Bias and Drift of the Medium-Range Decadal Climate Prediction System (MiKlip) validated by European Radiosonde Data

Abstract

Quality controlled and homogenized radiosonde observations have been used to validate decadal hindcasts of the MPI-Earth-System-Model for Europe (excl. some Eastern European countries). Simulated temperatures have a cold bias of 1 to 4 K, increasing with height throughout the free troposphere over Europe. This implies that the simulated troposphere is less stable than observed by the radiosondes over Europe. Simulated relative humidity is 10 to 40 % higher than observed. Part of the humidity bias, 10 to 25 % relative humidity, is due to the simulated lower temperature, but the remainder indicates that modelled water vapour pressure is too high in the free troposphere above Europe. After full-field initialization with oceanic state, the atmospheric temperature bias changes over the first couple of years, with a relaxation time of 5 years near the surface (850 hPa) and less than 1 year near the tropopause (200 hPa). Anomaly correlations, mean-square error and logarithmic ensemble spread score indicate small improvements in hindcasted tropospheric temperatures over Europe when going from ocean anomaly initialisation to ocean anomaly initialisation plus full field atmospheric initialisation, and then to full field ocean initialisation plus full field atmospheric initialisation. In the stratosphere, these changes have little effect. For humidity, correlations and skill scores are much poorer, and little can be said about changes over Europe due to different initializations.