Landslide susceptibility assessment in Tongguan District Anhui China using information value and certainty factor models

Abstract

The study of susceptibility to geological hazards geological hazards susceptibility is crucial not only for local risk assessment but also for understanding global patterns in disaster-prone regions. Geological hazards such as landslides and subsidence are a common threat worldwide, affecting millions of people and causing significant economic losses annually. Landslides are a major geological hazard in the Tongling City, Tongguan District, Anhui, China, posing significant risks to local infrastructure and human activity. This study assesses the susceptibility of landslides using seven influencing factors, including elevation, slope, aspect, and distance to faults. Both the information value and certainty factor (CF) models were applied to evaluate the region’s landslide susceptibility, resulting in a classification of the area into five hazard levels. While landslides are the primary focus, ground collapses were also observed, though to a much lesser extent. This study focuses on landslides as the primary geological hazard of interest in Tongguan District. The study found that: (1) landslides in the study area are predominantly concentrated within 300 m of the faults and along the cut slopes adjacent to mountain roads and buildings. This distribution indicates that both the faults and human construction activities are the primary factors contributing to the frequent occurrence of landslides in the region; (2) the proportion of the area classified as high-prone to landslides in Tongguan District, Tongling City is significant, indicating a need for focused mitigation efforts. (3) both the information value and CF models can effectively evaluate the susceptibility to landslides in the region. The area under the curve (AUC) and receiver operating characteristic (ROC) curves were used to evaluate the models’ performance, with the CF model demonstrating superior evaluation accuracy. The study emphasizes areas in Tongguan District susceptible to landslides, offering critical insights for hazard mitigation strategies and decision-making for prevention, treatment, and emergency response in regions with similar conditions.

Introduction

Geological hazards are a common type of natural disaster that easily threaten human life and property and cause damage to the natural ecological environment. The causes of this phenomenon are complex and result from the combined effects of multiple factors, including natural factors and human activities¹. In recent years, geological hazards have become a major concern for communities worldwide, causing widespread destruction and loss of life. Landslides alone account for thousands of fatalities annually, with major events recorded across Asia, Europe, and the Americas. The importance of accurately assessing susceptibility to such hazards is critical for effective disaster management and mitigation strategies. Numerous studies have been conducted globally using various models to predict and assess geological hazards².

Landslides are a significant geologic hazard in the United States, occurring in every state and causing substantial fatalities and economic losses. Estimates suggest that landslides result in 25–50 deaths annually and contribute billions of dollars in economic damages. Notable events include the 2014 Oso landslides in Washington, which caused 43 fatalities, and the 2018 Montecito debris flows in California, which resulted in 23 fatalities^3,4. Additionally, states like Kentucky report annual repair costs for landslides damage ranging from 10 to 20 million dollars. These figures underscore the importance of updated assessments of both direct and indirect losses, particularly in the context of climate change and increasing human encroachment into landslide-prone areas^3–6. Tongguan District, located in the southern part of Tongling City, Anhui Province, is characterized by a diverse topography and complex geology dominated by carbonate rocks. These geological conditions, combined with the area’s subtropical monsoon climate, contribute significantly to its susceptibility to geological hazards, particularly landslides. Frequent and intense rainfall during the summer months often leads to high soil saturation, which substantially increases the likelihood of landslides. Moreover, the impact of human activities, such as extensive urban development and construction on steep slopes, has further destabilized the region, making it prone to severe geological disasters.

For instance, Guzzetti et al. (2006) demonstrated the effectiveness of probabilistic models in Italy⁷, while van Westen et al. (2008) applied similar techniques in the Caribbean, underscoring the universal applicability of these methods⁸. This study builds on this global body of work by applying the information value and certainty factor models to the Tongguan District in China, providing insights that could inform hazard assessments in other regions with complex geological environments.

The main types of geological hazards include landslides, avalanches, mudslides, and ground collapses. Ground collapses, in this study, refer to sudden surface failures typically associated with karst development or human engineering activities, rather than gradual subsidence processes. In 2023, there were 3, 668 geological hazards reported in China, comprising 925 landslides, 2,176 avalanches, 374 mudslides, and 193 ground collapses (Ministry of Natural Resources, Beijing, China, 2024). In recent years, rapid human development has increasingly transformed the natural environment, exerting unprecedented pressure on it. Notable examples include a major mudslide in Zhouqu, Gansu Province in 2010, and a significant landslide in Shuicheng, Guizhou Province in 2022, both of which caused severe casualties and economic losses, seriously impacting human life and economic development⁹.

