Data Driven Model Development to Predict Soil Temperature and Freeze-Thaw Cycles in Pavement Foundation Systems

Abstract Road infrastructure is central to economic development of the United States. However, particularly in cold climates, the soil temperature and annual freeze-thaw cycles significantly impact road conditions, therefore it is important to accurately predict the number of freeze-thaw cycles and their duration. While prior data-driven studies have predicted soil temperature at daily or monthly timesteps, none have combined sub-daily temperature prediction with direct estimation of freeze-thaw cycle count and duration in a form deployable by practicing engineers. This study addresses that gap using two years of 15-minute soil temperature and climate data collected at six locations along the MNROAD pavement test facility in Minnesota. Fourth-order nonlinear regression model was developed using two time variables and six climate variables selected through stepwise regression, with two post-processing filters applied to ensure prediction validity. Results show that both models can accurately predict the soil temperature at varying depths, with improved performance at greater depths, along with the count and duration of freeze-thaw cycles. The performance of Model 1 with filters was superior to Model 2. This framework gives engineers a practical way to predict freeze-thaw timing without detailed soil property data or modelling expertise and plan maintenance schedules and spring load restrictions accordingly.