A new scoring scheme for verification of fixed-width, credible interval temperature forecasts, which penalizes forecasts of low probabilities for their uncertainty, has been developed. This scoring scheme encourages the forecaster to maximize the proportion of verifying forecasts and to provide probabilities that reflect his true confidence. Results of the use of this scoring scheme in the University of Michigan Forecast Game indicate that a forecaster should strive first for accuracy of the temperature forecast and second for having well-calibrated associated probabilities.

This content is only available as a PDF.