Fire risk to structures in California's Wildland-Urban Interface

Area of Science:

• Wildfire science
• Environmental risk assessment
• Machine learning applications

Background:

• Wildfires increasingly impact Wildland-Urban Interface (WUI) areas, threatening lives and property.
• Structure vulnerability is influenced by flames and embers, with combined factors poorly understood.
• Accurate risk assessment and mitigation strategies are limited by unquantified combined effects.

Purpose of the Study:

• To analyze structure loss patterns and mitigation effectiveness in major California WUI fires.
• To quantify the combined effects of factors influencing structure survivability.
• To improve risk assessments and mitigation strategies for WUI communities.

Main Methods:

• Machine learning (ML) analysis of post-fire data, remote sensing, and fire reconstruction.
• Examination of five major California WUI fires (Tubbs, Thomas, Camp, Kincade, Glass).
• Utilized an XGBoost classifier to predict structure survivability.

Main Results:

• Structure spacing is critical; arrangement significantly influences fire risk.
• Fire exposure, structure hardening, and defensible space interact to mediate risk.
• ML model achieved 82% accuracy in predicting structure survivability.

Conclusions:

• Structure hardening and defensible space are effective mitigation strategies.
• Hypothetical 52% reduction in structure loss is achievable with these measures.
• Community-level mitigation efforts are crucial for reducing future WUI fire losses.