
==== Front
Heliyon
Heliyon
Heliyon
2405-8440
Elsevier

S2405-8440(24)12831-9
10.1016/j.heliyon.2024.e36800
e36800
Research Article
Evaluation of landslides susceptibility in Southeastern Tibet considering seismic sensitivity
Yeqi Zang a
Yonggang Guo 1960373107@qq.com
ab⁎1
Guowen Wang a
Shengjie Wu a
a College of Water Conservancy and Civil Engineering, Tibet Agriculture and Animal Husbandry University, Linzhi, China
b Research Center of Civil, Hydraulic and Power Engineering of Tibet, Linzhi, China
⁎ Corresponding author. Research Center of Civil, Hydraulic and Power Engineering of Tibet, Linzhi, China. 1960373107@qq.com
1 Yonggang Guo, Professor, doctoral supervisor, mainly engaged in strong earthquake safety monitoring of water conservancy and hydropower engineering, seismic risk safety evaluation of hydraulic structure and disaster prevention research.

23 8 2024
30 9 2024
23 8 2024
10 18 e3680019 2 2024
30 7 2024
22 8 2024
© 2024 The Authors
2024
https://creativecommons.org/licenses/by-nc-nd/4.0/ This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Southeastern Tibet features a complex geological environment and a high incidence of earthquakes. Earthquake-induced chain disasters pose a great threat to engineering construction and public safety in this area, and landslides are among the most frequent postearthquake disasters. To investigate the impact of earthquakes on landslides, this study constructed a comprehensive database for landslide susceptibility analysis based on various factors, including elevation, slope, slope direction, distances from roads and rivers, proximity to faults, land use patterns, rainfall patterns, and seismic parameters. By integrating the frequency ratio (FR) model with the analytic hierarchy process (AHP) model, this work delineated landslide susceptibility zones in southeastern Tibet. Subsequently, the susceptibility zoning layer was overlaid with the magnitude sensitivity layer and validated using ROC curve analysis to identify the earthquake magnitudes that exerted the greatest influence on landslides. Finally, by incorporating the distance from earthquake epicentres into our refined model framework, different monitoring levels for landslide susceptibility zoning were established. The AHP results show the relative importance of the landslide-influencing factors in southeastern Tibet can be ranked as follows: elevation, slope direction, distance from road, land use, distance from river, slope, and rainfall. The ROC values of the landslide models with seismic sensitivities of 1, 2 and 3 are 0.876, 0.883 and 0.877, respectively, indicating that earthquakes of magnitude 4 and above have a great influence on landslides in the study area. Through the overlay of the landslide susceptibility zoning map and the vector map of distance from seismic focal points, a correlation with the distance between landslide-prone areas and seismic focal points is identified. Within the extremely high susceptibility area and within 40 km of the focal point, there are 220 landslide points, accounting for approximately 34 % of the total landslides in the study area. Additionally, 133 landslide points are located in the extremely susceptible area and within 40–80 km of the focal point, representing approximately 20 % of the total landslides in the study area. The susceptible areas were assessed based on grades, resulting in the production of 4 maps depicting different levels of monitoring for landslide-prone areas. These maps are valuable tools for implementing landslide disaster prevention measures within the study area.

Keywords

Landslides
Earthquake
Evaluation of susceptibility
Sensitivity analysis
==== Body
pmc1 Introduction

As the most widely distributed and frequent natural disaster in nature, landslides pose a great threat to human life safety and economic development, so disaster prevention and reduction has become a hot topic in human development [1]. The factors causing landslide mainly include human factors and natural factors.Among the natural factors, earthquake is one of the main causes of landslide [2].The energy from the earthquake changes the geological structure of the landslide.The change of geological structure affects the stability of the mountain, so it is easy to cause landslides [3,4].To select a group of suitable influencing factors for landslide susceptibility evaluation, it is necessary to understand the main causes of landslide in advance.Analyzing the relationship between these main factors and the resulting landslide and mapping the landslide prone area can effectively prevent the loss caused by landslide.

China is one of the countries with the highest incidence of geological disasters worldwide [5]. By investigation, approximately 4772 geological disasters occurred nationwide in 2021, among which approximately 2335 were landslide disasters, accounting for approximately 49 % of the total number of geological disasters [6]. Therefore, landslides represent one of the main types of frequent geological disasters in China. The geological environment in Tibet is complex, and disasters are frequent. Southeastern Tibet includes the Nyingchi and Qamdo regions located in eastern and southern Tibet, which are seismically active and have poor slope stability [[7], [8], [9]]. The region has a monsoon climate related to the Indian Ocean, frequent seasonal rainstorms, and glaciers, and abundant water sources, which provide favourable conditions for landslide disasters [10,11].Earthquake is one of the main causes of landslide in southeast Tibet [8,12].The main causes of the Chayu landslide, the 102 landslide and the 2017 Yarlung Zangbo landslide were all earthquakes.Therefore, considering the seismic factors, mapping the landslide prone areas in southeast Tibet can effectively predict the occurrence of landslides, do a good job in preparing for landslides, and reduce the losses caused by landslides in the study area.

Landslide susceptibility evaluation was first proposed in the 1960s, mainly by using geological maps to determine the spatial probability of landslide occurrence base on the environmental conditions of a certain region, that is, “a certain region is prone to landslides”, to provide an effective method for disaster prevention and reduction [[13], [14], [15]]. Since the beginning of the 21st century, scholars around the world have conducted in-depth studies on landslide susceptibility analysis using geographic information system (GIS) data, but for the construction of evaluation models of earthquake-induced landslide disasters, the rainfall-induced landslide model has been used as a reference [[16], [17], [18]]. However, the distinction between the models is not obvious, and there are few studies on the model's sensitivity to earthquakes [17,19].Lee C T et al. [20]studied shallow earthquake-induced landslides in central and western Taiwan by using deterministic and multivariate statistical methods. Chyi-Tyi Lee et al. [21] mapped the landslide areas induced by earthquake, and mapped the landslide susceptibility in these areas. At present, the main focus of related research is to determine the accuracy of landslide prediction results based on different evaluation factors and different prediction models [22,23]. Lee and Talib [24] collected landslide-influencing factors in Penang, Malaysia, by using GIS tools and calculated the contribution rate of each class of factors to landslide occurrence by the frequency ratio. Pradhan and Lee [25] used a GIS tool to conduct a regional susceptibility study in the Malaysian Highlands to determine the extent to which multiple factors contributed to landslide occurrence in the study area. Rohan Kumar and other scholars [26] used machine learning methods to map the landslide susceptibility of their study area, introducing the application of machine learning in geological hazard assessment, and combined it with the subjective analytic hierarchy process (AHP) model to explore the model evaluation results. The research presented above shows that the results of landslide susceptibility in different study areas are affected by the geological environment and data background, and there are great differences in model prediction results.In addition, as seismic factors are inducing factors, few studies have introduced seismic factors alone. Therefore, the magnitude and distance from the seismic focal point, acting as seismic factors, are introduced separately in this paper. The frequency ratio and AHP model were used to map the landslide susceptibility of the study area, analyse the sensitivity of landslide susceptibility to earthquake magnitude, and connect the distance from the earthquake focal point to the landslide susceptibility region to construct a map of the earthquake-related landslide susceptibility region in southeastern Tibet. Therefore, in order to solve the impact of geological environment and data on the research results, this paper adopts the factor correlation analysis method to process the data.In order to solve the problem of accuracy of the results of different models, this paper considers the influence of seismic factors and combines frequency ratio method and analytic hierarchy process to study the landslide susceptibility mapping.

