Investigation of systematic errors by relaxation experiments

The systematic errors of the ECMWF forecast model have been investigated by relaxing forecasts towards analyses in selected regions. This method proved to be a useful tool to study the initial error growth in a certain region and the subsequent error propagation. The forecast fields in the tropics were relaxed towards the analysis by adding a contribution proportional to the current forecast error to the tendencies of temperature and momentum. From the comparison of control forecasts with tropically relaxed forecasts it was possible to separate large errors generated in the extra‐tropics from comparatively small errors penetrating from the tropics into the extra‐tropics. Relaxation in the major mountain regions showed different results for the Rocky Mountains and the Himalayas. the errors originating from the Rocky Mountains are substantial contributors to errors of both transient eddies and stationary eddies. the baroclinic waves in the Atlantic storm track are weakened and the stationary eddies develop a wave‐train‐like error pattern which is typical for many operational forecasts in the winter season. The errors in the Pacific are much less dependent on errors from the mountainous region. the error structure of the transient and stationary eddies, as revealed from experiments in which the Asian mountains are excluded or included from the relaxation, shows little similarity to operational forecast errors.