akar.earthGround truth: Definition
What is ground truth in remote sensing and carbon MRV? Why field data is used to train and check satellite models, and how to collect it well.
Ground truth is information collected on the ground, such as tree measurements, land cover observations, photos and GNSS positions, used to train, calibrate and check remote sensing data. It is the reference that tells you whether a satellite map or biomass model is right.
Why it matters
Satellites measure light or radar signals, not trees or carbon. Without ground truth, a map is an untested guess. Carbon standards require models to be calibrated and validated against field data, and verifiers ask for evidence of how well remote sensing estimates match the ground. For EUDR, ground observations help confirm what a plot actually contains when imagery is ambiguous.
How it is collected
- Field plots: measured trees (diameter, height, species) in plots of known size and location.
- Land cover points: observations with photos at known coordinates, for classification training and accuracy assessment.
- Precise positioning: GNSS, ideally RTK GNSS, so the plot lines up with the right pixels.
- Timing: collected close to the image date so the two describe the same conditions.
Ground truth is split into training data, used to build the model, and validation data, held back to test it.
Indonesia and Southeast Asia angle
In Indonesian agroforestry, satellite images struggle to tell shade coffee or cocoa from secondary forest, or young plantations from shrubland. Local field teams, often recruited from the community, can collect structured data on phones far more cheaply than outside survey crews. That also builds local capacity and supports community participation now emphasised under Permenhut 6/2026.
Good ground truth is also documented ground truth. Each record should carry the date, the collector, the device, its accuracy and photos, so that a verifier can audit it later.
Common mistakes
- Placing plots only near roads, which biases the sample.
- Poor GPS accuracy under canopy, so plots match the wrong pixels.
- Reusing training data to report accuracy.
- Long gaps between field and image dates.
Read ground truth vs satellite and see our field data campaigns.
Frequently asked questions
How much ground truth data is enough?
It depends on the variability of the landscape and the accuracy you need. Use a sampling design with enough points or plots in every class or stratum, and keep some aside for independent accuracy checks.
Can high-resolution imagery be used as ground truth?
For land cover classes it is often used as reference data. For biomass or species it is not a substitute for field measurement.