Key Takeaways:
A University of Houston professor of civil and environmental engineering is using artificial intelligence, specifically large language models, to bring together roadway crash data.
The Texas Department of Transportation funded research assesses pavement structure, surface condition, road geometry and crash records from police reports.
Research results can help transportation agencies select candidates for pavement-safety projects by identifying pavement conditions associated with elevated crash risk.
A University of Houston professor of civil and environmental engineering is using artificial intelligence to make roads safer by connecting information that is usually analyzed separately. In a study funded by the Texas Department of Transportation, Lu Gao used AI to analyze large-scale roadway condition data, including pavement structure, surface condition, roadway geometry and crash records, especially police crash narratives.
By bringing data sources together, the study helps identify roadway segments where pavement or roadway conditions may be linked to higher crash risk.
“A case study using over 24,000 police crash narratives linked to a pavement management dataset of approximately 180,000 data records demonstrates strong associations between friction/texture measures and wet-pavement crash mechanisms,” said Gao, whose research results are published in the journal Accident Analysis and Prevention . “These results can help transportation agencies select candidate pavement-safety projects by identifying pavement conditions associated with elevated crash risk and prioritizing targeted, cost-effective countermeasures.”
The goal of the research is to help transportation agencies better determine which road segments are most in need of maintenance or safety improvements, so limited resources can be directed to places where they may have the greatest impact on improving pavement conditions and reducing crash risk.
Gao used large language model-based crash narrative analysis to identify and quantify pavement-related crash risk. The LLM component converts unstructured police narratives into structured, mechanism-specific labels (e.g., hydroplaning, curve-related loss of control), which enables outcomes that are directly linked to pavement and roadway-surface conditions that are often missing from conventional structured crash fields.
LLMs help decipher crash details which are often embedded in free-text narratives, making large-scale extraction difficult using manual review or simple keyword rules.
“Recent work in traffic safety has explored large language models for crash narrative understanding and structured information extraction, which provides a useful foundation for narrative-based analyses in this study,” said Gao.
The study focused on pavement conditions, like roughness and skid severity, which play critical roles in influencing both crash occurrence and severity. Prior research found that highly rough pavement significantly contributes to increased crash frequency and that skid resistance has a strong negative correlation with crash occurrence, particularly under wet conditions.
“Understanding how pavement conditions relate to crash outcomes can help inform segment screening and treatment selection, especially by identifying high-risk segments where elevated crashes are more strongly associated with pavement-related conditions and may therefore be responsive to pavement-focused treatments,” said Gao.
Integrating pavement condition records with LLM-based crash narrative analysis for pavement safety assessment
1-Sep-2026