Abstract Manager

Pavements in cold regions are subjected to pronounced seasonal temperature variations that strongly influence freeze–thaw behaviour, structural response, and long-term performance. Accurate prediction of subsurface temperature within the granular base course (GBC) and subgrade is therefore critical for frost assessment and climate-resilient pavement management. Although machine learning techniques have demonstrated strong predictive capability, many existing approaches lack transparency and do not yield explicit mathematical relationships that can be readily adopted in engineering practice.

This study presents a symbolic regression–based framework using Gene Expression Programming (GEP) and Multiple Expression Programming (MEP) to develop closed-form empirical equations for predicting subsurface pavement temperatures in Edmonton, Alberta. A full-scale instrumented pavement test section was constructed at Edmonton Waste Management Centre in 2022; these measurements are used to develop the proposed models. The temperature sensors were installed at multiple depths within the pavement structure, providing high-resolution subsurface temperature data spanning multiple freeze–thaw cycles. Key climatic and temporal variables, including air temperature, solar radiation, and day of the year, are used as model inputs to derive interpretable temperature prediction equations for the granular base course (GBC) and subgrade layers.

The performance of the GEP- and MEP-derived models was evaluated using multiple statistical metrics and demonstrated strong agreement with measured field temperatures. The resulting formulations effectively captured nonlinear interactions between environmental drivers and depth-dependent thermal responses while maintaining a compact, physically interpretable mathematical structure. A comparative assessment indicated that both GEP and MEP provide robust alternatives to black-box machine learning models, with MEP offering enhanced equation simplicity and computational efficiency.

The proposed GEP and MEP framework enables the development of transparent, data-driven empirical temperature prediction equations directly applicable to pavement engineering analyses. These findings support improved estimation of frost and thaw depths and contribute to more informed, climate-responsive pavement management in cold regions.