Evaluation of Landslide Susceptibility in Tekes County, Yili Prefecture Based on the Information Quantity Method

Abstract

In order to scientifically and rationally evaluate the susceptibility to landslide hazards in Tekes County, Yili State. This paper takes Tekes County in Xinjiang as an example, on the basis of a comprehensive analysis of the regional geological environment conditions and the distribution pattern and formation conditions of geological disasters, using the data of geological disaster points (landslide center points), and through the correlation matrix calculation of the evaluation factors, the nine evaluation factors with larger absolute values of correlation coefficients were determined to construct the evaluation system of the susceptibility to landslide geological hazards in Tekesi County. Combining the information quantity method and the entropy value method, using the weights determined by the entropy value method, the information quantity method is used to calculate the information quantity value of each factor within the factor, calculate the susceptibility index of landslide geological disasters within the territory of Tekes County, and then carry out the landslide susceptibility evaluation. The susceptibility of landslide disasters was evaluated by ArcGIS. The results show that the landslide disaster susceptibility level in Tekes County can be divided into four levels: high susceptibility, medium susceptibility, low susceptibility, and not susceptible, with areas of 491.3276 km², 1181.5171 km², 1674.7609 km² and 5295.2976 km² accounting for 5.68%, 13.67%, 19.38% and 61.27% of the total area of Tex County, respectively. The AUC number obtained by the success curve method (ROC) is 0.8736, reflecting the evaluation accuracy of 87.36%, indicating that the model method used in this paper is effective. The results are expected to provide practical data support for landslide disaster control in Tekes County and provide a reference for geological disaster monitoring, early warning and engineering prevention and control deployment in Yili Valley.

1. Introduction

China boasts a large landmass, a variety of topography and landforms, intricate geological structures [1], and delicate geological environments [2]. In addition, with the continuous development of modern science and technology and socio-economic development, human exploitation of resources and natural pioneering activities have led to a decrease in the stability of the earth’s surface, which has created conditions for the breeding of disasters. Extreme weather occurrences and increased human engineering operations have led to a significant increase in geological disasters in recent years [3]. China is now among the nations in the world where geological disasters are widely distributed and have a significant impact [1,4].

As a result, incidents of major economic losses have occurred from time to time, and the situation of geological disaster prevention and control remains critical. In 2022, China’s natural disasters were dominated by floods, droughts, wind and hail, earthquakes and geological disasters. According to the summary of geological disasters in China in 2021, a total of 4772 geological disasters occurred nationwide, resulting in 80 deaths, 11 missing, and direct economic losses of CNY 3.2 billion. In the distribution of geological disasters released in China in 2021, there were 2335 landslides, 1746 avalanches, 374 mudslides, 285 ground collapses, 21 ground fissures, and 11 ground subsidence. In terms of disaster level, there were 35 extra-large geological disasters, 27 large geological disasters, 328 medium geological disasters, and 4382 small geological disasters [5]. The distribution of geological disasters in 2021 shows that landslide disasters are the main ones, followed by avalanche disasters, reflecting that among the various types of geological disasters, China’s geological disasters are dominated by landslide disasters, which are of various types and have great dangers, and urgently need to be evaluated by scientific and reasonable methods of risk assessment. The disaster situation reflects that China’s geological disasters are still mainly small- and medium-sized, which is basically consistent with the situation worldwide. Landslides are a major geohazard, and economic losses due to geo-environmental degradation and geohazards in developing countries amount to more than 5 percent of the gross national product annually [6]. In China, among the environmental problems caused by disasters, losses due to geological disasters account for about 35 percent of the overall disaster losses, and about 55 percent of these losses are caused by avalanches, landslides, mudslides and human-engineering-induced shallow-surface geological hazards, which cause enormous difficulties in post-disaster reconstruction work.

Landslides are common geological hazards that take a heavy toll on human life and infrastructure every year, making it crucial to identify potential landslides [7]. The Yili Valley, with its varied topography and delicate and intricate geological environment, is situated in Xinjiang’s western region of the Tianshan Mountains. The area frequently experiences geological disasters like mudslides, landslides, and collapses, which pose a major threat to the protection of locals’ lives, property, and the ecological environment [8,9].

Geological hazard susceptibility assessment is imminent. In recent years, geological disaster risk assessment has become a research hotspot for many scholars (e.g., Academician Yin Yue Ping, Academician Peng Jian Bing, and Professor Xu Qiang in China), and fruitful research results have been obtained. However, from the current study, it seems that the complexity of the content of the geological hazard evaluation makes it important to study in depth how to transfer the qualitative evaluation to quantitative evaluation. Meanwhile, with the continuous development of computers and related technologies, how to better simulate the conditions of geohazard development and integrate various technologies for geohazard risk assessment is the current difficulty [10].

Landslides are one of the most widespread natural hazards that cause damage to both property and life every year in the Yili area. The landslide susceptibility evaluation is necessary for land hazard assessment and mitigation of landslide-related losses [11]. The combination of geological conditions and trigger events results in geological catastrophes, a natural occurrence that endangers people’s lives and property as well as the ecological environment. Geological hazard susceptibility assessment, also referred to as geological hazard sensitivity assessment [12], is a process that primarily uses the characteristics of past disasters and the study of geological conditions that are prone to disasters to predict the probability of geological hazards occurring in a given region and scope. The findings have a significant impact on urban land spatial planning as well as the mitigation and control of geological hazards in the area [13]. Numerous techniques exist for assessing a geological disaster’s susceptibility, but no one assessment standard has been established as of yet [14].

The evaluation methods of geological hazard susceptibility are varied, such as BP neural network analysis [11], statistic modeling and GIS spatial clustering analysis [15,16], the logistic regression model [17], the AHP and entropy comprehensive evaluation method [18], certainty factor and deep neural network [19], the evidential belief function model [20], logistic regression and the MaxEnt model [21], learning a deep attention dilated residual convolutional neural network [22], knowledge mapping [23], the information gain [24], multi-source data [25], the occurrence mechanism [26], VW-AHP-IV model [27], a hybrid variable-weight theory approach of hierarchical analysis and multi-layer perceptron [28], a coupled informative–logistic regression model [29], and the SHAP-XGBoost model [30]. As of right now, the analytic hierarchy process, information quantity method, evidence weight method, logistic regression method, and multiple method coupling methods are among the frequently employed techniques [31].

