Topographic location and connectivity to channel of earthquake- and rainfall-induced landslides in Loess Plateau area

Abstract

The position of landslides on a slope plays a crucial role in determining landslide susceptibility and the likelihood of landslide debris interacting with the fluvial system. Most studies primarily focus on shallow landslides in the bedrock weathering zone or large-scale bedrock landslides, but the relevant work about the location and connectivity to channels of loess landslides is limited despite their potential to provide insights into slope stability and material transport in loess regions. In this study, we explored differences in landslide location and connectivity to channels between 2013 Mw5.9 Minxian earthquake-induced (EQ) landslides and 2013 Tianshui rainfall-induced (RF) landslides in the Loess Plateau area, China. The result shows that more than 37% of EQ landslides occur in the vicinity of ridges and ~ 30% are concentrated near river channels. Landslide locations of the Minxian earthquake not only occur in ridge crest areas but also exhibit clustering near the channels. We attribute the former cluster to seismic shaking along the ridge crest, and the latter cluster to dynamic changes in pore pressure within saturated lower hillslopes due to nearly a month of rainfall prior to the Minxian earthquake. Compared to EQ landslides, RF landslides are more evenly distributed across slopes. However, due to heavy rainfall and river erosion, landslides are more concentrated in the middle and lower slope areas, especially near the river channels. Moreover, the connectivity of landslides to channels indicates that RF landslides exhibit stronger connectivity with river channels compared to EQ landslides, which may be related to the concentration of EQ landslides near ridge areas. Furthermore, due to the smaller scale of EQ landslides compared to RF landslides, larger landslides are more likely to be located closer to river channels. This may contribute to the lower observed connectivity index between EQ landslides and river channels.

Introduction

Loess, a loose sediment deposited by wind in arid climates since the Quaternary period, is extensively distributed across the northern region of China with a total area of 640,000 km^1,2. Because of its porous and weakly cemented structure, loess exhibits unique physical and mechanical properties. When exposed to water saturation or intense ground motion, the structural characteristics of loess are significantly damaged, resulting in the occurrence of landslides^2,3,,4. In mainland China, about one-third of landslides occur in the Loess Plateau area, and earthquakes and rainfall are the two main controlling factors for landslides^5,6. Therefore, it is essential to study the location and formation mechanism of loess landslides triggering by these two controlling factors in the Loess Plateau area^7,8.

The location of a landslide on a slope significantly impacts landslide susceptibility, as it directly correlates with the connectivity to river channels^9,10. The relevant studies indicates that storms and earthquakes can impact different areas of a mountainous terrain, potentially leading to variations in the location of landslides triggered by seismic and rainfall events^9,11. Consequently, several studies have been conducted to compare the influencing factors and locations of landslides triggered by earthquakes and rainfall^12,13,,14. Bai, et al.¹⁴ compared the influencing factors and spatial distribution of RF and EQ landslides in the Upper Minjiang catchment, China. The results indicate that lithology and topography play a significant role in determining the distribution of landslides triggered by both rainfall and the 2008 Wenchuan earthquake. RF landslides tend to occur predominantly in the lower sections of slopes, whereas EQ landslides exhibit a more even distribution. Huang and Montgomery¹⁵ utilized a 5 m resolution digital elevation model (DEM) to analyze the topographic positions and dimensions of landslides triggered by the 1999 Chi-Chi earthquake and 2001 Toraji typhoon in the Tachia River basin of Taiwan area. The findings indicate that landslides induced by earthquakes tend to cluster near ridgetops, while those triggered by typhoons occur more frequently at lower elevations on slopes. Qiu, et al.¹³ selected three RF landslide databases and three EQ landslide databases globally to compare the relative locations of RF landslides and EQ landslides on slopes, and obtained similar conclusions.

