Comparing the effects of Euclidean distance matching and dynamic time warping in the clustering of COVID-19 evolution: a spatiotemporal analysis of COVID-19 dynamics across Europe

A comprehensive understanding of COVID-19’s spatiotemporal progression is crucial for devising effective containment strategies, yet existing mapping approaches often fail to capture the pandemic’s multifaceted evolution. This study addresses that gap by focusing on how well different time-series clustering methods handle the complexities of misalignment in outbreak trajectories. Specifically, we investigate the spatiotemporal dynamics of COVID-19 cases across NUTS 2 regions in Europe during the second pandemic wave in the winter of 2020–2021. We employ time series clustering using Euclidean Distance Matching (EDM) and Dynamic Time Warping (DTW) to identify distinct patterns of pandemic progression. The hierarchical clustering results, visualized through heatmap dendrograms, chorochromatic cluster maps, and mean time series plots, reveal heterogeneous spatiotemporal patterns at different levels of clustering granularity. DTW outperforms EDM in capturing the temporal dynamics, yielding better-defined clusters in terms of temporal similarity. While DTW generally shows higher spatial contiguity values, EDM maintains statistically significant spatial coherence across clusters, especially at higher numbers of clusters. We discuss the trade-offs between optimizing for temporal similarity and spatial contiguity, and their implications for understanding the spatiotemporal dynamics of COVID-19. The findings highlight the importance of considering both temporal and spatial aspects when analyzing the spread of infectious diseases.