2). State Key Laboratory of Earthquake Dynamics, Institute of Geology, China Earthquake Administration, Beijing 100029, China
The number, scale, and distribution of co-seismic landslides are affected by earthquake characteristics (epicenter, seismogenic faults, PGA, seismic intensity, etc.), topography (elevation, slope aspect, slope position, slope curvature, etc.), and geology (lithology, fault, slope structure, etc.) (Xu Chong, 2018a). In the past 20 years, with the increasingly developed remote sensing and GIS technology, it has become possible to completely interpret co-seismic landslides triggered by strong earthquakes and establish earthquake landslide databases. Furthermore, the numbers, scales, and spatial distributions of numerous co-seismic landslides have been analyzed, such as the 2008 Wenchuan MS8.0 earthquake, the 2010 MS7.1 Yushu earthquake, the 2010 Haiti MS7.0 earthquake, and the 2015 MS8.1 Nepal earthquake (Du Peng. et al., 2020; Xu Chong et al., 2013, 2014, 2018a, b). In the Loess Plateau area around Ordos Block, Substantial M8 earthquakes have occurred (Gu Gongxu, 1983). According to historical seismic records, a large number of landslides are triggered by earthquakes. In our field investigation, a relatively complete distribution of earthquake-induced landslides in the loess area is observed (Xu Chong, 2018c, Xu Yueren, 2020a). Due to the preservation of the integrity of the landslide form, loess landslides, which have occurred for hundreds or even thousands of years, can be identified accurately. Therefore, a database of landslides triggered by strong historical earthquakes in these areas can be further established, and the focal parameter characteristics of these earthquakes can be clearly understood. Due to the fact that images cannot be used for comparative interpretation before and after earthquakes, those landslides induced by rainfall and human factors should be carefully excluded in the interpretation process.
At present, the magnitude of strong historical earthquakes and the scope of polar regions are roughly determined by the damage degree and casualties reported in historical records (Yuan Daoyang et al., 2004). The establishment of a landslide database of strong historical earthquakes is beneficial for effective evaluation and revision of their focal parameters; thus, landslide databases can be used as an effective research method in addition to paper review and research regarding active faults (Li Weile et al., 2015; Zhuang Jianqi et al., 2018; Xu Chong, 2018b; Xu Yueren et al., 2020a). However, determining the range of landslides triggered by a strong historical earthquake with a possible overlapped meizoseismal area by other earthquakes remains challenging. For example, there is a considerable debate regarding landslides near the Tongwei River, which is the epicenter area of the 1718 Tongwei earthquake and also within the influence range of the 1920 Haiyuan earthquake. Are the landslides in this area mainly induced by Haiyuan earthquake or Tongwei earthquake?Similar problems still must be carefully considered when constructing a database of landslides induced by strong historical earthquakes. Thus, these information should be analyzed in detail with a combination of historical seismic records.1 GEOLOGICAL BACKGROUND
It has been 100 years since the Haiyuan earthquake occurred in Ningxia on December 16th, 1920, which killed at least 234 000 people. The seismogenic fault is a left-lateral strike-slip fault that forms an 240 km earthquake surface rupture. This earthquake has triggered a large number of landslides. The Haiyuan earthquake is a typical strong historical earthquake with the longest history and extremely in-depth investigations in China. In 1921, the International Hunger Relief Association and the national government separately assigned scientific investigation groups to investigate the disaster situation (Close U. et al., 1922; Ningxia Seismological Bureau, 1980). This investigation is the first earthquake-related scientific research in China. Weng Wenhao, Xie Jiarong, and six other experts set out from Lanzhou and arrived in Guyuan via Huining, Jingning, and Longde. After conducting an investigation in Guyuan, they traveled from Guyuan to Pingliang and returned to Lanzhou via Huating, Zhangjiachuan, Tongwei, and Lintao. Weng and his team distributed questionnaires throughout the route and obtained information about the casualties, livestock deaths, building collapses, landslides, and real estate losses. With these information, the first intensity map of an earthquake in China was drawn (Fig. 1(b)). Furthermore, numerous descriptions of the Haiyuan earthquake-induced landslides are collected (Table 1) (Close U. et al., 1922; Ningxia Seismological Bureau, 1980). Owing to the rich and detailed historical records, this event serves as a suitable research subject to explore and determine the complete distribution of landslides induced by historical earthquakes. The intensity map obtained from the post-earthquake investigation (Fig. 1(a), (b)) is quite different from the intensity maps obtained in subsequent studies (Fig. 1(c), (d)). This is because those focus primarily on immediate geological disasters, and subsequent studies combined with the study of seismogenic structures could generate a supplement to the intensity map. The post-earthquake intensity map shows that the seriously damaged area of the Haiyuan earthquake is the region with a northwest direction from Haiyuan to Tongwei.