Overall, most studies indicates that EQ landslides are more likely to occur in the mid to upper sections of slopes, while RF landslides tend to be more concentrated near river channels^9,13. Previous studies mainly focused on shallow landslides in the bedrock weathering zone or large-scale bedrock landslides. However, due to the unique mechanical characteristics of loess, such as the development of vertical joints, collapsibility, and large pores, to our knowledge, there is no relevant study on the location and connectivity to channels of loess landslides. Therefore, further study is needed to determine whether the same phenomenon exists in the location of loess landslides triggered by rainfall and earthquakes. In addition, Lin, et al.¹⁶ compared the positional changes of RF landslides before and after the Chichi earthquake, and the results showed that the locations of RF landslides before and after the earthquake changed significantly, and the locations of RF landslides shifted from mid-slope positions before the earthquake to ridge tops after the seismic event. For loess areas, do long-term rainfall events before earthquakes affect the changes in the locations of coseismic landslides?

To address these two issues, this study focuses on the Mw5.9 Minxian earthquake that occurred on July 21, 2013 and the Tianshui heavy rainfall event on July, 2013 in the southwestern region of the Loess Plateau area. We compared the landslide locations of earthquake and rainfall events using detailed landslide inventories. Additionally, we quantified the landslide-channel connectivity of different types of landslides in terms of landslide count, area, and volume. Finally, we discussed the effects of earthquakes and rainfall on the location and connectivity index of two different types of landslides. This study is beneficial for understanding the formation mechanisms and effectively mitigating the potential hazard of loess landslides in the loess plateau area.

Study area

The study area is located in the central part of Gansu Province and situated at the junction of the Loess Plateau and the Qinling Mountains. This area is characterized by strong tectonic activity and is also the area with the most severe geo-hazards in the Loess Plateau area^17,18. The study area is mainly characterized by three main active faults, namely the Northern margin of West Qinling fault (NWQLF), the Lixian-Luojiapu fault (LLF), and the Lintan-Dangchang fault (LDF). Additionally, the area has experienced several major historical earthquakes, such as the 1929 M8.5 Haiyuan Earthquake, the 1970 M7.5 Lanzhou Earthquake, and the 1654 M8.0 Tianshui Earthquake (Fig. 1).

Fig. 1

Map showing the historical earthquakes, tectonic characteristics, and distribution of 10 days accumulated rainfall before the Minxian earthquake of the study area; The rainfall data is downloaded from the Integrated Multi-satellitE Retrievals for GPM (IMERG)( https://gpm.nasa.gov/data/imerg), The data of active faults is from Xu, et al.¹⁹; The Peak ground motion (PGA) contours of Mw5.9 Minxian earthquake is downloaded from the Shakemap published by United States Geological Survey (USGS); NWQLF: Northern margin of West Qinling fault; LDF: Lintan-Dangchang fault; LLF: Lixian-Luojiabu fault. The figure was created using the ArcGIS 10.2 and CorelDRAW Graphics Suite X8.

The elevation distribution of the study area ranges from 700 to 2600 m, with an average elevation of about 1500 m. The terrain shows a southwest-high to northeast-low trend. The average temperature is 5.7 °C, with a maximum temperature of 32 °C and a minimum temperature of approximately − 26.3 °C. In July and August, the area has the highest temperatures, with average maximum and minimum temperatures of 23 °C and 10 °C, respectively. The study area has a temperate monsoon climate, with an annual average rainfall of about 600 mm. The rainy season mainly occurs from June to September, with these three months contributing to approximately 50% of the total annual rainfall, with July experiencing the highest precipitation, averaging around 150 mm per month.

In July 2013, the study area experienced two catastrophic events with the long-lasting rainfall event and the Mw5.9 Minxian earthquake. The heavy rainfall event lasted for 38 days from June 19 to July 26, during which four short-duration heavy rainfall events occurred in Tianshui area (Fig. 2). Under extreme rainfall condition, the Tianshui area occurred massive landslides, predominantly characterized by elongated and small-scale shallow landslides, loess collapses and mudflows^20,21. On July 22, the Minxian earthquake occurred along the Lintan-Dangchang fault (LDF), exhibiting a NWW-NW direction with characteristics of both southward thrusting and sinistral strike-slip motion²². The epicenter was located at 34.5°N, 104.2°E, with a focal depth of 20 km and a maximum PGA exceeding 0.12 g. The quake-affected area is located in the southwestern Qinling Mountains and covered by thick layers of loess (Fig. 1). Additionally, the area experienced continuous rainfall for nearly a month before the earthquake (Fig. 2), leading to the triggering of numerous shallow loess landslides, collapses, falls, and liquefaction under the action of ground motion^23,24.

