A Novel Framework for Assessing Unpaved Road Condition for Asset Management Using Satellite Imagery

Unpaved roads play a crucial role in Low and Middle-Income Countries (LMIC), with poor road conditions restricting mobility and severely affecting socio-economic development. Road condition is crucial to planning and prioritising maintenance, but Local Road Authorities (LRAs) struggle to collect this data using conventional means. A gap analysis showed that optical satellite imagery is not currently used to provide condition information on unpaved roads, confirmed by a review of stakeholder requirements. This research aims to determine how optical satellite imagery can provide information useful for road asset management sustainably and cost-effectively. It explores the feasibility of measuring unpaved road condition in LMICs using satellite earth observation and how the results can benefit LRAs. A novel framework for unpaved road condition assessment using satellite imagery is proposed to support the implementation of this research in LRAs. Trials were conducted in Tanzania using 83 roads, totalling 131.7 km. Conditions were predicted using pixel variation, and the results were compared to a ground-truth audit. Machine Learning was employed to explore automation and test lower resolutions with the aim of reducing imagery costs. Using pixel variation, overall accuracies in predicting condition of up to 79.8% were achieved, with individual accuracy for bad-condition roads reaching up to 87.3%. Machine Learning accuracy results exceeded those of pixel variation, and tests on lower-resolution imagery showed the potential to reduce capital costs by up to 70%. The framework was developed to measure the condition of unpaved networks rapidly, objectively, and reliably. The framework can accommodate different strategies and data requirements, allowing LRAs to define how to use the data most effectively for asset management and maintenance prioritisation. Automated road condition assessment using satellite imagery would benefit LRAs by avoiding the barriers of conventional visual surveys and enabling them to transition towards a more modern asset management approach