Autoregressive modelling of the tropical stratospheric quasi‐biennial oscillation

Autoregressive time series models are used to make forecasts of stratospheric zonal wind and temperature at Kwajalein Island near the equator in the northern Pacific Ocean. The tropical stratosphere is dominated by the quasi‐biennial oscillation (QBO). Because the QBO has a varying period, an autoregressive model is more appropriate than a mixture of sinusoids. The data for 11 levels are efficiently represented by principal components, and univariate and multivariate autoregression models are fitted to six of the principal components. The autoregressive models are used for estimation of variance density spectra and for forecasting. The multivariate model is found to perform better in both detecting and forecasting the quasi‐biennial oscillation than either univariate models of the principal components or autoregressive models of the data set itself.