Fig. 2

Rainfall data for every 12 h from June 7 to July 30 at rainfall station 57,014 in the Tianshui area. The location of the rainfall station is shown in Fig. 1.

We obtained the rainfall data from national rain gauge stations near the Tianshui area affected by the rainfall event. The results indicate that the rainfall event lasted for 38 days from June 19 to July 26, with a cumulative rainfall of nearly 500 mm. During this period, there were four short-term heavy rainfall events and occurred from June 19th to 21st, July 8th to 10th, July 21st to 22nd, and July 24th to 25th respectively (Fig. 2). The Minxian earthquake occurred after the third heavy rainfall event. Additionally, the IMERG satellite rainfall data from July 13th to 22nd indicates that the rainfall in the study area shows a trend of low in the west and high in the east (Fig. 1)^20,25. Overall, in the western area of Minxian earthquake occurred, the rainfall in the ten days prior to the earthquake was relatively low, ranging from approximately 10 to 50 mm. In contrast, the eastern Tianshui area experienced higher precipitation of over 60 mm, with the maximum precipitation of nearly 200 mm.

Materials and methods

Visual interpretation and landslide inventories

RF landslides associated with the Tianshui event

The landslide inventory was visually assessed by comparing Google Earth images taken before and after a continuous rainfall event. This comparison confirmed that the identified landslides were caused by the rainfall⁶. The pre-rainfall imagery was acquired in April 2012, and the post-rainfall imagery was obtained in October 2013. The landslide inventory of the Tianshui rainfall event indicates that this event triggered approximately 54,000 landslides^6,20. Among them, the largest landslide area reached 100,000 m², the smallest landslide area is only 30 m², and the average landslide area is 2,000 m². Most landslides have an area of less than 1,000 m², accounting for approximately 67% of the total number of landslides. We computed the landslide number density (LND) by Gaussian kernel density function and a radius of 2.5 km moving window. The LND map shows that the landslides are predominantly concentrated in two areas including the Niangniangba area and the Shetang area (Fig. 3)²⁰. The Niangniangba area is characterized as the most densely-developed landslides, with the highest LND of 100 /km². This area is mainly composed of Devonian mudstone, shale with thin interlayers of limestone, and the predominant landslide types include shallow rockslides, collapses, and loess-mudstone landslides. The Shetang area exhibits the highest LND of 130 /km² and is mainly composed of Quaternary loess and slope residual deposits, and Neogene mudstone. Most landslides in this area are primarily loess landslides and mudflows (Fig. 3).

Fig. 3

Map showing the spatial distribution of RF landslides, lithological units and landslide abundance index of Tianshui area; (a) The landslides distribution of the Tianshui rainfall event; (b) The landslide number density (LND) map; (c) The distribution of lithological units in Tianshui area. Table S1 shows the detailed lithological descriptions of different lithological units; The figures were created using the ArcGIS 10.2 and CorelDRAW Graphics Suite X8.

Coseismic landslides associated with the minxian earthquake

By comparing remote sensing images before and after earthquakes,the Min County earthquake landslide database can be obtained based on visual interpretation methods ^23,24. The pre-quake imagery primarily consists of SPOT images with a resolution of 10 m and THEOS images with a resolution of 15 m, focusing on the dates of May 29, 2012, and June 28, 2013. The post-quake imagery mainly includes Pleiades images with a resolution of 0.5 m and ZY-3 images with a resolution of 2.1 m with the post-event dates primarily being August 2, October 11, and October 24, 2013 ^23,24. The landslide inventory of the Mw 5.9 Minxian earthquake shows that this seismic event triggered at least 8,500 coseismic landslides, with a total area of approximately (2.2 km² )²⁴. The maximum area of the landslides was 46,300 m2, and the minimum area of a single landslide was only 4 m². Among them, there were 234 landslides with an area greater than 1,000 m², and the 3,156 landslides were small-scale landslides with areas less than 100 m². The landslide types are mainly shallow loess landslides and mudflows. In terms of the spatial distribution of landslides, the majority of landslides are primarily situated along both sides of the LDF and are predominantly concentrated within the area where the PGA exceeds 0.08 g (Fig. 4). Additionally, over 65% of the landslides are primarily located within the Quaternary loess (Q) and Jurassic sandstone (J) (Fig. 4). Due to weak consolidation of these two lithological units, landslides are more prone to occur. Conversely, only 5% of landslides occur in other lithological units with robust mechanical properties²⁴.

