Detection and interpretation of local surface deformation from the 2018 Hokkaido Eastern Iburi Earthquake using ALOS-2 SAR data

Mapping linear surface displacements using SAR interferograms

The Geospatial Information Authority of Japan (GSI) reports detailed ground surface deformation through analyses of ALOS-2 interferograms for disaster response and mitigation purposes. In this study, we processed ALOS-2 data using GSISAR software as described in Fujiwara et al. (2016).

Fujiwara et al. (2016) called the lineaments ‘linear surface ruptures’, but as it is difficult to determine whether actual ruptures existed, here we refer to these features as ‘linear surface displacements’; they show sharp contrast in phase changes. These were derived from linear phase discontinuities and/or offsets showing displacement in the interferograms. It should be noted that these features are not the same as fault ruptures because we derived them from the form of displacement. Since one of our purposes was to find hidden faults, we attempted to remove displacements that showed intricately bent lines and/or circular-shape lines (such as landslides). Landslides were fairly easy to identify because they tended to occur as circular-shaped mass movements originating on a hillside or hilltop and moving downslope.

2.5-D analysis

The combination of SAR interferogram observations from both east and west enabled us to describe two-dimensional displacement parallel to LOS planes at all points in the SAR interferograms (Fujiwara et al. 2000). Accordingly, we divided the displacement vector into quasi-upward and quasi-eastward directions. We utilized the GNSS Earth Observation Network (GEONET), which has been deployed nationwide, and is operated by the GSI, to give absolute changes and correct distortion contained in the interferograms. As the 2.5-D analysis has errors due to mis-unwrapping, we mapped the lineaments in the interferograms and used the 2.5-D analysis to check them.

Coherence analysis

In SAR interferometric analysis, coherence is the degree of similarity between a pair of SAR images (Hanssen 2001; Watanabe et al. 2016). In general, where the ground surface has been changed (such as by landslides and/or liquefaction), the coherence shows lower values than the surrounding area. The absolute coherence values can be affected by baseline (geometric) decorrelation, thermal (system noise) decorrelation, temporal decorrelation and other sources (Hanssen 2001). Therefore, we used normalized coherence change based on two sets of coherence images: one pair obtained before the earthquake and another pair obtained before and after the earthquake (Watanabe et al. 2016). This method can detect small surface changes but is too sensitive to detect large changes (such as landslides). Therefore, we also use normalized phase noise deviation, a measure of the degree of phase noise deviation in a filtered interferogram. This method can be used to mask large phase noise areas for the purpose of smooth phase unwrapping in the GSISAR software. The normalized phase noise deviation is not coherence, but it is useful for distinguishing large phase noise areas from good correlation areas. Unfortunately, however, since the phase noise deviation method uses filtered interferograms, the resolution of the image decreases.

Linear surface displacements

The SAR interferograms representing co-seismic deformation are shown in Figs. 2, 3 and 4. The ascending (from the west, Fig. 4) and descending (from the east, Fig. 2) interferograms were used to construct quasi-upward and quasi-eastward displacement maps (Figs. 5, 6). The elevation angle from the south for the quasi-upward component was 83° and the quasi-eastward direction was true east. Details of the crustal deformation and the simulated fault model derived from the SAR interferograms are given in Kobayashi et al. (2019).

Three series of linear surface displacements were derived from linear phase discontinuities and/or offsets showing displacement (Figs. 2, 3, 4, 5, 6): a longer one through the middle of the study area (B) and shorter ones to the northeast (A) and southwest (C). A different, independent closed displacement area was also identified (D) that moved 5–10 cm westward (Figs. 5, 6).

Both of the figures, the interferograms (Figs. 2, 3, 4), and the 2.5-D analysis (Figs. 5, 6), have advantages and disadvantages in detecting linear surface displacements. For example, around the middle part of B (cross section b2), the interferogram (Fig. 3) shows the same blue color on both the west and east side of the linear surface displacement. However, they have one cycle (12 cm) difference in LOS displacement. For this area, the 2.5-D analysis gives correct displacement (b2 in Fig. 7), while for some other areas, the 2.5-D analysis shows mis-unwrapping, which are due to decorrelation; however, we have checked that at least around A, B, C and D, the 2.5-D analysis most likely gives correct displacement.

Liquefaction in coastal areas

In the coastal town of Mukawa, in the southern part of the study area, several localized areas of hundreds of meters in diameter moved up by 50 cm (Fig. 9). Unfortunately, the image quality of Fig. 9b is poor and the resolution is lower than that of Fig. 9a because of multi-look filtering to reduce noise. The ground deformation shown in Fig. 9a, b generally has opposite values (site 2 in Fig. 9b is not clear owing to decorrelation and for site 2′ it is easy to recognize the values), making it most likely that horizontal displacement in the east–west direction was dominant and vertical displacement was small in this area.

