Probability Statements in Weather Forecasting

Abstract

Subjective rain probability forecasts issued by the Weather Bureau Forecast Center at San Francisco during four winter seasons are evaluated. It was found that the observed occurrence of rain tended to be greater than forecast in the 0 to 40 per cent forecast categories and less than forecast in the 50 to 100 per cent forecast categories with the average difference for all categories balancing out to ?0.6 per cent. The average absolute difference between the forecast probability and the observed occurrence in all categories was 5.8 per cent. The frequency distribution curve of the forecast probability categories used showed a major concentration of forecasts centered at the 0 per cent category with a second and much flatter peak centered near the 60 to 70 per cent categories. There was a minimum of forecasts near the climatological probability of 30 per cent and also at the 100 per cent category. Skill in assigning rain probabilities, as measured by Brier's P score, decreased with increasing time from 0.18 for the 12-hr period centered 6 hr after forecast time to 0.35 for the 12-hr period centered 42 hr after forecast time. This may be compared to a score of 0.40 which would have been attained with a constant forecast of the climatological probability of rain.