With the rapid development of GIS technology, mathematical theory analysis and other means, scholars at home and abroad have developed a variety of models and methods for evaluating the susceptibility to geological hazards, but have not formed a set of systematic and effective methods for comparing the evaluation results and a unified evaluation standard [12]. Among them, the informativeness model has been widely used in the evaluation of geological disaster susceptibility due to its clear physical meaning, operability, practicality and reliability [14], and its ability to avoid subjective judgement, and the results are more objective and reasonable [15], which are better in line with the reality [32]. However, the informativeness method only reflects the impact of each evaluation factor on the occurrence of disasters when combined and does not actually take into account the degree of variability, i.e., the contribution/weighting, of each influence factor on the occurrence of disasters. And there is a strong subjectivity in the selection of the weights of the research into disaster-causing factors, and the different judgement standards of each person lead to differentiation in the results. The entropy method is adopted to objectively give the weights among the factors and the informativeness method is used to calculate the internal value of each classification level within a single factor so as to grade the sensitivity of geological disasters in the area. The advantage of the entropy method is that it can objectively give the weight of each factor, but the disadvantage is that the weight of each classification within a single factor needs to be “subjectively given”; however, the information quantity method is the opposite, and its advantage is that it can well calculate the internal value of each classification within a single factor (the amount of information), but the disadvantage is that the weight of each factor needs to be “subjectively given”.

The rapid development of “3S” technology has effectively improved the evaluation accuracy of the regional geological disaster sensitivity evaluation model, especially the combination of different modeling methods based on GIS has been widely used in geological disaster sensitivity evaluation and achieved good results. And most scholars believe that the evaluation effect of multiple methods is better than that of a single method. Wang Ningtao et al. (2012) used the geostatistical informativeness method in Wufeng County, Hubei Province as an example, selected and analyzed each disaster-causing factor from the disaster-causing mechanism based on GIS technology and RS technology, and completed the quantitative evaluation of geologic disaster susceptibility zoning using GIS spatial analysis technology [33]. Zhang Xiaodong et al. (2018) evaluated the susceptibility of Yanchi County in Ningxia through the quantitative information quantity method and the coefficient of determination method coupled with logistic regression, respectively, and concluded that the results of the coupled models were both better than the results of the single model evaluation [34]. Jin Chao et al. (2021) used the informativeness model and logistic regression model to establish the informativeness model and the informativeness method and logistic regression model for the evaluation of geological disaster susceptibility in the Utopia District of Shiyan City, Hubei Province, as an example [35], Lin Jinhuang et al. (2021) talked about risk assessment and its influencing factors in the analysis of geological hazards in typical mountain environments [36], and Song Yong et al. (2023), Yang Dehu et al. (2023) and Xiao Haiping et al. (2023) used the informativeness method to carry out a comparative study of geological hazard risk assessment pairs and concluded that the validity of the method was better [37,38,39]. Chowdhuri Indrajit et al. (2021) applied machine learning and statistical methods to assess the susceptibility of torrential-rainfall-induced landslides in the eastern Himalayas [40]; they compared and analyzed the results of the evaluation, and through the test of significance, it was concluded that the method was able to objectively respond to the susceptibility to geological disasters with better results. Pal Subodh Chandra et al. (2019) used the integrated analytic hierarchy process and GIS technique to analyze potential landslide vulnerability zonation of the Upper Rangit Catchment Area, West Sikkim, India [41].

Deep learning evolved from simple perceptron artificial neural networks, but perceptron machines have difficulty solving more complex nonlinear problems. Although the BP (backpropagation) neural network is generated through the error backpropagation algorithm of a multilayer perceptron with a nonlinear continuous transformation function, the BP neural network model uses the backpropagation method to update the weights of neurons according to the error rate, and adjusting the weight values based on the local gradient descent is prone to the gradient diffusion phenomenon, and this phenomenon increases with the increase in network layers. This problem restricts the development of neural networks. Deep learning algorithms support their learners by extracting high-complexity data table evidence. This data extraction mechanism provides raw data for machine learning and enables these “learners” to automatically discover appropriate data representations, while deep learning algorithms produce better results in different learning applications. Deep learning is an important achievement in artificial intelligence and machine learning, enabling simple data models to process knowledge gained from large amounts of data with complex representations: automatically extracting data representations, using different data types, and gaining semantic and relational knowledge from raw data with higher-level representations.

However, significant challenges related to deep learning remain unresolved, especially in the area of big data analytics. There are still some challenges and shortcomings in deep learning-based landslide susceptibility assessment methods. Existing deep learning methods for assessing landslide susceptibility mainly focus on algorithmic improvement, but using algorithmic improvement, it is sometimes difficult to obtain the ideal assessment of landslide susceptibility. This is because the effectiveness of landslide susceptibility modeling depends not only on the basic quality of the algorithms used, but also on the quality of the landslide inventory, the factors affecting it, the strategy for selecting positive and negative samples of landslides, and the methods for dealing with missing values, noise and erroneous data. The construction of the dataset in the study of landslide susceptibility based on deep learning relies too simply on the geological environment information of a single unit point in the region, which can lead to the problem of incomplete spatial information on landslides and discontinuous information on geological attributes, which seriously restricts the accuracy of the output variables of the deep learning model, thus restricting the improvement of the accuracy of deep learning in evaluating the susceptibility of landslides.

And so far, there are relatively few studies on landslide susceptibility assessment based on the informativeness method for typical loess landslide areas in Yili Prefecture. The novelty of this study is to address the hot issue of geological disaster risk evaluation; the complexity of the evaluation requires knowing how to transfer the qualitative evaluation to a quantitative evaluation. In this study, on the basis of collecting a large amount of detailed first-hand field data, field research and multivariate data analysis and information extraction were carried out to compare and analyze the advantages and disadvantages of a variety of evaluation methods and to determine the geohazard risk evaluation method based on the informativeness model [42,43] using the entropy method to calculate weights, as well as the combination of entropy theory and the informativeness method, and the geohazard risk evaluation method was used to carry out a disaster risk evaluation of Techex County of Yili, Xinjiang, China [9]. After field verification and survey unit feedback, the results of this evaluation were found to be highly consistent with the actual situation in the field, and it is an effective method for the evaluation of geological disaster risk in the Yili Valley, which has a better prospect of promotion and application to predict the probability of occurrence of geologic disasters in Tekes County, which is of great significance for the prevention and control of geologic disasters, as well as for the urban territorial spatial planning [11].

