Transferability of pavement performance models: Evaluation of direct and updated Markov model transfer

• Methodology for assessing transferability of pavement performance models • Application to real pavement condition data from Cambodia • Transferability is lower at deterioration initiation and higher during propagation • Covariate similarity improves transferability, while region closeness does not • Bayesian updating enhances model accuracy with limited recipient data Ensuring the reliability and longevity of infrastructure assets requires effective asset management supported by robust performance modeling, which constitutes a core process of asset management systems. However, reliable statistical estimation of such models generally demands large datasets, which are not always available in regions with limited resources or small asset inventories. Model transfer provides a potential solution by applying parameters estimated from data-rich donor regions to data-scarce recipient regions. This study proposes a methodology to evaluate the transferability of pavement performance models formulated as Markov hazard models, and applies it to national roads in Cambodia. In the analysis, two grouping strategies are examined: region-based and region-covariate-based, the latter incorporating traffic loading, structural number, and rainfall. Direct transferability and transfer effectiveness with Bayesian updating are analyzed using the 1-Wasserstein distance and the Wilcoxon signed-rank test. The analysis shows that, under region-based grouping, transferability is limited at the initiation stage of deterioration but higher in the propagation stage (89% of cases (32/36) within one-year life-expectancy error), and geographical closeness has no clear influence. Under region-covariate-based grouping, higher transferability is observed across regions with similar covariates, especially structural number. In such cases, Bayesian updating enables more accurate approximation of ground truth when recipient data are scarce. Even where direct transferability is low, updated donor models can yield closer approximations under very limited recipient data. These findings confirm the usefulness of the proposed methodology and its applicability to real-world pavement data, providing practical insights into data sharing and inspection planning in infrastructure asset management.