11:15 AM - 11:30 AM
[STT35-03] A Practical Recipe for Applying the Split Spectrum Method to L-band Interferograms
Keywords:InSAR, SSM, Ionosphreic Compensation
L-band InSAR images play a crucial role in detecting crustal deformation in densely vegetated areas. However, they are significantly impacted by ionospheric disturbances, hindering the accurate estimation of small amplitude and long-wavelength surface displacements.
The Split Spectrum Method (SSM, e.g., Gomba et al. 2016) provides a solution for estimating ionospheric effects by leveraging the dispersion nature of radio waves. Utilizing the SSM involves extracting two separate frequency bands from the original SAR data. However, interferograms of each sub-band often become noisy when acquired with a narrow bandwidth mode, leading to challenges in phase correction. In our pursuit of a more robust application of the SSM under severe conditions, we propose a practical workflow based on a series of systematic experiments. The key considerations include (i) outlier removal and (ii) interferogram smoothing.
We applied the SSM to ALOS-2 ScanSAR images (bandwidth: 14 MHz) acquired after the 2015 Gorkha Earthquake in the Himalayas, which exhibited post-seismic deformation. The details of our approach are outlined below:
(i) Outlier Removal: We compared two methods. (1) Introducing a polynomial as a reference surface, we removed pixels deviating beyond a fixed threshold. We used a coherence-weighted least-squares estimate of the polynomial coefficients. (2) Employing a moving median, we defined a reference surface and removed pixels deviating beyond a threshold calculated from local coherence (Gomba et al., 2016). Method (1) proved more robust, retaining many pixels while successfully eliminating outliers.
(ii) Smoothing (Filtering): In instances where the spatial gradient of the interferometric phase is large, the coherence-weighted Gaussian filter (Gomba et al., 2016) introduces artifacts along the image edges. To mitigate these artifacts, we took the following steps: (1) Subtract the polynomial surface estimated in (i) from the original data, (2) apply the coherence-weighted Gaussian filter to the residual, and (3) add the polynomial removed in (1) back to the smoothed residual in (2).
Regarding filtering (ii), there exists a trade-off relationship between the number of multi-looks used for ionospheric estimation and the filter size. A systematic study of this relationship reveals that the error is minimized when the product of the number of multi-looks and the filter size takes a certain value, except when the number of multi-looks is extremely large or small.
Most of the corrected images exhibit post-seismic displacements consistent with the results of Wang and Fialko (2018), who approximated the ionospheric phase using polynomial functions.
The Split Spectrum Method (SSM, e.g., Gomba et al. 2016) provides a solution for estimating ionospheric effects by leveraging the dispersion nature of radio waves. Utilizing the SSM involves extracting two separate frequency bands from the original SAR data. However, interferograms of each sub-band often become noisy when acquired with a narrow bandwidth mode, leading to challenges in phase correction. In our pursuit of a more robust application of the SSM under severe conditions, we propose a practical workflow based on a series of systematic experiments. The key considerations include (i) outlier removal and (ii) interferogram smoothing.
We applied the SSM to ALOS-2 ScanSAR images (bandwidth: 14 MHz) acquired after the 2015 Gorkha Earthquake in the Himalayas, which exhibited post-seismic deformation. The details of our approach are outlined below:
(i) Outlier Removal: We compared two methods. (1) Introducing a polynomial as a reference surface, we removed pixels deviating beyond a fixed threshold. We used a coherence-weighted least-squares estimate of the polynomial coefficients. (2) Employing a moving median, we defined a reference surface and removed pixels deviating beyond a threshold calculated from local coherence (Gomba et al., 2016). Method (1) proved more robust, retaining many pixels while successfully eliminating outliers.
(ii) Smoothing (Filtering): In instances where the spatial gradient of the interferometric phase is large, the coherence-weighted Gaussian filter (Gomba et al., 2016) introduces artifacts along the image edges. To mitigate these artifacts, we took the following steps: (1) Subtract the polynomial surface estimated in (i) from the original data, (2) apply the coherence-weighted Gaussian filter to the residual, and (3) add the polynomial removed in (1) back to the smoothed residual in (2).
Regarding filtering (ii), there exists a trade-off relationship between the number of multi-looks used for ionospheric estimation and the filter size. A systematic study of this relationship reveals that the error is minimized when the product of the number of multi-looks and the filter size takes a certain value, except when the number of multi-looks is extremely large or small.
Most of the corrected images exhibit post-seismic displacements consistent with the results of Wang and Fialko (2018), who approximated the ionospheric phase using polynomial functions.