Toward Systematic Modeling of Volcano Deformation Sources Using Automatically-Generated InSAR Products

Abstract

Volcano deformation measured through Interferometric Synthetic Aperture Radar (InSAR) is ideal for volcano monitoring in many regions due to its global coverage, characteristic spatio-temporal patterns, and modeling insights. Routinely acquired and processed Sentinel-1 InSAR datacubes provide the first opportunity to systematically catalog, model and compare volcano deformation globally. Here, we present a framework (GBIS-BULK) to systematically pre-process and model volcano deformation signals, designed to be applied to routinely processed InSAR data sets. This requires a robust (semi-) automated approach to estimate signal locations and footprints for effective pre-processing and modeling. Our approach combines (a) filtering and clustering to locate the signal center; (b) noise reduction using Independent Component Analysis (ICA); and (c) image classification using Otsu thresholding to delimit the signal footprint. We invert for the best-fit point source model using constraints from existing global volcano deformation catalogs. First, we examine the influence of downsampling schemes, image noise and coherence using synthetic interferograms, showing nested-uniform downsampling is more suited to automated processing than quadtree methods which typically require manual tuning. Then, we validate the approach using Sentinel-1 deformation images from the East African Rift System (EARS). The pre-processing steps reasonably locate the signal at 15/16 of the EARS volcanoes, and the signal footprint at 14/16. ICA reduces or approximately maintains the image RMS in all cases. Our systematic point source estimates showed consistency when directly compared with previous (bespoke) modeling studies. This approach has the potential to be integrated with existing toolkits for routinely processing and analyzing Sentinel-1 InSAR data and hence applied globally.