Seasonal-to-Interannual Prediction Skills of Near-Surface Air Temperature in the CMIP5 Decadal Hindcast Experiments

Abstract

Abstract This study explores the seasonal-to-interannual near-surface air temperature (TAS) prediction skills of state-of-the-art climate models that were involved in phase 5 of the Coupled Model Intercomparison Project (CMIP5) decadal hindcast/forecast experiments. The experiments are initialized in either November or January of each year and integrated for up to 10 years, providing a good opportunity for filling the gap between seasonal and decadal climate predictions. The long-lead multimodel ensemble (MME) prediction is evaluated for 1981–2007 in terms of the anomaly correlation coefficient (ACC) and mean-squared skill score (MSSS), which combines ACC and conditional bias, with respect to observations and reanalysis data, paying particular attention to the seasonal dependency of the global-mean and equatorial Pacific TAS predictions. The MME shows statistically significant ACCs and MSSSs for the annual global-mean TAS for up to two years, mainly because of long-term global warming trends. When the long-term trends are removed, the prediction skill is reduced. The prediction skills are generally lower in boreal winters than in other seasons regardless of lead times. This lack of winter prediction skill is attributed to the failure of capturing the long-term trend and interannual variability of TAS over high-latitude continents in the Northern Hemisphere. In contrast to global-mean TAS, regional TAS over the equatorial Pacific is predicted well in winter. This is mainly due to a successful prediction of the El Niño–Southern Oscillation (ENSO). In most models, the wintertime ENSO index is reasonably well predicted for at least one year in advance. The sensitivity of the prediction skill to the initialized month and method is also discussed.