From machine learning a precious help to understand the Neapolitan volcanoes
Elaborate and classify the composition of the volcanic rocks of the Neapolitan area through artificial intelligence. This is the objective of the study Machine learning applied to rock geochemistry for predictive outcomes: The Neapolitan volcanic history case just published in the journal ‘Journal of Volcanology and Geothermal Research’. The research is the result of the work of two researchers from the italian Istituto Nazionale di Geofisica e Vulcanologia (INGV).
The work represents a new ‘starting point’ for the development of petrological analyzes using the databases already in the possession of the researchers.
“The chemical composition of the rocks”, explains Monica Piochi, researcher at the Vesuvian Observatory of the INGV, “consists of the content of numerous elements present in the rock such as silicon, calcium, potassium, strontium, lead, sulfur, arsenic, uranium, barium and son on. This chemical set can be almost constant or vary both within the single volcanic deposit and during the various eruptive events, in response to the specific dynamics of the magma reservoir”.
“Each Neapolitan eruption”, continues the researcher, “has produced deposits with its own chemical composition, so we can deduce the eruption and the characteristics of the magmatic supply reservoir; it is like tracing the identity of an individual from his somatic characteristics and from the set of blood chemistry parameters.
Due the small and large existing diversities and the numerous parameters that describe such diversities, the identification of rocks, as well as individuals, is a long and complex operation. However, knowing its typology is necessary for the knowledge of the volcano’s behavior and its impact on the territory as well as, similarly, recognizing an individual is useful for establishing, for example, his lifestyle and his state of health”.
“Machine learning”, explains Alessandro Pignatelli, “is a common tool in the scientific field and is acquiring ever greater usage in various fields of research, medical, economic, social, and there are attempts to apply it also in the petrological field”.
To evaluate the potential of artificial intelligence, the two researchers collected and grouped the enormous amount of chemical data present in the literature in a single database (54 variables for 9800 samples), first of all deducing the abundance of data for some eruptions and the shortage for others. In particular, the database is very extensive for Campi Flegrei and Vesuvius while it was lacking for the volcanic activities of Ischia and Procida. In addition, they looked for the optimal algorithm for the study’s objectives.
“For a correct evaluation” explains Alessandro Pignatelli, researcher at INGV, “we used different machine learning techniques and, for each, we evaluated the ability to correctly classify the sample”.
“The results of our study”, Monica Piochi and Alessandro Pignatelli add, “indicate that on the basis of the existing database it is possible to obtain a first, rapid classification of compositional data of Neapolitan volcanic rocks using artificial intelligence. This classification has the advantage of being quick and free from the operator’s discretion. The machine learning, in fact, has a capacity of about 98% to “center” the attribution of a rock of unknown origin – but still identified in the Neapolitan context – to one of the volcanoes, about 90% to the eruptive period and at least 70% to the eruptive formation.
In addition, the AI (artificial intelligence) has proved to be capable of “handling” petrological data quickly thanks to computational skills superior to those of a human being.
Our artificial intelligence application to the Neapolitan case creates the premise for fast and reliable analyzes for future data and for development of automatic checking systems on large dataset related to the whole Italian volcanism – if not at a global scale”.
“The attribution of a rock deposit to a certain eruptive event”, concludes the researcher, “is very useful information in defining the areal distribution of magmatic products and the magnitude of the eruption itself, the effects on the territory and on climate change as well as on the mobility of living species. In archeology, for example, it can be useful to determine the place of extraction of construction materials and commonly used materials, such as millstones, and to reconstruct commercial traffics”.