Abstract
An automated system for forecasting the conditional probability of frozen precipitation was put into operation by the National Weather Service in November 1972. The Model Output Statistics (MOS) concept was used to develop the system, and both teletypewriter and facsimile products have been distributed to field offices twice daily. In this paper, guidance forecasts from this system are compared to subjective (local) forecasts prepared at Weather Service Forecast Offices. The local forecasts have been archived since September 1973 as part of a combined aviation/public weather forecast verification program within the National Weather Service. The comparative verification between the guidance and locals for two different data samples shows the guidance has produced better forecasts for the 18, 30, and 42 h projections.
In attempting to improve the operational system, we experimented with the Regression Estimation of Event Probabilities (REEP) screening technique. The operational system had been developed with the logit model, but our logit computer program does not objectively screen predictors as REEP does. A comparison of the REEP and logit systems on independent data shows logit to be better.
We used the logit model to develop a new operational system. Five winters of developmental data were used for the new system; the old system was developed from three winters of data. A comparison between the new and the old systems on independent data shows that they are equally accurate for the short-range (12 h) projection but that the new system is more accurate for the longer-range (36 h) projection.
The predictors in the new system include the 850 mb temperature, boundary-layer potential temperature, 1000–500 mb thickness, and 1000–850 mb thickness. These variables are forecast by the National Meteorological Center's primitive equation model. The new system was made operational during the winter of 1975–76 and provides forecasts for the 12, 18, 24, 30, 36, 42, and 48 h projections twice daily.
