Quantitative risk assessment of landslides over the China-Pakistan economic corridor

https://doi.org/10.1016/j.ijdrr.2021.102441Get rights and content

Abstract

The China-Pakistan Economic Corridor (CPEC) connects southwest China and runs through Pakistan, a regional connectivity framework. However, priority construction of roads, railways, and other transport infrastructure is seriously threatened by landslides. Hence, we put forward a systematic evaluation index system to carry out a quantitative risk evaluation of landslides in CPEC. The evaluation process based on the index system mainly includes four main steps. In step 1, Based on landslide area density (LAD) and landslide point density (LPD) on topographic, geological, and geomorphic indicators, the landslide susceptibility in CPEC has been analyzed. In step 2, the landslides hazard assessment model was established using the susceptibility assessment result and the rainfall erosion intensity. In step 3, Combined with traffic facilities, population density, and property factors, the vulnerability evaluation index system was organized. In step 4, the risk assessment of landslides was proposed according to the disaster environment, hazard distribution, and vulnerability degree. Five levels of risk are classified: low, slight, moderate, high, and extremely high. 31.5% of the research areas are at the level of moderate risk or above. The research can further help transportation planning, resettlement region selection, and mitigation measure formulation in the landslide-prone areas.

Introduction

“The Belt and Road Initiative” runs through Asia, Europe, and Africa, covering dozens of countries. There are considerable differences in the natural environment and various types of disasters along the route. The landslides with frequent activities can cause severe harm and the “one belt and one road” countries and regions. Landslides result from coupling the earth's internal and external dynamics processes and represent a familiar geohazard, causing people's and economic loss by damaging major infrastructure projects, buildings, and transportation systems [1].

In general, the combination of natural and trigger conditions can lead to landslides. Natural conditions include relative altitude, lithology hardness, topographic relief, weathering, drainage density, and geological structure. In contrast, prolonged rainfall, snowmelt, seismicity, infiltration and runoff, temperature variation, volcanic eruption, and human engineering activity are trigger conditions. Landslides can travel long distances over slopes damaging buildings and traffic facilities that lie in their ways, while others are less destructive and confined [2]. Subject to scientific and technological conditions, the potential damage from dangerous landslides was enormous in developing countries. Casualties and property losses due to landslide instability are more significant in developing countries than in the industrialized world [3]. Both China and Pakistan belong to developing countries, so landslide risk assessment is urgently needed in China-Pakistan economic corridor construction. More than a dozen strong earthquakes of Mw.7 or above had occurred over the past 20 years. The Belt and Road countries have a twice relative than the worldwide average disaster losses based on the global disaster database. Chiefly eastern and southern Asia is ten times higher than the worldwide average [4]. Therefore, it is vital to carry out a landslide risk assessment of the China-Pakistan Economic Corridor (CPEC).

Rapid risk assessment of landslides could provide valuable information in the emergency response phase. In the current evaluation system, regional-scale risk mapping units are constituted by pixels [5]. In landslide risk analysis, the risk is expressed as a function of the probability of landslide occurrence hazardous process, the exposed elements at risk, and their vulnerability [6]. A landslide occurrence probability condition represents the predicted probabilities of a landslide for each pixel in the presence of a set of evaluation factors. Lots of statistical methods have been used to carry out landslide susceptibility in different countries. In Japan, Ayalew and Yamagishi [7] produce two landslide susceptibility maps by combining the analytical hierarchy process and logistic regression method. Peng et al. [8] applied the support vector machine method to generate landslide susceptibility maps in China. Trigila et al. [9] compared the logistic regression and random forest models for landslide susceptibility evaluation in Italy. Pourghasemi and Kerle [10] create a landslide susceptibility map by the random forest approaches in Iran. In Pakistan, Khan et al. [11] draw the landslide susceptibility map of northern Pakistan using frequency ratio. Sajid et al. [12] determine a landslide susceptibility index by the weighted overlay method in Karakoram Highway of Pakistan. In Malaysia, Tien Bui et al. [13] using the support vector machine and index of entropy models to draw landslide susceptibility mapping in Cameron Highlands.

