Norway’s high-voltage power grid stretches across 130,000 kilometers of forested land. Every winter, trees fall onto the lines — knocking out power for households and straining the utilities responsible for keeping them on. The challenge isn’t finding the trees. It’s knowing which ones will fall. Traditional inspection methods can flag problem areas, but they can’t reliably single out the individual tree most likely to come down in the next storm. A new study asks whether a physics-based wind model, fed with detailed drone scans of each tree along a powerline, can do exactly that.
A forest full of suspects, a grid under pressure
Norway’s forests cover 38% of the country’s land area, and the high-voltage power grid runs through much of it — 130,000 kilometers in total. According to the study, 75% of power outages in Norway are linked to weather and vegetation, including wind-driven tree falls. Stronger storms and heavier wet-snow events in recent years have extended outages and placed growing financial pressure on utilities.
Global electricity demand has nearly doubled over the past two decades. Powerline grids are projected to expand by 16 million kilometers between 2020 and 2030 — more lines mean more forest edge, more exposure, and more trees capable of bringing a section down. Ground-based patrols are becoming both too slow and too expensive for the scale of vegetation management now required.
Scanning the forest from above: how drone LiDAR maps individual trees
The study mapped 18,689 individual trees along 25 powerline sections in southern and western Norway, using UAV laser scanning collected during the summers of 2020 and 2021. Flights at approximately 60 meters above ground produced point clouds with a ground sampling distance of near 1.41 centimeters per pixel — enough resolution to capture each tree’s height, species, crown shape, and exact position relative to the powerline.
From that raw data, researchers derived several tree-level variables. Diameter at breast height was estimated using species-specific equations pairing UAV-derived height measurements with field-based DBH data. Crown depth and stem volume were calculated using Norwegian National Forest Inventory equations.
One variable stands out as genuinely new: crown asymmetry. By measuring crown radii in eight cardinal directions, the team calculated how far a tree’s center of gravity was displaced from its trunk apex. A lopsided crown adds a bending moment that increases wind vulnerability — and including this measure gave the model a more realistic picture of each tree’s mechanical exposure.
ForestGALES: calculating the wind speed a tree cannot survive
ForestGALES is a hybrid mechanistic model, originally developed in the United Kingdom and now used across several European countries. Its core output is the Critical Wind Speed — the threshold at which a given tree will break or uproot. The model accounts for tree size, wood properties, root anchorage, soil type, and the competitive environment created by neighboring trees.
For this study, the model was applied using a version re-parameterized for Norwegian species: Norway spruce, Scots pine, and birch. Researchers tested three seasonal scenarios — summer, fall, and winter — the last including maximum recorded crown snow loads for 2020–2021. Snow loading adds weight to crowns and shifts each tree’s mechanical balance.
Validated against 180 recorded tree falls from that same winter, ForestGALES alone achieved an AUC of 0.67. The authors describe this as adequate but below the conventional threshold of 0.70 for good discrimination. The winter was mild, with no major storm recorded, which made this a harder test than a high-wind season would have provided.
Where machine learning takes over
The research team fed ForestGALES outputs into an XGBoost machine learning algorithm, testing two input configurations. The first combined the Critical Wind Speed with topographic exposure data (Topex) and wind climate information. The second bypassed ForestGALES entirely, relying on raw tree characteristics alone.
Both configurations outperformed ForestGALES in isolation, with the combined model producing the strongest results. Feature importance analysis identified local wind speed and the topographic exposure index as the two most influential predictors. Critical Wind Speed ranked third — reflecting the familiar risk equation: Risk equals Vulnerability multiplied by Exposure.
Even without wind data, the tree-characteristics-only model reached an AUC of 0.763. Drone-derived tree data alone carries substantial predictive signal, which matters for areas where wind climate records are sparse or nonexistent.
What the fallen trees revealed — and what the model missed
The 180 damaged trees showed a consistent pattern. They tended to be smaller in both height and diameter than surrounding trees, with shorter crowns relative to their total height — and they were generally not the tallest trees nearby. That finding complicates the intuition that the biggest tree always poses the biggest risk.
The model couldn’t account for everything. Root rot caused by Heterobasidion fungus — widespread in Norwegian spruce forests — likely weakened some trees in ways no remote scan can detect. A tree with compromised root anchorage may fall at wind speeds well below its theoretical Critical Wind Speed. The study acknowledges that some recorded falls may have been driven by root decay rather than wind alone.
From research tool to precision forest management
The results point toward a practical shift in how utilities could manage vegetation risk. Rather than clearing wide swaths along powerline corridors, managers could target only the trees the model flags as highest risk — reducing the operational footprint while helping utilities comply with local regulations.
The method could also inform new powerline routing decisions, with ForestGALES combined with drone surveys evaluating wind risk across candidate corridors before infrastructure is built. The authors envision an operational pipeline in which routine drone surveys feed directly into AI-assisted vulnerability maps, giving vegetation managers a continuously updated picture of which trees to watch.