De Bernardo, Andrea

Algorithms based on artificial intelligence (AI) have had a strong development in recent years in different research fields of earth science such as seismology and volcanology. In particular, they have been applied to the study of the volcanic eruptive products of the recent activity of Mount Etna volcano. This work presents an application of the self-organizing map (SOM) neural networks to perform a clustering analysis on petrographic patterns of rocks of Somma–Vesuvius and Campi Flegrei volcanoes, in the Neapolitan area. The goal is to highlight possible affinity between the magmatic reservoirs of these two volcanic complexes. The SOM is known for its ability to cluster data by using intrinsic similarity measures without any previous information about their distribution. Moreover, it allows an easy understandable data visualization by using a two-dimensional map. The SOM has been tested on a geochemical dataset of 271 samples, consisting of 134 samples of Campi Flegrei eruptions (named CF), 24 samples of Somma–Vesuvius effusive eruptions (VF), 73 samples of Somma–Vesuvius explosive eruptions (VX), and finally 40 samples of “foreign” eruptions (ET), included to verify the neural net classification capability. After a pre-processing phase, applied to have a more appropriate data representation as input for the SOM, each sample has been encoded through a vector of 23 features, containing information about major bulk components, trace elements, and Sr isotopic ratio. The resulting SOM identifies three main clusters, and in particular, the foreign patterns (ET) are well separated from the other ones being mainly grouped in a single node. In conclusion, the obtained results suggest the ability of SOM neural network to associate volcanic rock suites on the basis of their geochemical imprint and can be consistent with the hypothesis that there might be a common magma source beneath the whole Neapolitan area.