Keywords

1 Introduction

In many earthquakes, landslides have often been responsible for more damage to lifelines than other seismic hazards combined. In the 2016 Kumamoto Earthquake, Kyushu, Japan, about 750 locations along conventional railway lines reportedly suffered severe damage, of which 73% were caused by landslides, debris flows, rockfalls, etc. (Kyushu Railway Company, 2017). All are beyond the direct jurisdiction of lifeline management organizations. Kobayashi (1981) reported that more than half of deaths in earthquakes with magnitudes larger than 6.9 in Japan between 1964 and 1980 were caused by landslides.

An earthquake-induced landslide is just one scene of ever-evolving mountain topography. The fact indicates that there are always pros and cons to living in these areas. The cons are obviously landslide disasters, while the pros are geographical and hydrogeographic benefits the people have long been enjoying. Therefore, rehabilitation of lifelines and civil infrastructures in the landslide-devastated regions should not simply restore existing facilities but proceed with the build-back-better policy.

From this viewpoint, it pays to know rock fold dimensions associated with steadily evolving tectonic movements can often govern landslides’ representative sizes and distributions in active folding zones.

To deal with rehabilitation problems in a scientific manner, a three-year research program (2005–2007), “Earthquake damage in active-folding areas: the creation of a comprehensive data archive and suggestions for its application to remedial measures for civil-infrastructure systems,” was set up getting the Special Coordination Funds for Promoting Science and Technology, Ministry of Education, Culture, Sports, Science and Technology, MEXT for short. This program was unique among various MEXT-funded research programs in that an advisory panel was set up under the Japan Society of Civil Engineers (JSCE Active folding Project, 2008). This framework allowed researchers and experts from authorities such as the Niigata prefectural government, Ministry of Land, Infrastructure, Transport and Tourism (MLIT), etc., to join. All members were concerned about rehabilitation affairs. The idea was to facilitate real-time research information sharing among relevant organizations for rational rehabilitations. This article reviews some notable findings obtained through this three-year research program and even beyond it.

2 The 2004 Mid-Niigata Prefecture Earthquake

The 2004 Mid-Niigata Prefecture Earthquake, also known as the “Chuetsu” Earthquake, occurred in Niigata Prefecture, Japan, at 17:56 local time (08:56 UTC) on Saturday, October 23, 2004 (Fig. 1). The hypocenter of the mainshock was located at 37.29° N, 138.87° E, in mid-Niigata Prefecture, at a depth of 13 km. The acceleration reached its maximum value of 1500 cm/s2 at K-net Ojiya station, about 7 km west of the epicenter (Earthquake Research Institute (ERI), University of Tokyo, 2004). A series of strong aftershocks followed the mainshock in rapid succession. These strong earthquakes had focal mechanisms of reverse fault type with the compressional axes oriented in NW/SE direction. The orientation is consistent with the historical information of large earthquakes in this area.

Fig. 1
figure 1

Epicentral area of the Mid-Niigata Prefecture Earthquake of October 23, 2004: the earthquake of magnitude 6.8 was followed by many aftershocks (open circles in the figure), with four of magnitude 6 or greater in rapid succession (M6.3 at 18:03 JST, M6.0 at 18:11 JST, and M6.5 at 18:34 JST), resulting from complex multi-segment fault geometry (Hikima and Koketsu 2005). The earthquake hit the active-folding area with anticlines (red broken lines) and synclines (blue broken lines) trending in an NNE-SSW direction. Both Shinano and Uono Rivers flowing through the area meander remarkably, controlled by the folding structure

Full size image

This continual tectonic movement has formed NNE-SSW-trending geological folds of sedimentary rocks in the source region (Fig. 1). Up-folded rocks along anticlines expand and crack over a long geological time. Anticlines, thus, have often their crests deeply worn away. Consequently, many debris deposits rim the deeply eroded hollows called anticlinal valleys while down-folded synclines remain as ridges. Thus, the anticlinal valleys and the synclinal ridges often make up an inverted relief. However, the folding is still young, about 3 million years old or younger (Geological Survey of Japan (GSJ), 2007) in the quake-hit area, preserving remnants of the initial folding structure. Namely, a steep cliff or escarpment on one side and a gentle dip or back slope on the other feature the landform of this area; the landform is called “Cuesta.” Thus, large-scale landslides occur even on gentle mountainsides dipping towards synclines. Therefore, the site is one of the most landslide-prone zones in Japan.

