Name Surname
Melt inclusions snapshot of pre-eruptive magma reservoir conditions
Main Supervisor: Hélène Balcone Boissard (Sorbonne Universite Paris)
Co-Supervisors: Ferenc Borondics (SOLEIL) | Charles Le Losq (Institut de physique du globe de Paris)
Objectives
This PhD aims at leveraging machine learning to gather data on the concentration, partitioning and distribution of volatile elements in melt inclusions and crystal hosts. Melt inclusions are drops of magma trapped in crystals during their growth and are unique witnesses of deep processes in volcanic systems. Several methods allow gathering different data: Raman and Infrared spectroscopy allow quantification of water, CO2, SO2 and halogens (F, Cl, Br, I) concentrations, by SIMS, EPMA, and X-ray absorption spectroscopy data regarding the oxidation state of the melt inclusion. Those data are usually analyzed separately, and the final information is brought together by researchers. In this project, we propose to leverage machine learning for analyzing hyperspectral maps of melt inclusions. Infrared, Raman and X-ray absorption maps will be acquired at the SOLEIL synchrotron, with the help of the SMIS and LUCIA beamlines. Different algorithms will help process the spectroscopic data, and decipher the chemical mapping in the melt inclusions. Regression algorithms will link spectral features to concentrations of elements. Clustering and dimensionality reduction algorithms will help explore the hyperspectral dataset and leverage the data acquired with different methods.
Epected Results
- We will focus our effort on alkali-rich magmas such as time series eruptions of Vesuvius and Campi Flegrei;
- We will also focus on calk-alkaline magmas from Mt Pelée (subduction zone) which as repetitive plinian eruptions reactivating a priori the same superficial reservoir.