A Statewide Adverse Weather Forecasting Model in Montana Using Road Weather Information System Measurements Improved Forecasting Accuracy by 32 Percent.
Integrated Icy Road Detection and Alert Tool in Montana Used data from Road Weather Information System and drone-based ice detection technology.
Date Posted

Icy Road Forecast and Alert (IcyRoad): Validation and Refinement Using MDT RWIS Data

Summary Information

Forecasting inclement weather is a common practice overseen by State Departments of Transportation (DOTs) to plan for anti-icing activity to ensure safety on public roads. This effort was meant to improve precipitation detection capabilities to reduce the number of icy road vehicle crashes and fatalities. The Montana DOT (MDT) refined a statewide IcyRoad Detection and Alert system during the winter of 2020-2021 using Road Weather Information System (RWIS) and drone-based ice detection technology. A forecast algorithm was demonstrated by collecting hourly historical observations which was fed into a weather model to forecast temperature, clouds, rainfall information, and the physical scheme of the road surface. Integrating the forecast results, road physical scheme, ground observations, cloud computing and data visualization and mining techniques, and remote sensing technology, IcyRoad aimed to predict the road ice conditions for any road across the US with a 24-hour lead time. 


The RWIS site observations were collected through a data access webpage from January 1 to March 31, 2020 and October 2020 to March 2021 from 72 sites across Montana. The data collected were stored in a cloud-based database which enabled web-based automatic data analysis for all RWIS sites. The RWIS sites’ performance for the IcyRoad forecast was analyzed via programming software tools. A statistical analysis was conducted by comparing the National Oceanic and Atmospheric Administration (NOAA) North American Mesoscale Forecast System (NAM) model-based IcyRoad forecast with the 72 RWIS sites. Two variables were assessed: ground surface 2-meter air temperature and ice road status. Additionally, drone-based remote sensing technology was used to detect black ice through a hyperspectral camera launched on an unattended aerial vehicle (UAV).


  • The statewide IcyRoad model improved the average hourly road ice forecast accuracy from 62 to 82 percent (a 32 percent increase), helping assure that de-icing activities took place during the winter season and therefore reducing the possibility of vehicle crashes.
  • Only three sites had an accuracy below 70 percent: Bozeman (62.56 percent), Gary Cooper Bridge (67.83 percent), and Monida Pass (69.76 percent). The high accuracy rates for all the sites proved that the forecast model was reliable for any region.
  • The validation analysis showed that the forecasted two-meter surface air temperature had high accuracy, more than 0.8 for the correlation coefficient for more than 70 percent of the sites. The icy road status had a correlation coefficient of 0.64 for all sites. These model validation measures implied that the model could be used again with a new dataset, an important advantage for weather forecasting.
  • Research on black ice detection in this study was very limited due to a lack of black ice identification and funding availability.
Results Type
Deployment Locations