Identification of Spatio-Temporal Outliers through Minimum Spanning Tree

The paper presents a technique to discover spatio-temporal outliers from meteorological data. The main advantage of this work is the early detection of events of particular intensity, like cloudbursts that can cause damages to people and things. We considered as input data the temporal evolution of meteorological values at each sampling station and we characterised the data with a graph where nodes are the sampling station and the links are given according information measurements that take into account both the mutual information contents of neighbours nodes. The algorithm is based on the search of the minimum spanning tree that allow to retain the strongest connections among nodes while minimising a cost function at a global optimisation.