A wildfire risk score is a number, usually with a category such as Low, Moderate, High, Very High or Extreme, that rates how likely a specific location is to be affected by wildfire. UrbanStat’s score runs from 0 to 100 and comes with a long-term probability, the wildfire history of the location and real-time alerts.
What goes into a wildfire risk score
Traditional wildfire models rely on a handful of variables, typically slope, fuel (vegetation) and access. These describe how a fire spreads once it has started. They say little about where fires start.
A tree or a steep slope cannot start its own fire. Most fires start because of ignition sources such as power infrastructure, and are made more likely and more severe by climate. UrbanStat’s model uses more than 30 variables, including historical climate data, power distribution networks, land use, solar radiation, hydrology, drought indices, lightning strikes and wind characteristics, alongside the usual slope, access and fuel.
In UrbanStat’s models, none of the variables most commonly used by other risk map providers ranked in the top five by importance. The most important features concern temperature, lightning strikes, solar radiation and rainfall.
How to read the score
| Output | What it is |
|---|---|
| Score, 0 to 100 | A single number for the location, with a categorical level. |
| Probability | The long-term probability of wildfire occurring at the location. |
| History | Historical wildfire events at the location. |
| Real-time alert | Short-term risk level based on near-term climatic trends such as rising temperatures and decreasing precipitation. |
A score is long-term by design. The statistical probability of wildfire at an address does not jump because last season was bad; the alert is the part that moves with current conditions.
Why a non-linear model
Rule-based scores add up fixed weights. In practice the same factor behaves differently from place to place: slope has a different effect in different geographies. A model that learns non-linear relationships can capture that, which is one reason generalized national models perform poorly in regions where wildfires have different characteristics, frequency and severity.
How to judge a wildfire model
The honest test is a blind one: publish the map before the season, then compare it with what burned. The usual measure is the share of burned area that the map had already rated High or worse.
That number needs a baseline. If a map rated the whole state High, it would score 100% and be useless. About 35% of California is rated High or worse by UrbanStat’s map, so a result above 85% means the burned areas were concentrated in a minority of the state. Season-by-season results are collected on the wildfire model performance page.
Common questions
What is a good wildfire risk score?
Lower is safer. On UrbanStat’s 0 to 100 scale each score maps to a level from low to Extreme; carriers set their own thresholds for what they accept, refer or decline.
Does a bad wildfire season raise the risk score?
Not by itself. The long-term probability at an address does not change because of what happened last year, with the exception of recently burned areas, which are less likely to burn again soon. Models are updated each year with new data, systematically rather than reactively.
Which areas does UrbanStat’s wildfire map cover?
The proprietary Wildfire Risk Map covers California, Nevada, Arizona, Washington and Oregon. The platform also offers wildfire risk maps from other organizations covering the entire US, Canada and Europe.