2. Geological Environmental Conditions

Tekes County has complicated landforms, undulating terrain, steep slopes, and deep and narrow river basins. It is situated in the western part of the Tianshan Mountains. Human engineering activity is mostly focused on river valleys and valley slope areas, and these regions have active plate tectonics and well-developed folds and faults. Geological disasters are growing, and environmental geological issues are prevalent. There are four types of geological disasters: landslide, collapse, debris flow and ground collapse. Eight townships in the county contain geological catastrophe sites, which are primarily found in the Qiaolak Tiereke Township, the Qile Wuzeke Township, the Karadara Township, the Mongolian Township, and Tosba Ranch (Figure 1).

2.1. Topography and Landform

With its complex and varied geological processes, Tekes County is situated at the meeting point of the Central and Southern Tianshan Mountains. In general, the south has higher terrain than the north. The southern boundary is the main ridge of the Southern Tianshan Mountains. The northern boundary is the southern slope of the Alahar and Yeshkelik Mountains. The Tekes Basin, which stretches from east to west, is the center portion of the region. The Tekes River flows through the region, where it is continuously eroded by rivers to create a complicated topography with deep valleys. With the southern mountains having a maximum elevation of 4891 m, the valley plain tilts from west to east. At an elevation of roughly 869 m, the lowest altitude is situated in the middle northern region of the plains on both banks of the Tekes River. Landforms are classified into five main geomorphic units based on the processes that led to their formation: ice-eroded high mountains, eroded middle–high mountains, eroded middle mountains, eroded accumulation hills, and accumulation plains (Figure 2).

2.2. Meteorology and Hydrology

The basic characteristics of Tekes County’s continental desert climate are pleasant summers, erratic spring temperatures, rapid autumn cooling, and long, very cold winters. With an average yearly temperature of 5.3 °C, statistical data from the county meteorological station spanning more than thirty years indicate a notable variation in temperature throughout the county. Low topography and high temperatures are found in the eastern portion of the river valley; as the terrain rises from the east to the west, the temperature falls. The agricultural regions on each side of the river valley do not significantly differ in temperature. With an annual extreme high temperature of 36.2 °C and an annual extreme minimum temperature of −33.4 °C, January is the coldest month and July the hottest. The earth freezes for a lengthy time during the frigid winter months. Late October is when it normally begins to ice, and it thaws by the end of March the following year. There is more precipitation in the western portion of the valley than in the eastern portion, and there is more precipitation in the mountainous areas than in the valley itself. At an elevation of 2000–3000 m on the northern slope of the Tianshan Mountains lies the maximum precipitation zone. May and June see the most precipitation. The county has abundant precipitation and developed water systems. The Tekes River, Kuoksu River, Kuoktierek River, and Qiaolaktierek River are the county’s four main rivers. Approximately 6 billion m³ of surface water runoff occurs annually in the county, making up one-third of the Ili region.

The three main categories of water types are bedrock fissure water, clastic rock fissure water, and Quaternary loose rock pore water. A depression basin of the Mesozoic and Cenozoic ages is the Tekes Basin. Groundwater is regulated by well-developed faults, depressions, and uplifts that result from neotectonic motions. Thick, pore-filled Quaternary loose rocks have been accumulated in the Tekes River Valley, where abundant loose rock-pore water exists: the upper half of the alluvial fan on both sides of the basin is a single pebble layer with only pore water produced; the core portion of the basin has developed into a multi-layer structure with pressured (self-flowing) water. Only Tertiary clastic rock fissure pore water, also known as bedrock fissure water, is found west of Tekes County.

2.3. Stratigraphic Lithology

Proterozoic Sinian (Z), Jixian (Jx), Changcheng (Ch), and Qingbaikou (Qn) systems; Paleozoic Silurian (S), Carboniferous (C), Permian (P); Mesozoic Jurassic (J) and Cenozoic Paleogene (E), Neogene (N), and Quaternary (Q) are among the relatively complete strata found in the research region. The principal ridges of mountain ranges are frequently formed by well-developed intrusive rocks. Its long axis orientation essentially aligns with the principal line of the structure. Batholiths are the primary kind of intermediate and late intrusive rocks found in Hualixi. Only east of Tekes is the Luliang intrusion, which is primarily made up of acidic granite and a trace proportion of granodiorite.

2.4. Geological Structure

The research area’s primary structure, which is a part of the Tianshan latitudinal tectonic system, stretches east–west. Among other rock blocks, it is made up of a number of east–west composite folds and faults, northeast and northwest-trending torsional faults, east–west tracing faults, and north–south trending tensile faults. The primary fault in the region is the East trending fault zone at the north of Tekes River, which stretches from northeast to southwest and is made up of several faults with varying dimensions and characteristics. The two extremities are enlarged and widened while the center is squeezed and narrowed. It stretches along the Tekes River for around 100 km. Tekes County has had several earthquakes in the past and is a component of the North Tianshan Earthquake Belt.

