MULTIVARIATE STOCHASTIC PREDICTION OF THE GLOBAL MEAN SEA LEVEL ANOMALIES BASED ON TOPEX/POSEIDON SATELLITE ALTIMETRY

摘要

Forecasting global mean monthly sea level anomalies (SLA) based on (1) monthly SLA data taken from TOPEX/Poseidon (T/P) satellite altimetry, and (2) global monthly sea surface temperature (SST) data NOAA 01.v2 SST has been performed by means of multivariate autoregressive models of order p (AR(p)). The bivariate AR(2) (chosen by means of the Schwartz Bayesian Criterion (SBC)) has been fitted to the differenced data yielding a satisfactory goodness-of-fit. We also applied the Akaike Information Criterion (A1C) to select the order of an AR model for the differenced univariate SLA time series. The chosen AIC order has been assumed to be the order of a bivariate AR fitted to the differenced bivariate SLA and SST time series. This yields an AR(ll) for which a goodness-of-fit is much better than in case of an AR(2) chosen by means of SBC. The updated models (SBC1 and AIC based) have been used to build the mean monthly SLA predictions in the period Jan. 2003-Dec. 2003. We have also shown that the accuracies of the predictions are better for the model selected by the AIC. The 1-month and 2-month forecasts are acceptable however the long-term predictions tend to be inaccurate. The need for combining SLA with SST in forecasting SLA has been indicated by (1) the wavelet time frequency spectra and coherence analyses, and (2) the comparison between the bivariate and univariate models yielding better accuracies of predictions in case of the multivariate one.