Asphalt Pavement Temperature Prediction Based on Thermo-Physical and Machine Learning Modeling: Insights from the SMARTEP Facility

Accurate temperature prediction in asphalt mixtures is essential for evaluating pavement performance and durability. This paper presents a comparison between two traditional thermo-physical models (Barber and Park) and a CatBoost-based machine learning model trained on an annual database collected at the SMARTEP (Sustainable Model and Renewable Thinking Energy Parking) experimental facility (Palermo, Italy). Literature and developed models were evaluated both on an hourly and daily basis. The results showed that, based on a few weather variables that are commonly available (air temperature, solar radiation, relative humidity, and wind speed), the machine learning approach achieved superior performance in terms of accuracy with respect to thermo-physical ones on both timescales. The CatBoost model achieved a reduction of more than 60% in terms of mean absolute error with respect to thermo-physical models, while keeping good interpretability. The analyses highlight how data-driven methods can effectively complement traditional approaches to achieve smarter road infrastructure management.