Name Surname
DeepEruptive: Artificial Intelligence for Eruptive Parameter Estimations
Main Supervisor: Paola Cinnella (Sorbonne Universite Paris)
Co-Supervisors: Chiara Paola Montagna (INGV), Guillaume Carazzo (Institut de physique du globe de Paris)
Objectives
The project aims at (i) producing the first meta-model that couples petrological and rheological models of magmas to constrain fluid dynamics simulations (e.g., magma modelling before attaining the fragmentation threshold or gas flow modelling after magma fragmentation) and solve the forward problem, (ii) generating hundreds of synthetic datasets that then will be provided to a machine learning algorithm aimed at retrieving source and conduit parameters from the observation of tephra maps. We will apply this to eruptions at Mt Pelée (Martinique, France), an active volcano monitored by IPGP. This volcano typically shows transitions between plinian eruptions, pyroclastic flows and lava dome-forming eruptions, making it a perfect case study. The petrological model will be built ad hoc, using existing data on andesitic-to-rhyolitic compositions typical of the Mt Pelée.
Epected Results
- The development of new ML-assisted computational flow dynamic (CFD) simulations that combine petrological and rheological knowledge to solve the forward problem;
- The development of a machine learning algorithm aimed at retrieving source and conduit parameters from the observation of tephra maps;
- Use the results of the modelling to improve the hazard assessment and possibly communicate risk mitigation strategies to stakeholders and decision-makers.