Finding Small Transient Deformations in Noisy InSAR Time Series

Thursday, 18 December 2014: 11:05 AM
Howard A Zebker, Stanford Univ, Stanford, CA, United States and Jingyi Chen, Stanford University, Stanford, CA, United States
InSAR is a radar remote sensing technique that retrieves crustal deformation to mm/cm accuracies at m-scale resolution over 100 km wide regions. Here we address the imaging and characterization of transient events that are too subtle to be easily detected in interferogram time series. We show that it is possible to attain mm-level accuracy in temporal histories using small baseline subset analysis (SBAS) with certain constraints on spatial continuity and temporal variation.

Radar interferogram accuracy is limited by changes in atmospheric phase delay due to variability in water vapor, leading to roughly 2 cm rms noise signals in the image. If a deformation source is continuous and produces the same signal over time, then we can “stack” many interferograms to visualize the signal. Since the atmospheric water vapor distribution changes over time, its noise averages out and the steady signal remains. The same approach will not work for transient events because a transient deformation is not steady with time and will not be reinforced when interferograms are averaged.

We illustrate our approach by studying the signature of a small earthquake and a slow slip event (SSE) that occurred over Kilauea volcano in Hawaii in 2010. We analyze data from the TerraSAR-X and Cosmo Skymed X-band radar systems, and show that we can image small events even when the size of the signal is smaller that the atmospheric noise. Our SBAS InSAR algorithm with constraints is suitable for extracting both transient and secular ground deformation on the order of millimeters in the presence of atmospheric noise on the order of centimeters. The small earthquake is readily imaged in both space and time, while for the 2010 SSE we obtain high spatial resolution displacement estimates as well as secular motion at Kilauea. We also develop an L1-norm based sparse reconstruction algorithm to detect transient events in very noisy InSAR time series for cases when the time of a possible transient event is not known. We apply the L1 algorithm to the 2010 Kilauea SSE and solve for the time of occurrence, confirming that the largest jump detected in the TerraSAR-X InSAR time series is temporally and spatially correlated with GPS observations.