Modeling Viscosity of Volcanic Melts With Artificial Neural Networks

Abstract

The shear viscosity (η) of volcanic melts is of great importance for the transport dynamics of magmas and the eruptive styles of volcanoes (

Viscosity of anhydrous and hydrous peridotite melts

“…The addition of dissolved water leads to a marked and parallel reduction of both Tg and m, which can substantially affect flow behavior at high and intermediate viscosity. We compared our results with predictions from empirical models of volcanic melt viscosity: among long-established and more recent ones, the recently developed neural-network-based model (Langhammer et al, 2022) performed best. We also provide a calculator to derive the viscosity of peridotite melts as a function of temperature and dissolved water content.…”

“…The Hui and Zhang (2007) DSC measurements and literature data (Dingwell et al, 2004) on anhydrous and hydrous peridotite melts and the respective prediction using viscosity models. Considered models: GRD (Giordano et al, 2008), LDS (Langhammer et al, 2022) and HZ (Hui and Zhang, 2007). Error bars on measured data have been omitted for clarity.…”

“…Among those parameters that govern both magma transport from the reservoir to the volcanic vent and eruptive style, viscosity is arguably the most important (Cassidy et al, 2018;Di Genova et al, 2017a). Indeed, volcanologists have long pursued the experimental investigation of melt viscosity as a function of temperature and composition (e.g., Ishibashi and Sato, 2007;Kolzenburg et al, 2018;Whittington et al, 2009), as well as its subsequent numerical modelling by different approaches including classical linear regression and machine learning (Giordano et al, 2008;Hui and Zhang, 2007;Langhammer et al, 2022). The extent and quality of the dataset used to develop these models directly affect the reliability of their predictions, so that one should strive for a detailed experimental exploration of the largest possible compositional landscape.…”

“…Moreover, even when the required glassy samples can be synthesized, their limited stability during high-temperature measurements has been repeatedly demonstrated by postmortem measurements using Raman spectroscopy and high-resolution imaging (Di Genova et al, 2020b, 2017aKleest et al, 2020;Liebske et al, 2003a;Scarani et al, 2022): nanocrystallization of Fe-Ti-oxides can affect sample homogeneity and alter the composition of the residual melt, preventing derivation of the crystal-free viscosity. As a consequence, currently available viscosity models (Giordano et al, 2008;Hui and Zhang, 2007;Langhammer et al, 2022) have been trained in a relatively restricted volcanological domain of temperature and chemistry, with questionable applicability to compositional extremes such as ultrabasic melts and very H2O-rich matrices.…”

“…To overcome these limitations, we designed this work as a targeted investigation of melts that locate substantially outside the range of previously analyzed compositions, considering an unprecedented range of anhydrous and hydrous viscosities and testing the predictive performance of numerical viscosity models available in the literature (Giordano et al, 2008;Hui and Zhang, 2007;Langhammer et al, 2022). We therefore selected anhydrous and hydrous peridotite liquids: the viscosity of these materials was investigated only sporadically so far (Dingwell et al, 2004;Liebske et al, 2005) due to their extreme features, namely a SiO2 content < 50 wt% that leads to extreme structural depolymerization and a very strong tendency to undercooling-driven crystallization.…”