Data-Driven Road Condition Forecasting with High Spatial Resolution: Utilizing Tire-Centric Road Condition Monitoring Technology

Abstract

Forecasting road condition is important to realize a safe mobility society. Particularly, forecasting with high spatial resolution is desirable to provide the basic information for various tasks, such as issuing reliable advanced warnings, recommending optimal routes, improving efficiency in snow ploughing, and operating the minimum required salting with little damage to road structures. However, to our knowledge, owing to the difficulty in obtaining a sufficient accuracy of road condition detection with high spatial resolution, there is no report that addresses this important issue. This work attempts to address this issue through a tire-centric approach using tire-mounted sensors, which can accurately classify road conditions in real time without receiving other environmental effects, through tire acceleration monitoring. Since 2014, we have been using this technology for national highway maintenance services in Hokkaido prefecture (Japan), having accumulated reliable road-condition mapping records with high spatial resolution at all times of the day. The enriched operational records enabled us to construct forecasting models of road conditions with a data-driven method. Our results show that 70% accuracy for a 2-h-ahead forecast, with a spatial resolution of 100 m and covering a highway length of 20 km, is achievable with ensemble tree classifiers.