Comovement in Geo-referenced Time Series: A Copula-Based Approach for Clustering

Time series clustering plays a crucial role in managing and extracting knowledge from the vast and complex Earth Observation (EO) datasets such as satellite-derived temperatures, precipitation levels, or soil-related variables. This study explores copula-based clustering techniques that focus on temporal dependence structures among time series, rather than their marginal behavior, to detect patterns of comovement in environmental variables. Applied to summer maximum temperatures in Italy, the approach reveals spatially coherent clusters that reflect underlying climatic regimes. However, when applied to monthly maximum precipitation data, clustering based solely on temporal dependence yields fragmented and geographically inconsistent results. To address this, we introduce a method that incorporates spatial proximity via soft constraints, combining temporal and spatial-based dissimilarities through a tunable mixing parameter. Our results demonstrate that including spatial information can significantly improve cluster coherence and interpretability, particularly for variables with strong geographic variability. Applications are based on EO data from the Copernicus Climate Data Store.