Canopy Height Models: Global Products and Local Calibration

Updated 2026-11-02 · 4 min read · By the akar.earth specialist team

Canopy height model guide: how global 10 m and 30 m products are made, how drone and lidar CHMs differ, and how to calibrate them for Indonesian sites.

A canopy height model (CHM) is a map of vegetation height above the ground, and it is one of the most useful layers for scaling field carbon measurements across a landscape. Global CHMs now exist at 30 m, 10 m and even 1 m, but they are predictions with errors of several metres, so projects should calibrate them against local field or lidar data before relying on them.

Three ways a CHM is made

1. Direct measurement with lidar. Airborne laser scanning (ALS) or drone lidar measures both the canopy top and the ground under it. CHM = digital surface model (DSM) − digital terrain model (DTM). This is the most accurate source, with sub-metre detail.

2. Photogrammetry. Overlapping drone or high-resolution satellite images produce a DSM by structure-from-motion. The challenge is the DTM: under closed tropical canopy the cameras rarely see the ground, so the terrain model must come from elsewhere. See drone photogrammetry.

3. Model prediction from satellites. Spaceborne lidar such as GEDI gives sampled heights. Models then learn the relationship between those heights and wall-to-wall imagery (Landsat, Sentinel-2, sometimes Sentinel-1) and predict height everywhere.

Global products at a glance

These products differ in epoch, method and bias. Check the publication and data page for the version you use, and never assume a single-year map tells you anything about change.

Calibrating a CHM for your site

1. Collect reference heights. Measure tree heights in field plots with a hypsometer or laser rangefinder, or fly drone lidar over a sample of the area.
2. Match scales. Compare like with like: a 10 m pixel’s predicted height is not the same as the tallest tree in a 0.1 ha plot. Use plot-level metrics such as mean top height or the 95th percentile.
3. Fit a correction. A simple linear correction often fixes systematic bias, for example underestimation above 30 m.
4. Stratify. Calibrate separately for peat swamp forest, dryland forest, plantations, agroforestry and mangroves if the bias differs.
5. Validate on held-out reference data and report RMSE and bias.

How projects use CHMs

• Stratification. Height classes separate degraded from intact forest, which reduces sampling error. See plot sampling design.
• Biomass mapping. Height is a key predictor in remote sensing biomass models.
• ARR monitoring. Repeated drone CHMs show growth of planted trees in years where satellites still see mostly grass or shrubs.
• Degradation and HCV/HCS screening. Height and structure help distinguish secondary forest, scrub and plantation, which matters for high conservation value assessments and EUDR forest definitions.

Using CHMs to monitor change

A single CHM is a snapshot. Monitoring growth or degradation needs repeated CHMs from the same source and method. Drone or airborne surveys repeated with consistent flight settings and ground control can track tree growth in ARR sites and canopy gaps from logging. Global products are usually single-epoch, or their versions differ in method, so subtracting two of them to estimate change produces mostly noise. If you need change, plan your own repeat acquisitions over a sample of the area and calibrate trends against permanent plots.

Common mistakes

• Treating a global map as truth. A 10 m map that says 18 m may hide a 35 m dipterocarp stand.
• Using a photogrammetric DSM without a reliable DTM. The result is surface elevation, not tree height, and steep terrain makes it worse.
• Mixing epochs. Calibrating a 2020 map with 2026 field data in a growing plantation.
• Ignoring definitions. Lidar top height, field Lorey’s height and a modelled pixel height are different quantities.
• Assuming linear relationships across forest types. Mangroves, peat swamp and hill forest behave differently.

What good evidence looks like

A verifier should see: the CHM source and version; reference data with locations, dates and instruments; the calibration model and its fit statistics; validation results; and how the calibrated CHM feeds into stratification or biomass. If you produced your own drone or lidar CHM, include flight parameters, ground control or RTK positioning, and processing settings.

Indonesia specifics

Tall lowland forests in Kalimantan and Papua are exactly where global products tend to underestimate. Mangroves, by contrast, are relatively short and dense, and can be overestimated where sparse returns hit the canopy. Haze and cloud reduce clean GEDI shots, which thins the training data behind global maps in these regions. Local calibration is not optional here.

For field reference heights and drone campaigns, see our field data campaigns service. For the trade-offs between airborne options, read lidar vs photogrammetry, and for the wider system, the dMRV hub.