Multi-layer moisture prediction in unbound pavements under climatic extremes: a field-validated framework

Moisture dynamics in unbound granular pavements surfaced with thin asphalt seals critically influence long-term performance, particularly in high-rainfall regions such as Queensland, Australia. Excess moisture accelerates permanent deformation and structural deterioration, requiring accurate prediction of transient volumetric moisture content (VMC) variations for effective maintenance. While numerical and machine learning (ML) models have advanced, their adoption is limited by computational demands, calibration challenges and technical expertise. Existing empirical models often fail to capture dynamic, site-specific climatic effects. This study presents a novel, real-time predictive framework integrating key hydrological processes such as rainfall, drainage, capillary rise and seasonal oscillations, along with autoregressive memory effects. The model is calibrated using Dual Annealing (DA) separately for the base, subbase and subgrade layers. Validation using 12 months of field data from Marsden, Queensland, achieved mean absolute percentage error (MAPE) values of 5.06% (base), 4.42% (subbase) and 1.84% (subgrade). Sobol sensitivity analysis highlighted the dominance of depth-dependent parameters over material-specific ones. External validation using LTPP data from Miami, Florida, demonstrated the model's transferability effectively. Benchmarking against existing models supported the development of model selection criteria for various boundary and climatic conditions. This adaptive framework offers a reliable tool for managing pavement moisture under diverse environmental scenarios.