Impact of seasonal forecast errors and their interference on extratropical prediction skill

Operational seasonal forecasts exhibit an increasingly useful level of extratropical prediction skill. However, particularly in and around the Atlantic basin, features such as the North Atlantic Oscillation exhibit a phenomenon known as the signal-noise paradox; where ensemble means correlate more strongly (on average) with observations than individual ensemble members. A cause of the paradox is that modelled signals are too weak in amplitude given their correlation with observed signals. An improved understanding of the origin of this phenomenon would allow for significant improvement in climate predictions. Here we investigate the role of mean state and teleconnection biases in modulating the strength of predicted signals. We focus on winter (DJF) mean state biases in pressure at mean sea level and geopotential height and zonal wind at 300 hPa in the Met Office GloSea5 and ECMWF SEAS5 forecast systems, using a 24 year hindcast period and comparing to the JRA-55 reanalysis. Major teleconnections to ENSO and QBO show a variety of phase and amplitude errors. We consider how wave activity fluxes are affected by these biases, and how interference between climatological stationary waves and predictable teleconnection signals affects extratropical prediction skill and the magnitude of predicted anomalies.