Evaluation of the High-Resolution Rapid Refresh Model for Forecasting Roadway Surface Temperatures

Pavement surface temperatures are an important component to winter road conditions forecasts to determine when and how to treat the roadways with chemicals to prevent icy conditions. According to the Federal Highway Administration (FHWA), states and local agencies spend more than $2.3B on winter weather operations per year. Having accurate weather forecasts are essential for making the right call on a winter storm. This study evaluates an off-the-shelf weather model, the High-Resolution Rapid Refresh (HRRR), to determine pavement surface temperatures compared with the Road Weather Information System (RWIS). Six locations are used for ground truth in this study, two from each of three temperature zones in Indiana. The RWIS sites included in this study are Fort Wayne and Gas City (northern stations), I-74 at I-465 Indianapolis and US-31 at SR-38 (central stations), and Jeffersonville and Scottsburg (southern stations). Residuals between RWIS and HRRR are compared at each location in addition to the Mean Absolute Error (MAE) for four months, December 2018 through March 2019, and one storm that occurred on January 12^th, 2019. The data is analyzed without modification and with filters on a solar radiation threshold of 170 W/m² and precipitation for each forecast hour up to 18 hours in advance.

As can be seen in Figure 1, forecasts with narrow time windows and with solar radiation above 170 W/m² resulted in the highest errors (denoted by orange), while forecasts that coincided with precipitation result in the lowest errors (denoted by green). Given how important even a degree or two in pavement surface temperatures is to winter operations, this study demonstrates the need for improved pavement observations. Emerging connected vehicles that may have pavement sensors integrated into them are a particularly attractive source for a scalable, real-time calibration input. The Maintenance Decision Support System (MDSS) currently utilized by many Departments of Transportation (DOT) makes use of the HRRR to estimate pavement surface temperatures but a higher resolution dataset that is continuously calibrated by connected vehicles could further reduce errors in pavement surface temperature predictions, potentially leading to both improved treatment plans and reduced costs for DOTs.