A Classical-REEP Short-Range Forecast Procedure

Abstract

A statistical technique has been designed to assist in the forecasting of weather elements in the 0?12-h time period. The method uses only the last two surface hourly observations for a particular station as predictors, and thus is suitable for use in regional weather offices. Based on a dependent sample of 28 yr of hourly observations for each station, the technique can be viewed as a climatological aid to forecasting that gives a specific answer. The procedure follows a classical statistical formulation whereby relationships are developed between weather elements as predictands and a set of weather elements for an earlier time as predictors. All predictands and predictors are categorized, and the forecast output is expressed in terms of the probability of occurrence of each predictand category. To aid in the interpretation of the probabilities, an objective categorization procedure has been added to suggest the best forecast category for each predictand element. A special feature of the system is that the output format has-been tuned to the requirements of aviation terminal forecasting through refinements to the categories and by allowing ?best and second-best? category choices to mimic the use of the qualifiers ?variable? and ?occasional? in forecasts. The system design was based on numerous tests. It has been compared with a Markov-based system, and experiments were conducted to determine the best type of predictors, to determine the value of seasonal stratification, and to tune the categorization procedure for forecasts of each specific element. The final system was evaluated on a 3-yr independent sample and was shown to be superior to persistence at all time ranges.