Functional clustering by smoothing quantile regression

This work presents a functional clustering procedure applied to environmental time series of a physical parameter (the chlorophyll type A concentration) in the coastal area of the Adriatic Sea. The data for the classification analysis is formed by glob-colours data during the period 2002–2012 (monthly values, 11 calendar years) provided by the ACRI server (http://hermes.acri.fr/) using satellite data source combining information of MERIS, Seaways and MODIS optical sensors. The choice of a basis implies the type of features of the series that are to be enhanced or hidden in the representation. Our proposal combines time series interpolation with smoothing quantile splines and the agglomerative clustering algorithm, such as partitioning around medoids technique. Our final purpose is to obtain a classification of the coastal areas in to homogeneous zones in order to select areas at high impact of chlorophyll type A concentrations. The analysis was performed by R software. This approach permits to take into account the quantile of interest and to calculate a more robust clustering procedure respect to other classical methods