Interrelated impacts of seismic ground motion and topography on coseismic landslide occurrence using high-resolution displacement SAR data

Abstract

To date, a number of studies have been conducted to examine the relationship between seismic ground motion and coseismic landslides. However, the impacts of seismic ground motion on coseismic landslide occurrence are not fully understood owing to the poor spatial resolution of seismic ground motion data. Recently, seismic observation research has expanded with the use of satellite technology, as crustal deformation can be observed using pairs of SAR (synthetic aperture radar) satellite data. With this technique, obtaining information regarding the ground surface displacement induced by earthquakes is possible at a high spatial resolution, without the need for interpolation or extrapolation. In this study, we focus specifically on the interrelated impacts of seismic ground motion and topography on coseismic landslide occurrence, which has previously been difficult to detect. Using high-resolution ground surface displacement from SAR data, we examine these interrelated impacts in detail and assess coseismic landslide occurrence based on seismic ground motion and topography. Results show that the developed formula accurately reproduces coseismic landslide occurrence and that the impact behaviors of the two factors on landslide occurrence are different. Finally, based on the new formula, we suggest two different trends for the attenuation of seismic ground motion and topography related to coseismic landslide occurrence.

Appendix

Measurements of crustal deformation using SAR satellite

We used the spatial pattern of GSD caused by earthquakes based on pairs of SAR satellite data from the ALOS-2 satellite. The GSD data in three directions, up–down, north–south, and east–west used in this study were prepared by the GSI of Japan (Japan Society of Civil Engineers 2017), combining the following two calculation techniques.

The first is the DInSAR (differential SAR interferometry) technique, which calculates the amount of ground surface displacement using the phase contrast between pairs of reflections of microwaves obtained for the same area at different times (Massonnet and Feigl 1998). The accuracy of GSD measurements based on this process is assumed to be at the centimeter scale, as long as the spatial continuity of phases is preserved, i.e., phases have a high coherence. In contrast, there are cases where measurements are difficult for crustal deformation with a low coherence, owing to large-scale crustal deformation near a source fault (Kobayashi et al. 2009). Observations are also difficult for movements where displacement differs significantly at adjoining sites, such as for coseismic landslides where coherence cannot be maintained (Massonnet and Feigl 1998). To calculate GSD via the DInSAR technique for the data used in this study, pixel data on a 1.43 × 2.17 m grid were used. The data calculated using the DInSAR technique consisted of the component of displacement between the satellite and ground on the radar coordinates (i.e., the range offset component). The DInSAR technique often includes error owing to atmospheric phase delay and topographic components. To reduce this error, three processes are conducted. The error due to troposphere-related phase delay is often approximately 5–10 cm, but this error can be reduced to several centimeters by using the numerical weather model (Kobayashi 2016). The error due to ionosphere-related phase delay can be reduced from several tens of centimeters to a centimeter level by applying the split-spectrum method (Gomba et al. 2016). Moreover, the correction method using GEONET (GNSS Earth Observation Network System) data reduces the long spatial wavelength error induced by various other factors. This method reduces the error from several tens of centimeters to a centimeter level (Tobita et al. 2005; Kobayashi et al. 2011a, b). The phase change owing to the DEM error as the topographic component has almost no influence because the perpendicular baseline between satellite orbits, which produces a phase change proportional to elevation, is small enough. In this way, these errors are minimized to the greatest extent possible. Additionally, the influence of these errors is extremely small because this study focuses on deformation at a scale of tens of centimeters.

The second method is the pixel-offset technique, which is capable of calculating large-scale crustal deformation that is difficult to measure using the DInSAR technique. The pixel-offset technique is a method where crustal deformation is observed by measuring the residual offset of pixels in pairs of SAR amplitude images (Tobita et al. 2001). While the resulting accuracy of data obtained via this method is not as high as that for data obtained via the DInSAR technique, the measurement of phenomena is possible at a scale of several tens of centimeters (Kobayashi et al. 2009; Kobayashi et al. 2011a, b). Specifically, the displacement is evaluated by establishing a correlation window composed of multiple pixels, calculating the spatial correlation between corresponding correlation windows while their positions shift using two-period images, and finding the positions where the cross-correlation coefficient is the highest (Kobayashi et al. 2011a, b). For this study, the size of the correlation windows was 64 pixels (91.5 m × 138.8 m), with the displacement evaluated by shifting each window by 32 pixels (45.8 m × 69.4 m). The data measured by the pixel-offset technique consists of a component for the displacement between the satellite and ground on the radar coordinates (i.e., the range offset component) and a component for displacement that is parallel to the orbit of the satellite (i.e., the azimuth offset component). For the data used in this study, displacement data where the cross-correlation coefficient was calculated to be less than 0.2 were discarded.

Moreover, we examined the impact that deformation from the landslide itself has on the pixel-offset technique. In the case that coseismic landslides are sufficiently small for the correlation window, they only affect a few pixels within the window, suggesting that the influence of a landslide on the overall displacement in the correlation window should be relatively small. Landslides with sufficiently large movement can significantly disturb the ground surface in terms of the correlation window and can thus affect numerous pixels within the correlation window. Therefore, in these cases, surface displacement from landslides is considered to have an impact. However, if landslide activity is substantial enough to disturb the ground surface, the value of the cross-correlation coefficient is considered to be low, even in cases where the spatial correlation between corresponding correlation windows is high. Owing to the removal of data with poor correlation and large displacement, the probability that deformation due to coseismic landslides is reflected in surface displacement was considered to be limited.

We further demonstrate the specific displacement calculation process, which combines the above two analysis methods. First, the LOS component (line of sight direction) from the DInSAR technique and the LOS component and azimuth component (flight direction) from the pixel-offset technique were measured. Each displacement component was measured from two directions, i.e., northward and southward, along the tracks of the satellite. Next, the LOS component from DInSAR and pixel offset were composited at 32 pixels, i.e., the measurement size derived from the pixel offset. In addition, a total of four displacement components, i.e., the composited LOS components measured from two directions and the azimuth component from two directions, were transformed through the least-squares method into three-dimensional locations, followed by the calculation of the displacement in three directions, i.e., up–down, north–south, and east–west. Displacements in up–down and east–west directions have a centimeter-scale accuracy because they are mainly dependent on the LOS component from the result of DInSAR. Displacement in the north–south direction has an accuracy of several tens of centimeters because it is mainly dependent on the azimuth component from pixel offset. Finally, the data used in this study were fit to the pixel-offset grid size, which was converted to a decimal system grid size with a latitude of 0.0005 s (approx. 55 m) and longitude of 0.0005 s (approx. 46 m) as a rectangular plane from a radar coordinate system.

Comparisons of the GSD from the SAR satellite data and the results of emergent GNSS observations after the 2016 Kumamoto earthquake showed overall consistency (Geospatial Information Authority of Japan 2016). In addition, as the locally confirmed displacement was 2 m at maximum (Japan Society of Civil Engineers 2016), the maximum values were also consistent overall. As such, the overall trends in the GSD are consistent with actual crustal deformation, allowing the measurement of local displacement.