Roadway safety during active wildfires: A machine-learning analysis of crash occurrence and injury severity
Year: 2026
Type: article
Abstract: Wildfires create rapidly changing conditions for roadway operations and safety, yet how wildfire-related conditions relate to crash occurrence and injury severity remains insufficiently understood. Using data from 10 California wildfires between 2020 and 2025, this study examines crash occurrence at the roadway segment-day level and injury severity at the crash level through two separate machine-learning analyses. Across the two analyses, predictors included fire advancement toward roadway segments, fire-risk context, smoke exposure, historical Public Safety Power Shutoff exposure, and distance to the active fire perimeter, alongside conventional roadway, land-use, weather, traffic-exposure, and temporal characteristics. The severity analysis additionally incorporated driver, vehicle, and collision attributes. Five tree-based classifiers were evaluated using cross-validation, and predictor contributions were examined using complementary feature-importance measures and SHAP. Fire advancement toward roadway segments emerged as the strongest crash-occurrence predictor across all importance measures, while fire-risk context and smoke exposure also contributed to model predictions. In contrast, injury severity was characterized primarily by collision, vehicle-safety, driver, roadway, and built-environment factors, with collision type and airbag deployment among the strongest contributors. Most wildfire-related variables ranked below the leading crash-specific predictors in the severity model. Together, these findings highlight distinct predictor structures for crash occurrence and conditional injury severity, supporting their separate assessment when examining roadway safety during active wildfires.
Source: Accident Analysis & Prevention
Authors: Fateme Gorzin, Omid Nayeri, Ramin Shabanpour
Institution: University of Pittsburgh
Language: English
Topic: Traffic and Road Safety
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,150
OpenAlex ID: W7233574336
