Welcome to a new kind of hurricane season: one that no longer waits for utilities to catch up.
Across the power industry, organizations are coming to grips with extreme weather—storms that intensify faster, hit harder, and break historical patterns that storm-response planning has relied on for decades. In this new paradigm, utility leaders need to know not just whether a storm is coming, but know days in advance and down to the neighborhood, exactly what it's going to do to their grid.
So what are the options for utility leaders who want to tap into the granular, game-changing kind of data that can go beyond county-level forecasts and institutional memory? Technosylva is breaking it all down.
Their tech delivers outage and damage prediction at 2-kilometer resolution—sharp enough to inform real decisions about staging, staffing, andEstimated Time of Restoration (ETR)…the kinds of decisions that can either cost or save an org millions of dollars and countless hours of control room strategizing.
To understand how high-resolution predictive modeling is changing storm mobilization—and what it took to build a model that held up even against Hurricane Beryl, the most extreme storm in its training record—Energy Central spoke with Pavel Grechanuk, Technosylva Lead Data Scientist, and Brandon Hertell, Technosylva Product Lead for Extreme Weather Operational Products.
Here’s what we learned. ➡️
EC: Tell us about your roles at Technosylva and the top priorities you’re working on these days. What’s front of mind for you?
Pavel: At Technosylva, I lead the Data Science team. Historically, this encompasses creation of our risk and consequence metrics associated with wildfire, but more recently, our team has developed a range of storm-restoration predictive models in conjunction with CenterPoint Energy.
As Technosylva expands beyond wildfire, we’re leveraging our extensive experience in predicting microclimates that are critical to the impacts of extreme weather. Our team validates these models across a range of different types of extreme weather hazards, from large tropical storms to ice storms to lightning events.
As a science-based company, our focus is on improving the predictive capabilities of our models to support our customers’ decision making in those critical moments where they need to deploy resources and notify customers. Outage prediction is not necessarily new to utilities; they have been managing extreme weather risk to their systems since their inception. We are advancing the science in the industry in terms of spatial resolution, prediction accuracy, coverage across all hazards, and ability to put actionable information in the hands of operators.
Brandon: A large part of my career has been spent in the utility industry. Having worked both at a utility (former ConEdison meteorologist) and from the technology partner side, I bring diverse perspectives and depth in building decision-support software for the industry. At Technosylva, I lead the part of the Product organization that builds our Multi-Hazard Operations platform. Our Product team works with our Science and Engineering teams to build operational decision tools for utilities and agencies.
Front of mind for me with respect to our outage and restoration product line is translating our cutting-edge science to decision metrics that utilities can rely on to improve storm restoration for customers. Our software has to be accessible, intuitive, and help our customers drive an action that moves the needle in their business—decreased restoration times, effective staffing, and lower costs.
We want all our utility users, from front-line to C-suite, to have confidence in both the underlying science under the hood and the metrics we are delivering daily. That way, they can quickly assess and understand where the greatest impacts to their grid are going to be—days in advance of a potential event—and act accordingly. I love to talk about extreme weather preparedness for utilities, which is especially important right now given that we’re officially in hurricane season in many parts of the country.
EC: When a major storm is 72 hours out, what does mobilization planning typically look like today—and where does that process most often break down or fall short?
Brandon: In short, there are a bunch of factors at play that are highly dependent on the nature of the incoming hazard, the utility geography, and processes in place.
Let’s start with the weather situation. If it’s a major tropical system or winter storm, it’s all hands on deck. So, at 72 hours a utility is likely already in pre-planning mode for a multi-day or week restoration timeline.
If it's a more routine type of weather event (impactful, but not devastating), 72 hours is usually when confidence is high enough to start determining what the response will look like. These types of events, more often than not, are handled with internal resources and impacts don’t linger more than 24-48 hours.
Going a level deeper, every utility mobilizes for extreme weather somewhat differently depending on the size of their territory, location, the types of hazards they experience, and topography. Other factors, including their organizational maturity (OR experience) related to extreme weather, are important. The regulatory environment they operate in also plays a part in how utilities prepare and respond.
