Verification of route-based road weather forecasts

Abstract

Route-based road weather forecasting is increasingly becoming the standard methodology for winter maintenance decision making (i.e. whether or not to salt the road network) by the highway industry in the UK. Route-based forecasting requires, for the first time, the accuracy of forecasts around routes and away from sensor sites to be verified. This is essential so that end users have confidence in the models’ ability to accurately predict road surface temperature at every point around their road network. A new methodology for verifying route-based forecasts is proposed that uses clustering techniques to create clusters of forecast points with similar geographical and infrastructure characteristics. This facilitates the analysis of forecast statistics at the cluster level, which is found to improve statistical assessment of model performance since verification can be achieved at a much higher resolution than the current methodology allows. Furthermore, verification of the full spatial extent of a route-based forecast can be achieved with fewer forecast points since the majority of thermal variations around the road network are well represented by the clustering solutions. A new sampling strategy is proposed that potentially enables verification at the full spatial and temporal resolution.