The essence of landslide susceptibility research is the analysis of the spatial probability of landslides under the influence of the geological conditions, environmental conditions and human engineering activities [[27], [28], [29]]. The general methods of landslide susceptibility analysis include collecting historical landslide data, extracting environmental factors in the study area, constructing a susceptibility evaluation model, and evaluating the model accuracy [30]. At present, there are three types of landslide susceptibility evaluation models: statistical models, and coupling models [[31], [32], [33]]. The statistical models mainly include the analytic hierarchy process, frequency ratio method and logistic regression method [34]. Machine learning models include artificial neural network models, information content models and support vector machine models [33,35,36]. The commonly used landslide susceptibility evaluation models are shown in Table 1 [37] below. With the rapid development of “3s” technology, some geologists have applied this kind of technology to the analysis and prediction of geological hazards and combined “3s” technology with an evaluation model to study landslide susceptibility [38,39]. Shen Huaifei et al. [40] analysed regional landslide vulnerability in Gansu Province based on the AHP and the information content method and obtained the regional area and spatial distribution characteristics of vulnerability in Gansu Province. Jia Lin et al. [41] introduced geological environmental factors and carried out landslide susceptibility zoning of Nanzhang County city considering 9 other influencing factors. Wu Changrun et al. [42] coupled the frequency ratio and logistic regression model, carried out the landslide susceptibility division of Shuangbai County, and obtained the high landslide susceptibility area and its distribution characteristics in this region. In these studies, the landslide susceptibility mapping was carried out through the combination of factor selection and different model models in order to improve the accuracy of model prediction.However, in order to study the effect of earthquake on the accuracy of landslide susceptibility mapping, this paper takes the magnitude of earthquake and the distance from the seismic point as variables to discuss the effect on the accuracy of the model results.The frequency ratio method analyzes the single factor itself to determine the influence of different grades of single factor on landslide.The analytic hierarchy process regards a single factor as a whole and analyzes the influence of the whole factor on landslide.Therefore, this paper combines the frequency ratio method and analytic hierarchy process to solve the shortcomings of each method, so that the model prediction results are highly accurate.Table 1 The advantages and disadvantages of the susceptibility evaluation method.

Table 1Model type	Methods	Advantages	Disadvantages	
Statistical model	Analytic Hierarchy Process (AHP)	The weight of the impact factor is tested to reduce subjectivity	The judgement matrix is constructed with many indexes and a large amount of iterative calculation	
Frequency Ratio (FR)	The correlation between evaluation factors and landslide hazards is easy to judge	Pure mathematical statistics, lack of landslide development mechanism research	
Logistic Regression (LR)	The operation method is simple, the calculation process is not affected by subjective factors, and the physical significance of the evaluation results is clear	Based on the statistical law of large samples, it needs the support of a large amount of data	
Machine learning model	Artificial Neural Networks Model (ANNM)	It has strong ability of self-adaptation, self-learning and nonlinear mapping; can solve the problems of less data, poor information and uncertainty well; and is not limited by nonlinear models	It is difficult to select training samples and has issues with slow convergence speed, large number of iteration steps, easily falling into local minimization and poor global searching ability	
Information Model (IM)	Disaster regionalization with large number of units is more advantageous	It can reflect the possibility of disaster occurrence only under a specific combination of different impact factors, and the difference of contribution of each factor cannot be reflected. The accuracy of the analysis depends on the amount of data collected	
Support Vector Machine (SVM)	There is no need to make large adjustments to the input parameters; It has high computational efficiency and strong prediction ability and has advantages in solving small samples, high latitude and nonlinear problems, while avoiding human interference	The selection of the kernel function is difficult, and it directly affects the precision and time of the operation result	

In summary, coupling two different models to conduct landslide susceptibility analysis is currently a major trend in regional landslide susceptibility research, but few studies have considered earthquake factors to determine landslide hazards. Therefore, considering the sensitivity of seismic factors and other landslide influencing factors, this paper takes southeastern Tibet as the study area, selects the AHP and frequency ratio model, and uses the ArcGIS platform to study the susceptibility to seismic chain-induced landslide disasters in the study area. The accuracy of the susceptibility evaluation of different models and the relationships between them and seismic factors in southeastern Tibet were obtained to compensate for the lack of research on the relationships between seismic factors and landslide susceptibility in southeastern Tibet and to provide support for disaster prevention and mitigation in this region.

The purpose of this paper is to analyse the effect of earthquake magnitude and distance from earthquake point on the accuracy of landslide susceptibility results.The research area is southeast Tibet.The main research objectives are:（1）What are the main factors affecting the occurrence of landslide?（2）What are the effects of earthquake magnitude and distance from earthquake point on landslide susceptibility mapping?（3）What is the map of priority monitoring areas in the study area?

By solving the above objectives, this paper analyzes the earthquake magnitude and distance from the earthquake point as variables, combines the frequency ratio model and the hierarchical analysis model, and obtains the landslide-prone area mapping of different prediction models in southeast Tibet.The ROC curve was used to predict the results, and the model with the highest accuracy was obtained. The results of this study aim to provide reasonable landslide-prone zoning maps for local governments and people.

2 Overview of the study area

The study area is located in the southeastern part of the Tibet Autonomous Region, mainly the Nyingchi and Qamdo regions, with an average elevation of more than 4000 m. The geographical coordinates are 92°9' ∼99°56′ east longitude and 27°33' ∼32°43′ north latitude, and the region features an Indian Ocean monsoon climate. The area has dense river systems and abundant surface water systems, including the Yarlung Zangbo River and three other rivers (Jinsha River, Lancang River and Nu River). The geological structure in the area is complex, tectonic fault zones exist, and the regional faults mainly include the Yarlung Zangbo River fault, Jinsha River fault, Nujiang fault and Jiali fault.

Based on the research group's previous field investigation and data collection and statistics, the impact factors of topography, geomorphology, geological environment, land use and rainfall in Tibet were obtained. Through the analysis of these data, it was concluded that the distribution of landslide disasters in Tibet had obvious spatial and temporal characteristics and was mainly distributed in the Nyingchi and Qamdo regions. Some dates show there have been approximately 659 landslide disasters in southeastern Tibet since 2009, and they are mainly densely distributed along the Sichuan‒Tibet Line and Sanjiang River. The geographical location of the study area and the distribution of landslide points are shown in Fig. 1.Fig. 1 Location map of the study area in China.

