Automated Nationwide Forecasts of Maximum and Minimum Temperature

Abstract

An objective method of forecasting maximum and minimum surface temperatures for periods from 12 to 60 hr in advance is discussed and illustrated. The method makes use of multiple regression equations derived for 108 cities in the United States and 11 cities in Canada from 16 years of daily data stratified by 2-month periods. The predictors are selected by screening (by pairs) the following parameters: a) 700-mb height and 700?1000 mb thickness observed at 67 grid points in North America about 12 hr before the valid time of the prognostic temperature; b) maximum and minimum temperatures observed at the network of 119 cities about 12 or 24 hr before the prognostic valid time; and c) the day of the year. On the average approximately ? of the temperature variance is explained by 5 variables, and the standard error of estimate is about 4F. The method has been applied in an iterative fashion twice daily at the National Meteorological Center at Suitland, Md., since September 1965. The first forecast, for 12 hr in advance, uses only observed values of height, thickness, and temperature as input to the multiple regression equations. Subsequent forecasts utilize numerical prognoses of height and thickness, as upper air input, and preceding automated forecasts of maximum and minimum, prepared by the system, as temperature input. Verification statistics are presented for a year's operation, and the resulting objective forecasts appear to be almost as good as subjective forecasts and superior to persistence or an older objective method.