In the past decades, the distribution characteristics of co-seismic landslides in the Haiyuan earthquake have been studied extensively based on image interpretation and field investigation (Table 2). According to previous studies, the following characteristics are identified: ① The distribution of landslides is not affected by epicentral distance and dense landslide distribution is observed in numerous areas. (Zou Jinchang et al., 1996; Li Weile et al., 2015; Xu Chong et al., 2018b). ②Most co-seismic landslides triggered by the Haiyuan earthquake have slopes of 5°-20° (Li Weile et al., 2015), 10°-25° (Peng Zhengkui, 2013), and 15°-20° (Xu Chong et al., 2018b). The results show that the main slope ranges from 5° to 25° and that the landslide density increases as slope increases. ③ Most of the back scarp of the landslides is located in the upper 1/5 and 2/5 of the slope. (Peng Zhengkui et al., 2013; Zhuang Jianqi et al., 2018). ④ The slope heights of developed landslides are 0-224 m, among which most slope heights are in the range of 15-100 m (Li Weile et al., 2015) and 70-150 m (Peng Zhengkui et al., 2013). ⑤ The landslides in Haiyuan area primarily exhibit a double-layer structure of loess-red mudstone landslide and inner loess landslide. Therefore, landslides are located in the strata range of Q, E, and N on the geological map (Zou Jinchang et al., 1996; Xu Chong et al., 2018b).
The following differences are also observed in previous studies: ① The concentrated distribution area of Haiyuan co-seismic landslides may include the area south of Tongwei, which impacts data collection. In early studies, the study area is regarded to have a dense distribution of co-seismic landslides following the Haiyuan earthquake (Zou Jinchang et al., 1996; Chen Xiaoli et al., 2003). According to contemporary research, however, most landslides in the area south of Tongwei are characterized as co-seismic landslides induced by the 1718 Tongwei earthquake rather than the Haiyuan earthquake (Xu Yueren et al., 2020b). At present, research on Haiyuan co-seismic landslides mostly focuses on landslides within the scope of the Ⅸ intensity zone, ignoring the area south of Tongwei. ② Whether the "back-slope effect" similar to that of the Wenchuan earthquake exists, is unclear. Some studies suggest that the Haiyuan co-seismic landslides also exhibit the "back-slope effect" (Peng Zhengkui et al., 2013), while others report a face-slope effect (Li Weile et al., 2015; Zhuang Jianqi et al., 2018). Additionally, no obvious dominant slip direction is observed (Xu Chong et al., 2018b). ③ The statistical range of landslide quantity in the database is inconsistent; records 1-4 in Table 2 are mainly based on field investigation in Xiji, Guyuan, and other landslide accumulation areas, whereas records 5-8 are interpreted using remote sensing images focusing on the Haiyuan earthquake intensity Ⅸ area and above.
The database of historical earthquakes is incomplete due to the following reasons. Firstly, several small-scale landslides are even more challenging to be identified with long-time erosion. Furthermore, it is difficult to define the distribution range of those landslides completely.
Fortunately, with the support of abundant historical records and data, more questions may be answered. For instance, can we separate the distribution range of the landslides triggered by the Haiyuan earthquake? Are the controversial landslides in Tongwei area affected by the distribution of the landslides triggered by the Haiyuan earthquake? Is there any relationship between the sliding direction and epicenter/fault/wave propagation?
To address these questions, this study uses the most complete database of earthquake-induced landslides in loess areas. Through the Google Earth image interpretation, we review the historical records regarding the landslide and casualty data of the Haiyuan earthquake, and combine these findings with ASTGTM DEM data. Finally, we analyze the landslides and, geomorphic parameters, as well as the density distribution of the landslides in the Haiyuan area, for the purpose of determining the co-seismic landslide distribution.2 METHODS 2.1 Remote Sensing Interpretation Method
In this study, the most complete database of seismic landslides in the loess area is used (Xu Yueren et al., 2020a, b). The landslide data from Haiyuan and its adjacent areas are selected for analysis (Fig. 2); these data include the influence areas of the 1920 Haiyuan earthquake, the 1718 Tongwei earthquake and their surrounding areas.
This study selects the interpretation method used for historical earthquake-induced landslides (Xu Yueren et al., 2020a). In detail, we interpret multiple high-resolution satellite images (2000-2016) through the 3-D display function on Google Earth. Taking advantage of a computer-aided visual interpretation method combined with previous experiences, we are able to identify the characteristics of the landslide body, including back scrap, front edge accumulation, side scrap, etc. Furthermore, polygons are used to represent the spatial range of the landslide body (Fig. 3(a)). The interpreted landslides are stored and processed in Google Earth and ArcGIS software as*. SHP and*. KML formats. In Google Earth, the 3-D terrain information reveals the length, width, top and bottom elevations of the landslide body. It also shows the ridge and valley elevations of the slope where the landslide body is located. The area of each sliding surface is calculated using geographic information calculation tool in ArcGIS, and the earthquake landslide database is established. The process of landslide interpretation and database logging is checked by other scholars in the research group to ensure the reliability of landslide data interpreted by remote sensing. Concurrently, field surveys are conducted for the verification of typical landslides (shown in Fig. 3).