2.5. Engineering Geological Rock Formations

Seven categories comprise the engineering geological rock components in the research domain. Composed of late Variscan intrusive rocks, the block-like hard intrusive rock group is spread throughout the center southern mountainous area, stretching east–west. Granite and fine-grained granite make up the majority of the lithology, which has consistent lithology and good integrity. Its geological performance is suitable for engineering and it is hard and dense like a block. Thick layered carbonate rock formations, ranging in hardness from relatively hard to medium, are extensively found throughout the county’s intermediate to high mountain regions, extending from the south to the north. Grayish green tuff coarse sandstone, limestone, siltstone, sandy gravel, clastic limestone gravel, and sandstone mixed with calcareous siltstone locally make up the majority of the lithology. The rock is hard and has good engineering geological conditions, generally with a layered structure. Collapses are easily formed in deep valleys and mountain fracture zones, and they serve as a source of debris for debris flows. The northern portion of the county and the Tekesdaban area are home to the blocky, layered, hard to weak clastic rock formations, which are primarily made of sandstone and are arranged in strips. With significant tectonic action, the lithology is primarily composed of sandstone, coarse sandstone, and calcareous siltstone. The majority of the rocks have been moderately to heavily worn, and the engineering geological conditions are medium. This area, which is vulnerable to earth collapse, is home to the majority of the county’s coal mines. The interbedded hard to weak clastic rock formations, which are primarily made of mudstone and sandstone, are mostly found in the central region of the county in the Qiongkushitai area on the south bank of the Kurdai River. They are made up primarily of mudstone or muddy cementation, which are relatively weak Paleogene Oligocene strata. Because of their high permeability, ease of softening and collapsing when exposed to water, and hardness following water loss, loess landslides are quite common due to geological factors. Mostly made up of mudstone, layered weak clastic rock formations are found in the lower and middle reaches of valleys in the northwest of the county, including Bahalek, Jorakmis, Suasu, and the Wurtamis gullies. They are made up of weak Neogene strata that have limited resilience to weathering, primarily formed of mudstone that is brownish red and brownish yellow in color. They are in the county’s high-risk area for landslides caused by loose soil. The majority of the aeolian deposits in loess are from the middle and upper Pleistocene of the Quaternary, and they are primarily pale yellow in color. Based on the distribution of loess, topographic considerations are crucial for the development of loess. It is primarily found in the valley’s foothills on both sides. The upper limit of terrain elevation is found in the woodland zone, which is typically between 1800 and 2200 m high. Connected to the valley plain already, the lower limit has a minimum elevation of 900~1000 m.

2.6. Human Engineering Activities

Human engineering activities in Tekes County are primarily focused on agriculture and water conservation and manufacturing and processing in the plain areas; animal husbandry, forest logging, and mining are concentrated in mountainous areas; transportation facilities engineering and tourism engineering are reflected in both plain and mountainous areas, with a high concentration of activities in the latter. As the economy grows, so do human socioeconomic activity and environmental harm, which in turn causes the occurrence of geological disasters in particular regions.

3. Research Methods

The theoretical basis of the information quantity method is information theory and engineering geological analogy, which is a data-driven statistical prediction method. Its main idea is to obtain the information quantity of each classification level by grading each disaster-causing factor through the existing geological disaster data information, to judge the degree of connection between each classification level and the occurrence of disasters by its size, and to discuss objectively the information quantity contributed by the influencing factors to the occurrence of disasters. In the field of geohazard evaluation research, the information quantity method is more scientific in its calculation of the information quantity of subclasses of evaluation factors, but in the process of calculating the total information quantity, it is only obtained by adding up the information quantities of evaluation factors in equal quantities, which has certain limitations. The entropy value method is an objective assignment method that uses information entropy to calculate its weight by calculating the entropy value of each evaluation factor and then determining the weight of each evaluation factor according to the discrete degree of the evaluation factor; the greater the degree of discrete degree of a certain indicator, the greater its influence on the comprehensive evaluation. Finally, we combine the entropy value method to determine the weights, weight and superimpose the informativeness value given to each classification level of the influence factors in GIS, and combine the natural breakpoint method to obtain the geological hazard sensitive level, which can completely overcome the shortcomings of the informativeness method. In view of this, this study combines the information quantity method and entropy value method, using the weights determined by the entropy value method and the information quantity method to calculate the information quantity value of each factor within the factor, to calculate the susceptibility index of landslide geohazards in the territory of Tekesi County and then carry out the evaluation of the susceptibility of landslides.

3.1. Information Quantity Model

The informativeness model is a statistical predictive analysis method that gives the weights of evaluation factors in the form of conditional probabilities. Compared with the general evaluation model, it has higher objectivity, can correctly respond to the basic laws of geological disasters, and is easier to implement on the GIS platform, making it a more scientific and practical algorithm. Yan [44] was the first to use the information quantity model, a statistical technique for analysis and prediction, in the assessment and forecasting of geological hazards. The distribution of landslide points is first extracted in order to assess the geological hazard sensitivity to landslides. Next, landslide impact factors are chosen and ranked according to the circumstances of the research region. Each factor’s measured values are transformed into information values that indicate how vulnerable geological risks are. The effect factors are analyzed and overlayed using GIS, and the larger the information value, the greater the vulnerability to geological hazards [45]. In real-world scenarios, the computation formula is:

$$I_i=ln\frac{N_i/N}{S_i/S}$$

(1)

where S is the total number of grid units in the study area; N is the number of grid units of existing geological hazard points in the study area; I_i is the information quantity value of geological hazard factor classification i; N_i is the number of geological hazard grid units corresponding to the classification of geological hazard factors i; S_i is the total number of grid units corresponding to the classification of geological hazard factors i.

3.2. Determine the Weight of Each Evaluation Factor

Each evaluation factor’s entropy value is computed using the entropy technique, which then establishes the weight of each component according to how much it has changed. Each assessment factor has a larger weight when the estimated entropy value is smaller. This article chooses to utilize the entropy approach to calculate each evaluation factor’s contribution to the occurrence of landslide disasters due to their distinct contributions.

(1) Use the following formula to determine the entropy value:

$$E_{\mathrm{j}}=-K{\displaystyle \sum _{\mathrm{i}=1}^m{Y}_{ij}}\ln \left(Y_{ij}\right)$$

(2)

where i is the secondary evaluation factor, j is the first-level evaluation factor; Y_ij is the normalized value of the density of disaster points within the secondary evaluation factor area of each primary evaluation factor; E_j is the entropy value of the first-level evaluation factor of item j; K is a constant, and K = 1/ln (n), where n is the number of secondary evaluation factors.

(2) Use the following formula to determine the assessment indicators’ weight:

$$W_j=\frac{\left(1-E_j\right)}{N-{\displaystyle {\sum }_{j=1}^N{E}_j}}$$

(3)

where W_j is the weight of each first-level evaluation factor in the calculation and N is the number of first-level evaluation factors.

