A performance comparison of coupled and uncoupled versions of the Met Office seasonal prediction general circulation model

ABSTRACT We compare the performance of the Met Office's ocean–atmosphere coupled general circulation model (CGCM) seasonal prediction system with that of an atmosphere‐only system (AGCM). The CGCM and AGCM systems share the same atmospheric component and the performance comparison therefore provides insight into the skill benefits available from coupling atmosphere and ocean models. In this study, the AGCM is forced with predicted sea surface temperature (SST) based on persistence of prior observed SST anomalies. The analysis uses 43‐yr, nine‐member ensemble hindcast data sets generated with both systems as part of the European Union project DEMETER. Results are focused on global and regional comparisons of long‐term skill for probabilistic prediction of 2‐m temperature in the upper tercile, and on selected case studies for the tropics and Europe. Performance assessments using relative operating characteristic scores, Brier skill scores and the resolution and reliability terms of the Brier score decomposition are contrasted. The largest CGCM benefits are found in tropical regions, where benefits to both resolution (essentially ‘event detection’) and to reliability (essentially ‘calibration’ of the forecast probabilities) are demonstrated. Improvements to reliability are found to be substantially greater than improvements to resolution. Regional assessments show benefits, as expected, in the tropical east Pacific, from improved prediction of SST variability associated with the El Niño Southern Oscillation (ENSO). However, substantial benefits are also seen throughout the tropical belt in seasons associated with the peak and decay of ENSO activity. Such benefits appear associated with representation of lagged teleconnection responses to ENSO in the tropical Atlantic and Indian Oceans. In the extratropics, CGCM improvements to reliability are also substantial, although benefits to resolution (assessed over large regions) appear negligible. Two classes of benefit are described. First, advantages from improved ENSO predictions appear to benefit skill in the North Pacific and North American regions, through teleconnection responses. Secondly, there is evidence of benefits from representation of coupled processes over the North Atlantic. In particular, CGCM skill benefits for prediction of spring season temperature in the European region appear to derive, in part, from coupled model representation of linkage between a well‐documented tripole pattern in North Atla