Lessons Learned in Developing an Automated Road Surface Classification Tool using Machine Learning for the New York State Department of Transportation

Timely identification of current road surface conditions and prediction of future conditions can increase traveler safety and promote operational efficiency of transportation agencies. In this study, various data sources are leveraged to enhance the understanding of road surface conditions using machine learning models. Cameras capture images of the roadways, providing an image dataset that depicts the true road surface conditions. Additionally, High-Resolution Rapid Refresh (HRRR) forecasts provide valuable datasets of weather conditions that can impact road surface conditions. These two datasets - image data and HRRR data - have been trained to predict the road weather surface conditions on New York State roadways. This presentation will focus on lessons learned from the ongoing development of a tool that automatically predicts road surface conditions in collaboration with the New York State Department of Transportation (NYSDOT). Insights gained from a machine learning perspective will be shared, including a discussion on the skill of different algorithms and datasets. Co-development with the NYSDOT stakeholders continues to be a critical step in designing a tool for operational use, and takeaways from the collaboration will be discussed, such as different applications of real-time vs. forecasted road surface conditions. The presentation will end with a discussion about future work with the NYSDOT and the potential for applications to other regions.