If I’m a bigger IOU, operating across a large footprint in populous areas of the country, 72 hours is usually an important threshold to make some critical mobilization decisions. If a large, impactful event is anticipated, internal coordination calls have been going on for a few days. Requests for additional resources may already be in place or are about to be made around this time period. If confidence is high, actual preparations to protect equipment are beginning, staging areas are being considered, inventories are being checked for critical supplies. EOCs may be getting activated to start coordinating. It’s about this time that utilities start spending a lot of money.
At a utility with a smaller footprint or customer base, some of these same things are likely happening on a smaller scale. Things like pre-staging crews may not be necessary because the territory is small. However, smaller utilities still need to make sure they have the right number of resources to address what does happen within their territory.
Where things fall short or maybe don’t go according to plan: There are things the utility can control and there are things they cannot. For the things they can control, that’s making sure they have storm response plans and the inputs to that process are clear and standardized. Having the right tools at their disposal—like weather information and modeling of potential outcomes so they can understand the full aspect of the response from customers and outages to damage types and ETR and resource projections—is essential.
At the end of the day, utilities cannot control the weather or the ambiguity and uncertainty of the actual impact. But, they can implement decision tools to ensure that teams are more informed and defensible decisions can be made. These tools help frame or scope the bounds of an effective response.
For a hurricane, assuming the utility territory is in its path, it's always an all-hands-on-deck situation. These events are potentially devastating to utility infrastructure and customers and the stakes are high to get the power back on as quickly as possible. Utilities also have to communicate ETR expectations early and deal with the damage to their own systems. These efforts can also be hampered by things out of their control, like access issues due to flooding, tree damage, or other factors.
Given the size of these types of storms, multiple utilities will be within the potential path of the hurricane and are looking to staff up with as many internal and external resources as possible. This “pool” of mutual assistance from other utilities or specialized contractors is not unlimited. The ability to acquire and then stage the right number of resources days in advance is a multi-million dollar decision.
Utilities start paying for these resources as soon as they request them—even if they are hundreds of miles away from their territory. It is not uncommon for a large utility to spend $1B+ over the course of a large event.
Further, the resources available directly translate into ETR planning and communication. The faster this can be determined with a reasonable amount of confidence and communicated to the public the better. This is why just having an outage prediction model is not good enough. When this type of storm is coming, utilities already know they are going to get hundreds—if not thousands—of outages. The more important piece is connecting that number to the type of damage and then resource decisions that will dictate the ETR they are hoping to reasonably achieve.
Pavel: What’s counterintuitive in this whole process is that, regardless of storm size and severity, the weather forecast doesn’t necessarily directly translate into impacts to the grid. The greatest breakdown that we see is heavy reliance on institutional knowledge of historical storms based on the weather forecast alone, which may not reflect likely grid impacts.
EC: There’s a meaningful difference between knowing a hurricane is coming and knowing what it’s going to do to the grid and your specific service territory. How does high-resolution predictive modeling give utilities the information they need to make better calls?
Brandon: High-resolution predictive modeling is used by utilities to right-size storm response, decrease the costs of storm recovery, better estimate ETRs, and improve communication with customers. The only way to accomplish that is with a tool that can see the weather within the service territory and connect it to the assets and outage data.
It’s almost like looking at the moon from earth with the naked eye versus with a telescope. You know the moon is there and it’s round, but without a telescope you can’t see any of the finer details on the surface. A utility knows a storm is coming and it’s going to be impactful.
Without a set of tools from a predictive model tuned for their territory, they can’t “see” the actual details of the impact. We all know the weather can vary greatly down to a neighborhood level. A high-resolution model puts predictive data points in that neighborhood so it can “connect” the weather to the assets it’s closest to.
Pavel: Where many solutions provide county-level predictions, that more general resolution doesn’t align with the way that utilities mobilize, and also doesn’t provide the resolution necessary for detailed operational decisions such as pre-staging of crew and equipment.