The images used for interpretation are captured 80-100 years after the Haiyuan earthquake, exhibiting legible morphological characteristics of the landslides. Additionally, numerous barrier lakes formed after the Haiyuan earthquake (Fig. 3(b), 3(c)) are assisted for the verification process. The images indicate that the shape of the landslide is complete, and the image accuracy meets the requirements of interpretation; the distribution of earthquake landslides is very dense, and the entire gully exhibits a high landslide density distribution from the top of the ridge to the bottom of the valley (Fig. 3(a)). Field investigations reveal farmlands on the slope and landslide body (Fig. 3(d), (e)). The rear and side edges of the large landslide are still clearly identifiable. Thus, we are able to identify large landslides, but it is challenging to identify small-scale landslides after the reconstruction process.2.2 Geomorphic Parameter Information Extraction
Topography (relative height difference), stratum lithology, and slope are the main factors controlling a landslide formation (Qiao Jianping et al., 2001; Guo Fangfang et al., 2008). The earthquake-induced landslides exhibit concurrent directionality (Xu Qiang et al., 2010). To analyze the relationship between the distribution of earthquake-induced landslides and geomorphic features, the study uses ASTGTM DEM data (30 m resolution) and extracts the slope, aspect, and relief data of the study area through the ArcGIS platform. Then, the influence of geomorphic factors on the distribution of the Haiyuan earthquake-induced landslides are analyzed. The tools in ArcGIS are used to extract geomorphic parameters, "Spatial analyst tools-Surface-Slope" tool is used for slope extraction, "Spatial analyst tools-Surface-Aspect tool" is used for slope aspect extraction, and "Spatial analyst tools-Neighborhood-Focal Statistics tool" is used for relief extraction. The scale of the valley should be considered when calculating the relief. The best statistical unit of topographic relief in Ordos block and its surrounding areas is 2 km × 2 km (Guo Fangfang et al., 2008). As a reference, in this study, 75 m× 75 m (search radius of 2 250 m) neighborhood statistical unit is used to calculate the maximum and minimum DEM data in the neighborhood, and the difference between them is taken as the relief data in adjacent areas. In the subsequent analysis, the "Spatial analyst tools-Extraction-Extract multi values to points: tool" is used. The slope, aspect, and relief information of the corresponding positions are extracted using landslide points, and statistical analysis is conducted.2.3 Density Analysis
Several M≥7 earthquakes have occurred near the study area (e.g., Haiyuan M8
The co-seismic landslides of the Haiyuan earthquake are mainly distributed in the area where the Haiyuan and Liupanshan faults intersect (Fig. 2). The point density of the landslide corresponds to the ratio of landslide points to the area, and the density of the landslide surface is the ratio of the landslide area to the total area. The area with a high density of landslide points corresponds to a large number of landslides in the region, while the region with a high area density corresponds to a large number of landslides; meanwhile, the average area of landslides in the region is large. Based on the results of screening, the point density distribution of the landslide is consistent with the area density distribution range, but some differences are observed in Figs. 4, 5. In this study, regions A-M with a point density greater than 1/km2 and area density greater than 4% A-I (Fig. 5) are extracted. Combined with the point density and area density maps of landslides (Fig. 4 and Fig. 5), the co-seismic landslides of the Haiyuan earthquake are found to be mainly distributed in the east and south of Haiyuan County, In addition, the southeast end of the Haiyuan fault intersects with Liupanshan fault, showing multiple dense distribution areas, which is consistent with the previous research conclusions (Zou Jinchang et al., 1996; Li Weile et al., 2015; Xu Chong et al., 2018b). According to the statistical rule of the farthest distribution distance of a landslide induced by an earthquake (Keefer D.K., 1984; Delgado J. et al., 2011), the distance between the farthest landslide induced by the Haiyuan M8
Area H is located in the southeast of Haiyuan County, which is to the west of the Liupanshan fault and north of the Haiyuan fault. The shape and scope of the point density and area density of the landslide in this area are basically consistent, which is in agreement with the location of the macro-epicenter. In the area to the west of Liupanshan fault and south of Haiyuan fault, the shape and size of the point density and area density of the landslide are quite different, and the surface density of the landslide is larger than that of the landslide point density (a > C+D, c > E). Area F is located in the southeast of Guyuan, which is in the eastern area of the Liupanshan fault; here, the density of the landslide surface is larger than that of the landslide point (f > F). The density of the landslide surface is approximately equal to the point density in area G, which lies to the east of Guyuan. Area I locates in the north of Guyuan, and the density of the landslide point in this area is larger than that of the landslide surface (i < I). Although the area from Haiyuan to Jingtai is located in the high-intensity area of the Haiyuan earthquake, no significant landslides have been observed. Area M is situated in the south of Tongwei and is characterized by high density of both the landslide point and surface.4 DISCUSSION