However, the distribution of displacements was complex. At site 1 in Fig. 9, the displacement was ~ 10 cm westward, at site 2 it was ~ 50 cm eastward, and at site 3 it was ~ 5 cm eastward. In addition, the coherence of the interferograms around sites 4 and 5 was low and the form of displacement was not uniform, suggesting that relatively random displacements occurred. In an aerial photograph taken after the earthquake (GSI 2018a), traces of jetted sand were found in a baseball field close to site 1. As this location is near the coast, there is a possibility that lateral spread occurred owing to liquefaction caused by the seismic motion.

Liquefaction in fill land valleys

In a residential area within the southern part of the city of Sapporo, far from the epicenter, damage including road subsidence and tilted houses was caused by earthquake-induced liquefaction (Konagai et al. 2018). These deformation areas were detected based on the SAR interferogram (Fig. 10a) and the normalized coherence-change map (Fig. 10b). Surface displacement detected by comparisons of SAR observations before and after the earthquake mostly shows subsidence of several centimeters, but estimates from aerial photographs (GSI 2018a) suggest displacement amounts exceeding 1 m in some places where the coherence of the interferograms was greatly reduced. In Fig. 10b, the areas most affected (as determined through interpretation of aerial photographs) coincided with the area where the largest normalized coherence changes were observed.

By reconstructing the past topography (altitude) from aerial photographs, we obtained changes in topography due to human cut/fill operations (Fig. 10c) during a period of urban development beginning in the 1960s. The recent deformation areas coincided remarkably with the filled areas (ex-valleys), implying that earthquake-related liquefaction likely occurred in these areas (Konagai et al. 2018). These results clearly indicate that mapping topography changes due to man-made operations can be applied to urban planning and/or safety assessment in other areas.

Landslides in mountainous areas

Strong seismic motions triggered many landslides in the mountainous area surrounding the epicenter (Figs. 11, 12). The Ministry of Land, Infrastructure, Transport and Tourism (2018) estimated that the total affected area as ~ 13.4 km², exceeding the 1891 Nobi and the 2004 Niigataken Chuetsu earthquakes, the largest to occur in inland Japan in the last 150 years. Yamagishi and Yamazaki (2018) pointed out that most landslides slid over unstable pumice and ash layers deposited by past large eruptions of the Shikotsu caldera and the Mt Eniwa-dake and Mt Tarumae volcanoes.

For example, a large landslide at the foot of a small mountain caused the collapse of many houses (Fig. 12a, b). Although coherence-change maps clearly show high values in this area (Fig. 12c, d), in other areas they are likely over-estimated. Since the coherence-change maps have many mis-estimations in no landslide areas, it is difficult to fully understand the situation. The normalized phase deviation map (Fig. 12a) shows that, although fine changes could not be captured owing to the spatial filter, the distribution of the overall landslide area was well-extracted in Figs. 11 and 12a because of small over-estimation in contrast with the coherence-change maps. Therefore, coherence-change maps can be used to analyze detailed distributions in surface change related to such events, and normalized phase deviation maps can be used to estimate overall affected areas for rapid response or mitigation.

Connection between linear surface displacements and the seismic fault

The depth of the main earthquake’s hypocenter (obtained from the seismic wave) was ~ 35 km (Earthquake Research Committee 2018). Furthermore, the seismic source fault model obtained from the crustal deformation showed that the depth of the fault plane’s top was ~ 16 km and it had a dip angle of 74° (Kobayashi et al. 2019). Based on the surface displacements shown in Figs. 5 and 6, the linear surface displacement features represent the movement of the reverse fault along its low dip angle. In addition, the horizontal distance from the linear surface displacements to the epicenter was ~ 10 km (Fig. 8). The known active faults on the north side of the linear surface displacements had low dip angles at 3 km depth and almost zero at 2 km depth (Earthquake Research Committee 2010; Ikeda et al. 2002; Yokokura et al. 2014).

Given these facts, if the linear surface displacements are considered to be directly connected to the hypocenter, the fault’s low dip angle at shallow depth must change to a high dip angle at around 30 km depth. Therefore, it is unlikely that the linear surface displacements directly connect with the focal seismic fault. Furthermore, since few aftershocks were observed at depths < 10 km (Earthquake Research Committee 2018), even if the linear surface displacements were structurally connected to the hypocenter, it is almost impossible that the motion of the linear surface displacements was directly linked to the fault movement at the hypocenter.

