Computer Prediction of Precipitation Probability in the United States

Abstract

The “perfect prog” method of combining numerical and statistical weather prediction is applied to develop an automated system for forecasting the probability of precipitation at 108 cities on the mainland of the United States during daytime and nighttime periods from 12–60 hr in advance. Multiple regression equations are derived from a 4–5 year sample of data by seasons for each city by screening twice-daily geographical arrays of the following predictors: initial 850-mb height, initial 850–700 mb mean dew-point spread, and previous 12-hr precipitation at the network of surface stations. Each of the three predictor fields contributes about equally toward explaining the variance of the observed precipitation, but considerable geographical variation is exhibited by the equations. The forecast system is applied in an iterative fashion in 12-hr time steps by using as input numerical predictions of height and moisture at standard grid points as well as prior values of precipitation. The resulting computerized forecasts of precipitation probability, when applied on an operational basis in real time, may offer valuable guidance to the local weather forecaster.