At present, the evaluation of landslide susceptibility is shifting from traditional qualitative evaluation to quantitative evaluation or a combination of qualitative and quantitative evaluation¹⁰. With the rapid development of remote sensing (RS) and geographic information system (GIS) technology, the data collection and data processing capacity in the field of landslides research has been enhanced. Very high-resolution remote sensing (VHRRS) images, digital elevation models (DEMs), and convenient data processing methods provide powerful conditions for the study of landslide susceptibility evaluation. The methods include the analytic hierarchy process (AHP)^11–14. Quantitative evaluation methods mainly include logistic regression analysis^15–18, neural network (NN)^19–21, support vector machine (SVM)^22–24, random forest (RF)^25–27, information value mode^28–31, and certainty factor (CF) model^32–35, etc.

Among them, the qualitative method has the advantage of simple calculation, but it is affected by subjective factors. The information value and CF models in statistical modeling determine the relationship between landslides and individual influencing factors based on the priori knowledge. These methods can eliminate the problems such as the quantification of the same interval between complex factors and the monitoring errors. However, it is difficult to determine the relationship between each factor and the geological hazards sites in the high-dimensional space. The method represented by SVM in machine learning is widely used in text classification, remote sensing image processing, etc. It can achieve high classification accuracy in high-dimensional data spaces with fewer training samples. However, practical applications often encounter issues such as non-uniform complexity factor magnitudes, which may hinder the application of SVM in landslide susceptibility evaluation³⁶. While machine learning models like SVM and RF have been widely applied in landslide studies, their reliance on large datasets and computational resources limits their applicability in regions with sparse data. In contrast, the Information Value and CF models are more suitable for areas like Tongguan District, where interpretability and computational efficiency are prioritized.

This paper conducts a statistical analysis of landslides in Tongguan District, Tongling City, Anhui Province, examining their spatial distribution and occurrence patterns. Based on these findings, suitable influencing factors for assessing the susceptibility of landslides in the study area are identified. Subsequently, hierarchical maps are generated according to each evaluation factor. Based on the results from the information value and CF models, it is highly practical to delineate landslides prevention and control zones in Tongguan District. This will provide relevant departments with essential reference sites for landslides prevention and mitigation, as well as for future production and construction activities in the area. It is important to note that while the primary focus of this study is on landslides, ground collapses were observed in limited instances during the investigation.

Materials and methods

Study area

The study area is situated in the Tongguan District, Tongling City, in the Southern region of Anhui Province, China (Fig. 1A–B). Tongguan District is located in the Southwestern part of Tongling City, surrounded by Yi’an District to the East, North and Southeast, and across the river from the suburbs to the west. As of the end of 2021, Tongguan District had a resident population of 452,000, and the urbanization rate of the resident population was 97.5% (Tongling Statistical Bureau, Tongling, China, 2022). Tongguan District has a complex and varied topography, mainly consisting of plains, with mountains and hills in the southern part of the district. The overall terrain is high in the southeast and low in the northwest of the hilly terrain. The study area has developed carbonate rocks and geological hazards are widely distributed³⁷.

Fig. 1.

Schematic of the location of the study. (A) Tongling City in Anhui Province; (B) Tongguan District in Tongling City and (C) historical geological hazard sites (landslides, collapses) and faults distribution in Tongguan Distric.

Tongguan District experiences a subtropical humid monsoon climate, with an annual average temperature of approximately 18.2 °C. The coldest month is January (average temperature ~ 5.6 °C), and the hottest month is July (average temperature ~ 30.0 °C). Annual average precipitation is about 1122.5 mm, with the majority occurring in spring (March to May) and summer (June to August), contributing 30.74% and 39.69% of the total yearly precipitation, respectively. July is the wettest month, with an average rainfall of 462 mm. The seasonal precipitation pattern significantly impacts landslide occurrences. During the summer, high temperatures and heavy rainfall increase soil saturation, reducing shear strength and making slopes more susceptible to failure³⁸.

In addition, human engineering and construction activities on the cut slopes of the mountain have led to the destabilization of the mountain³⁹. Yet there is a lack of effective support and protection measures. Due to multiple factors, geological hazards occur frequently during the rainy season, destroying the natural environment and bringing danger to human production activities and even life safety. There are 36 historical geological hazards sites in the area, which are concentrated in the hilly area in the south of Tongguan District, and a very small number of them are distributed in the plains (Fig. 1C).