Fig. 1

3 Establishment of the impact factor database

3.1 Data sources

The data source and accuracy directly affect the results of landslide susceptibility evaluation. The research group cleaned the collected data, conducted field visits in the early stage, and collected a total of 659 landslide points. Based on the ArcGIS platform, the regional vulnerability of seismic landslides in southeastern Tibet was evaluated by considering topographic and geomorphic factors (slope, aspect of slope, elevation), road factors (distance from fault, distance from river, distance from road) and inducing factors (land use, rainfall) combined with seismic factor sensitivity. The specific data sources and data information are shown in Table 2.Table 2 Data sources.

Table 2First-order factor	Second-order factor	Data source	Data type	Precision	
Topographic and geomorphic factors	Slope	National Geospatial Data Cloud (www.gscloud.cn)
National Geological Data Center (www.ngac.org.cn)	Grid	30 m	
Aspect of slope	Grid	30 m	
Elevation	Grid	30 m	
Road factor	Distance from fault	1:100,000 national basic geographic information data	Vector quantity	1:50000	
Distance from river	Vector quantity	1:50000	
Distance from road	Vector quantity	1:50000	
Inducible factor	Land use	Center for Resources and Environmental Science and Technology, Chinese Academy of Sciences	Vector quantity	1:50000	
Rainfall	Grid	30 m	
Distance from source	Vector quantity	1:50000	

3.2 Factor correlation analysis

In this paper, a total of 9 factors were selected as the impact factors, and these 9 impact factors are divided into two categories: constant factors and variable factors. The constant factors include elevation, slope, slope direction, distance from roads, distance from rivers, distance from faults, rainfall and land use. The variable factor is the seismic factor, which is divided into two dimensions, distance from the earthquake focal point and magnitude, to analyse the relationship between earthquakes and landslides.

In this paper, the distance from the focal point is considered a variable factor, and the landslide susceptibility of the study area is studied by integrating other important factor data. To avoid the possibility of a significant correlation between important factors, the multicollinearity analysis method was used to analyse the interfactor correlation of factors other than the variable factors in the SPSS platform. The results of the factor correlation analysis are shown in Table 3. A tolerance TOL>0.1 and variance inflation factor (VIF) < 5 indicate that there is no linear correlation between the factor and other factors. The results of factor multicollinearity analysis indicate (Table 3) that the TOL values of the eight important factors, including slope, slope direction, elevation and distance from fault, are all greater than 0.1, and the VIF values are less than 5. Therefore, there is no collinearity among the factors selected in this paper, and the factor selection is reasonable.Table 3 Factor multicollinearity analysis.

Table 3Factors	Slope	Aspect of slope	Elevation	Distance from fault	Distance from river	Distance from road	Land use	Rainfall	
TOL	0.927	0.99	0.613	0.846	0.648	0.636	0.685	0.758	
VIF	1.079	1.01	1.631	1.183	1.544	1.571	1.459	1.319	

4 Analytical methods

Frequency ratio and hierarchy analysis are landslide susceptibility analysis methods based on mathematical statistics. The frequency ratio determines the influence of each factor associated with landslides by calculating the contribution rate of each evaluation factor to landslides of different grades. In contrast, the frequency ratio method ignores the fact that each factor has a different contribution to landslide occurrence. Therefore, this paper combines the frequency ratio with the hierarchical analysis method and first calculates the contribution rate of each factor to landslide occurrence in different grades by using the frequency ratio. Then, the contribution rate of each factor to landslide occurrence is calculated via the AHP to carry out landslide susceptibility regionalization in southeastern Tibet.

4.1 Frequency ratio analysis

4.1.1 Analytical methods

The frequency ratio method represents the contribution rate of each factor to landslides of different grades rely on the frequency ratio (FR), and the contribution rates of the different factors to landslides are determined via the landslide susceptibility evaluation index (LST). An FR equal to 1 indicates an average value, implying a moderate correlation between the factor and landslide disasters of a specific grade. When the FR is greater than 1, there is a high correlation between the factor and landslide disasters of a certain grade. Conversely, when the FR is less than 1, there is a low correlation between the factor and landslide disasters. The calculation formulas for the FR and landslide susceptibility evaluation index are as follows:(1) FR=ni/Nsi/S

(2) LST=∑FRij

where FR represents the influence degree of evaluation factors at various levels on landslides, ni represents the number of landslides in the classification, N represents the total number of landslides in the study area, si represents the classified area, S represents the total area of the study area, and LST represents the evaluation index of landslide susceptibility.

4.1.2 Factor grading

Through the ArcGIS platform, eight major factors, such as slope, slope direction, elevation and distance from the fault, were classified into grades, and the number of landslide points and frequency ratio in different grades of each factor were calculated. The data of different grades of each factor are shown in Table 4. Landslides are more likely to occur under the following conditions: when the slope is less than 13° and between 13° and 23°; when the aspect of slope is south, southwest and west; when the elevation is < 1371 m, 1371–2435 m, 2435–3305 m, 3305–3949 m; when the distance from the road is < 3000 m; when the distance from the fault is < 3000 m, 3000–6000 m; when the distance from the river is < 3000 m, 3000–6000 m and 6000–9000 m; when the rainfall is 288–524 mm and 766–846 mm; and when the land use type is ploughland, forest land and meadow. The landslide susceptibility index parameters from highest to lowest values are elevation, aspect of slope, distance from road, land use, distance from river, slope, rainfall and distance from fault.Table 4 General factor frequency ratio classification table.

Table 4Influence factor	Factor grading	Number of landslide	Proportion of landslides	Grid ratio	FR	LSI	Sort	
Slope(°)	<13	145	22.14 %	15.83 %	1.40	5.53	6	
13–23	184	28.09 %	23.79 %	1.18	
23–32	139	21.22 %	25.51 %	0.83	
32–41	132	20.15 %	21.67 %	0.93	
41–54	48	7.33 %	11.19 %	0.65	
54–90	7	1.07 %	2.01 %	0.53	
Aspect of slope	north	62	9.51 %	11.65 %	0.82	9.62	2	
northeast	84	12.89 %	13.56 %	0.95	
east	73	11.20 %	12.87 %	0.87	
southeast	77	11.81 %	12.44 %	0.95	
south	82	12.58 %	11.47 %	1.10	
southwest	110	16.87 %	13.31 %	1.27	
west	92	14.11 %	12.53 %	1.13	
northwest	72	11.04 %	12.18 %	0.91	
Elevation (m)	<1371	33	5.04 %	4.81 %	1.05	9.74	1	
1371–2435	82	12.52 %	5.18 %	2.42	
2435–3305	139	21.22 %	8.10 %	2.62	
3305–3949	263	40.15 %	14.37 %	2.79	
3949–4447	128	19.54 %	24.33 %	0.80	
4447–4916	10	1.53 %	26.33 %	0.06	
4916–7427	0	0 %	16.88 %	0	
Distance from fault (m)	<3000	447	67.83 %	27.00 %	2.51	5.32	8	
3000–6000	132	20.03 %	15.69 %	1.28	
6000–9000	43	6.53 %	9.46 %	0.69	
9000–12000	16	2.43 %	6.08 %	0.40	
12000–15000	12	1.82 %	4.54 %	0.40	
>15000	9	1.37 %	37.23 %	0.04	
Distance from river (m)	<3000	390	59.18 %	11.99 %	4.93	8.21	5	
3000–6000	87	13.20 %	11.18 %	1.18	
6000–9000	71	10.77 %	10.33 %	1.04	
9000–12000	26	3.95 %	9.03 %	0.44	
12000–15000	21	3.19 %	7.62 %	0.42	
>15000	64	9.71 %	49.84 %	0.19	
Distance from road (m)	<3000	500	75.87 %	11.00 %	6.90	9.33	3	
3000–6000	52	7.89 %	8.70 %	0.91	
6000–9000	18	2.73 %	7.23 %	0.38	
9000–12000	24	3.64 %	6.04 %	0.60	
12000–15000	14	2.12 %	5.16 %	0.41	
>15000	51	7.74 %	61.89 %	0.13	
Rainfall (mm)	288–524	211	32.26 %	19.17 %	1.68	5.49	7	
524–605	169	25.84 %	28.29 %	0.91	
605–692	120	18.34 %	24.75 %	0.74	
692–766	82	12.54 %	14.06 %	0.89	
766–846	71	10.86 %	8.84 %	1.23	
846–1186	1	0.15 %	4.89 %	0.03	
Land use	ploughland	47	7.18 %	1.62 %	4.44	8.79	4	
Garden plot	279	42.60 %	45.13 %	0.94	
Forestland	287	43.82 %	37.54 %	1.17	
meadow	32	4.89 %	2.29 %	2.13	
commercially	0	0.00 %	0.02 %	0	
Transportation land	10	13.40 %	13.40 %	0.11	