The landslide hazard is evaluated as the multiplication of temporal probability and spatial probability of a landslide occurrence [14]. Temporal probability is the annual probability that landslides will hit a given place within a specific hazard class. Spatial probability is the spatial distribution density of landslides in the susceptibility class. Several methods have been used for landslide hazard assessment, such as the rational method, decision tree method, empirical probability, and artificial neural network [15]. However, the most common approach for landslide hazard analysis was the susceptibility map of landslides multiplied with a trigger condition such as rainfall at different frequencies. In most cases, researchers consider susceptibility maps of the landslide as the only component of landslide hazard assessment because of data scarcity [16,17].

Landslide vulnerability is defined as the destroying degree caused by a landslide to a specific object or an element at risk at a particular scale [6]. The vulnerability assessment model can be obtained with the study area's social, economic, physical, and environmental conditions. At present, many landslide vulnerability assessment methods have been performed by several studies. Quan et al. [18] develop physical vulnerability curves for debris flow based on the dynamic run-out model. Fotopoulou et al. [19] make an analytical method for assessing the vulnerability of low-rise reinforced concrete buildings subjected to an earthquake. Silva et al. [20] used building resistance and the landslide magnitude method to evaluate buildings' physical vulnerability in the North of Portugal. Lin et al. [21] create a threshold curve by an empirical relationship, estimating potential landslide vulnerability.

The term element at landslide risk covers many parameters, including the population, environment, project, and properties affected by the landslide occurrence [22]. The amount of risk varies based on the elements in the area affected by the landslide. Therefore, data on landslide risk elements must be gathered for specific fundamental spatial units, including administrative division units, major tectonic elements, or matrix cells with comparable attributes concerning a hazard. Landslide risk assessment can be defined as the expected number of people injured, life lost, infrastructure damaged, or financial disrupted due to landslides for a designated area and a reference period [6].

In this study, many landslides were located in the China-Pakistan economic corridor area resulting in severe financial losses. Therefore, a detailed inventory database and accurate risk evaluation of landslides support the transportation infrastructure construction, ecological restoration, and mitigation projects construction in the China-Pakistan economic corridor area.

The China-Pakistan economic corridor starts in Kashgar in China's Xinjiang province. It ends in the port of Darfur in Pakistan, which has a length of about 3000 km and belongs to an essential part of the silk road in the twenty-first century (Fig. 1). Highways, railways, oil trade corridors, and cable channels were widely scattered in the study area. CPEC has a significant north-south topography gap, Khunjerab Pass is approximately 4700 m, Islamabad is around 500 m, and the Kashgar urban area is about 1300 m above sea level. The massive drop between the north and the south leads to the enormous vertical difference of elevation along the CPEC. Pakistan's annual rainfall is so intense that it can cause a lot of geological disasters. The drop of valley and ridge along CPEC can reach 3000 m, and the steep topographic conditions provide favorable dynamic conditions for landslides and debris flow [23].

At 11:30 a.m. on January 4th, 2010, a massive landslide occurred in Attabad village, located across the China-Pakistan Road in the Hunza River valley in northern Pakistan, and blocked the Hunza River from a weir dam. According to the Pakistan National Disaster Management Authority data, in July 2010, over 115 houses had been destroyed, twenty people were missing, and the Karakoram highway was buried nearly 16 km. The vast landslide barrier lake blocked the China-Pakistan Road for more than four years, severely affecting economic and trade ties between the two countries [24,25].

On October 8th, 2005, an Mw. 7.6 earthquake struck Pakistan, killing 73,000 people and leaving 3.5 million people without their homes [26,27]. With a focal depth of 26 km, the epicenter was located on the main boundary thrust about 18 km northeast of Muzaffarabad. This earthquake affected about 30,000 km2, where highway systems, infrastructure, cropland, and electricity and communications services were severely damaged. Simultaneously, many landslides in the epicentral area near the Muzaffarabad and Balakot were triggered by the Kashmir earthquake. The Hattian Bala landslide was the most massive landslide with a volume that can reach 68 × 106 m3, destroying a village and killing more than 1000 people [2]. Kashmir earthquake also produced lots of cracks that made the slope very unstable and might lead to landslides in the future [4].