Nature of linear surface displacements

Furthermore, Kobayashi et al. (2019) pointed out that a ΔCFF estimate suggests that the static stress change due to the 2018 Hokkaido Eastern Iburi Earthquake could have encouraged reverse slip in the area around the linear surface displacements of several dozen kilopascals. Given this context, the linear surface displacements were likely produced by secondary non-seismogenic fault motions passively generated by the 2018 Hokkaido Eastern Iburi Earthquake.

As for the independent horizontal movement in area D (Figs. 5, 6), other independent movements were found in the Mukawa area (Fig. 9). Because the latter occurred in flat coastal areas, they can be considered as lateral spread related to liquefaction. Although the movement of area D was similar to that in Mukawa, the former was situated on a slightly elevated area and seems not to have moved by lateral spreading. Togo (2000) pointed out that the surface expression of a reverse fault tends to move forward (the lower side of the fault), generating a new fault plane, and so area D may be a fault expression shaped by erosion.

Alternatively, area D could have remained in place while the surrounding area moved east (similar to lateral spreading). In this scenario, since the whole area should have shifted fairly uniformly to the east, a uniform horizontal slip must have occurred at a certain depth over a wide area of ~ 15 km N–S and ~ 5 km E–W. Fully understanding the complex movements of area D requires further study.

Use of linear surface displacements to recognize active faults

Although there is no evidence that the linear features considered in this study generated strong seismic waves at the time of the main earthquake, there is a possibility that these represent surface traces of active faults and could cause large earthquakes in the future. At present, we can provide no concrete answer to this question.

As shown in Fig. 8, the linear surface displacements appear to be geographical extensions of the known active faults. In addition, their strikes and displacements match surrounding geological features and fault-like structures were found just beneath them by a seismic reflection survey (Yokokura et al. 2014). Such circumstantial evidence sufficiently indicates that these features are most likely the traces of active faults.

Some of the linear surface displacements recognized after the 2016 Kumamoto Earthquake included those that had been identified as active faults by their topography, but the number was limited (Fujiwara et al. 2016). However, the linear surface displacements detected after the 2018 Hokkaido Eastern Iburi Earthquake had not been identified as active faults. One of the possible reasons why previous studies could not map these lineaments as ‘active faults’ is that the lineaments were not straight lines owing to the low dip angle. Even in this situation, our results support the existence of hidden faults, of which the linear surface displacements are most likely surface traces. On this basis, the detection of linear surface displacements can be used for active fault identification in the future.

However, it is also likely that those previously recognized ‘active faults’ from their topographical features did not generate seismic waves, either. Therefore, it is important to recognize the diversity of ‘active faults’ for earthquake hazard assessment.

Coherence analysis in mountainous and urban areas

Figures 10b and 12c, d show normalized coherence-change maps obtained by the same method. In Fig. 10b, fluctuations associated with liquefaction in urban areas can be clearly detected. Although Fig. 12c, d clearly shows the landslide distribution in the mountains, it is possible that these features were excessively extracted (see also Additional file 1: Fig. S4). There are two possible reasons for this: the sensitivity of the normalized coherence change could be too high, or the original coherence was too low in the mountains. In urban areas, the presence of buildings and pavements means that changes in the surface itself are relatively small, but in mountainous areas sediment moves more freely, and thus surface conditions can change more significantly. Although normalized coherence change is capable of detecting minute deformations, it does have a tendency toward saturation. Thus, in mountainous landslide-prone areas, it is possible that small displacements or changes occurring near actual landslides could be detected and classified as actual landslides.

Although the normalized coherence-change method is inaccurate in areas with low coherence, it can be corrected by masking those areas using a coherence filter for the pre-earthquake image (Watanabe et al. 2016). We used masks of 0.5 or less in the urban area (Fig. 10b) and of 0.2 in the mountainous area (Fig. 12c, d). In the latter, adopting a mask of 0.5 or less could improve the accuracy, but since the coherence was extremely low, values could not be obtained in most places using the coherence filter. Therefore, it was difficult to detect coherence changes in mountainous areas, and other methods such as multi-polarimetry (Watanabe et al. 2016), should be considered.

On the contrary, in areas masked using the normalized phase noise deviation (Figs. 11, 12a, and Additional file 1: Fig. S4b), the degradation area (large phase variation) of the SAR interferogram roughly agreed with the landslide area interpreted by the aerial photographs. Consequently, the normalized phase noise deviation can be used to roughly estimate the spatial distribution of landslide concentration areas for rapid response or mitigation.