Quantifying the ground displacement’s acceleration by using the Failure Forecast Method during the ongoing unrest of Vulcano (Italy) in 2021-2022.

Heraklion, Greece

We present a new use of the well-known Failure Forecast Method (FFM) to track the evolution of a volcanic unrest by modeling the acceleration of monitoring data. In particular, we analyze the temporal rates of GPS data collected by INGV network on Vulcano Island in 2021-2022. The FFM interprets monitoring signals, typically ground displacement or seismicity, as possible eruptive precursors, and provides quantitative forecasts through a nonlinear regression of their temporal rate X: dX/dt=AXα. We make the assumption that the nonlinear trend of the signals observed in the previous weeks will continue in the future, and accelerate in the same way as brittle materials subject to a constant stress while approaching their rupture. Therefore, under this assumption, the FFM calculates a time limit for the continuation of the observed nonlinear acceleration, called failure time. Our forecasts are expressed in terms of the time left, i.e. the waiting time before the volcanic system of Vulcano reaches a potentially critical state. We calculate the mean, the 5th and 95th percentile values of the estimate, because of the uncertainty affecting the model parameters A and α. However, the system, during its future evolution, can either strengthen its accelerating trend and shorten this time limit, or weaken the acceleration and extend it. For this reason, we performed a daily retrospective analysis of the FFM estimates in the last year. In this way, the method tracks the potential waiting time and highlights the most critical phases and their temporal evolution. During the acme of the unrest crisis, in mid-October 2021, the minimum waiting time estimated was of the order of 10 days for the station that registered the largest displacement. We compared the horizontal and vertical displacements, and the areal dilatation of the ground, different GPS stations and time intervals of the nonlinear regression.