Fig. 4

Map showing the distribution of coseismic landslides and lithological units of Minxian area; (a) The distribution of active faults, PGA contours and landslides of Mw 5.9 Minxian earthquake; (b) The distribution of lithological units of Minxian area, Table S2 shows the detailed lithological descriptions of different lithological units. The figures were created using the ArcGIS 10.2 and CorelDRAW Graphics Suite X8.

Characterizing landslide locations

To examine the differences in occurrence of EQ and RF landslides, we quantified landslide locations of the Minxian earthquake and the Tianshui rainfall events based on the method of Meunier, et al.⁹. The landslide position in the slope can be described by two parameters including (1) the relative distance from the landslide scarp to the mountain ridge, and (2) the distance from the landslide toe to the river channels. Firstly, we generated the distribution of ridges and river channels in the study area based on 12.5 resolution ALOS PLASAR DEM data and TopoToolbox tool (https://github.com/BCampforts/topotoolbox). Then, we determined the relative position of individual landslide with respect to the mountain ridge ($\:{d}_{top}$) and the river valley ($\:{d}_{st}$) by Eq. 1. Landslides occurring along the x-axis usually originate near or at the crest, whereas those along the y-axis are linked to a channel. Landslides located far from both axes are categorized as mid-slope occurrences⁹.

$$\:{|d}_{st,top}|=\frac{{d}_{st,top}}{{d}_{st}+{d}_{top}}$$

(1)

Where $\:{d}_{top}$varies from 0 to 1 for a grid, 0 represents proximity to ridges, while 1 represents proximity to river channels. $\:{d}_{st}$ varies in the opposite way.

Connectivity to channel

Landslide mobility can influence the likelihood of landslide debris deposition within river systems. Within river channels, landslides can induce erosion by generating transportable sediment and further impact sediment dynamics^10,26. This sediment deposition, termed “connectivity” is a function of landslide runoff pathways and their intersections with river channels¹⁰. This study employs the indicator of landslide connectivity to channels to quantify the number, area and volume of landslides connected to channels. This indicator is also a crucial measure for assessing the connectivity between landslides and river channels^10,27. Specifically, we used gradient-upstream contributing area curve to divide the study area into five geomorphic process domains of hillslopes, valley heads, colluvial, bedrock and alluvial²⁸ (Fig. 5). The point of inflection on the gradient-upstream contributing area plot indicates a transition in terrain morphology under varying drainage conditions, thereby facilitating the delineation of location types within the study area. In this study, a drainage area of 0.2 km² is set as the threshold to represent a minimum upriver area for channelization (Fig. 5). When the landslide falls within this range, it is considered connected to channels. Drawing upon the channel connectivity index proposed by Li, et al.²⁶, which measures the proportion of landslides within the river channel, we further calculates the connectivity of EQ and RF landslides with river channels.

Fig. 5

Map showing the upstream contributing area versus hillslope gradient of the Tianshui and Minxian area respectively; The red dots represent the gradient-upstream contributing area of Tianshui area, and blue squares represent the gradient-upstream contributing area of Minxian area.

In addition, we calculated the volume of each landslide based on the established “volume–area” power law relationships. The study area is located in the loess area, and the majority of landslides are shallow soil landslides. Therefore, we adopt the proposed landslide area-volume scaling method for estimating soil landslide of Larsen, et al.²⁹ to estimate the volume (V) of each landslide. The scaling relationships—power-law equation is

$$\:\text{V}={\alpha\:A}^{\gamma\:}$$

(2)

where $\:\alpha\:$ and $\:\gamma\:$ are scaling parameters and A is the landslide area. The $\:\alpha\:$ and $\:\gamma\:$ of soil-based landslides is 0.146 and 1.145 respectively.

