Utility Vegetation Management at Scale with Earth Intelligence

Modern computer vision models and new tooling make open and proprietary Earth observation data practical for observing risk remotely, at scale. With Wherobots RasterFlow, teams and AI agents can build risk measurement solutions on demand from these datasets in a couple of hours. The solutions run as an extension of your existing data platform. Pricing is usage-based and predictable, on a platform built for physical world data, scale and interoperability.

Companies traditionally book environmental risk as a cost line they cannot control, and this line keeps growing. According to Swiss Re, insured losses from natural catastrophes reached $107B in 2025, the sixth straight year above $100B. The Los Angeles wildfires alone cost $40B. Wildfire, flood and extreme weather risks compound with geopolitical and social change, and the economic costs follow suit.

Nature carries risk, and so do human activities. But both have elements that can be observed. Teams can monitor flammable vegetation that encroaches on power lines and fuel levels, to predict the damage a wildfire is likely to cause. They can also monitor changes in land development and use. Each of these needs Earth observation (EO) data, images of the ground taken from satellites or aircraft, to observe conditions remotely. It also needs tools that measure physical world conditions over time.

Open and proprietary EO data is growing more available. Wherobots RasterFlow lets teams experiment with this data and computer vision models for imagery quickly, and ship production-ready solutions in days, without building infrastructure first.

Key takeaways

• Wherobots mapped the trees that could reach overhead transmission lines in King County, Washington and Shasta County, California, from open aerial imagery and Meta’s canopy height model.
• Of the trees that could reach the lines, 25.7% stand 10 to 20 m from the line in King County, compared to 5.6% in Shasta County.
• King County’s lines run through far more tall trees within falling distance than Shasta County’s.
• Claude Code ran each county’s analysis in under an hour, at a Wherobots cost of about $40 for King County and about $30 for Shasta County at public pricing.
• The result can be a critical input to a risk model. It does not measure wildfire or wind risk on its own.

Utility vegetation management: which trees can reach a power line?

Utility vegetation management is the work of keeping trees and other plants clear of power lines. A utility with transmission lines through forest needs to know how tall the trees near each line are, to keep service safe and reliable. Trees and other vegetation, if close enough, can come in contact with a conductor, the wire that carries the current, particularly when wind moves them. Those trees are fire fuel and an outage risk, and the risk grows with the number of homes nearby.

From a single prompt, Claude Code exercised Wherobots and measured the trees along the overhead transmission lines of two counties in different states: King County, Washington and Shasta County, California. Wherobots processed open data, and the full analysis completed in under an hour. The method was the same in both counties, so the differences in the results represent differences in the state of the canopy and its distance to the power lines.

1. Start with open imagery and asset data. The power lines came from an Open Infrastructure Map export of 105,147 US transmission lines. We kept each county’s overhead lines and cut each line into segments of about 100 m. The imagery is each county’s latest National Agriculture Imagery Program (NAIP) flight at 60 cm per pixel: 2023 for King County and 2022 for Shasta County. Overture Maps supplied the building footprints.
2. Prepare the imagery. RasterFlow prepared the NAIP imagery for the model. It stitched the NAIP scenes along each county’s lines into one mosaic, a single image with no seams between scenes. Newer, higher-resolution imagery with more coverage can replace NAIP. We chose NAIP because it is free and its resolution is high enough for Meta’s canopy height model.
3. Run the model. RasterFlow ran Meta’s Canopy Height Model (CHM v1) across each mosaic on GPUs. The output is a canopy height raster: a grid with the height of the tree canopy for every 0.6 m pixel. The maps below show it.
4. Score each line segment. WherobotsDB drew four strips of ground, called zones, on each side of every line segment: 0 to 10, 10 to 20, 20 to 30 and 30 to 45 m from the line. For each zone, RS_ZonalStatsAll measured how tall the trees in it are on the canopy height raster. This is a raster-to-vector join: the zones are vector shapes, and the heights come from the raster. A zone is flagged when its trees are tall enough to fall onto the line. Each segment then got a risk score that rises with its number of flagged zones and the number of buildings within 100 m. The segments were ranked by that score.
5. Inspect the results. A Wherobots GL map layers the NAIP mosaic, the canopy height raster, the flagged zones, and each segment colored by its risk score. The maps for King County and Shasta County are below.

Tree risk to power lines: King County vs. Shasta County

King County’s lines run through far more tall canopy within falling distance of the power lines. Of the trees that could reach the lines in King County, 25.7% stand 10 to 20 m from the line, compared to 5.6% in Shasta County.

Wildfire and wind risk

This comparison measures where trees could reach a line today. It does not measure the risk of a wildfire or a windstorm. But it can be used as a critical input to a risk model. You could take the next step by layering in live wildfire risk with RasterFlow, by measuring NDVI, a measure of how green and dense vegetation is, ground surface temperature, and other elements. Or you could easily incorporate a prevailing wind filter to hone in on the power line segments that are at risk of tree-fall from high, prevailing winds.

Wildfire and severe wind may become more frequent west of the Cascades as the climate changes. If they do, King County’s lines start from a position with roughly four times Shasta County’s share of reachable trees. Many of those trees stand near homes. From this foundation, you can identify the highest risk and highest impact opportunities to tackle next. You can also put it into operation with fresh, high-resolution imagery with the coverage you need.

Earth Intelligence for utilities in days

Claude Code ran each county’s analysis in under an hour. At public pricing, the Wherobots cost was about $40 for King County and about $30 for Shasta County. It also shows how companies can use AI with Wherobots to build Earth Intelligence solutions like this one quickly, without building and running their own data and AI infrastructure. Their existing AI agents drive Wherobots directly. That cuts the time to a working Earth Intelligence solution, and the results go into the data platforms you already use.

How to get started

Want to talk about infrastructure risk management?
Reach out to schedule a conversation with the experts.

Get started at $0 with the Professional Edition
The plan includes every capability builders need to ship solutions from spatial data. Start at $0 with a 14-day free trial that covers up to $95 of usage. In that time you can connect a catalog, run the canopy height model over your own lines, and write the first Iceberg table back to Databricks, AWS Glue or another S3-based lakehouse.