Developing Forward-looking Probabilistic Road Weather Forecasts at the Met Office

Surface transport forecasting has traditionally focused on deterministically categorising the state - for example, dry, damp, or icy - of a road surface, to provide decision support to winter gritting services. However, with the emerging Connected Autonomous Vehicle (CAV) sector, the ability to accurately describe a broader range of weather conditions, at a much smaller temporal and spatial scale, is becoming equally important to ensuring public safety. Localised conditions such as road spray, rain and fog can all degrade the performance of CAV sensors, whilst settled snow also has the potential to impede navigation by obscuring road markings. The difficulty of deterministically forecasting such precise localised conditions requires us to quantify uncertainty, driving a move towards probabilistic, risk-based forecasting. Therefore, the Met Office is currently developing a new Surface Transport Forecasting (STF) post-processing system, designed to accommodate these future user requirements.

The new STF post-processing system is centred on the Joint UK Land Environment System (JULES); a community model used as the land-surface component of the Met Office Unified Model (UM), but which can also be used – as we do here – as a stand-alone land-surface model driven by forecast output from Numerical Weather Prediction (NWP) models. To produce probabilistic forecasts for locations within the United Kingdom, we are using output from the Met Office regional ensemble model MOGREPS-UK to drive JULES. MOGREPS-UK is a 2.2km resolution 18-member ensemble which provides forecasts out to 5 days. It is generated by time-lagging over 6 hours, initialising three new ensemble members every hour from perturbed initial conditions. By running JULES for each ensemble member, we create a set of possible road forecast outputs. Considering these predictions in aggregate allows us to generate probabilistic forecasts of road weather conditions.

Our ensemble-driven STF system has been verified using standard ensemble verification techniques, including rank histograms and reliability plots. Initial analysis of results, using data from approximately 300 locations in the United Kingdom for which good quality road weather observations are available between 2015 and 2021, indicate that observed road surface states are generally captured within the spread of predictions. The new system has been compared with the current deterministic STF system.  In particular, in challenging meteorological conditions (for example, where there is variable cloud cover, scattered showers, or on marginal temperature nights) the probabilistic approach allows us to quantify the uncertainty in the road state forecast in a way that the deterministic approach does not. Further work will focus on the most effective way to communicate probabilistic forecasts to end-users, ensuring they are able to apply the output to enhance their decision-making.