5:15 PM - 6:45 PM
[STT35-P01] Detection and correction of phase unwrapping error for more efficient InSAR time series analysis
Keywords:InSAR time series analysis, ALOS-2
Geospatial Information Authority of Japan (GSI) conducts InSAR time series analysis with SBAS method using ALOS-2 observations accumulated over a long period of time (Kobayashi et al., 2018). In 2021, InSAR time series analysis was introduced to monitor the volcanic regions in Japan (Ichimura et al., 2021), and in 2022 we have started the nationwide analysis covering all of Japan except for some remote islands.
InSAR time series analysis is a method of statistically processing a large number of SAR interferograms to reduce noise such as tropospheric correction errors. InSAR time series analysis requires the SAR interferograms to be phase-unwrapped. Generally, it is difficult to stably unwrap a two-dimensional interferogram image that contains noise. At GSI, all of the thousands of SAR interferograms processed in InSAR time series analysis are checked, and unwrapping errors are corrected manually.
Advanced Land Observing Satellite-4 (ALOS-4), which will be launched in the near future, has four times the observation range, and will increase observation frequency by a factor of five over ALOS-2. High-frequency observation will improve the accuracy and temporal resolution of InSAR time series analysis. On the other hand, the large number of interferograms produced would make manual verification and correction of unwrapping errors difficult.
Therefore, we have been studying automated detection and correction of unwrapping errors to improve the efficiency of InSAR time series analysis.
Unwrapping errors can be classified into several patterns. In this study, among the several patterns, we examined a method for detecting and correcting unwrapping errors that occur in images including some islands. Fig. 1(a) shows an SAR interferogram observed in Goto Islands, Nagasaki Prefecture, Japan by ALOS-2, while Fig. 1(b) shows the unwrapped image of (a). In Fig. 1(a), a phase ramp is seen in the entire image. This does not indicate crustal deformation, but it is due to an observational noise, such as the effects of the ionosphere between the primary and secondary observation dates. Unwrapping errors are likely to occur when phase variations exceeding -π to π. In Fig. 1 (b), unwrapping errors occur in the circled region.
The method proposed in this presentation corrects for unwrapping errors, in which wrong integer offsets are added to each island, by assuming that the entire image rides on a slope that can be expressed by a simple polynomial surface. First, the island that occupies the largest number of pixels in the image is searched (in Fig. 1(b), the island indicated by the arrow (Nakadori Island)). Assuming that the slope seen at this island is a reflection of that of the entire interferogram, a polynomial surface (in this case, a first-order plane) fitting to the slope of this island is estimated. Next, the residuals between the island closest in distance in pixels from the Nakadori Island and the estimated plane are calculated, and the integer offset is determined so that the median of the residuals is in the range from -π to +π. The polynomial surface is then re-estimated by adding the island with the integer offset determined. By repeating this operation, the integer offsets of all islands in the image are determined. Fig. 1(c) shows the image for which integer offsets were determined well using this method, and Fig. 1(d) shows the slope of the entire image that was estimated. In Fig. 1(c), it can be seen that the unwrapping errors seen in Fig. 1(b) has been improved.
In this presentation, we would like to provide details on our proposed method for the correction of unwrapping errors.
Acknowledgments: ALOS-2 data were provided pursuant to a cooperative research contract between GSI and JAXA. The ownership of ALOS-2 data belongs to JAXA.
InSAR time series analysis is a method of statistically processing a large number of SAR interferograms to reduce noise such as tropospheric correction errors. InSAR time series analysis requires the SAR interferograms to be phase-unwrapped. Generally, it is difficult to stably unwrap a two-dimensional interferogram image that contains noise. At GSI, all of the thousands of SAR interferograms processed in InSAR time series analysis are checked, and unwrapping errors are corrected manually.
Advanced Land Observing Satellite-4 (ALOS-4), which will be launched in the near future, has four times the observation range, and will increase observation frequency by a factor of five over ALOS-2. High-frequency observation will improve the accuracy and temporal resolution of InSAR time series analysis. On the other hand, the large number of interferograms produced would make manual verification and correction of unwrapping errors difficult.
Therefore, we have been studying automated detection and correction of unwrapping errors to improve the efficiency of InSAR time series analysis.
Unwrapping errors can be classified into several patterns. In this study, among the several patterns, we examined a method for detecting and correcting unwrapping errors that occur in images including some islands. Fig. 1(a) shows an SAR interferogram observed in Goto Islands, Nagasaki Prefecture, Japan by ALOS-2, while Fig. 1(b) shows the unwrapped image of (a). In Fig. 1(a), a phase ramp is seen in the entire image. This does not indicate crustal deformation, but it is due to an observational noise, such as the effects of the ionosphere between the primary and secondary observation dates. Unwrapping errors are likely to occur when phase variations exceeding -π to π. In Fig. 1 (b), unwrapping errors occur in the circled region.
The method proposed in this presentation corrects for unwrapping errors, in which wrong integer offsets are added to each island, by assuming that the entire image rides on a slope that can be expressed by a simple polynomial surface. First, the island that occupies the largest number of pixels in the image is searched (in Fig. 1(b), the island indicated by the arrow (Nakadori Island)). Assuming that the slope seen at this island is a reflection of that of the entire interferogram, a polynomial surface (in this case, a first-order plane) fitting to the slope of this island is estimated. Next, the residuals between the island closest in distance in pixels from the Nakadori Island and the estimated plane are calculated, and the integer offset is determined so that the median of the residuals is in the range from -π to +π. The polynomial surface is then re-estimated by adding the island with the integer offset determined. By repeating this operation, the integer offsets of all islands in the image are determined. Fig. 1(c) shows the image for which integer offsets were determined well using this method, and Fig. 1(d) shows the slope of the entire image that was estimated. In Fig. 1(c), it can be seen that the unwrapping errors seen in Fig. 1(b) has been improved.
In this presentation, we would like to provide details on our proposed method for the correction of unwrapping errors.
Acknowledgments: ALOS-2 data were provided pursuant to a cooperative research contract between GSI and JAXA. The ownership of ALOS-2 data belongs to JAXA.