Physics-informed machine-learning models of multi-phase magma flow between reservoirs

点击翻译按钮获取中文摘要

Melt processing in volumetrically dominant mush domains beneath volcanoes leads to the formation of multiple magma reservoirs in the middle and upper crust; transport of magma between these reservoirs can occur through conduits or porous flow, leading to a deformation signal at the surface. While there exist some models of this flow behaviour, they generally fail to account for the multi-phase nature of the magma flow. Recent progress in Computer Science has led to the development of Physics-Informed Machine Learning (PI-ML) approaches, which can be applied to an increasing range of problems. In this project we will apply a PI-ML approach to model multi-phase flow within magma mush systems. Initially we will use simulated data from FEM modelling to train the PI-ML models. Subsequently, we will use real deformation data measured at volcanoes using radar interferometry (InSAR) and global navigation satellite systems (GNSS), to constrain them further. Objectives1. Develop a single-phase PI-ML model for magma flow between reservoirs;2. Integrate multi-phase flow into the model;3. Use the model to constrain magma plumbing system behaviour beneath volcanoes undergoing deformation;4. Publish three peer-reviewed journal articles.