ORCID

0009-0008-1414-4386

Date of Award

Summer 2026

Language

English

Embargo Period

7-14-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

College/School/Department

Department of Atmospheric and Environmental Sciences

Program

Atmospheric Science

First Advisor

Kara Sulia

Committee Members

Kara Sulia, Christopher Thorncroft, Nick Bassill, Zheng Wu, Christopher Wirz

Keywords

Traffic camera imagery, Winter road maintenance, Artificial Intelligence, Social Science, New York State Department of Transportation

Subject Categories

Atmospheric Sciences

Abstract

Transportation agencies make critical operational decisions during hazardous weather events, including assessment of road conditions and resource allocation. In this study, machine learning models are developed to provide additional support for the New York State Department of Transportation (NYSDOT) by automatically classifying road conditions across the state. Convolutional neural networks and random forests are trained on NYSDOT roadside camera images and weather data to predict road surface conditions. This task draws critically on a labeled dataset of 22,000 camera images containing six road surface conditions (severe snow, snow, wet, dry, poor visibility, and obstructed) generated from an iterative hand-labeling process known as Quantitative Content Analysis. The robust labeling approach supports data consistency, which is important for both model development as well as for data reproducibility and replicability. Model generalizability is prioritized to meet the operational needs of the NYSDOT decision makers, including integration of operational datasets and use of representative and realistic images. By intentionally including imperfect, operational data over simple, idealistic images, the model is tasked with a complex classification problem to align with real-world reliability. The weather-related road surface condition model in this study achieves an accuracy of 81.5% on completely unseen cameras. The model is further analyzed on a large set of unlabeled data to mimic an operational environment, and verification against weather datasets and case studies provides evidence that the model is skillful in real-world conditions.

Recognizing that operational readiness extends beyond technical model development and verification, this study prioritizes end-user engagement and deeply integrates methods from social science. Informal collaboration with the NYSDOT throughout all stages of this work enhances the model's practical functionality and operational readiness, including additional features such as model forecasting and the inclusion of confidence metrics, as well as the development of an interactive dashboard that displays model output. Additionally, formal social science methods are employed through semi-structured interviews with 15 NYSDOT officials to understand end-user decision making, perceptions and trust of AI/ML, and perspectives on integrating the ML tool into their workflow. Qualitative analysis of the interview data reveals that officials foresee varied and individualized operational uses for the ML tool, but overall, they share a common sentiment of the importance of human oversight and hands-on, iterative use to build trust in it. While integrating these human elements adds complexity to the model development process, it is prioritized to ensure the resulting tool is operationally applicable and not solely a theoretical proof-of-concept. Overall, the convergence of atmospheric science, machine learning, social science, and transportation results in a co-developed ML tool that detects hazardous weather-related road surface conditions for operational application.

License

This work is licensed under the University at Albany Standard Author Agreement.

Download

Share

COinS