Prediction of the quasi-biennial oscillation with a multi-model ensemble of QBO-resolving models

A multi-model study is carried out to investigate the ability of models to predict the evolution of the quasi-biennial oscillation (QBO) up to 12 months in advance. All models are initialised from common reanalysis data, and forecasts run for a common set of 30 start dates over 15 years. All models have high skill in predicting the phase evolution of the QBO at 20–30 hPa, with slightly more variable results at higher and lower levels. Other aspects of the predicted QBO are of variable quality, and in some cases are consistently poor. QBO easterlies are too weak in all models at 20–50 hPa, while westerlies can be either too strong or too weak. This results in both a reduced amplitude of the QBO and a westerly bias in zonal-mean winds, notably at 30 hPa. At 70 hPa models tend to have reduced QBO amplitude and an easterly bias. Despite these failings, a multi-model ensemble of bias- and variance-corrected forecasts can be used to give accurate and reliable QBO forecasts up to at least a year ahead. Analysis of the zonal momentum budget during the first month of the forecast shows that large-scale forcing from Eliassen–Palm flux divergence and vertical advection are handled fairly well by the models, although vertical advection terms tend to be weaker than reanalysis estimates. Total tendencies show common errors, suggesting common failings in gravity-wave drag treatments. Teleconnections from the QBO to Northern Hemisphere winter circulation are also examined, and do not appear to be realistic beyond the first month. Analysis of initialised forecasts is a powerful tool for diagnosing the accuracy of model processes driving the QBO.