Effects of flooding on pavement performance: a machine learning-based network-level assessment

Catastrophic flood and hurricane events cause enormous adversarial impacts on roadways and their long-term performance. Despite the abundance of research attempting to understand the flood-induced damages, a critical knowledge gap in assessing roadway flood damage at a network-level exists. This study develops a holistic assessment model that evaluates network-level flood damage of roadways based on historic pavement distress data along with historic flood data. A rich volume of high-confidence historical pavement distress data was obtained from the Louisiana pavement management system. After a rigorous data pre-processing process by cross-referencing the flooded areas using the 2016 Louisiana flood map data, it was leveraged to analyze how flooding could interact with the pavement distress, thus affecting the overall performance of existing pavements. The study outcomes showed that the most flood-affected distress types include roughness and random cracking. Based on the findings from the analysis, this study developed a machine learning-based prediction method that can calculate future pavement performance after a flood event. After applying different algorithms for creating the prediction model, the eXtreme Gradient Boosting (XGB) classifier was selected because it represented the highest accuracy among other examined classifiers. Various datasets and scenarios were investigated with the developed prediction model to identify the most effective features and dataset combinations. The prediction model is expected to identify vulnerable pavement sections and facilitate network-level preventive maintenance of pavement to mitigate future flooding impacts by prioritizing resource allocations for maintenance and rehabilitation after a flood event.