Scalable Geospatial Machine Learning for Power-Line Asset Risk: Integrating Remote Sensing for Lightning and Vegetation Risk Modelling

Power lines are the invisible veins of modern life, carrying electricity from distant sources to our homes and businesses. Yet, these networks are constantly under siege by the natural world. Storms, shifting landscapes, and the very plants growing nearby can cause wires to snap or towers to fail, leading to blackouts that disrupt daily life and cost billions. For decades, utility companies have managed these risks by looking at the system as a whole, treating a thousand poles in a region as a single unit. But nature does not treat every pole the same way. A tree might be dangerous to one wire but harmless to another just a few hundred meters away, and a lightning strike might target a specific hilltop while sparing the valley below. To keep the lights on, engineers need to understand risk at the level of the individual asset, knowing exactly which pole is most likely to fail and why.

This is the challenge tackled by a team of researchers from SA Power Networks in Adelaide, Australia. They set out to build a new kind of digital map that could predict the likelihood of failure for every single power pole in their network, specifically focusing on two major threats: lightning and vegetation. Instead of relying on simple rules of thumb, they created a system that learns from the past. By feeding a computer a massive amount of historical data—combining records of when and where poles actually failed with detailed maps of the surrounding terrain, the density of nearby trees, and the history of lightning strikes in the area—they trained a machine-learning model to recognize the subtle patterns that precede a failure. The result is a framework that does not just guess; it calculates a specific probability of failure for each asset, turning vague concerns about "storm risk" into a precise, actionable list of which poles need attention first.

The researchers began by gathering a diverse collection of information about the environment around each power pole. They used satellite data to measure the height and roughness of the ground, the distance to the nearest coastline or body of water, and the proximity of buildings. They also looked at the health of the vegetation, using satellite images to track how green and dense the plants were in the immediate vicinity of the wires. Crucially, they paired this environmental data with the utility company's own records of past outages, linking every failure to the specific conditions present at that location. This allowed them to build a picture of what a "dangerous" environment looks like for a power pole. For instance, they found that poles located near water or in areas with dense, fast-growing vegetation were more likely to experience failures caused by trees falling or branches touching the lines. Similarly, poles situated on higher ground or in flatter, more open terrain were identified as being more vulnerable to lightning strikes.

To make sense of this complex web of data, the team used a sophisticated computer algorithm capable of finding non-linear connections between the environment and the failures. They did not try to force all the risks into a single, messy equation. Instead, they built two separate but connected models: one dedicated to predicting vegetation-related failures and another for lightning-related failures. This approach allowed them to isolate the specific factors that drive each type of risk. The system was designed to be efficient enough to handle millions of assets, a necessity for any utility company managing a vast grid. By processing the data in a cloud-based environment, they could quickly calculate risk scores for the entire network without getting bogged down by the sheer volume of information. The model was tested rigorously, splitting the historical data into different groups to ensure it could accurately identify failure patterns it had never seen before, confirming that the patterns it found were real and not just random noise.

The findings revealed a clear and logical picture of risk across the network. For vegetation, the most significant factors were the type of power line, how close it was to water, and whether trees were already hanging over the wires. Poles in urban and city-center areas showed a higher risk of vegetation failure, likely because strict regulations in these zones sometimes prevent the aggressive trimming of trees that might otherwise be necessary. For lightning, the model showed that poles on higher ground and in rural areas were at greater risk, while the presence of nearby buildings seemed to offer some protection, perhaps by altering how electrical charges move through the air. The system also confirmed that the type of terrain mattered; poles in rugged, hilly areas were actually less likely to be struck by lightning than those in flat, open landscapes. These insights are not just theoretical; they provide utility managers with a ranked list of assets, allowing them to prioritize inspections, trim vegetation, or reinforce poles where the risk is highest, rather than treating every part of the grid equally.

One of the most important aspects of this work is its flexibility. The system is built like a set of modular blocks, meaning that if new types of data become available—such as information about animal-related faults or new satellite sensors—the researchers can add them to the model without having to rebuild the entire system from scratch. This scalability is vital as the climate changes and new hazards emerge. The researchers acknowledge that their current model relies on historical data and does not yet predict the future with perfect certainty, nor does it account for every possible variable. However, the framework they have created offers a powerful new way to think about infrastructure resilience. By moving from a broad, average view of risk to a detailed, asset-by-asset understanding, utility companies can make smarter decisions today that will keep the power flowing tomorrow, even as the weather becomes more unpredictable. The work demonstrates that with the right data and the right tools, we can see the hidden vulnerabilities in our power grid and fix them before they cause a blackout.