This article describes how Technosylva chooses representative fire-weather scenarios for wildfire risk modeling.
Overview
Wildfire risk modeling estimates how fires would behave across the full range of weather a service territory experiences. Simulating every day in a multi-decade weather record, for every location, would be computationally impractical. Instead, Technosylva selects a representative set of weather days for each area, chosen to capture the conditions that matter most for risk while keeping simulation tractable.
The selected set is designed so that:
-
The most severe fire-weather days are always included.
-
A broad range of fire-conducive weather patterns is represented, not only the single most extreme type. This keeps the results from being dominated by one kind of event, such as high-wind days, while overlooking others, such as prolonged dry-heat conditions.
-
Days on which significant wildfires actually occurred are represented, grounding the selection in real fire experience.
-
Each pattern's real-world frequency is preserved, so downstream risk reflects how often each type of weather actually happens rather than treating every severe day as equally likely.
The result is a compact but comprehensive set of weather scenarios that spans everything from rare, catastrophic events to more common moderate conditions. Each simulated fire draws its weather from one of these days.
Fire Weather Zones
Weather can vary sharply over short distances. A hot, dry, windy day in one part of a territory may coincide with cool, damp conditions a hundred miles away. To account for this, the selection operates at the level of fire weather zones, geographic areas defined by the National Weather Service in which fire-weather conditions are expected to be broadly similar.
Each zone receives its own independently selected set of weather days. A day that counts as extreme in a coastal zone is judged against that zone's own climate, not against an inland desert zone. This ensures that local conditions drive local risk estimates.
Zone boundaries are sourced from National Oceanic and Atmospheric Administration (NOAA) data and intersected with the utility's asset footprint, so every asset is assigned to the appropriate zone.
Data Sources
The selection draws on more than 20 years of record across three categories of data:
-
Weather conditions: Gridded weather observations and reanalysis products supply daily wind speed/wind gusts and a composite Hot-Dry-Windy (HDW) index that combines temperature, humidity, and wind into a single measure of atmospheric fire potential.
-
Fuel moisture: Modeled moisture for dead and live vegetation indicates how readily fuels will ignite and carry fire. Days when fuels are too damp to support significant fire activity are set aside. In regions where live fuel moisture data are limited, a drought-based measure is used in its place. See the Technosylva Fuels Glossary for more information.
-
Historical wildfire records: Federal wildfire databases, including Wildland Fire Interagency Geospatial Services(WFIGS) and the USDA Forest Service Karen C. Short fire-occurrence database, identify the days when large fires actually burned in or near each zone.
Day Selection
Each day in the historical record is scored for overall fire-weather severity. The score is built so that wind, atmospheric dryness, and fuel dryness must all be elevated for a day to rank highly. A single extreme factor, such as high wind on a damp day, does not by itself produce a severe rating.
Days are then grouped by the type of fire-weather pattern they represent, and a representative set is assembled for each zone with several guarantees in place:
-
The most severe days in each zone are always included.
-
Days tied to significant historical fires are included, even when their weather metrics alone might not have ranked them at the top.
-
Every distinct weather pattern is represented, so no type of fire-conducive condition is left out of the risk calculation.
-
Each zone receives the same number of selected days, so simulation effort is balanced and zone-to-zone comparisons are not skewed by differing sample sizes.
Where a customer needs specific historical dates included, those can be added to the selection on request.
The final output is a uniform set of several hundred weather days for each fire weather zone. This scale is large enough to capture rare but consequential events while remaining practical to simulate across many thousands of ignition points.
From Simulation to Annual Risk
Simulation produces fire outcomes for each selected weather day, but a raw average across those days would overweight severe conditions, because severe days are deliberately oversampled in the selection.
To correct for this, each weather pattern's simulated results are weighted by how often that pattern actually occurs in the historical record. The outcome is a true expected annual risk that reflects real-world frequency, not an average over a set of mostly severe days. Portions of the year when conditions cannot support significant fire, such as wet or off-season periods, are accounted for as effectively zero-risk.
Quality and Validation
Several checks support the reliability of the selection:
-
Reproducibility: The process is deterministic. The same inputs produce the same selection, which supports independent verification.
-
Validation against fire history: Because historically significant fire days are included, the selection is cross-checked against real events. Well-calibrated weather metrics will already surface most major fire days before that step is applied.
-
Full-range coverage: Each zone's selection is reviewed to confirm it spans the full range of relevant weather conditions, with no significant type of condition overlooked.
-
Balanced comparison: Every zone carries an equal simulation budget, so results can be compared fairly across a service territory.