3 Tectonic Displacements

Interferometric Synthetic Aperture Radar (InSAR) is one of the most advanced technologies to measure elevation changes with high precision, and the RADARSAT-I satellite (C-band at 5.3 GHz) flew over the epicentral area at 20:45 (UTC). However, thick vegetation and thousands of landslides have severely blurred its InSAR fringe patterns in the epicentral area (Ozawa et al. 2005). Therefore, we have obtained precise digital terrain models (DTMs, hereafter) before (1975–1976) and after the earthquake (October 24, 2004) using stereoscopy and “Laser Imaging Detection and Ranging (LiDAR)” technology, respectively. The DTMs were then compared to detect elevation changes and translations of the topography. Lastly, all the changes in landforms due to landslides and artificial changes are removed to detect tectonic deformations of the ground surface, which can be useful in estimating the deformation of the whole mountains.

However, subtracting the pre-quake DTM from the post-quake DTM allows us to detect displacements only in the Eulerian description. Namely, the motion is described in terms of the spatial coordinates that do not follow the motion of soil particles. Discussions of earthquake-inflicted geotechnical issues require a more direct description of soil particle movements because soils are typically history-dependent materials. Konagai et al. (2009a) obtained three orthogonal components of tectonically induced surface displacement (tectonic displacement, hereafter) by assuming that the tectonic displacement varies gently in space. That is to say, three pixels of DTM arranged next to each other would undergo the same Lagrangian displacements because the DTM has a substantially high resolution of 2.0 × 2.0 m per pixel. Thus, the observed changes in elevation (Eulerian displacements) at three pixels arranged side by side could determine the three orthogonal components of the Lagrangian displacements common to these pixels. This calculation is repeatedly performed while moving over the target terrain to obtain all Lagrangian displacements. The method later underwent major upgrades by Zhao (2010) and Zhao and Konagai (2014) in a manner that detected tectonic displacements better conform to the measured displacements at triangulation points on the intact ground in the earthquake-hit area.

However, the obtained Lagrangian displacement components often show a remarkable scatter. The causes of the scatter are:

  1. (1)

    Quake-induced landslides,

  2. (2)

    Artificial terrain changes during the time between the two DTMs, and

  3. (3)

    the presence of some non-surface objects on digital surface models.

The spatial frequencies resulting from the causes mentioned above are remarkably higher than the tectonic displacement. Thus, the moving average method allows us to see the overall features of tectonic displacements. Assuming that the scattered values follow the Gaussian distribution within a square window, the most frequent value (mode) can represent the actual vector of the soil displacement in this area.

The size of the smoothing window has to be set based on its physical interpretation. The window size is desirable to be larger than the largest landslide in the target area to minimize the effect of the coherent landslide mass movements. At the same time, the size should not be too large to allow significant variation of the tectonic deformations within the smoothing window. The size of the square window was set at 1400 m by 1400 m for smoothing both vertical and lateral components of Lagrangian displacements in the quake-hit area.

Figure 2 shows the lateral components of surface tectonic displacements obtained through a 1400 m by 1400 m smoothing window. There are two clusters of large lateral displacements. Large displacement vectors reaching almost 0.5–1 m form a belt between the Kajigane syncline and the Higashiyama anticline in the southeastern cluster. The other cluster is northwest of the Higashiyama anticline. The former cluster along the Kajigane syncline is located around the surface extension of the hidden fault rupture planes for the mainshock and the largest aftershock of the Mid-Niigata Prefecture Earthquake (Hikima and Koketsu 2005).

Fig. 2
figure 2

Lateral components of surface tectonic displacement of the target zone on Zone VIII of the Japanese National Grid System. There are two clusters of large lateral displacements. Large displacement vectors reaching almost 0.5–1 m form a belt between the Kajigane syncline and the Higashiyama anticline in the southeastern cluster. The other cluster is northwest of the Higashiyama anticline

Full size image

Figure 3 shows the vertical components of surface tectonic displacements obtained through a 1400 m by 1400 m smoothing window. The vertical components show a hump in the southwestern part of the target zone. This hump is where the Uono River joins the Shinano River. Measurements of the benchmarks along the Uono River (Shinano River Office, Hokuriku Regional Bureau of the Ministry of Land, Infrastructure, and Transport), marked in Fig. 3, have shown that the area downstream of the Kajigane syncline has moved up. There is a cause-and-effect relationship between these tectonic displacements and flooding along the Uono River during heavy rains of June 2005 and August 2011. Blue polygons in Fig. 3 are areas along the Uono River inundated in the flood of June 2005.

