Quantifying the counteracting impacts of climate change on large-scale pavement infrastructure serviceability

This study quantifies competing climate-change effects, temperature rise versus reduced freezing, on pavement networks using over 35 years of records from more than 1,100 sections. We combine explainable machine learning (ML) with Monte Carlo simulation to propagate global climate model (GCM) projections to future infrastructure impacts, considering the two-layer uncertainty from climate ensemble and ML residuals. Results reveal substantial inter-GCM model differences and occasional opposing trends, underscoring climate projection uncertainty. Trained ML models accurately predict long-term pavement performance; the freezing index and air temperature are the two dominant drivers. Reduced future freezing tends to extend service life, partially offsetting warming's negative effects. Thus, climate change does not always accelerate pavement deterioration: in some regions (notably wet, freeze-prone zones) and for some time horizons or scenarios, net effects can be neutral or beneficial. In wet, freeze zones, pavement service life is being extended in nearly 65 % simulations under SSP585 by 2050–2060, whereas dry, freeze counterparts only show a figure of around 35 %. These findings indicate that pavement resilience assessments should consider both warming and changing freeze–thaw regimes rather than temperature alone under climate uncertainty and inform local adaptation decisions practically.