A landslide inventory database is an important instrument to carry out landslide susceptibility, hazard, vulnerability, and risk assessment [28]. Based on the interpretation of satellite images, field investigation, and previous publications' compilation, a detailed landslide database can be established [29]. SPOT-5 satellite image with a spatial resolution of 2.5 m was used to interpret the typical landslides carefully. Landsat-8 satellite images with a spatial resolution of 30 m were used to map the NDVI and residential in the study region (Table 1). The Digital Elevation Model (DEM) with a spatial resolution of 30 m was used to compute slope, aspect, elevation, distance to the river, and ground variability for the study area. The lithology and PGA factors have been extracted from the 1:50000 scale geological and seismo-tectonic maps [11,30].

As the landslide threat area gradually expanded, field investigation, remote sensing interpretation, and quantitative risk assessment are necessary to carry out [31,32]. Some typical landslides were identified by comparing the remote sensing images before and after the slide (Fig. 2). Combined with field investigation, a large number of landslides have posed a severe threat to transportation infrastructure (Fig. 3). According to the multi-source data coupling survey, 4518 landslides in the study area cover an area of 1.014 × 106 m2.

Through sorting out, collecting, and analyzing the research method at home and abroad [[33], [34], [35], [36]], we construct the landslide susceptibility evaluation system from the topographical, hydrology and geological structure. The slope, aspect, elevation, ground variability, lithology, PGA, distance to river, distance to fault, distance to road, and NDVI are evaluation factors in establishing the landslides susceptibility assessment model for CPEC. Different evaluation factors have different types: continuous (slope, elevation, ground variability, distance to river, distance to fault, distance to road, and NDVI) and discrete (aspect, lithology, and PGA). Based on previous classification standards and the Natural breakpoint method [37,38], we unify all the evaluation factors. In constructing the evaluation system of landslide susceptibility, although the evaluation factors closely related to landslides are selected, these factors may not be completely independent of each other. These factors may have a specific correlation with each other. Therefore, blindly pursuing more indicators without dealing with them will have the opposite effect. In this paper, Pearson correlation was used to analyze the independence of these evaluation factors. Using the Pearson method, 16 evaluation factors related to soil permeability coefficient were studied, and their correlation was determined [39]. Therefore, it is feasible to use Pearson correlation to judge the independence of the factor in this paper. If the pairwise correlation value is less than 0.5, it proves that there is good independence. Therefore, all the indexes met the requirement of being independent of each other (Table 2). Based on reclassification and spatial statistics, ten evaluation factors were divided into different grades. We calculated the landslide areal density (LAD) and the landslide point density (LPD) in different grades of each influencing factor (Fig. 4) [40]. The LAD was defined as the ratio between the coseismic landslide area and the total area under each class of ten factors, and the LPD interpreted as the number of coseismic landslides per square kilometer affected by an earthquake.

The slope factor usually refers to the ratio of the slope surface's vertical height to horizontal distance, directly reflecting the landslides' dynamic characteristics and potential energy. The number and area of landslides increase with the slope increase (Fig. 4, Fig. 5a). The aspect factor is the direction of the slope average's projection on a horizontal plane, which can regularly reflect the massif's basic topographic features [35,41]. The northeast and north part aspect have little vegetation coverage, which can severely produce soil erosion was likely to induce landslides (Fig. 4, Fig. 5b). A more significant altitude difference could obtain higher potential energy. The spatial distribution diagram between the landslides and elevation shown that it is prone to landslides in an elevation range of 2000 m–3500 m (Fig. 4, Fig. 5c). The ground variability is the primary natural landform factor that distorts the topographic surface, directly restricting surface materials and energy redistribution. The ground variability is divided into profile-type and plane-type according to their curvature [42,43]. Formula (1) shows the calculated method of the ground variability factor.KV=KP×KCKV represents the ground variability factor; KP represents the profile curvature; KC represents the plane curvature: the higher ground variability, the more frequent the landslide activity (Fig. 4, Fig. 5d). The PGA refers to the acceleration of ground motion during an earthquake and the basis for determining the intensity [42]. The diagram reflects PGA factors with values range from 0.39 to 0.45 can induce many landslides (Fig. 4, Fig. 5e). The lithology can directly reflect the necessary geological conditions, affecting landslides' distribution characteristics [44]. The research area is mainly composed of sandstone and gneiss, which can provide favorable geological conditions for the occurrence of landslides (Fig. 4, Fig. 5f). The distance from the river reflects the steepness of the canyon. With the river's distance increase, the LAD and LPD gradually decreased (Fig. 4, Fig. 5g). The landslide concentration decreases with distance from the seismogenic fault; the landslide concentration's highest value is located within a distance of 10 km (Fig. 4, Fig. 5h). The distance from the road reflects the intensity of human activities. Statistics show that landslides' area density is concentrated within 15 km on both sides of the road (Fig. 4, Fig. 5i). The surface vegetation coverage is less, which is conducive to the development of landslides. More landslides in the range of 0.5–0.6, accounting for 66.78% of the total (Fig. 4, Fig. 5j).