Numerous previous studies have investigated the distribution characteristics of co-seismic landslides of strike-slip faults. The fault itself if the center line and the landslides are distributed symmetrically on both sides of the fault with slight difference (Xu Chong et al., 2012). However, the distribution of co-seismic landslides in the Haiyuan earthquake is notably different on both sides of the fault. The generation and distribution of co-seismic landslides in the Haiyuan earthquake are controlled not only by the distance from the fault but also by strata and geomorphic factors (slope, relief, slope aspect).4.1 Relationship Between Landslide Distribution and Strata
In this study, a 1:500000 geological map of the study area is obtained, and the seismic intensity map and high density area of landslides are overlaid on the geological map (Fig. 6). In detail, the quaternary strata are mainly distributed in the Ⅸ degree intensity zone of the Haiyuan earthquake. The Neogene strata are distributed in the south of the study area around Tongwei and Zhuanglang. The Paleogene strata are distributed in the piedmont of Liupanshan Mountain and the intersection area of the Liupanshan and Haiyuan faults. The Mesozoic, Paleozoic, and Precambrian strata are distributed on both sides of the Haiyuan fault from Haiyuan to Jingtai. The main strata in the densely distributed area of landslides belong to Paleogene, Neogene, and Quaternary. In the area of Mesozoic-Precambrian strata, almost no landslides are observed. The Mesozoic-Precambrian strata are mainly distributed in the Liupanshan and Xihua Mountain areas. The landslides in the bedrock mountain area caused by earthquakes are mainly dominated by collapses, which is similar to the Wenchuan earthquake. Although the number and area of landslides caused by the earthquake are large, the volume of a single landslide tends to be small, and the surface morphology may be restored in a relatively short time. It is noted that substantial earthquake-induced landslides have not been identified in the Mesozoic to Precambrian strata. This may be because the number of co-seismic landslides is small and co-seismic landslides dominated by collapses are difficult to be identified. The results indicate that the strata affect the overall distribution of landslides.
In the area of densely distributed landslides, the differences in point density and area density distribution are mainly due to differences in the landslide area. In large landslide areas, the area is larger than the point density; in small landslide areas, the point density is greater than the area density. The degree Ⅸ area of the Haiyuan earthquake, where the area A, E, F's density is considerably greater than the point density, is dominated by the quaternary strata; the area I where the point density is significantly greater than the area density, is also in the quaternary strata distribution area. Other factors affecting the landslide distribution should be considered as well.4.2 Relationship between Landslide Distribution and Geomorphic Parameters
In this study, using ArcGIS, the geomorphic parameters (Figs. 7, 8) and some other attributes (Table 3, Fig. 9) of the Haiyuan earthquake intensity Ⅸ degree and the density of points in the adjacent area greater than 1/km2 (A-M) are extracted and counted. Fig. 7 and Fig. 8 show that co-seismic landslides induced by the Haiyuan earthquake mainly occurs in the area with slopes of 5°-25° and a relief range of 150-300 m. Adjacent regions are also selected in the study area, including area A, B, C, and D. Table 4 indicates that on the northeast side of the Haiyuan Fault, when area K, I, G, and J are close to the fault, the slope and relief of area K, I, G, and J are ordered as K > I > G > J (Fig. 7 and Fig. 8), the average area of landslide is ordered as K < I < G < J (Fig. 9), and the landslide density is ordered as K > G > I > J. Although the landslide density of G is greater than that of I, the landslide densities in these two areas are relatively close. Area H is closest to the Haiyuan fault at the northeast plate, and it exhibits the largest landslide density. The relief and slope in this area are medium (Figs. 7 and 8). Therefore, H is a landslide-concentrated area due to the minimum distance to the fault in the highest seismic-intensity area. The strata in area H are mainly tertiary mudstone and overlying loess. Under strong motion, loess may easily slide along the contact surface between itself and mudstone (Zou Jinchang et al., 1996). The average landslide area in area F is large, and the landslide density is small. When the relief and slope are not prominent, the aforementioned effects may be associated with lithological factors.
Area E is closest to the southwest plate of the Haiyuan Fault. Although the slope and relief therein are the smallest (Figs. 7, 8), this area exhibits the highest landslide density in the southwest plate of the fault (Table 3). Area C and D exhibits the smallest relief and slope after area E (Figs. 7, 8), but area C and D has the largest landslide area (Figs. 9) and the largest ratio of landslide area to total regional area. A large number of barrier lakes have developed in area C and D due to the high thickness of loess coverage (Zhuang Jianqi et al., 2018) and low-angle, high-speed, and long-distance landslides (Yuan Lixia, 2005). Areas L and A + B are near the fault, and the slope and relief in these areas are similar (Figs. 7, 8). Therefore, the landslide density and average landslide area of areas L and A + B are similar (Fig. 9), having a low density and small average area.
