error vs lidar · MAE 2.75 m

44,641 ground-truth hexagons. Mean absolute height error reduced by 2.67 m (−49%).

🇲🇽 Mexico — MEX_4

Model	MAE (m) ↓	vs lidar GT
CHM v.2026-08-31	1.69	best
CHM v.2026-07-03	2.93	−1.24 m worse

Ground truthlidar canopy height (darker = taller) on aerial

CHM v.2026-07-03error vs lidar · MAE 2.93 m

CHM v.2026-08-31error vs lidar · MAE 1.69 m

18,248 ground-truth hexagons. Mean absolute height error reduced by 1.24 m (−42%).

🇩🇰 Denmark — DEN_5

Model	MAE (m) ↓	vs lidar GT
CHM v.2026-08-31	2.20	best
CHM v.2026-07-03	2.67	−0.47 m worse

Ground truthlidar canopy height (darker = taller) on aerial

CHM v.2026-07-03error vs lidar · MAE 2.67 m

CHM v.2026-08-31error vs lidar · MAE 2.20 m

1,289 ground-truth hexagons. Mean absolute height error reduced by 0.47 m (−18%).

Summary

Country	Sample	CHM v.2026-07-03 MAE	CHM v.2026-08-31 MAE	Improvement
🇷🇺 Russia	RU_4	8.06	2.73	−66%
🇨🇭 Switzerland	SWISS_3	7.78	3.36	−57%
🇪🇪 Estonia	EST_473628	5.42	2.75	−49%
🇲🇽 Mexico	MEX_4	2.93	1.69	−42%
🇩🇰 Denmark	DEN_5	2.67	2.20	−18%

Mean absolute canopy-height error vs lidar ground truth · 5 m hexagons, max aggregation · absolute-error ramp 0–15 m. The US sample is omitted (a single, highly generalised tile).