The seismic sensitivity is taken as a variable factor, and its sensitivity is classified. If the seismic focal point magnitude ranges from 3 to 4, the corresponding landslide sensitivity is rated 1. For an earthquake with a focal point magnitude ranging from 4 to 5, the associated landslide sensitivity increases to level 2. For earthquakes exceeding a focal point magnitude of 5, the landslide sensitivity further increases to level 3. Moreover, when the earthquake's focal point magnitude surpasses 3, the resulting landslide sensitivity reaches its maximum at level 4. The Euclidean distance is used to grade seismic factors with different sensitivities, and the classification data are shown in Table 5.Table 5 Variable factor sensitivity classification of the distance from the source.

Table 5Earthquake focal point magnitude	Magnitude sensitivity	Seismic point ratio	Range sensitivity	Distance from source	Number of landslide	Slip ratio	Grid ratio	Frequency ratio	
Level 3 to 4	1	72 %	a	<40 km	373	56.60 %	41.75 %	1.36	
b	40–80 km	236	35.81 %	33.70 %	1.06	
c	>80 km	50	7.59 %	24.55 %	0.31	
Greater than grade 4	2	28 %	a	<40 km	198	30.05 %	18.16 %	1.65	
b	40–80 km	281	42.64 %	27.88 %	1.53	
c	>80 km	180	27.31 %	53.96 %	0.51	
All magnitudes	3	100 %	a	<40 km	403	61.15 %	44.76 %	1.36	
b	40–80 km	217	32.93 %	31.96 %	1.03	
c	>80 km	39	5.91 %	23.27 %	0.25	

4.2 Analytic hierarchy process

4.2.1 Analytical method

The AHP is a decision-making method based on mathematical statistics. The algorithm can divide the landslide evaluation factors into several intervals and calculate the weight of each influencing factor within the interval. Its model structure decomposes the factors related to decisions into a target layer, an index layer and a scheme layer. In this paper, each impact factor is taken as a division interval to calculate the weight of each factor's contribution to the occurrence of landslides. The hierarchical structure model is shown in Fig. 2.Fig. 2 Hierarchy model diagram.

Fig. 2

In line with the 1–9 scaling method proposed by Saaty, the matrix analytic hierarchy process quantification table (Table 6) is constructed, and the judgement matrix of the index layer is obtained by pairwise comparison of each factor in the index layer. The relative importance of each factor in the index layer on the occurrence of landslides, namely, weight ω, is determined by the judgement matrix combined with the arithmetic average method.Table 6 Quantitative table of the analytic hierarchy process.

Table 6Factor i compared to factor j	Quantized value	
Equally important	1	
Slightly important	3	
Stronger importance	5	
Strongly important	7	
Vital	9	
The median of two adjacent judgements	2,4,6,8	
Count backwards	aij = 1/aij	

After listing the judgement matrix, it is necessary to carry out a consistency test on the matrix to determine whether the weight is reasonable. When the consistency test result is CR < 0.1, the judgement matrix passes the one-time test. The consistency test formula is as follows:(3) λmax=∑i=1n[Aω]inωi

(4) CI=(λ−n)(n−1)

(5) CR=CIRI

where A is the index layer judgement matrix; ω is the weight after standardization; n is the number of indicators; CI is the consistency index; and RI is a fixed value, which can be obtained by looking up the table.

4.2.2 Factor weight analysis

The judgement matrix of the 8 factors in the paper is constructed, and the judgement matrix is shown in Table 7. The weights of the eight influencing factors, including elevation, slope, rainfall, distance from road, aspect of slope, land use, distance from road and distance from river, with respect to landslide occurrence were 0.33, 0.05, 0.03, 0.16, 0.23, 0.11, 0.07 and 0.02, respectively. The constructed judgement matrix is an 8th-order matrix, so RI = 1.41. The calculated consistency test result CR = 0.0298 < 0.1 passed the consistency test.Table 7 Judgement matrix of the indicator layer.

Table 7	Elevation	Slope	Rainfall	Distance from road	Aspect of slope	Land use	Distance from river	Distance from fault	
Elevation	1	4	1/2	1/3	5	3	2	6	
Slope	1/4	1	1/5	1/6	2	1/2	1/3	3	
Rainfall	2	5	1	1/2	6	4	3	7	
Distance from road	3	6	2	1	7	5	4	8	
Aspect of slope	1/5	1/2	1/6	1/7	1	1/3	1/4	2	
Land use	1/3	2	1/4	1/5	3	1	1/2	4	
Distance from river	1/2	3	1/3	1/4	4	2	1	5	
Distance from fault	1/6	1/3	1/7	1/8	1/2	1/4	1/5	1	

4.3 Coupling analysis of the FR and AHP

The calculated weight ω of each factor is coupled with the FR in different stages of each factor, and the coupling value of the FR in different stages of each factor based on the AHP is obtained, as shown in Table 8.Table 8 FR–AHP coupling values.