(3) Determine weighted information quantity:

The weighted information content I for the final GIS spatial analysis is calculated by multiplying the information quantity of each secondary evaluation factor determined by Equations (1) and (3) by the weight of each first-level evaluation factor. This is because the information quantity model does not take the contribution value of each evaluation factor to the occurrence of disasters into account. The following is the formula:

$$I=W_j\times I_i$$

(4)

To generate the comprehensive evaluation index of Tekes County’s geological hazard susceptibility to landslides, calculate the weighted information value I of each evaluation element and stack them using the ArcGIS grid calculator. The ultimate evaluation of Tekes County’s geological hazard susceptibility to landslides is produced following several debugging attempts. In Figure 3, the particular flowchart is displayed.

4. Analysis of Data Sources and Evaluation Factor Selection

4.1. Data Sources

The data used in this study are as follows. (1) DEM base map with 12.5 m resolution: used to extract topographic and geomorphic factors. (2) Geological disaster data: “Detailed Geological Disaster Survey Project Database of Tekes County, Yili Prefecture, Xinjiang (1:50,000)”, completed by the Geological Environment Monitoring Institute of Xinjiang Uygur Autonomous Region in 2022, is used to obtain relevant data of geological disasters. (3) Basic geological data: The “Geological Map of Xinjiang Uygur Autonomous Region (1:2.5 million)” completed by the Geological Survey Institute of Xinjiang Uygur Autonomous Region in 2023 is used to extract evaluation factors such as formation lithology and geological structure. (4) Basic geographic base map: The “Basic Geographic Information Database (2019)” of the National Basic Geographic Information Center is used to extract data of major rivers and administrative divisions.

4.2. Analysis of Evaluation Factor Selection

The formation and occurrence of geological disasters is the result of a combination of factors, and its occurrence is affected by a number of factors such as geological conditions and triggering factors, etc. However, too much selection of evaluation indexes will complicate the calculation process, and there is a correlation between some evaluation indexes, which will lead to bias in the evaluation results [46]. Reasonable selection of evaluation indexes is an important basis for the evaluation of geological disaster susceptibility [47].

Based on the research results of previous scholars and the geological environment of the study area, as well as the geological conditions and triggering factors of the disaster, Without considering the differences in the sliding mechanisms of different types of landslides on a case-by-case basis, this study, combined with the actual situation of Tekes County, selects the representative factors that have a greater impact on geological disasters as the preliminary factors and uses the multivariate analysis tool in the ArcGIS software (v.10.8.2) to calculate the correlation matrix of the evaluation factors so as to analyze the covariance of the preliminary factors and obtain the table of correlation coefficients of the various factors. The correlation coefficients of the two groups of factors are shown in the table, and the larger the absolute value of the correlation coefficients of the two groups of factors, the larger the correlation is. In the end, nine evaluation factors, including elevation, slope, slope direction, relief, distance from faults, engineering geological rock group, land use type, average rainfall over the years, and distance from rivers, were selected to construct the evaluation system of the susceptibility to landslides and geological hazards in Tekes County.

4.2.1. Elevation

One of the most significant topographic variables influencing the likelihood of catastrophes is elevation [48]. Elevation influences the stress condition within slopes as well as providing potential energy for geological disasters like slopes. Elevations, slopes and relief are spatially aggregated, and often where elevations are high, slopes and relief are equally high within a certain area, and there is a connection between them. The study area’s Digital Elevation Model (DEM) indicates that its elevation ranges from 869 to 4891 m. The research area’s elevation is divided into five categories using the natural discontinuity categorization method: 869~1606 m, 1606~2191 m, 2191~2799 m, 2799~3392 m, and 3392~4891 m (Figure 4). Geological hazards exhibit a trend of initially increasing and then decreasing in density as elevation rises. These risks are concentrated in the height range of 1606~2191 m, making up around 54.27% of the total.

4.2.2. Slope

Slope gradient is an important factor in the occurrence of geological hazards such as landslides, which are closely related not only to the thickness of the soil layer, climatic conditions, hydrological conditions, lithological conditions, and many other factors. And it is closely related to the physical and mechanical properties of the slope geotechnical body, which is not only one of the important factors affecting the distribution of slope geohazards, but also an important control index for the occurrence of loess landslide disasters. Slope determines the critical conditions and the degree of unloading development of a slope, thus affecting the internal stress distribution of the slope, as well as controlling the thickness of the accumulation of loose rock and soil bodies on the surface of the slope, the vegetation cover and the surface runoff. Previous studies have shown that most geological risk assessments use slope as one of the important independent variables. As the slope increases, the shear force, including gravity, increases, the range of the stress relief zone near the slope surface expands, the stress concentration at the foot of the slope increases, and the probability of landslides and other geological hazards under its influence also increases [49]. The slope can act as a free surface for the slope to deform, and the slope’s magnitude directly influences the sliding force’s magnitude, which in turn influences the slope’s instability [50]. The natural discontinuity categorization method divides the slope into five categories: 0~15°, 15~30°, 30~45°, 45~60°, and above 60° (Figure 5). The primary distribution of geological disasters is between approximately 0 and 15° and between approximately 15 and 30°.

4.2.3. Slope Orientation

Slope orientation refers to the direction of slope orientation, which affects landslides by determining the light time, water/heat ratio, and precipitation distribution on the hillside. It is an important indicator for determining the direction of sliding, and is an important guide for the prediction and avoidance of geological hazards. Slopes facing south are usually considered as quasi-sunny slopes and slopes facing north are considered as standard shady slopes. Due to the different orientations, there are regular differences in the minor environment of the slopes. In general, sunny slopes have longer sunshine hours, greater solar altitude, stronger solar radiation, more abundant heat, higher air and water temperatures, greater daily differences in temperature, and differences in the slope orientation of hydrothermal conditions than shady slopes, leading to regular differences in the natural geographic elements on slopes of different orientations. In general, it appears that shady and semi-shady slopes distribute more geohazards overall than sunny and semi-sunny slopes and that slope orientation affects evapotranspiration, intensity of solar radiation, and changes in vegetation and environmental ecology. The short sunshine hours and weak vegetation cover conditions on shaded slopes lead to reduced slope stability and susceptibility to geohazards. The development of slope flora, water evaporation, and weathering intensity are all impacted by the differing slope orientations’ exposure to solar radiation. As shown in Figure 6, there are nine directions that make up the slope direction: plane, northeast, east, southeast, south, southwest, west, northwest, and north. Geological disasters known as landslides are common on the research area’s slopes facing the northwest, northeast, west, and north.

