Autonomous extraction of millimeter-scale deformation in InSAR time series using deep learning.

Area of Science:

• * Geophysics
• * Remote Sensing
• * Tectonics

Background:

• * Understanding tectonic faulting requires characterizing slip behaviors, including slow and fast earthquakes.
• * Interferometric Synthetic Aperture Radar (InSAR) measures ground deformation globally but is hindered by atmospheric noise.
• * Current InSAR analysis demands expert interpretation and fault knowledge, limiting global deformation studies.

Purpose of the Study:

• * To develop an automated method for extracting ground deformation signals from noisy InSAR time series.
• * To enable global investigations of tectonic fault dynamics without prior knowledge of fault location or behavior.
• * To apply the method to identify slow earthquakes and volcanic deformation.

Main Methods:

• * Implemented a deep auto-encoder architecture to distinguish ground deformation from noise in InSAR data.
• * Applied the method to InSAR data from the North Anatolian Fault.
• * Tested the approach on deformation data from the Coso geothermal field, California.

Main Results:

• * The deep auto-encoder autonomously extracted deformation signals with 2 mm precision.
• * Revealed a slow earthquake on the North Anatolian Fault twice as large as previously known.
• * Demonstrated the method's applicability to inflation/deflation-induced deformation.

Conclusions:

• * Automated InSAR analysis using deep learning can overcome atmospheric noise limitations.
• * This approach facilitates global-scale characterization of fault slip and volcanic deformation.
• * The method enhances our understanding of earthquake physics and geothermal processes.