A Multivariate Probabilistic Graphical Model for Real-Time Volcano Monitoring on Mt. Etna

Real-time assessment of the state of a volcano plays a key role for civil protection purposes. Unfortunately, because of the coupling of highly non-linear and partially-known complex volcanic processes, and the intrinsic uncertainties in measured parameters, the state of a volcano needs to be expressed in probabilistic terms, thus making any rapid assessment sometimes impractical. With the aim of aiding on-duty personnel in volcano monitoring roles, we present an expert system approach to automatically estimate the ongoing state of a volcano from all available measurements. The system consists of a probabilistic model that encodes the conditional dependencies between measurements and volcanic states in a directed acyclic graph and renders an estimation of the probability distribution of the feasible volcanic states. We test the model with Mt. Etna (Italy) as a case study by considering a long record of multivariate data. Results indicate that the proposed model is effective for early warning and has considerable potential for decision making purposes.