4.2.4. Terrain Undulation

There is a correlation between the distribution of landslides and the undulation of the landscape, which represents surface fluctuations and changes [42]. The study area’s topography undulation is categorized into five categories using the natural discontinuity grading method (Figure 7), which are 0~9 m, 9~18 m, 18~30 m, 30~51 m, and 51~286 m, in that order. According to research, the region of 0~18 m is where geological disasters are most commonly found, which is favorable for disaster occurrence.

4.2.5. Distance from Fault

In addition to regulating the border and geographical location of landslides, fault structures also regulate the stability and rupture of adjacent rock and soil. We analyze the fault buffer zone in the study region using Arc GIS’s multiring buffer function. Then, we divide the buffer into six sections: <1000 m, 1000~2000 m, 2000~3000 m, 3000~4000 m, 4000~5000 m, and >5000 m (Figure 8). Studies have revealed that geological catastrophes are very developed and the number of disaster locations is comparatively high within the buffer distance of less than 3000 m.

4.2.6. Engineering Geological Rock Formations

The lithology of the strata provides the structural foundation for landslides, and various engineering geological rock formations have unique mechanical and physical characteristics that lead to various kinds of disasters. The soil body that constitutes a slope is not only subject to gravity, but also to shear stress, and the geotechnical soil that resists the effective stress determines both the likelihood of the occurrence of slope failure and the change in slope instability in time and the nature of the geotechnical body is of considerable importance to both the occurrence and the development of slope failure. The mechanical properties of the geotechnical body depend on the particle and mineral composition of the geotechnical body, the structural configuration of the geotechnical body, the nature of adsorbed cations and the nature of pore water, and the composition and structure of the geotechnical body depend on the cause of the geotechnical body and the historical processes it has undergone. The study area’s stratigraphic lithology is separated into seven categories of engineering geological rock formations, such as loess-like subsandy soil monolayer soil and interbedded, relatively weak sandstone conglomerate (Figure 9). The engineering geological rock formations in the study area with the most developed disasters were discovered to be the tougher layered volcanic clastic rock formations and the weakly interlayered sandstone conglomerate.

The middle mountainous area and the pile hilly area, with a slower topography, are important loess distribution areas, and loess landforms such as loess beams can often be seen. The structure of the soil is relatively dense, and slightly wet, with small wormholes containing calcareous nodules and conchoidal shells, and columnar joints are developed. From the results of the particle analysis of the soil, it can be seen that they are all powdery clay and clay, and in the east–west direction, there is a trend of the powder content increasing from west to east. The loess-like soil monolayer is the most widely covered geotechnical body in the landslide-prone area, and its lithology is light yellow-greyish-yellow chalk. Its structure is relatively loose, wormholes, large void structures, and vertical joints are relatively developed, and the shear strength is relatively high in the unsaturated state, but greatly reduced when saturated with water. Its main physico-mechanical indexes are as follows: natural water content (ω) in 8.14~21.3%, specific gravity (Gs) in 2.70~2.71, porosity (n) in 0.42~0.54, plasticity index (IP) in 7.68~13.23, compression coefficient (a_V) in 0.89 MPa⁻¹~1.06 MPa⁻¹, compression modulus (Es) in 2.05~2.42 MPa, high shear strength in the natural state, cohesion (c) in the range of 9.13~48.3 kPa, and angle of internal friction (φ) in the range of 13~37°, and the strength decreases rapidly after contact with water, with the cohesion (c) in the range of 3.7~20.1 kPa, and the angle of internal friction (φ) in the range of 11~33°.

4.2.7. Distance from River

The river’s cutting and erosion activity at the base of slopes will hasten the slope’s deformation. In addition, different levels of surface scouring and erosion on slopes might be brought on by the density of river development and distribution [51]. The primary river systems in the study area were subjected to a multi-ring buffer zone analysis. The river buffer zones were subsequently split into seven areas: <400 m, 400~800 m, 800~1200 m, 1200~1600 m, 1600~2000 m, 2000~2500 m, and >2500 m (Figure 10). The majority of the geological hazard sites in the study region are located fewer than a thousand meters from the river; the distribution of hazards decreases with increasing distance from the river.

4.2.8. Land Use Types

Vegetation cover is an important condition that leads to geological hazards. Vegetation can play a role in protecting slopes from soil erosion and has an impact on the evolution and stability of slopes. Tekes County has a variety of land uses (Figure 11), with bare rock gravel land, high coverage grassland, and medium coverage grassland areas being the most common locations for landslide disasters. This illustrates the connection between vegetation coverage and landslides, and it shows how important plant slope protection is in real-world engineering buildings. The occurrence of landslides in locations covered by glaciers and snow indicates that seasonal freeze–thaw conditions and cycles have a significant role in triggering landslides. Additionally, landslides are a major form of disaster in regions where seasonal frozen soil is distributed in high alpine regions.

4.2.9. Annual Average Rainfall

Rainfall is an important factor that induces landslides and triggers geological disasters. Geological disasters occur in close relationship with rainfall intensity. The number of rainfall disasters and rainfall is basically positively proportional to the relationship between the number of rainfall disasters and the amount of rainfall. From the area of the disaster, it can be seen with the increase in the amount of rainfall, the frequency of disasters increases. Due to the special characteristics of loess in the study area, surface water rapidly seeps into the deep part, and the shear strength of the soil body decreases, thus triggering landslides. One of the main causes of landslide tragedies is precipitation. The multi-year rainfall in the study area was analyzed and categorized into five categories: 0~3500 mm, 350~400 mm, 400~450 mm, 450~500 mm, 500~550 mm, 550~650 mm, and >650 mm (Figure 12). The typical rainfall range of 450~650 mm is mostly responsible for geological disasters, and this range is extremely compatible with the terrain and geomorphology.

5. Results and Discussion

5.1. Selection of Evaluation Elements

Tekes County, Xinjiang, is used as the research region in this article, and there have been 199 landslide catastrophes in total. Since most disasters are minor in scope, grid elements can be used to achieve the necessary evaluation accuracy. We chose 12.5 m × 12.5 m as the grid element size for this evaluation. The research region is divided into 55,314,582 grid elements.

