Disruptive influences of residual noise, network configuration and data gaps on InSAR-derived land motion rates using the SBAS technique

Abstract

The interferometric synthetic aperture radar (InSAR) small baseline subset (SBAS) technique can be applied to land with varying deformation magnitudes ranging from mm/yr to tens of cm/yr. SBAS defines a network of interferograms that is limited by temporal and spatial baseline thresholds that are often applied arbitrarily, or in apparently subjective ways in the literature. We use simulated SAR data to assess (1) the influence of residual noise and SBAS network configuration on InSAR-derived deformation rates, and (2) how the number of interferograms and data gaps in the time series may further impact the estimated rates. This leads us to an approach for defining a SBAS network based on geodetic reliability theory represented by the redundancy number (r-number). Simulated InSAR datasets are generated with three subsidence signals of linear rates plus sinusoidal annual amplitudes of −2 mm/yr plus 2 mm, −20 mm/yr plus 5 mm and −100 mm/yr plus 10 mm, contaminated by Gaussian residual noise bounded within [−2;+2] mm, [−5;+5] mm and [−10;+10] mm, corresponding to standard deviations of approximately 0.5 mm, 1.5 mm and 3.0 mm, respectively. The influence of data gaps is investigated through simulations with percentages of missing data ranging from 5% to 50% that are selected (1) randomly across the 4-year time series, and (2) for three-month windows to represent the northern winter season where snow cover may cause decorrelation. These simulations show that small deformation rates are most adversely affected by residual noise. In some extreme cases, the recovered trends can be contrary to the signal (i.e., indicating uplift when there is simulated subsidence). We demonstrate through simulations that the r-number can be used to pre-determine the reliability of SBAS network design, indicating the r-values between ~0.8 and ~0.9 are optimal. r-numbers less than ~0.3 can deliver erroneous rates in the presence of noise commensurate with the magnitude of deformation. Finally, the influence of data gaps is not as significant compared to other factors such as a change in the number of interferograms used, although the blocks of “winter” gaps in the SBAS network show a larger effect on the rates than gaps at random intervals across the simulated time series.