This paper has adopted the improved weight of evidence method for landslides susceptibility regionalization [45], which considers the landslide's occurrence probability under different evaluation factors grading intervals. The calculation formula is as follows:Wmj+=ln(Gmj1Gmj1+Gmj2Gmj3Gmj3+Gmj4)Wmj=ln(Gmj2Gmj1+Gmj2Gm4Gmj3+Gmj4)Cmj=Wmj+WmjP(mj)=exp{cmj+lnPL}where m-ten evaluation factors; j-different interval grades in ten evaluation factors; Gmj1- the j grading interval under the m evaluation factor, the grid quantity of landslides; Gmj2-outside the j grading interval under the m evaluation factor, the grid quantity of landslides; Gmj3- the j grading interval under the m evaluation factor, the grid quantity without landslides; Gmj4- outside the j grading interval under the m evaluation factor, the grid quantity without landslides; Wmj+-the positive correlation weight; Wmj -the negative correlation weight; Cmj- the degree of correlation between the j grading interval under the m evaluation factor and landslides; PL -the grid quantity of landslides divided by the grid quantity of research area, which is the prior probability; P(mj)- in the j grading interval under the m evaluation factor, the occurrence probability of landslides is the conditional probability. Parameter values of ten evaluation factors are shown in Table 3.

Many scholars have used the analytic hierarchy process (AHP) to research the weight of landslide susceptibility evaluation factors [46,47]. For example, according to the research of experts [48]; Irum et al., 2018; [49], the weight of the ten evaluation factors in this paper was determined by the analytic hierarchy process method. The analytical hierarchy process includes calculating the maximum eigenvalue, selecting the discriminant matrix, and standardization of the index weight [50]. (6), (7) were used to test the matrix's consistency, and the eigenvector was normalized.CI=λmaxnn1CR=CIRI

CI - the coincidence indicator; n - the order of the matrix; λmax - the maximum eigenvalue of the judging matrix; RI - the average random consistency index; According to the construction discrimination matrix (Table 4), the values of λmax Moreover, CR is 10 and 0, respectively. Generally, if CR < 0.1, the judgment matrix is considered to pass the consistency test; otherwise, it does not have satisfactory consistency. The calculation results show that the construction of the discriminant matrix is practical. After normalization, the weight of ten evaluation factors is shown in Table 4. Equation (8) was used to calculate the landslide susceptibility evaluation in the CPEC.Sm=m=1n(j=1mP(mj)×Wm)where m = 1, 2 … …10; Sm-the susceptibility value of landslides; Wm- weight of ten evaluation factors; According to the administrative division of CPEC [51], the smallest administrative unit in this paper is county-level administrative units, with 147 evaluation units. Based on the natural breakpoint method [52], five landslide susceptibility categories have been distinguished within the study's scope (Fig. 6). The receiver operating characteristic curve (ROC curve) is adopted to verify the sensitivity accuracy [53]. The area under the curve (AUC) can reflect the model's accuracy. The larger the AUC, the better the model. The AUC values of the improved weight of evidence methods were 0.915 (Fig. 6). Areas of moderate sensitivity and above are mainly concentrated in the northern part of Pakistan.

