Automatic Detection of Volcanic Unrest Using Blind Source Separation with a Minimum Spanning Tree Based Stability Analysis

Abstract

Repeated synthetic aperture radar (SAR) acquisitions can be utilized to produce measurements of ground deformations and associated geohazards, such as it can be used to detect signs of volcanic unrest. Existing time series algorithms like permanent scatterer analysis and small baseline subset are computationally demanding and cannot be applied in near real time to detect subtle, transient, and precursory deformations. To overcome this problem, we have adapted a minimum spanning tree based spatial independent component analysis method to automatically detect sources related to volcanic unrest from a time series of differential interferograms. For a synthetic dataset, we first utilize the algorithm's capability to isolate signals of geophysical interest from atmospheric artifacts, topography, and other noise signals, before monitoring the evolution of these signals through time in order to detect the onset of a period of volcanic unrest, in near real time. In this article, we first demonstrate our approach on synthetic datasets having different signal strengths and temporal complexities. Second, we demonstrate our approach on a couple of real datasets, one acquired in 2017-2019 over the Colima volcano, Mexico, showing the occurrence of previously unrecognized short-term deformation events and the other over Mt. Thorbjorn in Iceland acquired over 2020. This shows the strength of the deep learning application to differential interferometric SAR measurements, and highlights that deformation events occurring without eruptions, which may have previously been undetected.