On «Results of forecasting with the barotropic models»

Dear Sir, In a recent paper describing results of forecasting by numerical methods (STAFF MEMBERS, UNIVERSITY OF STOCKHOLM, 1954), it was stated that the correlation between observed and computed changes is not a satisfactory measure of the goodness of a forecast. Previously the same group (Staff Members, University of Stockholm, 1952) had raised the objection that a correlation coefficient measures the similarity in behaviour of two variables without any reference to systematic errors such as a difference in scale of the variations.