The estimation of latent temporal patterns in multivariate geolocated time series

When dealing with a large number of time series observed over extended geographic regions it is useful to identify groups of temporally coherent time series. This allows division of the regions into sub-areas which can then be studied by implementing local instead of global space-time models. In this paper, each coherent group (cluster) is characterized by a latent temporal pattern common to all the time series of the cluster. The estimation of both the number of clusters and the cluster membership is obtained using a novel model-based clustering approach while model estimation is carried out by means of the Expectation Maximization algorithm. The approach is used to cluster NO2 concentration time series for 2009 at the European level.