Factors influencing crash severity under adverse weather conditions: Insights from rain, snow, and fog

To identify key factors that exacerbate crash severity under adverse weather conditions such as rain, snow, and fog, and support effective crash prevention, this study develops a hybrid analytical framework combining the Gradient Boosting Decision Tree (GBDT) model, the interpretable Shapley Additive Explanations (SHAP) framework, and a Random Parameters Logit Model with Heterogeneity in Means (RPL-HM) using crash data from Tianjin, China (2018–2023). The results indicate that fog significantly affects the likelihood of injury crashes, with substantial heterogeneity in means. A visibility–lighting composite indicator is proposed, and the findings show that insufficient lighting exacerbates the likelihood of severe accidents under low visibility. Road classification and crash participants also have a significant impact on crash severity. Factors such as highways, crashes involving both motor vehicles and pedestrians, speeding, and failure to yield are all significantly associated with a higher occurrence of severe crashes. The hybrid modeling framework and findings provided in this paper offer a theoretical foundation and practical guidance for formulating targeted road traffic safety management policies under adverse weather conditions, contributing to a reduction in both the frequency and severity of crashes in such environments.