Abstract
The spatial autocorrelation of the error of the ECMWF 24-h forecast of 500 mb geopotential height during 1980–83, hereafter referred to as the “forecast error,” is studied. The leading EOF of the forecast error describes a “teleconnection” between the Himalayas and the Pacific/European sector. The third EOF resembles the analysis error. Approximations for the isotropic part of the autocorrelation are presented. They consist of two source terms having different spatial scales. A “random-like” term could be due to model errors in baroclinic processes or to random analysis errors. A “scale-dependent” term could be due to model errors in barotropic processes and over mountain areas, or to analysis errors over data-sparse areas or mountains. The terms do not contribute uniformly. The “scale-dependent” term is strongest over mountains The relative contribution of that term decreased from forecasts of 1980–81 to those of 1982–83.
