Predicting Road Conditions using Mobile and Remote Observations

Mobile weather observations taken from roadways can be a useful indicator of road conditions. When coupled with other types of remote observations, such as radar and fixed road side weather observations, a predictive model can be trained to provide real-time conditions for road segments. Commercial vehicle fleets provide an ideal platform for collecting mobile weather observations. Commercial vehicle fleets often operate 24 hours per day and cover wide geographical regions. When equipped with weather sensing systems that have the ability to transmit observations in real-time, this creates a wide reaching weather sensor network. This data is ideal for predicting road conditions at the moment of observation. Verification of this data is a challenging problem as road conditions are quite subjective and difficult to classify. Some Road Weather Information Systems (RWIS) provide a road condition parameter that can be used to assist in verification. This presentation will demonstrate the results of training regression models using mobile data from the Weather Telematics fleet that can be used to predict road conditions using mobile and remote observations. It will also cover issues related to data quality and verification of predicted road conditions using ESS road condition observations from RWIS.