For our utility customers, we provide industry-leading outage forecasts at extremely granular resolution, predicting line outages, customer outages, and likely type of grid damage at 2 kilometer resolution. We help them understand where and when the grid will likely experience impacts, the types of impact, and what they need to do in terms of staffing to hit restoration targets.
EC: How does high-resolution predictive modeling impact important outcomes like restoration times? And how are you seeing utilities factor that into the ongoing conversation about costs and affordability?
Brandon: The frequency and severity of extreme weather events is increasing: 2024 marked the 14th consecutive year on record with more than 10 separate billion-dollar events.
That's translating directly into utility bottom lines, and regulators are paying attention. Utilities are facing greater scrutiny around how they're using analytics to improve resilience, minimize costs, and restore power faster. Storm-restoration costs are often dominated by mutual assistance, contractor expenses, and overtime labor. And utilities need better information to size those correctly and lock in the right contracts.
That's where predictive modeling comes in, and it impacts the entire planning process. The first major decision a utility makes is whether to activate a full emergency response. Opening an Emergency Operations Center can cost upward of $1M—and even a more regionalized response can run into six figures. That decision happens multiple times during storm season.
Knowing when a storm can be managed within normal operations versus when full mobilization is truly necessary is one of the most direct levers utilities have for putting downward pressure on storm recovery costs.
As you move through the event, high-resolution modeling becomes critical to understanding resourcing needs. You can't get to an estimated restoration time without first modeling the type and distribution of damage—and that's downstream from predicted outage counts and locations. We see many utilities trying to achieve ETRs at a neighborhood level. So there’s the “global” ETR—usually when the majority 95–99%+ of an entire utility's customers are back on. Then there can be regional or sub-regional ETRs developed for areas of the utility territory. To do this, those in charge of ETR estimation require more granularity on the outages and potential damage to create that next-level restoration estimate. Getting to that sub-regional granularity requires more detail on what damage is actually occurring and where. Low-resolution models smooth over a lot of that; a high-resolution model surfaces it, providing more accurate predictions for both global and local ETRs.
EC: With climate patterns shifting, extreme weather is the new norm. Stronger storms, wider impact zones, faster intensification. How does your modeling approach handle scenarios that don't have strong historical analogues?
Pavel: On principle, most models are designed with the assumption that the future will look like the past, and unfortunately, we know that is not the case. The frequency and severity of extreme weather is increasing, which means that we have to design our models fundamentally differently.
Ours is built on the premise that the most damaging events are the ones least similar to anything in the record. For us, the question we keep asking is this: Can the model extend what it knows to a storm it has never seen?
The clearest test we have is Hurricane Beryl. Beryl was the strongest event in our entire training record, and because we validate with year-and-season cross-validation, the model was scored on it as a genuinely held-out event—it had never trained on a storm of that severity. It still came within about 12% of the four-day total: roughly 8,300 outages predicted against about 10,133 actual, with day-by-day tracking at an R-squared of 0.83. That's the number I'd point to. With a storm that broke the historical envelope, held out of training, the model still predicted outcomes with usable accuracy.
That approach works because the model learns from the physics of the storm—the actual wind, precipitation, and weather intensity driving the damage—rather than memorizing past events by name or date. So when an unprecedented storm pushes those physical drivers beyond anything in the record, the model extends the learned weather-to-damage relationship into that new territory instead of capping at the worst thing it has seen before. That's the extrapolation that matters as storms keep intensifying past the historical record.
A few things reinforce that. Rather than training a single model on all weather days equally, we train dedicated aspects of the model architecture on just the most extreme days so that rare, violent storms aren't averaged away by thousands of quiet ones. Every forecast also runs through a check that flags when a coming storm has no strong historical match. On those days, instead of projecting false confidence, the system widens its uncertainty range to reflect what it genuinely doesn't know. And we validate against the named storms that genuinely broke the mold—Beryl among them—to confirm the system handles them sensibly, rather than just performing well on average.
