Objective Estimation of the Conditional Probability of Frozen Precipitation

Abstract

A system is developed which produces objective forecasts of conditional probability of frozen precipitation for the conterminous United States. Development of the system consists of two basic steps, in each of which the MOS (Model Output Statistics) concept is used. First, for each of 186 stations, we find a ?50%? value for each of three variables predicted by the National Meteorological Center's Primitive Equation (PE) model: 1000?500 mb thickness, boundary-layer potential temperature, and 850-mb temperature. For instance, we find the value of the 1000?500 mb thickness which indicates a 50?50 chance of frozen precipitation at a particular station, provided precipitation occurs. These 50% values are determined by using the logit model to fit data from three winter seasons, September 1969 through March 1972. Secondly, the deviations from the 50% values are determined for each station for each variable; the relative frequency (for those cases when precipitation occurred) of frozen precipitation is then computed, again with the logit model, as a function of these new variables. In order to get stable results in this last step, data for all stations are combined. In addition to the meteorological variables, we also use the first harmonic of the day of year and station elevation as predictors. Separate logit equations are determined for each of the PE run times, 0000 and 1200 GMT, and for each of four projections, 12, 24, 36, and 48 hours. This system was put into operation by the National Weather Service in November 1972. Both teletypewriter and facsimile products are being distributed to field offices twice daily. A comparative verification on independent data for the 12- and 36-hr forecast projections shows the objective system produced better forecasts than those prepared subjectively at the National Meteorological Center.