Predicting pavement cracking performance using laser scanning and geocomplexity-enhanced machine learning

Transport infrastructure is vulnerable to crack formation and deterioration due to aging and repetitive loading. Accurate and timely crack assessment and prediction are crucial for effective road maintenance, but existing studies often rely on individual indicators such as crack types, attributes, and severity, which fail to capture the full complexity of crack deterioration. Furthermore, limited research has explored the long-term impacts of traffic, socioeconomic, and climate changes on crack progression, and existing machine learning (ML) models struggle to explain the contributions of individual predictors due to inherent complexities in such spatial data. This study develops a geocomplexity-enhanced ML (GML) approach to evaluate crack deterioration and predict cracks under various future scenarios in the Wheatbelt region of Australia. The study employs laser-scanning data to generate a novel cracking performance index (CPI) and integrates geocomplexity (GC) measures with random forest models to capture local spatial complexities. Results demonstrate that GML significantly outperforms standard ML models in predicting CPI-based crack deterioration. Crack predictions in future scenarios reveal that in the Wheatbelt region, changes in climate factors over time have a more substantial impact on crack progression than traffic and socioeconomic changes, and without effective maintenance, crack propagation rate will significantly increase. It provides empirical evidence for developing preventive maintenance strategies. The developed methods and findings can support the development of adaptive, climate-resilient infrastructure, and long-term road management strategies, enhancing the sustainability of transport infrastructure.