Table 8Influence factor	Factor grading	FR	Weight (ω)	FR–AHP coupling value	
Slope (°)	<13	1.40	0.05	0.07	
13–23	1.18	0.06	
23–32	0.83	0.04	
32–41	0.93	0.05	
41–54	0.65	0.03	
54–90	0.53	0.03	
Aspect of slope	north	0.82	0.23	0.19	
northeast	0.95	0.22	
east	0.87	0.20	
southeast	0.95	0.22	
south	1.10	0.25	
southwest	1.27	0.29	
west	1.13	0.21	
northwest	0.91	0.26	
Elevation (m)	<1371	1.05	0.33	0.32	
1371–2435	2.42	0.80	
2435–3305	2.62	0.86	
3305–3949	2.79	0.92	
3949–4447	0.80	0.27	
4447–4916	0.06	0.02	
4916–7427	0	0.00	
Distance from fault (m)	<3000	2.51	0.02	0.05	
3000–6000	1.28	0.03	
6000–9000	0.69	0.01	
9000–12000	0.40	0.01	
12000–15000	0.40	0.01	
>15000	0.04	0.00	
Distance from river (m)	<3000	4.93	0.07	0.35	
3000–6000	1.18	0.08	
6000–9000	1.04	0.07	
9000–12000	0.44	0.03	
12000–15000	0.42	0.03	
>15000	0.19	0.01	
Distance from road (m)	<3000	6.90	0.16	1.10	
3000–6000	0.91	0.15	
6000–9000	0.38	0.06	
9000–12000	0.60	0.10	
12000–15000	0.41	0.07	
>15000	0.13	0.02	
Rainfall (mm)	288–524	1.68	0.03	0.05	
524–605	0.91	0.03	
605–692	0.74	0.02	
692–766	0.89	0.03	
766–846	1.23	0.04	
846–1186	0.03	0.00	
Land use	ploughland	4.44	0.11	0.49	
garden plot	0.94	0.10	
forestland	1.17	0.13	
meadow	2.13	0.23	
commercially	0	0.00	
transportation land	0.11	0.01	

5 Results and analysis

On the basis of the magnitude at the seismic focal point itself, the seismic focal point magnitudes are divided into three sections: seismic focal points with all magnitudes, the seismic focal points with magnitudes 3–4 and the seismic focal points with magnitudes >4. This is the first dimension of the risk study, and the seismic sensitivities are 1, 2 and 3. With the earthquake focal point as the centre of the circle, the research area was divided into three sections: <40 km from the focal point, 40–80 km from the focal point, and >80 km by using ArcGIS, reclassification and other tools. This is the second dimension of risk research, and the sensitivity to distance from the focal point is a, b, and c.

5.1 Disaster factor layer processing

Through the ArcGIS platform, 8 constant factors, including elevation, slope, aspect of slope and distance from the road, were graded. The layers of factor grade division are shown in Fig. 3.Fig. 3 Impact factor thematic layers. a) Elevation grading. b) Slope grading. c) Distance from a river grading. d) Distance from a road grading. e) Aspect of the slope grading. f) Distance from a fault grading. g) Rainfall grading. h) Land use grading.

Fig. 3

5.2 Risk zoning

The calculated data values were assigned to the grid in the study area, and the grid calculator was used for calculation and reclassification. The study area was divided into five categories: very high-prone area, high-prone area, prone area, low-prone area and very low-prone area.

A zoning diagram of landslide susceptibility in the study area without superimposed seismic factors is shown in Fig. 4. The zoning results showed that the proportions of very low-prone area, low-prone area, prone area, high-prone area and very high-prone area to the total area were 34.40 %, 24.88 %, 21.48 %, 12.13 % and 7.11 %, respectively. In total, 536 landslide points fell in the high- and very high-prone areas, accounting for 82.21 % of the total. The areas with high and extremely high susceptibility accounted for approximately 19.24 % of the total area.Fig. 4 Landslide susceptibility zoning map.

Fig. 4

5.3 Results of risk zoning considering magnitude sensitivity

The landslide susceptibility zoning diagram of the study area with a magnitude sensitivity of 1 is shown in Fig. 5(a). The zoning results showed that the proportions of very low-prone area, low-prone area, prone area, high-prone area and very high-prone area to the total area were 2.10 %, 30.09 %, 30.21 %, 28.74 % and 8.56 %, respectively. In total, 605 landslide points fell in the high- and very high-prone areas, accounting for 92.79 % of the total. High- and very high-prone areas accounted for approximately 37.60 % of the total area.Fig. 5 Seismic sensitivity zoning map. (a) Landslide susceptibility mapping when the magnitude sensitivity is 1. (b) Landslide susceptibility mapping when the magnitude sensitivity is 2. (c) Landslide susceptibility map when the magnitude sensitivity is 3.

Fig. 5

The landslide susceptibility zoning diagram of the study area with a magnitude sensitivity of 2 is shown in Fig. 5(b). The zoning results showed that the proportions of very low-prone area, low-prone area, prone area, high-prone area and very high-prone area to the total area were 2.09 %, 32.22 %, 28.19 %, 28.65 % and 8.84 %, respectively. In total, 605 landslide points fell in the high- and very high-prone areas, accounting for 92.79 % of the total. High- and very high-prone areas accounted for approximately 37.60 % of the total area.

The landslide susceptibility zoning diagram of the study area with a magnitude sensitivity of 3 is shown in Fig. 5(c). The zoning results showed that the proportions of very low-prone area, low-prone area, prone area, high-prone area and very high-prone area to the total area were 2.14 %, 55.16 %, 21.85 %, 13.87 % and 6.97 %, respectively. In total, 605 landslide points fell in the high- and very high-prone areas, accounting for 81.60 % of the total. High- and very high-prone area accounted for approximately 20.85 % of the total area.

Reasonable landslide susceptibility evaluation results should meet the following two conditions: (1) as the regional susceptibility grade increases, the grading area decreases gradually; (2) as the susceptibility grade increases, the landslide ratio (the ratio of the number of landslide points of different susceptibility grades to the area of each susceptibility grade) gradually increases. Based on the analysis results presented above, the prediction results of the landslide susceptibility of the three models are highly similar in space, showing a zonal distribution, and the results have a certain degree of similarity. When the magnitude sensitivity is 1, the correct prediction results show that the number of landslide points accounts for 92.79 % of the total, and the predicted area accounts for 37.60 % of the total area; that is, the landslide ratio is 2.468. When the magnitude sensitivity is 2, the correct prediction results show that the number of landslide points accounts for 92.79 % of the total area, and the predicted area accounts for 37.49 % of the total area; that is, the landslide ratio is 2.475. When the magnitude sensitivity is 3, the correct prediction results of the model show that the number of landslide points accounts for 81.60 % of the total area, and the predicted area accounts for 20.85 % of the total area; that is, the landslide ratio is 3.914. The landslide ratio of the three models gradually increases, and when the magnitude sensitivity is 3, the landslide ratio is the largest, indicating that the prediction model results are the most accurate and reasonable. In the mapping of landslide susceptibility in the study area, the FR–AHP model with a magnitude sensitivity of 3 has the most accurate prediction results. The results of the magnitude sensitivity partition are shown in Table 9.Table 9 Magnitude sensitivity zoning.

Table 9Prone zoning	Sensitivity classification	Landslide point	Proportion of landslide (%)	Zone area (%)	
Very low prone area	1	7	1.07	2.1	
2	7	1.07	2.09	
3	7	1.07	2.14	
Low prone area	1	11	1.69	30.09	
2	11	1.69	32.22	
3	35	5.37	55.16	
Prone area	1	29	4.45	30.21	
2	29	4.45	28.19	
3	78	11.96	21.85	
High prone area	1	211	32.36	28.74	
2	210	32.21	28.65	
3	179	27.45	13.87	
Very high prone area	1	394	60.43	8.56	
2	395	60.58	8.84	
3	353	54.14	6.97	

5.4 Magnitude sensitivity accuracy of the risk zoning results

The area under the ROC curve (AUC) was used to quantify the accuracy of the landslide susceptibility evaluation results with different magnitude sensitivities. When the AUC is 1, the model is a perfect model; when the AUC is between 0.85 and 0.95, the model has high precision; and when the AUC is between 0.75 and 0.85, the model has good precision.

