Skip to content

WITHDRAWN -Training deep learning networks with models integrating complex rheologies in the magmatic system: An advance to forecast time series models of volcanic deformation

Camila Novoa Lizama, Andrew Hooper, Matthew Gaddes, Milan Lazecky, Shailza Sharma, Gopal Phartiyal, David Hogg, Susanna Ebmeier

Abstract

Over the past three decades, InSAR has revolutionized volcano monitoring, offering unmatched insights into global volcanic behaviors. Interferograms reveal the large variations in magmatic reservoir geometries, while time series data illuminate magma dynamics and complex subsurface processes occurring at depth. Despite these advances, the diversity of volcanic behaviors poses challenges for forecasting modeling. Recent machine learning and deep learning techniques have shown promise in identifying and localizing volcanic deformation in interferograms and have been useful to create alerts of volcano activity based on temporal or spatial changes observed in InSAR time series. However, these methods have not yet been applied to forecast the temporal evolution of surface deformation in volcanoes.  This study addresses this gap by integrating global observations of volcanic deformation with numerical simulations to train deep learning networks capable of forecasting volcanic uplift episodes. Using a decade of Sentinel-1 observations, the study identifies global patterns in the temporal evolution of volcanic uplift. Building on recent research highlighting the mechanical complexity of magmatic systems, it incorporates advanced physical models that simulate realistic deformation scenarios. These models account for viscoelastic and poroelastic processes to reflect the influence of high-temperature rock and fluid behaviors on volcanic deformation.  By combining InSAR data, dynamic modeling, and deep learning, the study develops a robust framework for forecasting volcanic deformation. This approach not only advances our understanding of volcanic systems but also enhances global volcanic hazard forecasting. Bridging monitoring and alert systems, it offers a transformative tool to improve safety and preparedness for communities worldwide.