Long Short-Term Memory Modeling of Road Surface Grip for Salt Application in Winter Roadway Maintenance

Road surface grip during snowstorms is critical for traffic mobility and safety. It is important to incorporate surface grip as a quantitative indicator for effective decision-making in winter roadway maintenance. However, the timing of salt application and its influence on grip are rarely considered in decision-making to improve the efficiency of winter maintenance. To address this gap, data-driven approaches were developed to predict road surface grip (friction) for salt application in winter roadway maintenance. An advanced recurrent neural network (RNN) model, long short-term memory (LSTM), was developed with hyperparameter tuning. The LSTM model considers sequential effects of surface temperature, atmospheric condition, and salt application on the time-dependent evolution of road surface grip. Sensitivity analysis results show that road surface grip changes to a higher level more quickly when a 50–100 lb/mi higher salt application rate is applied as compared to current practice. The interaction effects of salt application and climate condition on road surface grip were further analyzed. It was found that the grip change became more sensitive to salt application rate when road surface temperatures were 4°C lower. The use of the LSTM model enables event-based decision-making for salt application in winter roadway maintenance.