The ROC curves of the models with sensitivity magnitudes of 1, 2 and 3 are shown in Fig. 6. The AUC values of the three models in descending order are AUC magnitude sensitivity 3>AUC magnitude sensitivity 2>AUC magnitude sensitivity 1, indicating that the prediction accuracy of the model is the most accurate when the earthquake has a magnitude of 4 and above, and the contribution rate of earthquakes with magnitudes 4 and above is greater. The ROC curve analysis yielded the same results as the landslide ratio analysis presented in the previous section, indicating that when the magnitude sensitivity is 3 (that is, when an earthquake of magnitude 4 or above occurs), the prediction results of the landslide susceptibility prediction model are the most accurate and reasonable among the three prediction models.Fig. 6 ROC curves of different magnitude sensitivity models.

Fig. 6

5.5 Landslide monitoring area

Comparison of the three different magnitude sensitivity models suggests that when the magnitude is 4, the prediction model has the highest accuracy. The study area divided based on the distance from the focal point, corresponding to a (<40 km), b (40–80 km) and c (>80 km), are redivided into extremely vulnerable areas, nonprone areas, prone areas, high prone areas and extremely prone areas based on the prediction model when the magnitude sensitivity is 3. The prediction area division data are shown in Table 10. Tools such as intersection and superposition in the ArcGIS platform were used to obtain the key landslide prediction areas by analysing and processing the divided data. Monitoring grades 4, 3, 2 and 1 represent the extremely severe monitoring area, severe monitoring area, moderate monitoring area and light monitoring area, respectively.Table 10 Predicted area reclassification.

Table 10Sensitivity of distance from the source	Susceptibility zoning	Landslide points	Monitoring level	
a	Very low prone zone	0	0	
Low prone zone	20	0	
Prone zone	38	1	
High prone area	119	2	
Very high prone area	220	4	
b	Very low prone zone	0	0	
Low prone zone	6	0	
Prone zone	28	1	
High prone area	49	1	
Very high prone area	133	3	
c	Very low prone zone	7	0	
Low prone zone	9	0	
Prone zone	12	0	
High prone area	11	0	
Very high prone area	0	0	

When an earthquake occurs in southeastern Tibet, the landslide-prone area is predicted by a model with a magnitude sensitivity of 3. The prone area, high-prone area and extremely high-prone area identified by the model should be monitored. To discuss the relationship between the three monitoring areas and the distance from the seismic focal point, the data obtained from the superimposed layer were analysed (Table 10). The data showed that the number of landslide points in different grades from high to low was 220, 133, 119, 49, 38 and 28, and the greater the number of landslide points was, the greater the regional landslide susceptibility was. Therefore, the numbers of landslide points in the first three grades are 220, 133 and 119, respectively, and the numbers of landslide points in the remaining areas are not much different. The landslide area should be monitored at the same level. First, in the very highly prone area predicted by the model and within 40 km of the earthquake focal point, landslides easily occur, and the monitoring level is 4. The monitoring area is shown in Fig. 7(a). Second, the very highly prone area predicted by the model and within 40–80 km from the earthquake focal point is prone to landslides, and the monitoring level is 3. The monitoring area is shown in Fig. 7(b). Then, the highly prone areas predicted by the model and the areas less than 40 km from the earthquake focal point are prone to landslides, and the monitoring level is 2. The monitoring area is shown in Fig. 7(c). Finally, in the highly prone area predicted by the model and within 40 km–80 km from the earthquake focal point, landslides easily occur, and the monitoring level is 1. The monitoring area is shown in Fig. 7(d).Fig. 7 Regional monitoring. (a) Landslide susceptibility monitoring area map at monitoring level 4. (b) Landslide susceptibility monitoring area map at monitoring level 3. (c) Landslide susceptibility monitoring area map at monitoring level 2. (d) Landslide susceptibility monitoring area map at monitoring level 1.

Fig. 7

Based on the above four monitoring areas, it is found that the roads and rivers southeast of the rock reservoir in the landslide-prone area and the key monitoring area are distributed in a belt, as shown in Fig. 8. Some typical landslides in southeastern Tibet were sampled, and the geographical locations of major landslide events were imported into the susceptibility zoning map. The prediction results for major landslides are shown in Table 11.Fig. 8 Overall monitoring area distribution map of the study area. (a) Map of the monitoring areas along roads. (a) Map of the monitoring areas along rivers.

Fig. 8

Table 11 Typical landslide inspection in southeastern Tibet.

Table 11Name of disaster	Geographical position	Predicted susceptibility	
68 Daoban Landslide	Bangda town	Moderate monitoring area	
Yigong power station landslide	Tongmai Village	Extremely severe monitoring area	
Landslide in Damu Township	Damu township	Extremely severe monitoring area	
Landslide on Tongyi Road	Tongmai Village	Extremely severe monitoring area	
Landslide in Abba village	Aba Village	Moderate monitoring area	
Landslide group of Muxie Township, Gongjue County	Xieqiu Village	Severity monitoring area	

6 Discuss

In this paper, the ability of a coupling model involving the FR and AHP to predict spatial landslide susceptibility under the influence of seismic factors is studied. The resolution of the data is 30 m, and the study area is southeastern Tibet. Through field investigation, 659 historical landslide events were verified, 8 influencing factors, such as slope and slope direction, were determined to be major factors, and earthquakes were selected as the variable factor. The FR method and AHP method were used to map the landslide susceptibility of the study area. The results show that the coupled FR–AHP model with a seismic sensitivity of 3 has a high prediction accuracy; that is, earthquakes of magnitude 4 and above have a greater impact on landslides. The landslide susceptibility model, which uses an influencer with a seismic sensitivity of 3 as the variable factor, has the highest accuracy in predicting landslide susceptibility in southeastern Tibet. The relationships between earthquake magnitude and distance from earthquake focal points in the study area and landslide susceptibility zoning are analysed in order to provide better data support for predisaster prevention and postdisaster resettlement of landslides in the study area.

In this paper, the earthquake magnitude and distance from the earthquake point have been used as variables to map the landslide susceptibility in southeast Tibet.Considering the complex geological structure and active seismic zones in southeast Tibet, future studies should pay more attention to the occurrence of landslides under the coupling of multiple factors such as earthquakes and rainfall, and how these factors affect the scale, frequency and distribution of landslides.For landslides under earthquake action, attention will be paid to the regulation of fault activity on landslide development, and the quantitative relationship between fault type, activity rate and landslide susceptibility will be explored, so as to improve the evaluation model.

