[S-TT40] 合成開口レーダーとその応用

[STT40-04] 時系列InSARデータへの独立成分分析の適用による時系列DEM誤差補正：三宅島火山を例として

*重光 勇太朗¹、Gaddes Matthew²、Hooper Andrew²、田殿 武雄¹ (1.国立研究開発法人宇宙航空研究開発機構、2.リーズ大学)

キーワード：時系列干渉SAR解析、独立成分分析、DEM誤差

In general, one type of SAR satellite data is limited for the discussion of digital elevation model (DEM) error estimation in synthetic aperture radar interferometry (InSAR), and the Advanced Land Observing Satellite (ALOS) is the leading example. ALOS is an L-band SAR satellite operated by JAXA from 2006 to 2011. The accuracy of ALOS perpendicular baseline is about a few kilometres, so the cm-level DEM errors are included in the InSAR analysis. On the other hand, thanks to their sophisticated satellite orbit control technology, the perpendicular baseline values of ALOS-2 and Sentinel-1A, launched in 2014, have been greatly improved to an accuracy of about 500 m and 100 m, respectively. Therefore, in many previous studies, DEM errors have been considered negligible in time-series InSAR analysis using ALOS-2 and Sentinel-1. However, it is important to investigate the extent to which DEM error components are included in ALOS-2/Sentinel-1 InSAR data for more accurate surface deformation estimation and DEM generation. Here, we performed a time-series InSAR analysis using ALOS, ALOS-2 and Sentinel-1 data and checked whether DEM error components were included in each data set using independent component analysis. Independent component analysis is capable of separating different phase components contained in InSAR data without prior information. In this study, we used LiCSAlert (Gaddes et al., 2019) to calculate the independent components correlated with the DEM error components in the InSAR analysis data. We compared each extracted independent component with the perpendicular baseline distance time series (Liang et al., 2019) and the spatial distribution of the DEM error components using the method of Fatthai et al. (2013), and calculated the correlation coefficient for each. We also prepared three types of DEMs to remove the topography phase component in the time-series InSAR analysis of ALOS data, and confirmed the robustness of our method by checking whether the DEM errors contained in the time-series InSAR data were removed in each case.
Our results show that it is possible to extract DEM error components from the phase using independent component analysis for all ALOS, ALOS-2 and Sentinel-1 data. In particular, for ALOS it is possible to obtain different DEM error distributions for each of the three DEM types. However, the absolute values of the correlation coefficients, when compared in terms of temporal perpendicular baseline changes and spatial distribution of DEM errors, were 0.91, 0.87, 0.83, and 0.86, 0.85, 0.53, respectively, from highest to lowest. This is likely to be due to the different amount of DEM error components remaining in the InSAR data due to the different times at which each DEM was produced. In addition, the absolute value of the correlation coefficient for ALOS-2 was 0.94, while for Sentinel-1 it was around 0.5. We expect that the values of our correlation coefficients are affected by the formation of the time series network, masking and unwrapping processes in the InSAR analysis. We hope to improve the correlation coefficients by changing the InSAR analysis method in the future.