Graph-based spatial pattern multi-type change detection

Abstract

Long-term spatial pattern analysis becomes an important and effective way of monitoring the interactions between land covers and associated evolutions over time, which removes disturbance and uncertainty and has considerable benefits for dynamic, cyclical geographical processes. The time series images with time–space continuity hold unique advantages for tracing spatial pattern evolution and discovering the evolutionary regulations and mechanisms underlying in history archive. However, current research is inadequate in characterizing spatial patterns with spatial relations and is limited in change types mining from evolving spatial patterns. This study hence proposed a novel algorithm for detecting multi-type changes in spatial patterns (MTCD) with graph change analysis. Specially, 1) we designed a frequent graph cell extraction method to construct the spatial pattern evolution graph (SPEG) based on time-series land cover images; 2) a graph coding method was developed to represent SPEG digitally and convert it into the graph-encoded time series; 3) a graph change mining method was proposed to detect the changing types of graph-encoded time series for stable, cyclic, and dominant change. The Landsat land cover time series of Datong, Shanxi Province, China, was adopted to validate the MTCD algorithm. Results show our algorithm's effectiveness in simultaneously capturing three typical types of spatial pattern change at different spatial scales. The MTCD provides an essential step toward analyzing changing geographical patterns and graph structure.