A spatial generalized extreme value framework for traffic conflict using max-stable process approach.

Area of Science:

• Traffic safety engineering
• Spatial statistics
• Extreme value theory

Background:

• Traditional extreme value theory (EVT) models for crash risk assessment overlook spatial dependencies across different zones.
• Proactive traffic safety management requires accurate crash risk estimation that accounts for geographical variations.

Purpose of the Study:

• To develop and validate a spatial EVT modeling framework using max-stable processes (MSP) to analyze traffic conflicts and estimate crash risk.
• To investigate the spatial dependence of extreme traffic conflicts and its impact on crash risk estimation.

Main Methods:

• Utilized vehicle trajectory data from the NGSIM dataset on U.S.101, employing time to collision (TTC) as the conflict indicator.
• Applied three types of MSP models (Schlather, Brown-Resnick, Smith) with various correlation and link functions, estimated using pairwise composite likelihood.
• Employed the extremal coefficient to quantify spatial dependence and estimated crash risk for different zones.

Main Results:

• Significant spatial correlation was found in extreme traffic conflicts, diminishing with distance.
• The Schlather model with a powered exponential correlation function demonstrated the best goodness-of-fit, outperforming other MSP models.
• Inner lanes exhibited lower crash risk than outer lanes, and entrance ramps showed higher crash risk than exit ramps.

Conclusions:

• The proposed spatial EVT model effectively captures spatial dependence in extreme traffic conflicts, improving crash risk estimation.
• The findings highlight the importance of incorporating spatial covariates for accurate traffic safety analysis.
• The model's results align with TTC heatmaps, validating its reliability for identifying high-risk traffic zones.