According to the characteristics of strata, relief, slope, landslide area, and distance to fault in each of areas A-L, it is found that the distribution of landslides is the result of the combined influence of various factors. The average area of the landslide is small in the area with a large slope and relief. The density of the landslide distribution is related to the distance of the area to the fault and is affected by strata distribution. The point density of the landslide in the loess distribution area in the south plate of the Haiyuan Fault show slight difference. However, as the distance to the fault increases, the area density of the landslide decreases. In the tertiary-strata-dominated north plate of the Haiyuan Fault, as the distance to the fault increases, the point density increases while the area density decreases. In addition to fault distance, the area density concentration area of the landslide is affected most significantly by strata distribution. In the loess area, the area density accounts is large. Therefore, the distribution characteristics of surface density (E, H) in the Xiji-Jingning landslide area (C, D) is larger than that in the Haiyuan earthquake macro-epicenter area.
Area M is the farthest from the southwest plate of the fault, and it is the landslide concentration area of the 1718 Tongwei earthquake (Xu Yueren et al., 2020b). This area has the largest relief and slope (Fig. 7 and 8) as well as the largest landslide density, area, and proportion of landslide area. The anomalies in area M verify that the vast majority of landslides in area M are associated with the Tongwei earthquake rather than the Haiyuan earthquake. The post-earthquake investigation also identifies several landslides south of Tongwei. However, by evaluating the scale, density, and landslide area ratio, we distinguish landslides in the Tongwei concentrated area from those in the Haiyuan high-density earthquake area. The concentrated area (M) of the Tongwei earthquake is mainly characterized by the co-seismic landslides of the Tongwei earthquake, being a co-seismic landslide area with a steep slope and large relief. Thus, the topography of this region is conducive to new landslides or re-sliding on existing landslides. Affected by the far-field effect of the Haiyuan earthquake, area M exhibits a number of co-seismic landslides following the Haiyuan earthquake. Most of the landslides induced by the Haiyuan earthquake in area M formed by the far-field effect are usually small in area and density, revealing why landslides with a small area have occurred in area M compared with regions C and D when the average area is large, and why the landslide area accounts for a large proportion of the total area.4.3 Relationship between Landslide Distribution and Population Death Distribution
Co-seismic landslides often lead to substantial casualties. More than 230 000 people died during the Haiyuan earthquake (Liu Baichi et al., 2003), and the main cause of the death in loess areas was house collapse (Wang Lanmin et al., 2017). At that time, residents in the area of the Haiyuan earthquake mainly lived in caves. The earthquake occurred at night, and almost all residents were in their homes. As a result, house collapses caused by the earthquake led to a large number of deaths (Close U. et al., 1922). Furthermore, the distribution of earthquake-induced landslides also affects total deaths. The investigation records after the earthquake indicate that landslides have resulted in the death of all residents in the village (Table 1).
To analyze the relationship between the earthquake-induced landslide and death toll, this study uses the death statistics of each district and county (township) in the historical records and counts the geographical location points along with the number of deaths(Lanzhou Institute of Seismology, 1984; Seismological Bureau of Ningxia Hui Autonomous Region, 1989). Using the nuclear density tool of the ArcGIS software, we create a density distribution map of the Haiyuan earthquake death population (Fig. 10). The areas with larger landslide surface density and point density are overlaid on the map. Figure 10 shows that the area with the highest population death density overlaps area H which has the most intensive landslide distribution and is the macro-epicenter area of the Haiyuan earthquake. The death population density exhibits a decreasing trend as Haiyuan > Xiji > Jingning > Tongwei. Although there are many death toll records and detailed data on the region near Tongwei, no areas of abnormal population death density are observed. Additionally, the density of population death is approximately symmetrical along the faults. However, in the position where the density of the landslide surface protrudes southward (area A), the density curve of the death toll also protrudes southward.
By analyzing landslide density distribution and death population density distribution, we can see that landslides aggravate death population. However, from Haiyuan to Tongwei, the trend gradually decreases. The Tongwei area is not an abnormally distributed area. The distribution of landslides refers to the distribution trend of death population density.4.4 Slope Direction Effect of Landslide Distribution
A back-slope effect is observed in the sliding direction of the earthquake-induced landslides (Huang Runqiu et al., 2008), that is, the landslide development density on the back-slope side of seismic wave propagation is significantly greater than that on the face-slope side. However, the co-seismic landslides of the Haiyuan earthquake show the opposite phenomenon, in which the landslide development density on the face-slope side is greater than that on the back-slope side (Zhuang Jianqi et al., 2018). Zhuang Jianqi et al. (2018) examine all landslide bodies when analyzing landslide sliding direction and do not consider the influence of the spatial distribution of landslides on the sliding direction. Since the landslides induced by the Haiyuan earthquake are mainly distributed on both sides of the river valley, in this study, the landslide distribution area is divided according to the drainage basin; thereby, these seven basins (A-G) can be identified. Because the number of landslides in E-G basins is small and the river flow direction is consistent, E-G basins are combined in our analysis. As a result, five regions are analyzed. There are 1 586 landslides in drainage basin A, 1 567 in B, 115 in C, 3 746 in D, and 536 in E-G. The slope aspect information is determined through DEM data, and the slope direction information is extracted by determining the sliding directions of the landslide points. The number of landslides in each area and the directional distribution of the accumulated area of landslides in each direction are statistically significant (Fig. 11).
