Application of Optical Satellite Data in Pavement Management: Assessing Its Value in Partially Observable Stochastic Environments

Abstract

This study evaluated integrating multispectral satellite imagery and machine learning algorithms in monitoring and managing pavement conditions. The research used XGBoost, an advanced machine learning technique, to classify pavement conditions into three categories—good, fair, and poor—based on International Roughness Index (IRI) data using multispectral images from 2018 to 2023. A partially observable Markov decision process (POMDP) framework was applied to optimize maintenance and monitoring decisions, considering the inherent uncertainties in pavement condition assessments. The study area included of four Interstate Highways for which satellite imagery was analyzed. The findings showed that the XGBoost model achieved a classification accuracy of 69%, demonstrating substantial potential for satellite data in pavement condition assessment despite challenges posed by resolution and mixed pixel issues. The POMDP model indicated that incorporating satellite monitoring can reduce the life-cycle costs of pavements by approximately 0.1%. This research highlights the potentials of combining remote sensing, machine learning, and decision analysis techniques in enhancing pavement maintenance strategies, paving the way for more-efficient infrastructure management.