Automated Weather Classification for Decision Support using Roadside Imagery

Transportation departments (DOT's) require accurate and reliable observations of the current weather and road conditions over their networks to determine the need for road treatment to increase safety and mobility. Road weather information system (RWIS) sensors provide a great deal of point observation data that indicate the current weather and road conditions, but having visual confirmation of current conditions is still a primary tool used in maintenance planning and decisions to reduce variable speed limits, close roads, or require chain-up. The camera systems installed on stationary and mobile RWIS provide eyes on the ground but require human monitoring which can be challenging during changing weather conditions. The rapid growth of mobile RWIS will increase the effort required to monitor camera imagery for making localized treatment and maintenance decisions. Advances in machine learning (ML) and computing technologies have made automated weather and road condition classification possible using RWIS imagery. The NSF National Center for Atmospheric Research has integrated road weather and condition classification models into the Pikalert maintenance decision support system (MDSS). This presentation will discuss some of the modeling techniques used, performance obtained, lessons learned, and the use of the ML models in the MDSS.