5.2. Calculation of Information Quantity of Evaluation Factors

Using the information quantity model as a guide, we calculate each evaluation factor’s information quantity values. Next, we determine each evaluation factor’s weight using the entropy approach. Lastly, we use Equation (4) to determine the weighted information quantity values for each secondary evaluation factor (

Table 1. Weight and weighted information of each influencing in a landslide.

Primary Impact Factor	Secondary Impact Factor	Disaster Spot (Location)	Area (km2)	Disaster Spot Density (km2)	Weight	Weighted Information Quantity
Eevation	869–1606	85	1735.1870	0.0490	0.3558	0.26862
	1606–2191	108	1973.0442	0.0547		0.30812
	2191–2799	4	1745.8658	0.0023		−0.82102
	2799–3392	2	1733.8125	0.0012		−1.06517
	3392–4891	0	1454.9939	0.0000		0.00000
Slope	0–15	113	3077.3652	0.0367	0.0577	0.02693
	15–30	67	2855.9755	0.0235		0.00108
	30–45	16	2132.4848	0.0075		−0.06470
	45–60	2	503.5000	0.0040		−0.10139
	>60	1	73.5780	0.0136		−0.03042
Slope orientation	Plain	0	170.2992	0.0000	0.0366	0.00000
	Northeast	29	1112.3845	0.0261		0.00455
	East	18	1148.7739	0.0157		−0.01409
	Southeast	10	1099.4645	0.0091		−0.03399
	South	10	862.0239	0.0116		−0.02509
	Southwest	21	936.3509	0.0224		−0.00096
	West	34	1137.3806	0.0299		0.00956
	Northwest	46	1234.6095	0.0373		0.01762
	North	31	941.6163	0.0329		0.01309
Relief amplitude	0–9	121	3338.7381	0.0362	0.0571	0.02590
	9–18	59	2548.4577	0.0232		0.00031
	18–30	14	2005.3339	0.0070		−0.06814
	30–51	4	641.3795	0.0062		−0.07458
	51–286	1	108.9942	0.0092		−0.05254
Fault buffer zone	0–1000	63	3173.4739	0.0199	0.0466	−0.00691
	1000–2000	37	2066.5031	0.0179		−0.01172
	2000–3000	12	1289.6542	0.0093		−0.04222
	3000–4000	15	850.8013	0.0176		−0.01244
	4000–5000	19	503.0083	0.0378		0.02307
	>5000	53	759.4627	0.0698		0.05167
Engineering geological rock formations	A	32	250.8534	0.1276	0.0818	0.14005
	B	22	1383.5292	0.0159		−0.03028
	C	46	5075.4648	0.0091		−0.07627
	D	27	1259.8809	0.0214		−0.00587
	E	48	514.3091	0.0933		0.11448
	F	10	105.2644	0.0950		0.11594
	G	14	53.6016	0.2612		0.19867
Land-use type	Paddy field	0	1.2348	0.0000	0.1637	0.00000
	Dry land	21	433.5838	0.0484		0.12173
	Forest land	22	1033.6438	0.0213		−0.01287
	Shrub	3	24.6384	0.1218		0.27264
	Sparse forest land	0	91.5045	0.0000		0.00000
	Other forest lands	0	8.7031	0.0000		0.00000
	High coverage grassland	112	2862.3880	0.0391		0.08681
	Medium coverage grassland	36	2191.7117	0.0164		−0.05529
	Low coverage grassland	0	49.7048	0.0000		0.00000
	River and canals	2	37.7741	0.0529		0.13632
	Lakes	0	4.3225	0.0000		0.00000
	Glacier snow cover	0	525.3338	0.0000		0.00000
	Shoal	0	0.3092	0.0000		0.00000
	Urban land	0	6.6989	0.0000		0.00000
	Rural residential land	0	34.0231	0.0000		0.00000
	Bare Rock Gravel Land	2	47.4873	0.0421		0.00000
	Other	1	1289.8416	0.0008		−0.55512
Distance from river	<400	83	486.6303	0.1706	0.0892	0.17863
	400–800	31	459.9100	0.0674		0.09581
	800–1200	15	436.4616	0.0344		0.03573
	1200–1600	13	420.1498	0.0309		0.02636
	1600–2000	7	400.2408	0.0175		−0.02453
	2000–2500	8	465.9930	0.0172		−0.02618
	>2500	42	5973.5180	0.0070		−0.10581
Annual average rainfall	0–350	4	101.2805	0.0395	0.1116	0.06022
	350–400	16	453.2650	0.0353		0.04769
	400–450	10	581.2325	0.0172		−0.03252
	450–500	61	1320.6922	0.0462		0.07769
	500–650	86	1382.7242	0.0622		0.11090
	>650	22	4803.7091	0.0046		−0.18022

Note: The engineering geological rock group consists of the following: A is an interbedded hard to weak clastic rock group that is primarily made up of sandstone and mudstone; B is a hard intrusive rock formation; C is a relatively hard–medium thick-layered carbonate rock formation; D is a multiple-layer system of sand, gravel, and silt; E is a layered weak clastic rock formation that is dominated by mudstone; F is a single-layer soil that contains pebbles and gravel; G is a block, hard- to weak-layered clastic rock formation that is dominated by sandstone.

).

5.3. Evaluation of Landslide Susceptibility

The ArcGIS grid reclassification function is used to input the weighted information of each secondary evaluation factor based on the grading map and weighted information of each evaluation factor. The grid calculator is then used to superpose each evaluation factor to obtain the Tekes County landslide geological hazard susceptibility evaluation index map. Tekes County’s landslide geological hazard susceptibility evaluation index map is divided into high-susceptibility, medium-susceptibility, moderate-risk, and low-risk zones using the natural breakpoint grading method (Figure 13).

5.4. Precision Inspection

Statistics on the area, percentage of area, number of disaster points, and density of disaster points in the prone areas were measured out based on the results of the susceptibility evaluation, as indicated in

Table 2. Zoning statistics of susceptibility to landslide.

Prone Zoning	Area	Area Proportion (%)	Number of Disaster Points (Location)	Proportion of Disaster Spots (%)	Disaster Spot Density (Location/km2)
No	5295.2976	61.27%	9	4.52%	0.0016
Low	1674.7609	19.38%	34	17.09%	0.0203
Moderate	1181.5171	13.67%	66	33.17%	0.0558
High	491.32765	5.68%	90	45.23%	0.1831

.