Fig. 11 demonstrates that the number of landslides in each region is basically consistent with the direction of the accumulated landslide area. The landslides in basins A and B have obvious dominant sliding directions. The dominant sliding direction of basin A is 250°-310° and that of basin B is 20°-290°; the epicenter of the Haiyuan earthquake is located in the west of Haiyuan County. Therefore, the dominant direction of landslides in basins A and B exhibits the face-slope effect, which confirms the research results of Zhuang Jianqi et al (2018). However, the dominant slip directions of C, D, and E-G are not obvious, indicating that the slope effect of the Haiyuan earthquake landslide is obvious in the southwest plate rather than the northwest plate of the fault.
The number of landslides and the sliding direction of the landslide accumulation area in the entire Haiyuan earthquake region and the landslides concentration area of the Tongwei earthquake recounted. Finally, the distribution rose diagram of the sliding direction is drawn (Fig. 12). As shown in Fig. 12(a), the dominant direction of the number of landslides is 270°-0°, and the dominant directions of the landslide area are 250°-300° and 70°-90° for the co-seismic landslides in the Haiyuan earthquake. The river channel of basin A is mainly oriented along the north-south direction, which determines the dominant direction of the landslide area in the Haiyuan earthquake area. In the Tongwei landslide concentration area (Fig. 12(b)), the dominant landslide direction is 250°-320°, and the dominant direction of the accumulated landslide area is 50°-200°. The number of landslides and the dominant direction of the accumulated area of landslides are notably inconsistent. The Tongwei area is located in the influence area of the Haiyuan earthquake, and the southernmost landslide point after the earthquake is located near Gangu in the south of Tongwei (Fig. 4), indicating that the Tongwei area is located in the influence range of the Haiyuan earthquake, which has subsequently resulted in the co-seismic landslide of the Haiyuan earthquake. Combined with the analysis of the slope effect of co-seismic landslides from the Haiyuan earthquake, the dominant direction of landslide number and area in the Tongwei area is the "slope facing" direction of the Haiyuan earthquake, and small-area landslides are developed in this direction. These landslides may have been induced by the Haiyuan earthquake. According to the comprehensive characteristics of slope, relief, average landslide area, and landslide area proportion, it is suggested that the main composition of the landslide area is dominantly contributed by the co-seismic landslides of the Tongwei earthquake of 1718. Under the existing technical conditions, it is necessary to search other means to distinguish the co-seismic landslides of the two earthquakes more accurately in the same area. However, the landslide formed under the far-field effect may not have a significant impact on the analysis of the relevant parameters of the Haiyuan earthquake.
Based on multi-temporal high-resolution satellite image interpretation in Google Earth, field investigation, and seismic record reanalysis, combined with several analysis tools in ArcGIS, this paper presents the attribute characteristics of the Haiyuan co-seismic landslides. The following main conclusions are obtained:
(1) The landslides induced by the Haiyuan earthquake are mainly distributed in the area where Haiyuan fault intersects with Liupanshan fault, exhibiting multiple dense distribution centers.
(2) The distribution of landslides following the Haiyuan earthquake is determined by distance to the fault, topographic relief, slope, lithology, etc. As the distance to the fault decreases, the density of the landslide surface increase. With the slope and relief increase, and the density of landslide points increase while the average area of the landslide decreases. In the quaternary loess area, the average area of an individual landslide is larger and the density of landslide is lower. In the tertiary strata distribution area, the average area of landslide is small, and the landslide density is high.
(3) The death toll density curve following the Haiyuan earthquake can be used as a reference for the distribution of co-seismic landslides. The Tongwei landslide concentration area includes a small number of co-seismic landslides in the Haiyuan earthquake, but the main type of landslides in this area is the co-seismic landslides of Tongwei earthquake in 1718 rather than those triggered by the Haiyuan earthquake.
(4) The co-seismic landslides of the Haiyuan earthquake show a face-slope effect in the southwest plate of the fault and develop in the dominant sliding direction toward the fault and epicenter. In the northeast plate of the fault, the slope effect is not obvious, and no clearly dominant direction of the sliding is observed.
Chen Xiaoli, Ye HongChen Xiaoli, Ye Hong. Application of GIS for earthquake landslide research[J]. Earthquake Research in Shanxi, 2003(2): 17-19 (in Chinese with English abstract).
Close U., McCormick EClose U., McCormick E. Where the mountains walked[J]. National Geographic Magazine, 1922, 41(5): 445-464.
Delgado J., Garrido J., López-Casado C., Martino S., Peláez J.ADelgado J., Garrido J., López-Casado C., Martino S., Peláez J.A. On far field occurrence of seismically induced landslides[J]. Engineering Geology, 2011, 123(3): 204-213. DOI:10.1016/j.enggeo.2011.08.002
Deng Longsheng, Fan WenDeng Longsheng, Fan Wen. Deformation breakage characteristics and development mechanism of loess landslide triggered by Haiyuan M8.5 earthquake in Ningxia[J]. Journal of Catastrophology, 2013, 28(3): 30-37 (in Chinese with English abstract).
