Ensemble member forecast spread and its implication for road weather forecasting

This case study illustrates several issues that need additional investigation. It was surprising to see such large discrepancies between the weather models in predicting air and dewpoint temperature, wind speed, cloud cover, and precipitation. This result was repeated many times throughout the winter season. None of the models consistently outperformed the others for any parameter. All the models were too dry (low dewpoint temperatures) and most had difficulty predicting 100% cloudy conditions. The models also had a low wind speed bias overall. Some of these deficiencies can be traced to the fact that the Denver area often experiences shallow, moist frontal systems that are not well captured by the models. The fronts often arrive hours before they are predicted bringing along moist, cool air and thin, shallow cloud layers. These findings support the conclusion that an intelligent data fusion system should be used to optimize an ensemble of forecasts. The RWFS was able to demonstrate more skill overall than any of the individual forecast members. There is ongoing research on how to best configure the RWFS to make it more responsive to rapidly changing conditions and how to select the appropriate type and number of forecast members. The findings also support the concept of presenting weather prediction results in probabilistic terms because there are clearly times when the atmosphere is more predictable than others. Users should be made aware of the certainty of specific predictions that are important and relevant to their operations and decision making. More research is required to determine the best approaches to use to present this uncertainty to end users.