The biggest thing AI adds to a forestry GIS is new layers: canopy height, tree crowns, species, disturbance and volume, estimated from satellite, aerial and LiDAR data at a scale no field crew could measure. A smaller addition is help running analyses, where AI agents turn questions into GIS steps. Each new layer comes with an error, and that error decides what it can be used for. A layer good enough to show where to send a crew may not be good enough to value a sale.
What the usual answer says
The descriptions that rank for this question list AI capabilities for GIS: automated mapping, species classification, change detection, predictive analytics and plain-language queries. All of them exist. None comes with a measure of how accurate the result is, which is the first thing a forest manager needs to know before using it.
New layers, with stated error
Canopy height is a good example of a layer AI now adds. Jamie Tolan and colleagues produced canopy height maps from very high resolution satellite imagery, published in Remote Sensing of Environment in 2024. They trained a model against aerial LiDAR and tested it on held-out data. They note that "Very high resolution satellite imagery (less than one meter (1m) Ground Sample Distance) makes it possible to extract information at the tree level while allowing monitoring at a very large scale." Their maps were a large step up in resolution from "the ten meter (10m) resolution of previous" worldwide canopy height maps. The model produced "an average Mean Absolute Error (MAE) of 2.8 meters".
Whether 2.8 metres is good enough depends on the decision. For finding gaps, regrowth or storm damage across a large estate, it is useful. For estimating the volume of a stand for sale, height is only the start.
Volume is harder to predict than height. FLORA, a June 2026 preprint by Emilie Vautier and colleagues, predicted forest attributes directly from airborne LiDAR on 32,052 inventory plots. It "achieves an rRMSE of about 12.3% (R2 = 0.88) for dominant height and 39% (R2 = 0.74) for total volume". In our reading, the further an attribute is from what the sensor measures, the larger the error tends to be.
Matching the layer to the decision
For each AI-derived layer, record its published or tested error beside it in the GIS: the metric, the value and what it was tested against. Then list the decisions it might feed, and ask for each whether an error of that size would change the decision.
Targeting field work can live with large errors, planning with moderate ones, and valuing a sale with small ones only. The same canopy height layer can be right for the first and wrong for the third.
Layers also age. A canopy map built from imagery of a given year describes that year, and harvesting, storms and growth change the forest after it. Record the date of the imagery with the error, and treat an older layer's error as larger than the figure published for it. In our view, a layer updated each year can be worth more than a sharper one that is never refreshed.
Test it on your own plots
Published errors were measured where each model was trained. Applied to another forest, the error is unknown until it is measured, so a published figure beside a layer can give false comfort. Local calibration changes the picture. A 2025 study in Forests by Sparks and colleagues, read from its abstract, trained LiDAR models on local plots and checked stand volumes against logs "scaled at the processing mill". The LiDAR estimates came within −8 to 6% of scaled volume, similar to the field cruise at −16 to 6%. Before any AI layer feeds more than targeting, test it against your own plots or scale records.
Help with analysis
The second addition, AI agents that run GIS analyses from plain-language requests, is earlier, and title 34 sets out how often they get the answer exactly right. Quarri's own work on map questions is in design, not delivered, and this piece describes the field.
When it doesn't apply
Layers used only to decide where to look, such as a first pass for storm damage, need little accuracy, because the field visit does the measuring. And some GIS work, such as boundary maintenance and legal mapping, is not helped by AI estimates at all.
Quarri for forest management is built around how a forest operation runs, from the cruise to the settled account.
Sources
- Tolan et al., "Very high resolution canopy height maps from RGB imagery using self-supervised vision transformer and convolutional decoder trained on Aerial Lidar", Remote Sensing of Environment 300, 113888, 2024 (arXiv 2304.07213): arxiv.org
- Sparks, Corrao, Keefe, Armstrong and Smith, "Comparison of Field Sampling- and Airborne Laser Scanning-Derived Stand-Level Inventories in a Mixed Conifer Forest and Volume Validation Using Log Scaling Data", Forests, 7 May 2025 (abstract read via Crossref): doi.org
- Vautier, Mallet and Vega, "FLORA: A deep learning approach to predict forest attributes from heterogeneous LiDAR data", arXiv preprint 2606.32023, 30 June 2026: arxiv.org
Quarri is an AI-native data platform for the timber supply chain. It connects buying, production, sales and inventory for forest management, sawmill, wood products and pulp, paper and packaging operators.