Secure and Accurate Road Weather Services - The Belgian SARWS project

For the generation of accurate warnings for dangerous road conditions, road weather models typically depend on observations from road weather stations (RWS) at fixed locations along roads and highways. Observations at higher resolution in space and time have the potential to provide more localized, real-time weather warnings. The rise of connected vehicles with onboard sensing capabilities opens up exciting new opportunities in this field. For this purpose, a heterogeneous group of industrial stakeholders and researchers consisting of more than thirty partners from seven countries including Belgium, initiated the CELTIC-NEXT project "Secure and Accurate Road Weather Services" (SARWS). The goal of SARWS is to provide real-time weather services by expanding observational data from traditional RWS sources with data from large-scale vehicle fleets. The Belgian consortium consists of Verhaert New Products & Services, Be-Mobile, Inuits, bpost, imec - IDLab (University of Antwerp) and the Royal Meteorological Institute of Belgium (RMI). Within the Belgian consortium, the focus is on the use of vehicle data to enable real-time warning services for potentially dangerous local weather and road surface conditions. The vehicle fleet consists of cars of the Belgian Post Group (bpost) in the region around Antwerp, and will consist of 15 cars by the end of summer 2021. Data on vehicle dynamics, such as wheel speed, are gathered from the vehicle's CAN bus, while an additional installed sensor box collects air temperature, relative humidity and road surface temperature observations. Data on wipers and fog light activation, and camera images are also collected.

We present the Belgian SARWS setup, data flow, and the developed data distribution platform. We discuss validation results for 2021, comparing car sensor observations to close RWS and weather stations, focusing mainly on air temperature, humidity and road surface temperature, and show the need for calibration and bias correction. We also demonstrate an experimental version of the RMI road weather model that provides short-term road weather forecasts for 50-meter road segments, using car sensor data for initialization, and compare with road weather forecasts at nearby station locations. We also demonstrate machine learning approaches that are explored to detect weather information from the vehicle dynamics.