Evaluation of a Nonprobabilistic Weather Forecasting Experiment

Abstract

The results of experimental real-time nonprobabilistic weather forecasting at the University of Virginia during the period 1983?1988 are described. During the fall (September to December) and spring (January to May) semesters, daily forecasts of surface maximum and minimum temperatures and precipitation amount out to two days in advance were prepared. Departure from normal temperature average and precipitation amount during the period three to five days in advance were also predicted. All these forecasts were valid for Charlottesville, Virginia (CHO). Skill relative to climatological control forecasts is demonstrated by consensus (average of all forecasts) for predictions of maximum and minimum temperatures and (on days with measurable precipitation) of precipitation amount. The precipitation-amount forecasts were also skillful at categorically predicting precipitation occurrence, that is, distinguishing wet days from dry days. All these skills decreased with increasing forecast range. The three-to-five-day temperature and (to a lesser extent) precipitation outlooks were skillful relative to persistence control forecasts. Among the various types of forecasts, only the three-to-five-day temperature outlook exhibits a statistically significant upward trend with time in skill, possibly reflecting the impact of improvements in medium-range forecast guidance during the period of these experiments.