Predictive Temperature Modeling in Granular Base and Subgrade Layers Using Machine Learning: A Smart Road Study in Edmonton, Alberta, Canada

Temperatures in the pavement layers are critical for preserving pavement structural integrity especially during freeze and thaw cycles. Better prediction of these temperatures, based on factors such as air temperature, incoming solar radiation, and day of the year, is essential for effective pavement management. In recent years, machine learning (ML) has emerged as a powerful tool for predictive modeling, particularly for addressing nonlinear and complex problems in pavement analysis. This study applies three ML models, namely, decision tree, random forest, and adaptive boosting, to predict temperatures in the granular base course and subgrade layers of pavement. Temperature data were collected from embedded sensors, while air temperature and incoming solar radiation were obtained from meteorological sources in Edmonton, Alberta, Canada, between June 2023 and May 2024. These measurements were used to train, test, and validate the models. Models’ performance was assessed using statistical metrics, with results applied to forecast frost penetration depths and the onset of thawing. The models outperformed similar studies conducted in Edmonton, achieving coefficient of determination (R2) values ranging from 0.99 to 1.0. Additionally, freeze–thaw cycles showed a 1-day buffer between predicted and actual dates, demonstrating high accuracy. This research underscores the effectiveness of ML models for predicting subbase and subgrade temperatures, offering valuable insights for pavement load management and maintenance strategies for the local and regional agencies.