A multivariate hierarchical regionalization method to discovering spatiotemporal patterns

Abstract

In GIScience, the regionalization method is widely used for geographical data mining, spatiotemporal pattern discovery, and regional studies. An ideal regionalization method should consider spatial contiguity, temporal contiguity, and attribute similarity. Existing regionalization approaches mostly focus on spatial contiguity and attribute similarity while ignoring the temporal contiguity characteristics of geographic phenomena. We propose a multivariate spatiotemporal regionalization (STR) method that considers spatiotemporal contiguity and attribute similarity. We design a bottom – up unsupervised multivariate hierarchical clustering algorithm with constraints using spatiotemporal proximity rules, enabling the automatic regionalization of spatiotemporal data. To test the performance of the STR method, we applied it to a synthetic dataset and a real-world dataset (Chinese air pollutant data) and achieved ideal results. Such a method offers a spatiotemporal perspective to address regionalization or clustering problems, potentially supporting other applications in spatiotemporal data analysis, remote sensing, urban planning, and social science.

Keywords

• regionalization
• spatiotemporal cube
• geographical boundary model
• spatial clustering
• air pollution