Bayesian Estimation of Cost-Effective IRI Intervention Thresholds for Pavement Repair Actions Under Latent Repair Conditions

Determining appropriate intervention thresholds for pavement surface repairs is a fundamental challenge in pavement management. In practice, these thresholds are often defined using inspection conditions that are not observed at the time of repair, rather than the actual pavement condition at that time. This distinction is important because repair activities commonly occur between inspection dates, meaning that pavement condition indicators such as the International Roughness Index (IRI) immediately before and after repair are not directly observed. Consequently, repair decisions are typically based on estimated conditions or engineering judgment. This study develops a Bayesian latent-condition framework to estimate pavement conditions at the time of repair and determine cost-effective IRI intervention thresholds for pavement surface repairs. Using historical pavement inspection and maintenance records, a no-repair deterioration model is first estimated to infer the latent pavement condition immediately before and after repair. Repair effects are then modeled using bounded restoration fractions, allowing action-specific improvements to vary across latent pre-repair IRI intervals while preserving physically realistic post-repair conditions. The proposed framework evaluates repair intervention thresholds using multiple criteria, including absolute IRI reduction, relative restoration, the probability of achieving good pavement condition, and cost-effectiveness. The results indicate that although the largest absolute IRI reductions occur under severe deterioration, the most practical serviceability-based intervention range for patching, partial overlay, and full overlay is approximately 4–5 m/km. When repair costs are incorporated, earlier intervention at around 3–4 m/km yields greater benefits per unit cost. Overall, the framework provides an uncertainty-aware basis for defining repair intervention thresholds using imperfect inspection-based maintenance records.