Due to the lack of rainfall data at different frequencies in Pakistan, the hazard of landslides under different recurrence periods is difficult to obtain. However, some average rainfall data was collected from the research area's weather bureau (Fig. 7a). Therefore, this paper adopted rainfall erosion intensity as the landslide average hazard evaluation factor. Rainfall provides necessary hydrological conditions for loose solid matters and acts on the surface in the form of raindrop splashing and runoff erosion, which leads to soil erosion and affects the stability of the slope body to a certain extent. The rainfall erosion intensity was computed using the empirical model of Wischmeier [54], which reflects the impact of rainfall intensity on soil erosion and can reflect the time probability of landslide occurrence [55], as shown in formula (9).F=i=112{1.735×10(1.5×lgpi2p0.8188}

F - rainfall erosion intensity, P - average annual rainfall (mm), Pi - average monthly rainfall (mm). The distribution map of rainfall erosion intensity can reflect the multi-year average rainfall probability in time and indicate the soil erosion probability in spatial (Fig. 7b). The hazard assessment of landslides in CPEC is evaluated to multiply landslide susceptibility categories and rainfall erosion intensity (Fig. 8). The extremely high and high landslide hazard areas were distributed in the northern part of Pakistan, while the plain's low hazard areas were distributed. Areas of high hazard tend to be areas of high rainfall erosion.

Landslide vulnerability to a particular phenomenon should be evaluated when performing the landslide risk assessment. Landslide vulnerability is defined as the degree of damage to a given element within the area affected by the landslide hazard. The vulnerability evaluation indicators generally recognized by the international community can be divided into two categories, including material vulnerability, social vulnerability [6]. Material vulnerability mainly refers to analyzing the relationship between tangible assets' distribution characteristics in the research area affected by landslides and the extent of damage [56]. The research area's material vulnerability factors consider the traffic facilities, divided into three levels based on the traffic grade (Table 5). The value of each traffic level was counted (Fig. 9a). Social vulnerability mainly considers population factors' influence in the study area, and population density is the best indicator. The higher the population density, the more vulnerable the population was to the impact of landslides. Therefore, the distribution of population density was calculated (Fig. 9b).

The vulnerability can be defined as a scale ranging from 0 to 1, where 0 represents no damage, and 1 represents complete damage. In this paper, the natural breaks method is used to classify the vulnerability. It identifies breakpoints by picking the class breaks that best group similar values and maximize the differences between classes. The features are divided into classes whose boundaries are set where there are relatively big jumps in the data values. The vulnerability map was created after normalization and divided into five classes, namely, low, slight, moderate, high, and extreme high (Fig. 10). The areas with high and extremely high vulnerability to landslides in the whole region of CPEC are mainly located near Pakistan's capital and in the central plain's open area. The plain's open area is suitable for human habitation with high population density, so the vulnerability is naturally higher.

The quantification of the elements at risk is an essential link to landslide risk evaluation. Elements at risk are one of the primary spatial data layers required for a total risk calculation [6]. Thus, landslides risk assessment is a comprehensive research of landslide's threat level, resilience, and exposure. Measures of exposure can include the number of people or types of assets in an area. These can be combined with the specific vulnerability and capacity of the exposed elements to any particular hazard to estimate the quantitative risks associated with that hazard in the area of interest. In general, the more developed the economy is, the higher the income will be, and the higher the loss will be when threatened by the landslide [56]. Therefore, the asset (GDP) of 147 units was used as an index to represent the exposure factor (Fig. 9c). Therefore, this paper combined the hazard, vulnerability, and exposure to calculate the landslide's risk value.R=H×V×E

In the formula, R - the landslides risk in CPEC; H - the landslides hazard; V - the landslides vulnerability; E - the exposure; Landslide risk is divided into five grades in CPEC (Fig. 11). The AUC value is 0.931, proving that the risk model can be effectively applied in the landslide risk assessment of CPEC. The results show that the extremely high and high risk areas are mainly distributed in Pakistan's northern part. The landslide has seriously threatened the traffic facilities and the safety of people and property in these areas. The risk map can guide the route selection of future roads and other transportation facilities in time.