Result

The Landslide frequency-area distribution (FAD) curves can unveil the occurrence patterns of landslides^30,31. We compared the landslide area statistics for Tianshui RF landslides with Minxian EQ landslides³² (Fig. 6). Overall, the two landslide inventories exhibit distinct distribution characteristics in terms of landslide areas. For EQ landslides, the majority occur within the area range of 10 and 100 m², accounting for 45.6% of the total landslides. In contrast, RF landslides primarily concentrate within the area range of 100 and 1,000 m², constituting 62% of the total landslides, indicating that the scale of RF landslides is slightly larger than that of EQ landslides. According to the FAD curves, the two landslide inventories conform to the inverse gamma distribution proposed by³¹. As landslide area increases, the corresponding FAD curves initially rise to a peak and then exhibits a decreasing trend. Furthermore, the absolute scaling parameter (β) is calculated using the method proposed by³³. For the EQ landslide database, the absolute scaling parameter is -2.47, whereas for the RF landslide database, the absolute scaling parameter is -3.19. A higher absolute value of the scaling exponent implies that a lower proportion of large-scale landslides contributes to the total landslide number³⁰. The β value indicates a more concentrated distribution of landslide area for RF landslides, where fewer, larger landslides account for a greater proportion of the total area. In contrast, coseismic landslides, which have a lower scaling exponent, exhibit a more dispersed distribution, with a larger number of smaller landslides contributing to the overall area.

Fig. 6

Landslide area statistics for Tianshui RF landslides and Minxian EQ landslides; (a) Number of landslide within a area bin versus landslide area; (b) The distribution of frequency density versus landslide area.

Figure 7 illustrates the frequency density and landslide area density (LAD) of EQ and RF landslides at different hillslope gradient intervals. The results indicate that the majority of EQ landslides are concentrated in the range of 30° to 40°, while RF landslides predominantly occur in the range of 20° to 30°. Overall, compared to RF landslides, EQ landslides are more likely to occur in areas with larger hillslope gradient. In terms of LAD, LAD increases with the rise of hillslope gradient, with the highest LAD occurred in the range of 65° to 70° with the LAD value of 2%. However, RF landslides have the highest LAD between 20 and 55°, maintaining around 2.5-3%. In areas with hillslope gradient < 15° and > 55°, the LAD is relatively low.

Fig. 7

Map showing the distribution of Tianshui RF landslides and Minxian EQ landslides in different hillslope gradient ranges; (a) The frequency density of landslides in different hillslope gradients; (b) The frequency density of landslides in different hillslope gradients.

Figure 8 depicts the topographic location of RF landslides with areas greater than 1,000 m² and greater than 5,000 m², respectively. The results indicate that RF landslides exhibit a roughly uniform distribution across slope areas, but there is a cluster of landslides i.e. red area adjacent to rivers (relative distance to the river less than 0.2) (Fig. 8a). Specifically, over 36.6% of landslides are concentrated in the vicinity of rivers, while 30.5% of landslides are distributed in mid to lower slope positions with a relative distance to the river ranging from 0.2 to 0.4. It indicates that RF landslides predominantly occur in mid to lower slope areas with their deposition areas situated near river channels. This phenomenon is even more pronounced for large-scale landslides with area > 5000 m², where over 51.4% are located in areas adjacent to rivers (relative distance to the rivers less than 0.2) (Fig. 8b).

Fig. 8

Topographic location of the RF landslides triggered by 2013 Tianshui rainfall event in relation to the ridge crest and rivers. (a) The topographic locations of the landslides with an area larger than 1,000 m²; (b) The topographic locations of the landslides with an area larger than 5,000 m².