Fig. 3
figure 3

Vertical components of surface tectonic displacement of the target zone on Zone VIII of the Japanese National Grid System. The vertical components show a hump exceeding 1 m in the southwestern part of the target zone. Since this hump is where the Uono River joins the Shinano River, the upper reach (blue polygons) of the Uono River was flooded in the torrential rain of June 28, 2005

Full size image

4 Underground Stress Field Associated with the Earthquake

Kazmi et al. (2013) further conducted an inversion analysis of the Lagrangian surface tectonic displacements to determine the spatial distribution of slips on fault rupture planes for known fault geometry, thus obtaining the co-seismic change in the stress field in the mountains (Kazmi et al. 2013). The fault geometry for this earthquake was complex such that three aftershocks with JMA magnitudes of 6.0 or greater occurred in rapid sequence. Hikima and Koketsu (2005) proposed a multi-segment fault model with five fault planes corresponding to the mainshock and the following four most significant aftershocks with magnitudes larger than 6 (Fig. 4). This fault model was adopted in this study. We need an appropriate kernel function for a half-space of stratified sedimentary rocks for the inversion analysis. Kazmi et al. (2013) used a robust and stable numerical scheme developed by Wang (Wang 1999; Wang et al. 2003). Material properties of each layer (material density, ρ, and two Lame’s constants λ and μ) are determined from the representative velocity structure of the source region (shown in Fig. 5), obtained by combining borehole data logging, geological information, and seismic exploration (Honda et al. 2005).

Fig. 4
figure 4

Illustration of multi-segment fault model (not to scale) beneath the epicentral area of the 2004 Mid-Niigata Prefecture Earthquake (based on Hikima and Koketsu 2005). Segments A and D have caused the mainshock and the largest aftershock, respectively

Full size image
Fig. 5
figure 5

Velocity structure model. \({v}_{p}\) and \({v}_{s}\) are the primary and secondary wave velocities, respectively (based on Honda et al. 2005)

Full size image

Besides the earthquake-induced stress changes mentioned above, it is necessary to incorporate the initial rock stresses in the whole mountains before the earthquake, which is particularly important to discuss the impact of quake-induced ground deformation on underground facilities such as tunnels. With less probability of measuring directly complicated stress conditions before the earthquake, coefficients of lateral rock pressure help estimate the stresses before the quake. We used some records of pre-quake tunneling through the epicentral area, and the records suggested that the coefficients are almost 1.0, excluding shallow locations near the tunnel mouses. This finding is consistent with what Matsumoto and Nishioka (1991) have shown for tunnels through sedimentary rocks excavated using the New Austrian Tunneling Method (NATM). The New Austrian Tunneling Method (NATM), which relies on the inherent strength of the surrounding rock mass, requires the installation of sophisticated measurement instrumentation. By virtue of this requirement, they could examine the measured convergence values of tunnel cross-sections in both the transverse and vertical directions. The coefficients of lateral rock pressure in soft rocks deeper than 50 m were about 0.9 to 1.0, with few exceptions (Fig. 6).

Fig. 6
figure 6

Coefficient of lateral pressure calculated from ground displacement data. Tunnels A and B are two tunnels with available convergence data within the study area

Full size image

The square root of the second principal invariant of the deviatoric stress tensor, \(\sqrt{{J}_{2}}\), can be considered as an index for rocks’ and soils’ deformability in the absence of reliable rock failure criterion and spatial coverage of soil/rock properties. The change in \(\sqrt{{J}_{2}}\) values associated with the earthquake was calculated at 75 m below the ground surface; 75 m was the average depth of the tunnels of the Joetsu-Shinkansen, a high-speed railway line connecting Tokyo and Niigata. \(\sqrt{{J}_{2}}\) values at this depth are also deemed to represent the internal driving stress field associated with earthquakes to evolve the active-folding terrain.

Figure 7 shows the spatial distribution of \(\sqrt{{J}_{2}}\). Remarkably, several stripes of large \(\sqrt{{J}_{2}}\) values appear parallel at a regular interval of about 4 km, with each stripe trending in the NE–SW direction. This figure also shows the locations of damaged tunnel cross-sections. East Japan Railway Company grouped the damaged sections into four categories based on their damage extents, with a higher number indicating more severe damage (East Japan Railway Company 2006). Almost all damaged tunnel sections deeper than 40 m (colored circles in Fig. 7) are found within these stripes (Kazmi and Konagai 2018).

