Independent Component Analysis (ICA) as a Superior Atmospheric Correction Method for InSAR Time Series

Abstract Satellites such as the European Space Agency's Sentinel‐1 constellation allow for the creation of unprecedented volumes of Interferometric Synthetic Aperture RaDAR data that contains both deformation and atmospheric signals. Correction methods have been developed to reduce these atmospheric signals, but they do not generally perform well on smaller‐scale studies of volcanoes. To address this, we have developed a novel atmospheric mitigation method termed LiCSAtmo that uses independent component analysis (ICA) to identify atmospheric signal patterns and subtract them from the time series. By using ICA, we are able to empirically estimate the spatial pattern and temporal behavior of atmospheric noise signals with no constraints other than that they are non‐Gaussian and statistically independent of other signals (such as deformation). Furthermore, as our algorithm can work in both space or time, we can mitigate a variety of signals that may not be non‐Gaussian and independent of other signals in one domain by working in the other. We applied our algorithm to Vesuvius, where displacement of up to mm/year is highly correlated with topography, and successfully corrected the apparent seasonal reversal of the long‐term subsidence caused by atmosphere. In a comparison with Global Navigation Satellite System‐derived measurements of deformation, our approach dramatically outperforms a state‐of‐the‐art correction method based on a weather model, reducing root‐mean‐square error by (0.006 m vs. 0.017 m). When these advances are used within our existing volcano monitoring algorithm, LiCSAlert, we detected subtle variations in deformation rate of mm/year.