A novel physics- and object detection-based approach for estimating volcanic SO2 degassing flux from satellite measurements of volcanic clouds

Abstract:

Long-term time series of SO2 degassing are considered critical elements of volcano monitoring and basic research into magmatic systems. Owing to their continuity, regularity, and global coverage, satellite SO2 data are ideal for such analyses, though many complexities are needed to match the data quality of ground-based, airborne, and UAV-based sensors. Methods to analyze SO2 flux from satellite data have advanced significantly to include rotation of plume data into a common wind direction for more unified analyses. This approach has two principal stages. First, a mean (wind-rotated) plume is made from a sequence of plume images with varied wind dispersal directions. Second, a theoretical SO2 distribution is fit to the mean plume, constraining geophysical parameters such as the mean SO2 flux and lifetime in the atmosphere.

Recent work has developed around the use of numerical weather prediction (NWP) winds to perform the plume rotation and again as a constrained parameter in the subsequent fitting. These works utilize a mixed empirical-physical model of the SO2 cloud to perform the fitting, sacrificing some of the clarity of a fully physics-informed fit. Because these approaches have used traditional SO2 data lacking altitude information, an altitude must be assumed to correctly interpret the SO2 data and NWP winds.

In adopting this general approach, we have made novel modifications to both stages which do not rely on prior knowledge of winds and therefore do not inherit errors associated with NWP. To perform the plume rotation, we modify a rudimentary computer vision algorithm designed for object detection in medical imaging to detect plume-like objects in gridded SO2 data. We use newly developed hyperspectral infrared SO2 data from the CrIS instrument aboard JPSS series satellites which includes SO2 altitude, hence height estimation errors are greatly mitigated. In the second stage we employ a fully physics-based fit from the theory of point source SO2 dispersal to estimate the SO2 flux and lifetime.

This process is repeated over many intervals of time, generating long-term time series of degassing flux and other variables which compare favorably with previous methods. Lastly, we explore the limits of time series resolution by evaluating changes in signal-to-noise ratio induced by adopting a multi-sensor approach.