Accurate and timely forecasts of road surface conditions are crucial for efficient winter maintenance, enhanced traffic safety, and the optimized use of de-icing agents. Road surface phenomena, in complex fields present challenges to traditional forecasting methods due to their nonlinear and localized nature. This study presents a machine learning framework predicting real-time road states (dry, wet, icy, snowy) across Bavaria, Germany. It integrates data from over 516 Road Weather Stations (RWS), thermal measurements from winter maintenance vehicles, and elevation data from the Open Elevation API. Data undergoes temporal alignment, spatial interpolation, and missing-value imputation. Decision Trees form the core model for interpretability and nonlinear pattern handling. Each RWS employs a localized model, while a generalized version covers unmonitored roads via spatial adjustments. With over 85% accuracy, the system facilitates dynamic winter maintenance and minimizes resource waste. Cyber-physical in smart mobility and transportation networks support improved real-time hazard responses. This approach shows how scalable infrastructure can be made resilient using machine learning.
Saldana et al. (Thu,) studied this question.