Forecasts of Monthly 700 mb Height: Verification and Specification Experiments

Abstract

Twenty-five years (1958-82) of monthly 700 ml, geopotential forecasts produced by the U.S. National Weather Service are verified and then used in a series of temperature specification experiments. The forecasts show skill with respect to climatology through the positive correlations between forecast and observed anomalies in all calendar months. Improvement over persistence is apparent in the root mean square error, the mean absolute error, and the S skill score. The verification statistics also show a temporal trend toward smaller errors during the 1958-82 period. Stepwise screening and EOF (empirical orthogonal function) regressions are compared as alternative strategies for the specification of surface station temperatures from 700 mb heights. When the statistical significance of the model order is comparable and the equations are applied to the developmental sample of 700 mb verification grids, the screening procedure outperforms the EOF procedure according to the mean absolute error, the fraction of described variance, and a 3-category skill score. However, when the same sets of equations are applied to an independent sample consisting of the most skillful 700 mb forecasts, the results of the screening procedure show considerably more degradation. Temperature forecasts derived fromthe two procedures have comparable 3-category skill scores, but the EOF-derived forecasts are characterized by smaller mean absolute errors and higher correlations between the forecast and observed temperatures. The tendency for less degradation of the EOF-derived forecasts is attributed to their smaller variances and to the ability of the EOFs to capture modest amounts of forecast skill regardless of the region in which the skill is achieved.