Evaluation of probabilistic forecasts in Uppsala and its potential use in winter road maintenance

Abstract: Efficient winter road maintenance is crucial for safety and societal function during the winter months in Sweden. This report aims to evaluate the MetCoOp ensemble system CMEPS and investigate its potential use as a basis for formulating criteria for snow removal that accounts for forecasted weather. Today the criteria for activating snow removal in Sweden are static, meaning they start after a set amount of snow and should end within a set time span. The verification metrics rank-histogram, continuously rankprobability score, reliability diagram, and Brier score were used to evaluate temperature and solid precipitation. Observations used as verification were taken at the measuring station Geocentrum in Uppsala during the winters of 2020/2021, 2021/2022, and November-December 2022. The analysis shows the temperature forecast to be under-dispersive and with a cold bias. The ensemble system is shown to be less reliable for predicting temperatures below 0 °C the first 24 hours after the forecast is issued. Still, the forecast generally performs better for short lead times. The forecast overestimates solid and liquid precipitation. The wet bias is greatest for short lead times and long accumulation times. Short lead times are most reliable regarding solid precipitation over 1mm and 3mm. The first 24-30 hours are most important for an application in winter road maintenance, and based on how the forecast system performs for these lead times in this study, it would need calibration. For larger amounts of snow, new criteria could help adjust the starting time and time limits. Before implementing such criteria, practical questions as if dynamic criteria would lead to an improvement and how high the probability threshold should be must be answered. The sample size is also found to be too small, and further analysis is required, especially with data allowing for evaluation of higher thresholds.