Evaluation of a Probabilistic Subfreezing Road Temperature Nowcast Using Machine Learning

An emerging need within the NWS is impacts-based decision support for road hazards as this is the leading cause of weather-related fatalities in the United States. An important discriminant for whether some road hazards exists is whether the roads are subfreezing or not. While some experimental products exist to aid in this part of the decision-support paradigm, they suffer from some inadequacies that prevent their implementation on a national scale. Herein, a machine learning algorithm for generating a gridded CONUS-wide probabilistic road-temperature forecast is presented. A random forest is used to tie a combination of HRRR model surface variables and information about the geographic location and time of day/year to observed road temperatures. This approach differs from its predecessors in that road temperature is not deterministic, but rather it provides a 0-100% probability that the road temperatures are subfreezing. Probabilistic guidance can account for the varying controls on road temperature that are not easily known or accounted for in physical models, such as amount of traffic, road composition, and differential shading by surrounding buildings, terrain, etc. The algorithm is trained using road temperature observations from the 2016/17 winter season, calibrated using observations from the following winter season, and evaluated on the 2018/19 winter season. Statistical evaluation for the predicted probabilities shows exceptional skill. The mean area under the receiver operating characteristics curve is 0.97 and the Brier Skill Score is 0.67 for a 2-hr forecast and only degrades slightly as lead time is increased. Additionally, the algorithm produces well-calibrated probabilities with maximums near both low-end and high-end probabilities, suggesting the algorithm provides clear and consistent discrimination between clearly subfreezing and above-freezing environments. Case-study analyses show the algorithm performs well for various scenarios and captures the temporal and spatial evolution of the probability of subfreezing roads reliably. This product has been integrated into the Multi-Radar Multi-Sensor (MRMS) system for evaluation and feedback.