Figure 9 illustrates the topographic location of EQ landslides with areas greater than 50 m² and greater than 1,000 m², respectively. The results reveal that EQ landslides tend to concentrate in ridge areas, where the relative distance to the ridge is less than 0.2. Approximately 37.8% of landslides occur in the ridge areas, primarily encompass small-scale landslides. Additionally, cluster of landslides is observed near river channels (relative distance to the river less than 0.2), accounting for ~ 30% of the total number of landslides (Fig. 9a). It suggests that under the influence of river erosion, slopes adjacent to channels are susceptible to landsliding under the ground motion in loess areas. For EQ landslides with areas greater than 1,000 m², over 50% of landslides occur in areas near river channels (relative distance to the river less than 0.2), while landslides near ridge areas comprise only 25.8% of the total number of landslides. Overall, compared to RF landslides, EQ landslides are more likely to occur in ridge areas, especially small-scale landslides. Due to topographic site effects, a large number of small to medium-sized landslides occur on ridges. However, this phenomenon is not observed for large-scale landslides, and the edge of deposited area may be closer to river channels (Fig. 9b).

Fig. 9

Topographic location of the coseismic landslides triggered by 2013 Mw 5.9 Minxian earthquake in relation to the ridge crest and rivers. (a) The topographic locations of the landslides with an area large than 50 m²; (b) The topographic locations of the landslides with an area greater than 1,000 m².

Based on the criterion that mapped landslide areas overlap directly with identified channels suggested by¹⁰, we can calculate the landslide-channel connectivity of EQ and RF landslides respectively. The result shows that for RF landslides, over 5% of landslides reach the vicinity of river channels. These account for 13% of the total landslide area and 16% of the volume of landslide deposition, as larger landslides are more prone to reaching channels. For EQ landslides, approximately 2.8% of landslides reach the vicinity of river channels, corresponding to 4.5% of the landslide area and 7.2% of the landslide volume. In terms of the landslide-channel connectivity index, RF landslides exhibit better connectivity with rivers compared to the EQ landslides, with the number of RF landslides near channels being twice that of the EQ landslides.

Discussion

Rainfall and earthquakes are recognized as primary triggers of landslides^5,34. The relevant studies have shown that EQ landslides tend to concentrate near ridge crests due to enhanced ground acceleration from topographic effect, whereas RF landslides are more concentrated downslope within river channels⁹. Our results essentially confirm the validity of this hypothesis in the Loess Plateau region. However, it should be noted that landslide locations of Minxian earthquake not only occur in ridge crest areas but also exhibit the signature of clustering near the channels (Fig. 9). Additionally, the relationship between the landslides and hillslope gradient indicates that earthquake-triggered landslides tend to occur in the areas of steep hillslope gradient and LAD increases with the rise of hillslope gradient (Fig. 7). Steep areas often correspond to ridge crests and hillslope toes. The analysis of landslide locations also indicates clustering of landslides in ridge crests and hillslope toes areas, with fewer landslide occurrences in the central part of the slopes. We attribute the former cluster to seismic shaking along the ridge crest, and the latter cluster to dynamic changes in pore pressure within saturated lower hillslopes¹¹. Due to nearly a month of continuous rainfall prior to the Mw 5.9 Minxian earthquake, significant increases in loess moisture content occurred in the low slope areas caused by rainfall infiltration and surface runoff, which led to the saturation of the loess²⁰. Ground motion increased pore-water pressure, causing liquefaction and triggering landslides in mid- and lower slope areas³⁵. The influence of long-term rainfall events before earthquakes on the locations of co-seismic landslides appears to be a crucial factor. In the Minxian earthquake, nearly a month of rainfall before the event altered the hydrological conditions of the slopes, saturating the loess and increasing pore pressure. This suggests that antecedent rainfall not only exacerbated slope instability but also influenced the spatial distribution of co-seismic landslides, with clustering observed near river channels and mid-downslope areas. Therefore, the combined effect of rainfall and seismic events should be considered when assessing landslide hazards in regions prone to both triggers.