Tekes County’s geological high hazard-prone areas cover 491.32765 km², or 5.68% of the total county area, as shown in Table 2. The density of disaster spots is 0.1831/km², and there are 90 landslide hazards spread over 45.23% of the total county area. The area of the moderate prone area is 1181.5171 km², or 13.67% of the whole area. The density of catastrophe locations is 0.0558/km², and there are 66 landslide disasters over 33.17% of the total county area. The low-prone area is 1674.7609 km², or 19.38% of the whole area. The density of disaster points is 0.0203/km², and there are 34 landslide disasters spread over 17.09% of the total county area. The size of the less prone area is 5295.2976 km², or 61.27% of the whole area. The density of disaster points is 0.0016/km², and there are nine landslide disasters over 4.52% of the total area. The sensitivity of landslides in the study area increases with increasing hazard class.

The ROC curve is simple, intuitive, can accurately reflect the relationship between the specificity and sensitivity of the analysis method used, is more practical for testing and evaluating the degree of fit of the model to the sample, has good experimental accuracy, and is an effective method of verifying the prediction model, and thus is widely used in the accuracy verification of geohazard risk and hazard assessment. The research area’s landslide geological hazards were subjected to accuracy testing using the receiver operating characteristic curve (ROC) (Figure 14). The cumulative percentage of the prone area from high to low is used as the horizontal axis, and the cumulative percentage of the number of disaster points is used as the vertical axis to create the ROC curve. The evaluation accuracy of the susceptibility results is determined by measuring the area beneath the curve. The accuracy increases with the area. The AUC of the area under the ROC curve is a good indicator to test the quality of the geohazard susceptibility zoning map. With an area underneath the research curve (AUC) of 0.8736, the evaluation accuracy is 87.36%. This suggests that the weighted information model for landslide susceptibility evaluation in Tekes County has strong applicability, as does using the information model as the foundation and the entropy approach to calculate weights. The outcomes should serve as a guide for Tekes County’s efforts to prevent and manage landslide disasters.

5.5. Review of Evaluation Methods

The study of the spatial distribution pattern of geological disasters and the evaluation of their susceptibility is a crucial study. With regard to the evaluation of the susceptibility and danger of landslide disasters, the existing landslide danger evaluation is still in the exploratory stage, and there has been extensive and in-depth research both at home and abroad, but there is no unified and complete understanding of the advantages and disadvantages of the various evaluation methods and a complete set of judgement standards has not been formed.

The current assessment of landslide susceptibility is mainly assessed using qualitative and quantitative methods. Qualitative evaluation is fast and simple and can achieve rapid analysis and evaluation of geological disaster susceptibility. Its limitations are mainly subjectivity in determining the weights of the factors involved in the model, while the applicability of the model is also limited by geographical constraints and lacks repeatability, and the accuracy of the physical model is affected by the geotechnical parameters of individual slopes or the study area, which makes it suitable for the assessment of small scopes or individual hazards only. Mathematical modeling for geohazard risk assessment requires accurate multivariate data such as elevation, slope, slope direction, relief, distance from faults, engineering geological rock group, land use type, multi-year average rainfall, and distance from rivers. So, differences in data selection and element hierarchy have a large impact on the output. The single evaluation model has a frequency ratio model, right of evidence model, and informativeness model; among them, the informativeness method obtains the evaluation factor weights by statistically analyzing the conditional probability. The evaluation is more objective, and the calculation results are easy to be understood in the evaluation of the susceptibility to geologic hazards has been widely used, and the evaluation process avoids the conversion of judgement, which is better in line with reality [32].

Through years of application practice, it is known that the above modeling methods have a certain degree of reliability, can accurately predict the occurrence of landslide geological hazards, and can greatly promote the construction of quantitative spatial early warning of landslide hazards, which is widely used in the evaluation of the susceptibility to geological hazards in Tekesi County and in engineering practice.

6. Conclusions

(1)

A relatively perfect geological disaster evaluation index system was established, and the relationship between different geological factors and evaluation factors was analyzed on the basis of geological disaster environment analysis. Taking Tekes County as an example, nine evaluation factors such as elevation were selected, and the information quantity model was adopted to carry out the susceptibility evaluation. The landslide disaster susceptibility level of the county was classified into four levels: high susceptibility, medium susceptibility, low susceptibility and not susceptible. The area of high susceptibility area was 491.3276 km², the area of medium susceptibility area was 1181.5171 km², the area of low susceptibility area was 1674.7609 km², and the area of not susceptible area was 2295.2976 km², which accounted for 5.68%, 13.67%, 19.38%, and 13.67%, respectively, of the total area of Texaco County. The zoning has been well applied in the geological disaster zoning area, the deployment of prevention and control works, and the construction of disaster prevention and mitigation early warning systems in Tekes County.

(2)

In this study, the evaluation results were examined by ROC curve, and the results showed that the AUC of the evaluation results of the geological disaster susceptibility in Tekes County = 0.8736, that is, the evaluation accuracy is 87.36%, which indicates that the use of the information quantity model as the basis, the entropy method of calculating weights, and the final use of the weighted information quantity model for the evaluation of the susceptibility to landslides in Tekes County has good applicability. It can provide a certain reference value for the prevention and mitigation of geological disasters in the district.

(3)

Based on the method of combining the information quantity method and entropy value method with GIS technology, the evaluation of the susceptibility to geological disasters in Tekes County is more in line with the actual investigation, and the modeling methods have a certain degree of reliability and are able to accurately predict the occurrence of landslide geohazards, which greatly promotes the construction of quantitative spatial early warning of landslide hazards and provides scientific bases for the evaluation of the regional riskiness and for the prevention and mitigation of disasters and other work. At the same time, this research method provides a practical, simple, and highly accurate prediction method for evaluating the susceptibility of geological hazards in China and other countries.

(4)

The spatial prediction of landslide geological hazards is a complex nonlinear process, and improving the accuracy of the model is of great significance to the task of landslide prediction. Exploring the distribution characteristics of the evaluation factors helps to understand the mechanism of landslide occurrence and it is necessary to establish a better system of landslide evaluation factors to quantitatively express the importance of the evaluation factors while exploring a more reasonable and reliable landslide prediction model.