Although the regional map of landslide susceptibility and the partial correlation between earthquakes and landslide occurrence in the study area were obtained in this paper and technical support was provided for predisaster prevention and postdisaster resettlement in the study area, there are still certain limitations in this study. The shortcomings of this paper are as follows:（1）The number of landslide points in the study area is limited, which limits the research model.（2）In this paper, only seismic factors are taken as variables, while rainfall is the main factor that induces landslides. Subsequent studies could adopt both earthquakes and rainfall as dependent variables.（3）The research models used in this paper are frequency ratio and hierarchical analysis models, which are relatively simplistic. In the future, multiple models, such as information content, neural network and decision tree models, could be introduced for accuracy comparisons to select the most suitable landslide susceptibility evaluation model for southeastern Tibet.（4）In this paper, southeastern Tibet is taken as the research area for model training, and data from other regions could be incorporated into the research model of this paper in the future to determine the accuracy of the model when it is applied in other regions.

7 Conclusion

In this paper, southeastern Tibet is taken as the study area. Considering the sensitivity of landslides to earthquakes in southeastern Tibet, the prediction accuracies of landslide susceptibility prediction models with different earthquake magnitudes are compared, and the predicted regions of the models and the distance from the earthquake focal point are superimposed to obtain the number of landslide points in different regions. Then, more accurate prediction of landslide-prone regions is carried out. Based on the above research, the following conclusions can be drawn:（1）Calculation of the frequency ratio of different factors in different grades shows that landslides are most likely to occur when the slope is less than 23°; the slope aspect is south, southwest, or west; the elevation is between 1371 and 2435 m; the distance from a fault is within 6000 m; the distance from a river is within 9000 m; the distance from a road is within 3000 m; the rainfall is between 288-524 mm and 766–846 mm; and the land use type is farmland, forestland, or grassland.（2）Through the analytic hierarchy process, the weights of the factors contributing to landslides in southeastern Tibet are as follows: elevation, slope aspect, distance from a road, land use, distance from a river, slope and rainfall.（3）Considering the influence of earthquakes on landslide occurrence, earthquakes of magnitude 4 and above have a great influence on landslides. The accuracy of the landslide susceptibility zoning map obtained by superimposing the original layer and earthquake data of magnitude 4 and above is the highest. Additionally, the number of landslide points correctly predicted accounts for approximately 81.6 % of the landslide points, and the predicted landslide-prone area accounts for 20.85 % of the total area.（4）The overall distribution map of the monitoring area is obtained by superposition of four monitoring areas of different levels. Landslides in southeastern Tibet are mainly distributed along water systems and roads.

The landslide susceptibility zoning map obtained from the study shows:Extremely high and highly prone areas are mainly distributed in the western part of Lang County, the central part of Milin County, the northern part of Metuo County, the junction of Bomi County and Bayi District, the central part of Zayu County, the southern part of Dingqing County, the northern part of Luojian County, the southern part of Karuo District, the northern part of Chaya County, the central part of Badu County and the northern part of Zuogong County. In the area along the Yarlung Zangbo River, Daqu, Zangqu, Nu River, Lancang River, Weiqu, Zhan Qu and Zaqu on both sides.

The study area was monitored according to the resulting susceptibility mapping:The test grades are 4, 3, 2 and 1 respectively from high to low, and the higher the monitoring grade, the greater the monitoring intensity.The results show that landslides are very easy to occur in the highly prone area predicted by the model and within 40 km from the earthquake point. The monitoring level of this area is 4.Secondly, landslide is relatively easy to occur in the highly prone area predicted by the model and within 40–80 km from the earthquake point, and the monitoring level is 3.Then, the landslide is moderately easy to occur in the highly prone area predicted by the model and the area less than 40 km from the earthquake point, and the monitoring level is 2.Finally, in the highly prone area predicted by the model and within 40 km–80km from the earthquake point, the landslide is relatively difficult. to occur, and the monitoring level is 1.

In general, this study combined earthquake disaster data and landslide disaster data to construct a landslide susceptibility evaluation model that is sensitive to earthquakes for southeastern Tibet. This work yielded estimates of the influence of earthquakes of different magnitudes on landslide occurrence in southeastern Tibet and the relationship between the distance from the earthquake focal point and the landslide susceptibility area. The model can well reflect the influence of earthquakes on landslides in southeastern Tibet, has good application prospects in terms of rapid assessment of landslide probability and susceptibility, and can provide theoretical support for predisaster prevention and postdisaster resettlement in southeastern Tibet.

Data availability statement

The data will be made available upon request.

CRediT authorship contribution statement

Zang Yeqi: Writing – original draft, Visualization, Validation, Supervision, Software, Resources, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Guo Yonggang: Writing – review & editing, Supervision, Software, Resources, Project administration, Funding acquisition, Formal analysis, Conceptualization. Wang Guowen: Validation, Supervision. Wu Shengjie: Validation, Supervision.

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.

Acknowledgements

The work described in this paper was funded by grants from the Major Science and Technology Project of Tibet Autonomous Region (XZ202201ZD0003G03 ), the Key Research and Development Project of Tibet Autonomous Region (XZ202201ZY0034G ).
==== Refs
References