Fig. 7
figure 7

Distribution of second principal invariant of the stress deviator tensor, \(\sqrt {J_{2} }\), over the entire target zone at a depth of 75 m from the ground surface, with locations of damaged tunnel sections (based on Kazmi et al. 2013). East Japan Railway Company grouped the damaged sections into four categories based on their damage extents as shown in the legend, with a higher number indicating more severe damage

Full size image

The first invariant of the Cauchy stress tensor \({I}_{1}\) and the square root of the second principal invariant of the stress deviator tensor \(\sqrt{{J}_{2}}\) were calculated at a regular interval for each tunnel. The examined points included the damaged sections. Assuming that the lateral pressure coefficient before the earthquake was unity as suggested above in Fig. 6, the initial confining pressure was set at its overburden pressure at each point. When the values of \({I}_{1}\) and \(\sqrt{{J}_{2}}\) are plotted on a scatter diagram (Fig. 8), points for the damaged sections (solid circles in Fig. 8) are found clustered along the upper bound of the entire cluster of points. This upper bound line evokes the Drucker–Prager yield criterion. This line’s gradient, 0.939, is relatively small for ordinary sedimentary rocks. Perhaps the value reflects the overall features of the sedimentary rocks with weak joints and cracks smeared over the whole domain (Kazmi and Konagai 2018).

Fig. 8
figure 8

Scatter diagram of \({I}_{1}\) and \(\sqrt{{J}_{2}}\) along the tunnels (based on Kazmi et al. 2013)

Full size image

Many landslides triggered by this earthquake appeared over these stripes of large \(\sqrt{{J}_{2}}\) values (white polygons in Fig. 9). This stripes’ pattern highlights the importance of the internal seismic stresses in causing and reactivating landslides. It is also noteworthy that all existing landslides identified as of 2000 (yellow polygons in Fig. 9, National Research Institute for Earth Science and Disaster Prevention 2000) form clearer clusters over the stripes of large \(\sqrt{{J}_{2}}\) values. Figure 9 thus suggests that similar co-seismic events repeatedly occurred over a long geological time, causing the active folding terrain to evolve (Kazmi and Konagai 2018).

Fig. 9
figure 9

Distribution of second principal invariant of the stress deviator tensor, \(\sqrt{{J}_{2}}\), over the entire target zone at a depth of 75 m from the ground surface, with locations of quake-induced and preexisting landslides (based on National Research Institute for Earth Science and Disaster Prevention 2000; Kazmi and Konagai 2018)

Full size image

5 Landslide Masses Perching on the cuesta’s Gentle Slopes

Many inhabitants in the quake-hit area depend upon rice and carp breeding for their livelihood, enjoying geographic and hydrogeographic benefits unique to this active-folding terrain. Figure 10 shows that ponds for irrigation and carp breeding are also found over the intense \(\sqrt{{J}_{2}}\) stripes. Therefore, relocating inhabitants in the quake-hit area is not always the best course of action. As long as they continue living in this area, a map showing the spatial distribution of landslides that moved a little downslope in the earthquake and still perch on the cuesta’s gentle slopes will help the inhabitants go along with these landslide masses.

Fig. 10
figure 10

Distribution of second principal invariant of the stress deviator tensor, \(\sqrt{{J}_{2}}\), over the entire target zone at a depth of 75 m from the ground surface, with locations of ponds for irrigation and carp breeding. This figure also shows the locations of the north mouth of Kizawa Tunnel and Azumagawa Check Dam (see Fig. 11)

Full size image

As said above, the Lagrangian displacement components obtained from pre-and post-quake DTMs often show a remarkable scatter, and a 1400 m by 1400 m smoothing window moved over the quake-hit area to filter out shallow disturbances such as landslides. We can filter out small-scale artificial terrain changes by reducing the smoothing window size as the next step. Then, by subtracting the once-obtained tectonic displacements from those extracted using the smaller window, we can detect shallower soil displacements caused by landslides. Sugawara et al. (2015) tuned the smaller window size so that the detected displacements better conform to the deformed alignment of the Kizawa Tunnel. Kizawa Tunnel is a shallow road tunnel going through a landslide mass that moved a little in the earthquake, causing the southern part of the tunnel embedded in the hidden landslide mass to move about a half meter in its transverse direction and expand by about 1.5 m (Konagai et al. 2009a, b; Zhao and Konagai 2014). Setting the window size at 200 m by 200 m, Sugawara et al. (2015) obtained displacements of hidden landslide masses, which moved a few meters downslope and are still perching on the quake-hit area’s mountain slopes. Figure 11 shows detected downslope movement vectors around Azumagawa Check Dam. Many of these vectors are deemed to have appeared in the quake. But they can include those caused by the other causes because the pre-quake DTM was from aerial photos taken in 1975 and 1976. Some reports say that some mudflows covered paddy fields in Yamakoshi in the torrential rain in July 2004, about three months before the earthquake. In any case, Fig. 11 indicates that unstable soil masses are still perching on slopes in the quake-hit area, and they can move again anytime.

