Abstract

Complex patterns in multivariate spatiotemporal data are often challenging to decipher due to the intricate relationships that exist across both space and time. Traditional methods often simplify the problem by viewing each location or time point independently, transforming the data into either a series of time graphs or a collection of maps. However, these approaches can result in missing out on the important connections between data points that are geographically and temporally close. To address this, we propose a framework utilizing geographically-constrained spectral clustering, enhanced by Delaunay triangulation and spatial continuity rules. By avoiding the assumption of stationarity across space or time, our approach dynamically segments the data into connected regions and continuous time intervals. This method not only captures complex spatiotemporal relationships more effectively but also helps to highlight unusual patterns or anomalies. The integration of these segments may offer a clearer understanding of multivariate spatiotemporal data, leading to deeper insights.