Abstract
A statistical method is performed on the 10-day forecasts by the European Centre for Medium Range Weather Forecasts (ECMWF) operational model for the 1979/80 to 1985/86 winter 500 mb geopotential heights in order to look for the spatial structures of the field that are best predicted by the model and to evaluate the quality of these optimum predictions by comparing with persistence predictions. It is shown that one index associated with a structure connecting the Pacific negatively to the Atlantic has better skill than the smoothed persistence of any linear combination of the initial state. The same investigations are then carried out on 3 sub-regions of the northern hemisphere: Atlantic-Europe, Asia, and Pacific-North America. The Pacific-North America area has some predictability at 10-day lag, the Atlantic-Europe area a marginal one, and the Asian area no predictability at all (referring to persistence as well as to a randon predictor). DOI: 10.1111/j.1600-0870.1988.tb00328.x