GNGTS 2017 - 36° Convegno Nazionale

GNGTS 2017 S essione 2.2 351 Hierarchical clustering methods are recently extended to both geographically referenced data and functional data in order to handle with the problem of classifying spatially correlated curves in groups of curves which are spatially homogeneous. In the more tradional setting, the spatial structure is taken in to account by mean of the variogram, that is the variance of the increments of the observed data, defined as function of the geographic distance; an efficient way to weigth dissimilarities between functions is the use of the trace-variogram (Giraldo et al., 2007) that is the variogram of the coefficients accounting for the temporal dynamics of the observed data. Waveforms clustering, based on cross-correlation measures between signals, may presents some limitations (see Adelfio et al., 2012) , so we refer to more recent contributes relating data- depth based clustering analysis. Statistical data depth is a robust technique providing an alternative way to find the “center” of multivariate data sets and is robust for clustering. Statistical depth functions provide an ordering of all points from the center outward in a multivariate setting, extending the concept of linear order induced in one-dimensional observations. But with the complication that for dimension greater then 2, there is no natural order. Fig. 1 - Initial allocation.