1 Ali R. Kuriqi A. Kisi O.J.C. Human–environment natural disasters interconnection in China: A review 8 4 2020 48
2 Gizzi F. Bentivenga M. Lasaponara R. Danese M. Potenza M. Sileo M. Masini N.J.G. Natural Hazards, Human Factors, and “Ghost Towns”: a Multi-Level Approach vol. 11 2019 1533 1565
3 Yang Z. Liu G. Wang L. Liu S. Fu X.J.B.o.E.G. Environment t. Evolution of hydro-mechanical behaviours and its influence on slope stability for a post-earthquake landslide: implications for prolonged landslide activity 80 11 2021 8803 8822
4 Cui S. Yang Q. Pei X. Huang R. Guo B. Zhang W.J.G. Geological and morphological study of the Daguangbao landslide triggered by the Ms. 8.0 Wenchuan earthquake China 370 2020 107394
5 Guangzhi R. Jiquan Z. Tiantao L. Weihua F. Risk assessment of extreme precipitation-induced geological disaster chain:A case study of ShuichengCounty,Guizhou Province J Journal of Catastrophology 37 4 2022 201 210
6 Zhou Y. Li Y. Liu Y.J.H.I. The Nexus between Regional Eco-Environmental Degradation and Rural Impoverishment in China vol. 96 2020 102086
7 Gong Y. Yao A. Li Y. Li Y. Tian T.J.E.E.S. Classification and distribution of large-scale high-position landslides in southeastern edge of the Qinghai–Tibet Plateau China 81 11 2022 311
8 Zhao B. Su L. Xu Q. Li W. Xu C. Wang Y.J.E.-S.R. A Review of Recent Earthquake-Induced Landslides on the Tibetan Plateau 2023 104534
9 Fan X. Scaringi G. Korup O. West A.J. van Westen C.J. Tanyas H. Hovius N. Hales T.C. Jibson R.W. Allstadt K.E.J.R.o.g. Earthquake‐induced chains of geologic hazards: patterns, mechanisms, and impacts 57 2 2019 421 503
10 Lu M. Yang S. Zhu C. Wang J. Lin S. Wei W. Fan H.J.J.o.C. Thermal impact of the southern Tibetan Plateau on the Southeast Asian summer monsoon and modulation by the tropical Atlantic SST 36 5 2023 1319 1330
11 Yang Z. Ni W. Niu F. Li L. Ren S.J.R.S. Spatiotemporal distribution characteristics and influencing factors of freeze–thaw erosion in the qinghai–tibet Plateau 16 9 2024 1629
12 Zhao B. Su L. Wang Y. Li W. Wang L.J.J.o.R.M. Engineering G. Insights into some large-scale landslides in southeastern margin of Qinghai-Tibet Plateau 15 8 2023 1960 1985
13 Ilia I. Tsangaratos P.J.L. Applying Weight of Evidence Method and Sensitivity Analysis to Produce a Landslide Susceptibility Map vol. 13 2016 379 397
14 Pourghasemi H.R. Teimoori Yansari Z. Panagos P. Pradhan B.J.A.J.o.G. Analysis and Evaluation of Landslide Susceptibility: a Review on Articles Published during 2005–2016 vol. 11 2018 1 12 (periods of 2005–2012 and 2013–2016)
15 Pradhan B.J.I.j.o.c.i.s. Application of an advanced fuzzy logic model for landslide susceptibility analysis 3 3 2010 370 381
16 Lee S. Ryu J. Min K. Choi W. Won J. Development and application of landslide susceptibility analysis techniques using geographic information system (GIS), IGARSS 2000. IEEE 2000 international geoscience and remote sensing symposium. Taking the pulse of the planet: the role of remote sensing in managing the environment Proceedings (Cat. No. 00CH37120) 2000 IEEE 319 321
17 Shao X. Xu C.J.N.H.R. Earthquake-induced landslides susceptibility assessment: a review of the state-of-the-art 2 3 2022 172 182
18 Süzen M.L. Doyuran V.J.E.g. Data driven bivariate landslide susceptibility assessment using geographical information systems: a method and application to Asarsuyu catchment Turkey 71 3–4 2004 303 321
19 Neighbors C. Cochran E. Caras Y. Noriega G.J.N.H.R. Sensitivity analysis of FEMA HAZUS earthquake model: case study from King County, Washington 14 2 2013 134 146
20 Lee C.-T. Huang C.-C. Lee J.-F. Pan K.-L. Lin M.-L. Dong J.-J.J.E.G. Statistical approach to earthquake-induced landslide susceptibility 100 1–2 2008 43 58
21 Shafapourtehrany M. Rezaie F. Jun C. Heggy E. Bateni S.M. Panahi M. Özener H. Shabani F. Moeini H.J.R.S. Mapping post-earthquake landslide susceptibility using U-Net, VGG-16, VGG-19, and metaheuristic algorithms 15 18 2023 4501
22 Chen W. Sun Z. Zhao X. Lei X. Shirzadi A. Shahabi H.J.I.i.j.o.g.-i. Performance evaluation and comparison of bivariate statistical-based artificial intelligence algorithms for spatial prediction of landslides 9 12 2020 696
23 Qin Y. Yang G. Lu K. Sun Q. Xie J. Wu Y.J.S. Performance evaluation of five GIS-based models for landslide susceptibility prediction and mapping A case study of Kaiyang County, China 13 11 2021 6441
24 Lee S. Talib J.A.J.E.g. Probabilistic Landslide Susceptibility and Factor Effect Analysis vol. 47 2005 982 990
25 Pradhan B. Lee S.J.L. Regional landslide susceptibility analysis using back-propagation neural network model at Cameron Highland Malaysia 7 1 2010 13 30
26 Kumar R. Anbalagan R.J.J.o.E.S.S. Landslide susceptibility zonation in part of Tehri reservoir region using frequency ratio fuzzy logic and GIS 124 2015 431 448
27 Sun D. Xu J. Wen H. Wang Y.J.J.o.E.S. An optimized random forest model and its generalization ability in landslide susceptibility mapping: application in two areas of Three Gorges Reservoir China 31 2020 1068 1086
28 Ozturk U. Pittore M. Behling R. Roessner S. Andreani L. Korup O.J.L. How Robust Are Landslide Susceptibility Estimates? vol. 18 2021 681 695
29 Hodasová K. Bednarik M.J.N.H. Effect of Using Various Weighting Methods in a Process of Landslide Susceptibility Assessment vol. 105 2021 481 499
30 Clerici A. Perego S. Tellini C. Vescovi P.J.G. A procedure for landslide susceptibility zonation by the conditional analysis method 48 4 2002 349 364
31 Reichenbach P. Rossi M. Malamud B.D. Mihir M. Guzzetti F.J.E.-s.r. A Review of Statistically-Based Landslide Susceptibility Models vol. 180 2018 60 91
32 Merghadi A. Yunus A.P. Dou J. Whiteley J. ThaiPham B. Bui D.T. Avtar R. Abderrahmane B.J.E.-S.R. Machine Learning Methods for Landslide Susceptibility Studies: A Comparative Overview of Algorithm Performance vol. 207 2020 103225
33 Azarafza M. Azarafza M. Akgün H. Atkinson P.M. Derakhshani R.J.S.r. Deep learning-based landslide susceptibility mapping 11 1 2021 24112
34 Park S. Choi C. Kim B. Kim J.J.E.e.s. Landslide susceptibility mapping using frequency ratio, analytic hierarchy process, logistic regression, and artificial neural network methods at the Inje area Korea Times 68 2013 1443 1464
35 Kavzoglu T. Colkesen I. Sahin E.K.J.L.T. Practice, Modelling, Machine Learning Techniques in Landslide Susceptibility Mapping: a Survey and a Case Study 2019 283 301
36 Goetz J. Brenning A. Petschko H. Leopold P.J.C. Geosciences, Evaluating Machine Learning and Statistical Prediction Techniques for Landslide Susceptibility Modeling vol. 81 2015 1 11
37 Wei T. Yue S. Metals J.W.N. Review on the Evaluation Methods of Landslide Geological Hazards Based on GIS vol. 21 2020 157 159
38 Yong C. Jinlong D. Fei G. Bin T. Tao Z. Hao F. Li W. Qinghua Z.J.S.E.R. Assessment R. Review of landslide susceptibility assessment based on knowledge mapping 36 9 2022 2399 2417
39 Chang Z. Catani F. Huang F. Liu G. Meena S.R. Huang J. Zhou C.J.J.o.R.M. Engineering G. Landslide susceptibility prediction using slope unit-based machine learning models considering the heterogeneity of conditioning factors 15 5 2023 1127 1143
40 Huaifei S. Yu D. Mei Y. Caiqi S. Xinke Y. Assessment on landslide susceptibility in Gansu Province based on AHP and information J Research of Soil and Water Conservation 28 6 2021 412 419
41 Lin J. CalJingsen Echuan Y. Qian C. Yi D. Assessment of landslide susceptibility in Nanzhang County basea on geological envIronen zoning %J Yangtze River 52 5 2021 86 94
42 Changrun w. Yuanmei J. Jinliang W. Hanhua X. Hua Z. Qiue X. Frequency ratio and logistic regression models based coupling analysis for susceptibility of landslide in Shuangbai County %J Journal of Natural Disasters 30 4 2021 213 224
