Spatial-temporal statistical modelling and prediction of environmental process
Hierarchical modelling for the environmental sciences: statistical methods and applications, Oxford University Press, Oxford, UK, (2006)Abstract
Environmental data usually have both spatial and temporal components. Therefore, it is essential to have statistical models to describe how the data vary across space and time. The techniques from geostatistics and time series are powerful tools to study spatial–temporal processes where the spatial and temporal structure can be modeled separately (separable models), and the spatial–temporal structure does not change with location and time (stationarity). However in real applications spatial–temporal processes are rarely separable and stationary. In this chapter, the current methods for space–time modeling are reviewed, and a new class of nonstationary, nonseparable
spatial–temporal models is proposed. We model the nonstationary spatial–temporal process as a mixture of local orthogonal stationary spatial–temporal processes, and we do not assume separability.We also present a general framework for combining disparate spatial–temporal data and for Bayesian spatial–temporal prediction. We apply the methodology proposed here to model and predict wind fields over the Chesapeake Bay. Our results show that improved wind field maps can be obtained by combining output from numerical weather prediction models with the observed data.