Fig. 11
figure 11

Detected displacements of hidden landslide masses, which moved a few meters downslope and are still perching on the quake-hit area’s mountain slopes (based on Sugawara et al. 2015)

Full size image

Azumagawa Check Dam was constructed to prevent Prefectural Route 24, one of the major arteries necessary to rehabilitate the quake-devastated area. The dam’s construction started in March 2005, the following year of the earthquake, waiting for the thaw, and was completed in December 2006. Precast concrete blocks were stacked one after another to form the whole dam body. A dam photo taken on November 13, 2013 (Fig. 12) shows that the dam body turned slightly up. We first suspected that the downslope movement of the landslide mass on the western river wall may have pushed the dam against the eastern river wall. But comparing post-quake DTMs did not yield any significant Lagrangian displacements of this landslide mass (Sugawara et al. 2015). According to the Yuzawa Sediment-Control Office, Ministry of Land, Infrastructure, Transport, and Tourism, the dam started to deform even during its construction period, probably because its abutments were immediately on the toe of the landslide mass. The toe was soft enough for heavy precast concrete blocks to sink. At any rate, it was wise to construct the check dam by stacking precast concrete blocks so that the whole dam body could conform to the ground deformation.

Fig. 12
figure 12

Azumagawa Check Dam at N 37.3300°, E 138.9021°. The center of the dam crest was 0.4–0.5 m higher than the abutments as of November 17, 2013

Full size image

Given that there are still many landslide masses remaining on mountain slopes, the rehabilitation went forward following the steps shown below:

  1. (1)

    Slipped major landslide masses were first stabilized, not allowing their toe parts to be eroded,

  2. (2)

    Roads were then rerouted and reconstructed, avoiding remaining unstable soil masses, and finally,

  3. (3)

    Houses, farmlands, and carp-breeding ponds were relocated to safer places or reconstructed at the same locations where stabilization works have been done.

6 Summary

Though serious, the damage caused by the Mid-Niigata Prefecture Earthquake of October 23, 2004, has given us a rare opportunity to study the landform changes caused by an earthquake that hit an active-folding mountainous terrain. This article reviewed some significant findings that have been obtained through a MEXT-funded research program, “Earthquake damage in active-folding areas: the creation of a comprehensive data archive and suggestions for its application to remedial measures for civil-infrastructure systems.” The findings highlighted that the deep-seated tectonic displacements had something to do with unique features of damage caused by the earthquake.

The points to be highlighted are:

  1. (1)

    The change in \(\sqrt{{J}_{2}}\) values (square root of the second principal invariant of the deviatoric stress tensor) associated with the earthquake was calculated at 75 m below the ground surface (average depth of the Joetsu Shinkansen Railway tunnels). \(\sqrt{{J}_{2}}\) values deep in the ground can represent the internal driving stress field associated with earthquakes to evolve the active-folding terrain. Several stripes of large \(\sqrt{{J}_{2}}\) values appear parallel at a regular interval of about 4 km, with each stripe trending in the NE–SW direction. Almost all damaged tunnel sections were found clustered within these parallel stripes.

  2. (2)

    Many landslides triggered by this earthquake also appeared over these stripes of large \(\sqrt{{J}_{2}}\) values.

  3. (3)

    It is also noteworthy that all existing landslides identified as of 2000 (National Research Institute for Earth Science and Disaster Prevention 2000) form clearer clusters over the stripes of large \(\sqrt{{J}_{2}}\) values. This finding suggests that similar co-seismic events repeatedly occurred over a long geological time, causing the active folding terrain to evolve.

  4. (4)

    Displacements of landslides that moved a few meters downslope in the earthquake and are still perching on the cuesta’s back slopes were extracted.

  5. (5)

    Given the presence of these landslide masses, slipped major landslide masses were first stabilized, not allowing their toe parts to be eroded. Then rehabilitation works followed.