Experiments in probability of Precipitation Amount Forecasting Using Model Output Statistics

Abstract

Modifications to current model output statistics procedures for quantitative precipitation forecasting were explored. Probability of precipitation amount equations were developed for warm and cool seasons in a region in the eastern United States. Twelve-term equations, which were simultaneously regressed for four precipitation categories, were compared to equations that were regressed independently for each of the categories. The effect of varying the number of terms in the independently regressed equations was also considered. The utilities of linear predictors not presently considered and of multiplicative predictors selected with the aid of a one parameter multiplicative model were investigated. All forecast equations were evaluated using threat scores and biases achieved upon verification for one year of independent data. The independently regressed equations generally achieved threat scores similar to the twelve-term simultaneously regressed equations, and usually required fewer terms to do so. These more compact equations could be more readily interpreted by individual forecasters than could the twelve-term equations, making it easier to develop techniques for local adjustments to the objective forecasts. The prediction of the higher precipitation amount categories may benefit from the inclusion of predictor variables not presently considered.