None of this makes an unprecedented storm predictable to the decimal. But Beryl shows the approach holds up where it counts—at the edge of, and beyond, what the data has seen.
EC: Pavel, from a data science standpoint: What inputs matter most for storm impact modeling? What data gaps do you most commonly find when you start working with a new utility?
Pavel: On inputs, weather is the obvious driver—wind, gusts, precipitation, the intensity measures that physically do the damage. But the part that's easy to overlook is consistency between what you train on and what you run on in production. We use a high-resolution 2-kilometer reanalysis built in the same configuration as our operational forecast. That matters more than people expect: If you train a model on one representation of the weather and then feed it something different operationally, you've introduced a gap the model was never calibrated for, and your validation numbers stop meaning anything. Matching the two means the relationships the model learns in training are the same ones it sees in the field.
The single most valuable input is high-quality outage data. The outage record is the source of truth you're learning from. Good outage data depends on line crews in the field entering the right information—the cause, the timing, the asset that actually failed—accurately and consistently, often in the middle of a storm response when that's the last thing on their mind. You can have the best weather inputs in the world, but if your target variable is noisy, the model is learning from noise.
The biggest gaps I find with data from a new utility are almost always on that side. Timestamps that don't line up with when the outage actually happened. Cause codes that are blank or defaulted. Outages that aren't tied back to a specific circuit or asset. Restoration times that were never really captured.
None of these are exotic—they're the byproduct of data being treated as paperwork rather than as an asset. Which leads to the thing I'd actually emphasize: this is a cultural problem before it's a technical one. The utilities with the best models are the ones with a data-conscious culture from top to bottom—where the crew entering a record understands it feeds something that matters, and leadership treats data quality as part of the job, not an afterthought.
EC: Brandon, on the product side: What does the workflow look like when a utility operations team is actually using this during a storm event?
Brandon: Our product is intended to be used daily by operations, meteorology teams, and others across the company. In fact, some leaders from our large IOU customers will tell you that our app is the first thing they check each morning. It’s not simply a tool to be saved for the largest events.
Instead, it is a part of the daily workflow for the utility, where each morning teams can see predicted outages and impacts 5 days out. This allows teams to quickly get a sense of potential system impacts and decrease “cycle time” dramatically. Instead of consulting multiple different data sources, compiling them, and providing a recommendation to leadership based on more generalized data, teams quickly have the insights they need, in a fraction of the time.
It’s worth noting that we are not the only data source that a utility consults, nor should we be. Utilities have extensive institutional knowledge, and it is important to triangulate data sources. That being said, we can be a part of a core workflow for a utility in determining: “Does it look like I’m going to have significant outage risk this week?” That way, operators can make the best possible decisions for their teams, resources, and communities and have the data to explain the decisions.
EC: Let’s say we’re looking at a utility that wants to improve its storm-mobilization capabilities but doesn't know where to start. What’s your advice to help them get to operational readiness?
Brandon: First, improve your outage-data tracking. Understanding current performance is critical for enhancing future performance. Further, the higher the quality of the data that goes into a model, the higher quality the output from that model. Spending time in the field to document outages correctly is a critical first step.
Second, develop clear, quantifiable goals for storm restoration that can be tracked. Is there a designated ETR for each storm type and level? What is the projected daily storm cost maximum that your team is trying not to exceed? Aligning around shared organizational goals and then capturing the correct data is critical to understanding current state and demonstrating improvement.
Third, consider tools that help your teams make appropriate storm-preparedness decisions earlier and minimize reliance on institutional knowledge. This includes predictive modeling of outages and grid impacts, but also broader situational awareness through data elements such as lightning detection and weather stations.
About Technosylva
Technosylva is the leading global provider of comprehensive wildfire and extreme weather risk-intelligence solutions for utilities, government agencies, insurance, fire services, and other industries. Founded in 1997, Technosylva’s AI and data--driven platform provides the confidence that helps large and small customers across the U.S., Canada and Europe to protect their communities in the most critical situations. For more information, visit at www.Technosylva.com or follow Technosylva on LinkedIn.