A clustering algorithm for spatially dependent functional data

We propose a new algorithm for unsupervised classification of functional data – indexed by the sites of a spatial lattice – that exploits spatial dependence by repeatedly generating random connectivity maps and by clustering, at each iteration, local representatives of neighbouring functional data. The algorithm supports different implementations – both with respect to the treatment of spatial dependence and to the employed clustering technique – consistent with different contexts arising in applications. The performance of the algorithm is tested on synthetic data. We then illustrate a case study where irradiance data are analyzed to investigate the exploitability of solar energy in different areas of the planet.