Department of Earthquake Disaster Prevention, State Seismological Administration. Catalogue of Historical Strong Earthquake in China[M]. Beijing: Seismological Press, 1995: 1-118 (in Chinese).
Du Peng, Xu Yueren, Tian Qinjian, Zhang Weiheng, Liu ShuangDu Peng, Xu Yueren, Tian Qinjian, Zhang Weiheng, Liu Shuang. The Spatial Distribution and Attribute Parameter Statistics of Landslides Triggered by the May 12th, 2008, MW7.9 Wenchuan Earthquake[J]. Earthquake research in China, 2020(1): 29-49.
Gu Gongxu. Catalogue of Chinese Earthquakes[M]. Beijing: Science Press, 1983: 1-116 (in Chinese).
Guo Fangfang, Yang Nong, Zhang Yueqiao, Meng Hui, Ye BaoyingGuo Fangfang, Yang Nong, Zhang Yueqiao, Meng Hui, Ye Baoying. GIS-based analysis of geomorphological factors for landslide hazards[J]. Journal of Geomechanics, 2008, 14(1): 87-96 (in Chinese with English abstract).
Huang Runqiu, Li WeileHuang Runqiu, Li Weile. Research on development and distribution rules of geohazards induced by Wenchuan earthquake on 12th May, 2008[J]. Chinese Journal of Rock Mechanics and Engineering, 2008, 27(12): 2585-2592 (in Chinese with English abstract).
Huang Xueqiang, Ma Junxue, Chen XiHuang Xueqiang, Ma Junxue, Chen Xi. Database of landslides triggered by 2015 Gorkha (Nepal) MW7.8 earthquake[J]. Seismology and Geology, 2018, 40(5): 1115-1128 (in Chinese with English abstract).
Keefer D.KKeefer D.K. Landslides caused by earthquakes[J]. GSA Bulletin, 1984, 95(4): 406-421. DOI:10.1130/0016-7606(1984)95<406:LCBE>2.0.CO;2
Lanzhou Institute of Seismology, State Seismological Bureau. Summary of Gansu Historical Earthquake Records[M]. Beijing: Seismological Press, 1984: 166-338 (in Chinese).
Li Weile, Huang Runqiu, Pei Xiangjun, Zhang Xiaochao. Historical co-seismic landslides inventory and analysis using Google earth: a case study of 1920 M8.5 Haiyuan earthquake, China. In: Lollino G., Giordan D., Crosta G.B., Corominas J., Azzam R., Wasowski J., Sciarra M. (Editors). Engineering Geology for Society and Territory. Cham: Springer, 2015: 709-712.
Liu Baichi, Zhang Junling, Wu Jianhua, Guo HuaLiu Baichi, Zhang Junling, Wu Jianhua, Guo Hua. Reevaluating on casualty in the Haiyuan MS8.5 earthquake on December 16, 1920[J]. Earthquake Research in China, 2003, 19(4): 386-399 (in Chinese with English abstract).
Ningxia Seismological Bureau. AD 1920 Haiyuan Great Earthquake[M]. Beijing: Seismological Press, 1980: 1-134 (in Chinese).
Peng Zhengkui, Pei XiangjunPeng Zhengkui, Pei Xiangjun. Relationship between landslides induced by Haiyuan earthquake Ⅸ degree and geological and geomorphological factors[J]. Journal of Chengdu University (Natural Science Edition), 2013, 32(4): 419-422 (in Chinese with English abstract).
Qiao Jianping, Zhao YuQiao Jianping, Zhao Yu. Review on rick degree regionalization of landslide[J]. Journal of Mountain Science, 2001, 19(2): 157-160 (in Chinese with English abstract).
Seismological Bureau of Ningxia Hui Autonomous Region. Summary of Ningxia Hui Autonomous Region Historical Earthquake Record[M]. Beijing: Seismological Press, 1989: 1-343 (in Chinese).
Wang Lanmin, Wu Zhijian, Xia KunWang Lanmin, Wu Zhijian, Xia Kun. Effects of site conditions on earthquake ground motion and their applications in seismic design in loess region[J]. Journal of Mountain Science, 2017, 14(6): 1185-1193. DOI:10.1007/s11629-016-3921-7
Xu C., Shyu J.B.H., Xu XXu C., Shyu J.B.H., Xu X. Landslides triggered by the 12 January 2010 Port-au-Prince Haiti MW=7.0 earthquake:visual interpretation, inventory compiling, and spatial distribution statistical analysis[J]. Natural Hazards and Earth System Sciences, 2014, 14(7): 1789-1818. DOI:10.5194/nhess-14-1789-2014
Xu Chong, Xu Xiwei, Yu GuihuaXu Chong, Xu Xiwei, Yu Guihua. Study on the characteristics, mechanism, and spatial distribution of Yushu earthquake triggered landslides[J]. Seismology and Geology, 2012, 34(1): 47-62 (in Chinese with English abstract).
