A Novel Spatio-temporal Clustering Approach by Process Similarity

Spatio-temporal data in earth science is usually of huge volume and high dimensionality. Clustering is usually preparation work for many applications in this field. Traditional clustering methods are of high complexity when applied to spatial-temporal data. Traditional methods neglect the changing process of the temporal data by treating data with consecutive timestamps independently and do not consider objects' spatial proximity which is important in earth science. An effective spatial-temporal tight clustering approach with domain knowledge is proposed for this field. The similarity measurement for the cluster method named Value- Process (VP) measurement estimates similarity of two objects from the view of their attributes value and the value changing process. The computation of the measurement adopts a filter-and-refinement strategy with a growing search window to lift the efficiency and guaranty the spatial proximity. Based on the VP similarity measurement, a tight clustering approach, which has a more strict cluster rule, is applied to the global climatic dataset and the promising result shows that it was an effective clustering method for the spatial-temporal data in earth science.