Sequence analysis of local indicators of spatio-temporal association for evolutionary pattern discovery

Abstract

The Local Indicators of Spatial Association (LISA) is one of the most widely used methods for identifying local patterns of spatial association in geographical elements. However, the dynamic trends of spatial-temporal (S-T) autocorrelation remain poorly understood, yet capturing these patterns is essential for analyzing the evolution of spatial processes. To fill the gap, we propose a novel S-T LISA methodology to automatically discover co-occurrences LISA subsequences over time by incorporating sequence analysis techniques. First, we extend the classical LISA to a dynamic context, and clarify the definition, properties, and classification of S-T LISA sequences. Second, we adopt an enhanced Hamming distance to quantify the similarity of LISA sequences, followed by hierarchical clustering to group similar LISA sequences. Next, an improved FP-Growth algorithm is applied to identify frequent patterns. Finally, we conduct experiments using grid-scale social media check-in records and city-scale carbon emission data to discover significant evolutionary patterns. The results verified the applicability of the proposed method in both human and physical geography. The proposed approach outperforms traditional S-T cube methods in its ability to automatically capture dynamic, complex, and transient S-T association trends as well as irregular outliers. The integration of sequence analysis with LISA statistics presented in this article provides an effective framework for identifying evolutionary patterns of S-T association.

Keywords

• Spatial-temporal autocorrelation
• local indicators of spatial association (LISA)
• dynamic evolutionary patterns
• sequence analysis
• hierarchical clustering
• frequent pattern mining