Adopting Advanced Technologies for Effective Wildfire-Induced Building Damage Assessment

In recent decades, there has been a significant rise in the frequency and severity of wildfires. Businesses operating in areas prone to wildfires should quantify the probability of experiencing wildfires and the likelihood of asset damage in such events. This information can be crucial for calculating potential monetary losses from property damage, assessing the probability of business disruptions, and creating appropriate backup plans.

To address the urgent need for quantifying wildfire risks, a machine learning model has been developed by PwC. This model utilizes a finely tuned gradient boosting binary classification system to differentiate between damaged and undamaged structures during wildfire events. The indicators of the model include a few key building features, such as exterior siding, roof type, windowpanes, and the surrounding land cover data. By analyzing these features, the model generates wildfire damage probability values, which indicate the likelihood of a structure being impacted by a wildfire event. The model was trained and assessed using a holistic dataset that included post-fire inspection records from historical wildfire events over the past decades. The results were highly promising with overall accuracy of 93% on the test set.

This machine learning model can be an effective solution to help predict the likelihood of future wildfire damages, which enables businesses to plan ahead to address and mitigate wildfire risks.