Improved prediction for a spatio-temporal model

Abstract

We investigate a framework for improving predictions from models for spatio-temporal data. The framework is based on minimising the mean squared prediction error and can be applied to many models. We applied the framework to a model for monthly rainfall data in the Murray-Darling Basin in Australia. Across a range of prediction situations, we improved the predictive accuracy compared to predictions using only the expectation given by the model. Further, we showed that these improvements in predictive accuracy were maintained even when using a reduced subset of the data for generating predictions