Skip to content

Optimizing lava flow simulations using a Markov Chain Monte Carlo approach

Francesco Zuccarello1, Giuseppe Bilotta1, Flavio Cannavò1, Annalisa Cappello1, Gaetana Ganci1

Abstract

Effusive eruptions in basaltic volcanoes pose significant risks especially to the human infrastructure. Accurately predicting lava flow dynamics and their likely paths remains a challenge, due to the numerous factors that influence their motion and emplacement. This study introduces a novel application of the Metropolis-Hastings algorithm, a Bayesian approach based on Markov Chain Monte Carlo (MCMC) method, for the real-time modeling of lava flows. The approach optimizes the prediction of lava flow behaviour by exploring the parameter space and deriving posterior distributions, based on multi-source satellite data and observed characteristics of the eruption. The method relies on the GPUFLOW model, a cellular automata-based tool, developed at INGV-Catania to simulate lava flow dynamics. Key input parameters include lava physical properties, topography, vent location(s), and TADR (time-averaged discharge rate) data. The Metropolis-Hastings algorithm refines the simulation over time, adjusting for uncertainties, such as the position of active vents, the distribution of TADR values, and the lava's water content, which influences its viscosity. The optimization process integrates satellite data, minimizing discrepancies between the simulated and observed flow field. This methodology has been applied to the effusive eruption that occurred at Mt. Etna (Italy) between 27 November 2022 and 6 February 2023, which produced a complex lava field with an extensive lava tube system. The results highlight the potential of the use of Bayesian approach to make more accurate predictions of lava flow paths and their extents, helping us to refine eruption monitoring, hazard assessment, and planning of volcanic risk management strategies.