A PROBABILISTIC FORECAST CONTEST AND THE DIFFICULTY IN ASSESSING SHORT-RANGE FORECAST UNCERTAINTY

Results are presented from a probability-based weather forecast contes t. Rather than evaluating the absolute errors of nonprobabilistic temp erature and precipitation forecasts. as is common in other contests, t his contest evaluated the skill of specifying probabilities for precip itation amounts and temperature intervals. To forecast optimally for t he contest, both accurate forecasts and accurate determination of one' s uncertainty about the outcome were necessary. The contest results in dicated that forecasters over a range of education levels produced ski llful forecasts of temperature and precipitation relative to persisten ce acid climatology. However, in this contest forecasters were not suc cessful in assessing the uncertainty of their maximum or minimum tempe ratures from day to day, as measured by the correlation of interval wi dth and absolute error. Though previous experiments have shown more op timistic results, the seasonal variation of forecast uncertainty can a ccount for much of the observed correlation, suggesting that day-to-da y assessment of forecast uncertainty may be more difficult than previo usly believed. It is argued that objective methodologies should be dev eloped to quantify uncertainty in forecasts.