The previous studies have shown that rainfall-triggered loess landslides often occur on hillslope gradients ranging from 30° to 50°³⁶. Similarly, the Tianshui rainfall-triggered landslides exhibited analogous characteristics, with the most susceptible range being 25–45°, where the LAD reached up to 3% (Fig. 7). This phenomenon aligns closely with the statistical result of landslide locations, as over 60% of rainfall-triggered landslides are concentrated in the middle to lower slope areas, where the hillslope gradients roughly range from 20 to 55° (Fig. 7). This can be attributed to the fact that areas with the gradient of less than 30° can maintain stability even in saturated shallow soil mass, while steep slopes or upright slopes in loess are rarely subjected to rainfall, and areas with gradient ranging from 30° to 50° are prone to rainfall infiltration³⁶. Continuous rainfall infiltration leads to soil softening and increased soil-weight, rendering landslides more likely to occur within this hillslope gradient range of 30 ~ 50°. Furthermore, prolonged rainfall intensified river erosion adjacent to channels, resulting in erosion at the foot of slopes and enlargement of the exposed faces of valley slopes. Consequently, landslides triggered by rainfall are more likely to cluster in close proximity to river channels. Overall, compared to EQ landslides, RF landslides are more evenly distributed on slopes. However, due to heavy rainfall and river erosion, landslides are more concentrated in the middle and lower slope areas, especially near river channels.

After a catastrophic event, a significant amount of loose landslide debris accumulates at slope positions, providing a material source for potential subsequent landsliding^27,37. The connectivity of landslides to channels can indicate the potential for landslide debris to directly interact with the fluvial system, thereby facilitating the quantitative analysis of material transport from slopes to channels^10,26. Our results indicate that compared to EQ landslides, RF landslides exhibit stronger connectivity with river channels, which may be related to the concentration of EQ landslides near ridge areas. However, using the same methodology, we calculated the connectivity of Wenchuan EQ landslides with river channels, revealing that about 13% of landslides reach the vicinity of river channels, corresponding to 26% of the landslide area and 39% of the landslide volume, which is significantly higher than that of Tianshui RF landslides (Fig. 10). We attribute this phenomenon mainly to the larger average landslide area associated with the Wenchuan earthquake (~ 5,800 m2), compared to Tianshui RF landslides (~ 2,000 m²) and Minxian EQ landslides (~ 300 m²). Larger landslides are more likely to reach channels¹⁰, thus resulting in higher connectivity between landslides and channels for the Wenchuan earthquake. In contrast, the small scale of landslides of the Minxian earthquake and their concentration in the middle to upper slope areas contribute to its lower connectivity index with rivers (Fig. 11).

Fig. 10

Proportion of landslide polygons connected with river channels as defined by critical upriver area of 2008 Mw7.8 Wenchuan earthquake; The coseismic landslide data of Wenchuan earthquake is obtained from the landslide inventory of Xu, et al.³⁸.

Fig. 11

Proportion of landslides connected with river channels as defined by critical upstream contributing area; (a) Tianshui rainfall-induced landslides; (b) Minxian earthquake-induced landslides.

Although there are lithological differences between the Minxian earthquake-induced (EQ) and Tianshui rainfall-induced (RF) landslides, both areas are located in the Loess Plateau, where the geological conditions and soil type are similar, with loess being the dominant soil. While the looseness and water-solubility of loess may influence slope stability, this study finds that the primary factors affecting landslide occurrence and connectivity to river channels are slope gradient and rainfall conditions, rather than lithology. Future research could reduce the influence of lithology by comparing landslides in areas with varying lithologies and integrating these findings with factors such as slope and rainfall to better understand their combined effects.

Conclusion

The aim of this study is to identify the landslide locations triggered by earthquakes and rainfall events in the Loess Plateau area, and to quantify the number, area, and volume of landslides connected to channels. The results show that the majority of EQ landslides are concentrated in the range of 30° to 40°, while RF landslides predominantly occur in the range of 20° to 30°.Compared to RF landslides, EQ landslides are more likely to occur in areas with larger hillslope gradient. The landslide locations of different types of landslides indicate that EQ landslides are more likely to occur in the mid to upper sections of slopes, while RF landslides tend to be more concentrated near river channels. In terms of the landslide-channel connectivity index, Tianshui RF landslides exhibit higher connectivity with rivers compared to Minxian EQ landslides. On the one hand, it is due to the clustering of coseismic landslides in the ridge area, and on the other hand, it is because the scale of Tianshui RF landslides is higher than that of Minxian EQ landslides. Larger landslides are more like to reach channels, which results in higher connectivity between landslides and channels.