Xu Chong, Xu Xiwei, Wu Xiyan, Dai Fuchu, Yao Xin, Yao QiXu Chong, Xu Xiwei, Wu Xiyan, Dai Fuchu, Yao Xin, Yao Qi. Detailed catalog of landslides triggered by the 2008 Wenchuan earthquake and statistical analyses of their spatial distribution[J]. Journal of Engineering Geology, 2013, 21(1): 25-44 (in Chinese with English abstract).
Xu ChongXu Chong. Landslide seismology geology:a sub-discipline of environmental earth sciences[J]. Journal of Engineering Geology, 2018a, 26(1): 207-222 (in Chinese with English abstract).
Xu Chong, Tian Yingying, Ma Siyuan, Xu Xiwei, Zhou Bengang, Wu Xiyan, Zhuang Jianqi, Gao Yuxin, Wu Weiying, Huang XueqiangXu Chong, Tian Yingying, Ma Siyuan, Xu Xiwei, Zhou Bengang, Wu Xiyan, Zhuang Jianqi, Gao Yuxin, Wu Weiying, Huang Xueqiang. Inventory and spatial distribution of landslides in Ⅸ-Ⅺ high intensity areas of 1920 Haiyuan (China) M8.5 earthquake[J]. Journal of Engineering Geology, 2018b, 26(5): 1188-1195 (in Chinese with English abstract).
Xu Chong, Tian Yingying, Shen Lingling, Ma Siyuan, Xu Xiwei, Zhou Bengang, Xu Chong, Wu Xiyan, Xu XiweiXu Chong, Tian Yingying, Shen Lingling, Ma Siyuan, Xu Xiwei, Zhou Bengang, Xu Chong, Wu Xiyan, Xu Xiwei. Earthquake-triggered landslides in the loess plateau and its adjacent areas[J]. Journal of Engineering Geology, 2018c, 26(S1): 260-273 (in Chinese with English abstract).
Xu Qiang, Li WeileXu Qiang, Li Weile. Distribution of large-scale landslides induced by the Wenchuan earthquake[J]. Journal of Engineering Geology, 2010, 18(6): 818-826 (in Chinese with English abstract).
Xu Yueren, Allen M.B., Zhang Weiheng, Li Wenqiao, He HonglinXu Yueren, Allen M.B., Zhang Weiheng, Li Wenqiao, He Honglin. Landslide characteristics in the Loess Plateau, northern China[J]. Geomorphology, 2020a, 359: 107150. DOI:10.1016/j.geomorph.2020.107150
Xu Yueren, Du Peng, Li Wenqiao, Zhang Weiheng, Tian Qinjian, Xiong Renwei, Wang Lin. A case study on AD 1718 Tongwei M7Xu Yueren, Du Peng, Li Wenqiao, Zhang Weiheng, Tian Qinjian, Xiong Renwei, Wang Lin. A case study on AD 1718 Tongwei M7. 5 earthquake triggered landslides——Application of landslide database triggered by historical strong earthquakes on the Loess Plateau[J]. Chinese Journal of Geophysics, 2020b, 63(3): 1235-1248 (in Chinese with English abstract).
Xu Yueren, Liu-Zeng Jing, Allen, Mark B., Zhang Weiheng, Du PengXu Yueren, Liu-Zeng Jing, Allen, Mark B., Zhang Weiheng, Du Peng. Landslides of the 1920 Haiyuan earthquake, northern China[J]. Landslides, 2020c.
Ye Ling. Distribution and motion characteristics of loess landslides caused by Haiyuan earthquake[D]. Chengdu: Chengdu University of Technology, 2013 (in Chinese with English abstract).
Yuan Daoyang, Lei Zhongsheng, Liu Xiaofeng, Liu Baichi, Cai Shuhua, Zheng WenjunYuan Daoyang, Lei Zhongsheng, Liu Xiaofeng, Liu Baichi, Cai Shuhua, Zheng Wenjun. Textual research on the historical data of the 138 AD Jincheng-Longxi M6
Yuan Lixia. The mechanism of loess landslide caused by earthquake in Haiyuan of Ningxia[D]. Xi'an: Northwest University, 2005 (in Chinese with English abstract).
Zhuang Jianqi, Peng Jianbing, Xu Chong, Li Zhenhong, Densmore A., Milledge D., Iqbal J., Cui YifeiZhuang Jianqi, Peng Jianbing, Xu Chong, Li Zhenhong, Densmore A., Milledge D., Iqbal J., Cui Yifei. Distribution and characteristics of loess landslides triggered by the 1920 Haiyuan Earthquake, Northwest of China[J]. Geomorphology, 2018, 314: 1-12. DOI:10.1016/j.geomorph.2018.04.012
Zou Jinchang, Shao ShunmeiZou Jinchang, Shao Shunmei. Characteristics of Haiyuan earthquake landslide and its distribution[J]. Inland Earthquake, 1996, 10(1): 1-6 (in Chinese with English abstract).