Improving Climate Prediction Using Seasonal Space Time Models

In this paper a class of seasonal space-time models is introduced for general lattice systems. Covariance properties of spatial rst-order models, including stationarity conditions, are studied. Procedures for examining spatial independence and symmetry of the models are developed. Estimation approaches in time series analysis are adopted, and forecasting techniques using the seasonal space-time models are discussed. The models are applied to 516 consecutive maps of monthly-averaged 500 mb geopotential heights over a 10 10 lattice in the extra-tropical Northern Hemisphere for the purpose of improving climate prediction. It is found that space-time models with instantaneous spatial component give better t than other models in terms of maximizing the conditional likelihood function, but their forecast ability is poor because of inverse problems. On the other hand, space-time models without instantaneous spatial component provide more accurate forecast values than univariate time series models and space-time models with instantaneous spatial component.