Section snippets

Conclusion and discussion

After a systematic analysis of susceptibility, hazard, and vulnerability, this paper constructs a landslide risk assessment model suitable for the CPEC research area (Fig. 12). The results prove that the landslide hazard level is generally higher in the northern part of the CPEC. The value of vulnerability is higher in the central and southern areas of the CPEC than in other regions. According to the statistics area of CEPC under different risk levels (Table 6), 31.5% of the research areas are

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledge

The research was supported by the Second Tibetan Plateau Scientific Expedition and Research (STEP) program (Grant No. 2019QZKK0902), the International Science & Technology Cooperation Program of China (2018YFE0100100), the National Natural Science Foundation of China (42077245), and Open fund of Key Laboratory of Mountain Hazards and Surface Process, Institute of Mountain Hazards and Environment, Chinese Academy of Sciences. A special acknowledgment should be expressed to the China-Pakistan

References (56)

  • U. Kamp et al.

    GIS-based landslide susceptibility mapping for the 2005 Kashmir earthquake region

    Geomorphology

    (2008)
  • H.X. Chen et al.

    A physically-based multi-hazard risk assessment platform for regional rainfall-induced slope failures and debris flows

    Eng. Geol.

    (2016)
  • F.C. Dai et al.

    Landslide risk assessment and management: an overview

    Eng. Geol.

    (2002)
  • T.L. Saaty

    A scaling method for priorities in hierarchical structures

    J. Math. Psychol.

    (1977)
  • C.T. Lee

    Statistical seismic landslide hazard analysis: an example from Taiwan

    Eng. Geol.

    (2014)
  • S.Y. Yang et al.

    Characterization of landslide distribution and sediment yield in the TsengWen river Watershed, Taiwan

    Catena

    (2019)
  • S. Lacasse et al.

    Landslide Risk Assessment and Mitigation Strategy

    (2009)
  • L. Peng et al.

    Quantitative risk analysis for landslides: the case of the Three Gorges area, China

    Landslides

    (2014)
  • A. Triglia et al.

    Comparison of logistic regression and random forests techniques for shallow landslide susceptibility assessment in Giampilieri (NE Sicily, Italy)

    Geomorphology

    (2015)
  • H.R. Pourghasemi et al.

    Random forests and evidential belief function-based landslide susceptibility assessment in Western Mazandaran Province, Iran

    Environ. Earth Sci.

    (2016)
  • A. Sajid et al.

    Landslide susceptibility mapping by using a geographic information system (GIS) along the China–Pakistan Economic Corridor (Karakoram Highway), Pakistan

    Nat. Hazards Earth Syst. Sci.

    (2019)
  • D. Tien Bui et al.

    Landslide detection and susceptibility mapping by Airsar data using support vector machine and index of entropy models in Cameron Highlands, Malaysia

    Rem. Sens.

    (2018)
  • J. Zhang et al.

    How size and trigger matter: analyzing rainfall and earthquake triggered landslide inventories and their causal relation in the Koshi River basin, central Himalaya

    Nat. Hazards Earth Syst. Sci.

    (2019)
  • M. Chang et al.

    Hazard assessment of debris flows in the Wenchuan earthquake-stricken area, southwest China

    Landslides

    (2017)
  • B. Quan Luna et al.

    The application of numerical debris flow modeling for the generation of physical vulnerability curves

    Nat. Hazards Earth Syst. Sci.

    (2011)
  • S.D. Fotopoulou et al.

    Vulnerability assessment of reinforced concrete buildings subjected to seismically triggered slow-moving earth slides

    Landslides

    (2013)
  • M. Silva et al.

    Assessment of physical vulnerability and potential losses of buildings due to shallow slides

    Nat. Hazards

    (2014)
  • Q. Lin et al.

    The vulnerability of people to landslides: a case study on the relationship between the casualties and volume of landslides in China

    Int. J. Environ. Res. Publ. Health

    (2017)
  • Cited by (0)

    View full text