Exploring multivariate spatio-temporal change in climate data using image analysis techniques

Spatio-temporal data from earth observation systems and models are increasing at astronomical rates in the climate domain. This results in a massive dataset that is increasingly difficult to navigate to find interesting time periods where the spatial pattern of a process changes. The ability to navigate to such areas can lead to new knowledge about the factors that contribute to a spatio-temporal process. This paper proposes a method to automatically characterize multi-variate spatio-temporal datasets using basic image processing techniques and an efficient distance measure. The approach uses a measure of local image entropy combined with edge detection to find naturally occurring boundaries in the dataset. Then a distance measure is used to track the change in these boundaries over time. The resulting measure of spatio-temporal change can be used to explore spatio-temporal datasets to find new relationships between the spatial pattern of variables over time. Experiments were performed on a real-world climate dataset and the results were promising in that new patterns emerged and interesting relationships between variables were found.