Understanding and forecasting polar stratospheric variability with statistical models.

The variability of the north-polar stratospheric vortex is a prominent aspect of the middle atmosphere. This work investigates a wide class of statistical models with respect to their ability to model geopotential and temperature anomalies, representing variability n the polar stratosphere. Four partly nonstationary, nonlinear models are assessed: inear discriminant analysis (LDA); a cluster method based on finite elements (FEM-VARX); a neural network, namely a multi-layer perceptron (MLP); and support vector regression (SVR). These methods model time series by incorporating all significant external factors simultaneously, including ENSO, QBO, the solar cycle, volcanoes, etc., to then quantify their statistical importance. We show that variability in reanalysis data from 1980 to 2005 is successfully modeled. FEM-VARX and MLP even satisfactorily forecast the period from 2005 to 2011. However, internal variability remains that cannot be statistically forecasted, such as the unexpected major warming in January 2009. Finally, the statistical model with the best generalization performance is used to predict a vortex breakdown in late January, early February 2012. [ABSTRACT FROM AUTHOR]