Neural forecasting of seismicity and ground displacements in different volcanic areas

Volcanic events have been, up to now, traditionally predicted through deterministic or probabilistic linear methods. Here a new non-linear approach is carried out, based on artificial neural networks (ANN), a non-linear physical mathematical model to interconnect data emulating animal brain behaviour. Being an artificial intelligence system, this model is able, once both the observed data records and starting values of the network’s parameters have been entered, to self-guide itself and supply the best outputs with respect to the input parameters. On varying the starting values, the exit values change too, the degree of improvement being measurable quantitatively. This model has been applied to four different volcanoes (Vesuvius, Phlegraean Fields, Etna, Hawaii) with different volcanic characteristics, in order to measure the effectiveness of the method as a general one. For this purpose, data regarding seismicity or ground displacements (vertical, radial, tangential components) were processed and predicted by means of ANN. The results are encouraging and, in many cases, in very fair agreement with the observed data.