Data sources

Through the field investigation and historical data, geologic hazards of landslides and collapse exist in the study area³⁹. Of these 36 historical geohazard sites, these are mainly shallow landslides in gravelly soils, and the formation mechanism is mainly traction type, closely related to anthropogenic slope-cutting activities. Beyond that, a small number of collapse sites also exist in the study area. The main data sources in this paper are: (1) 5 m resolution DEM data for extracting the topography and geomorphology of Tongguan District; (2) geological map of Anhui Province for extracting the distance to faults and the geological rock group of the strata; (3) vector data of the administrative boundaries of Tongguan; (4) data of the historical geological hazard sites in Tongguan; (5) the first 1-m resolution national-scale land-cover map of China created with the deep learning framework and open-access data (SinoLC-1, http://www.ncdc.ac.cn)⁴⁰ for extracting land cover of Tongguan District. Influencing factors with varying spatial resolutions were resampled to 10 m to acquire the necessary evaluation factors for assessing landslide susceptibility in Tongguan District (Table 1).

Table 1.

Classification of landslide susceptibility influencing factors in Tongguan District.

Influencing factor	Type	Source
Elevation	Raster data (5 m)	Calculation and extraction of DEM data based on ArcGIS 10.2 software
Slope	Raster data (5 m)
Aspect	Raster data (5 m)
Profile curvature	Raster data (5 m)
Geological rock group	Vector data	Geological map of Anhui Province
Distance to fault	Raster data (5 m)	Extracted using ArcGIS 10.2 software
Land cover type	Raster data (1 m)	SinoLC-1 data set

In assessing the susceptibility to landslides within Tongguan District, our study utilizes seven key influencing factors: elevation, slope, aspect, profile curvature, geo-logical rock group, distance to fault, and land cover type. Each factor was chosen based on its known impact on landslide occurrence and its relevance to the regional characteristics of Tongguan District. Elevation and slope contribute directly to the gravitational dynamics and potential energy conducive to landslide initiation. Aspect and profile curvature are critical in determining hydrological behaviors such as moisture retention and runoff, which are fundamental to landslide processes. The geological rock group delineates the inherent mechanical properties of subsurface materials, impacting their stability or susceptibility to failure. Proximity to fault lines is a crucial indicator of potential geological disturbances that could trigger landslides. Finally, land cover types are significant as they affect both hydrological properties and the physical stability of surface materials, influencing landslide susceptibility through modifications in water drainage and soil cohesion⁴¹.

In Tongguan District, human construction activities and the proximity to active faults significantly contribute to landslide occurrences. Construction activities, such as road building, excavation, and urban development, often alter the natural landscape and drainage patterns, creating instability in slopes and increasing erosion rates. These activities typically remove vegetation that stabilizes the soil, exacerbate the loosening of soil particles, and lead to increased landslide risks, particularly on steeper slopes. Furthermore, the district’s proximity to fault lines means that seismic activities can trigger landslides by shaking loose already unstable slopes and saturating soils with water from disrupted groundwater paths. This dual influence of human activity and geological vulnerability necessitates a comprehensive approach to landslide risk assessment and mitigation in the area.

Research methods

Process for evaluating susceptibility to landslide

In this study, the geological hazard sites in Tongguan District were modeled separately by the information value and CF models based on the seven selected influencing factors (Fig. 2). Finally, the landslide susceptibility mapping was obtained, and the susceptibility results were graded and evaluated. The area, proportion and landslide density within each classification were obtained, which facilitates the comprehensive evaluation of the applicability of the two models for landslide susceptibility.

1. Preparation of data sources for the study area, including historical hazard sites data, DEM data, administrative boundary vector data, geologic map data, and land cover data for Tongguan District.

2. Based on ArcGIS software, each influencing factor was extracted, and the information value and CF value were calculated and modeled for the geological hazard sites respectively. As a result, the magnitude of the information value and CF value within each pixel were obtained.

3. The raster calculation was utilized to accumulate the information value of each pixel with the CF value to obtain the landslide susceptibility zonation map.

Fig. 2.

Framework for data source, influencing factors and accuracy evaluation of the study.

Selection and evaluation of models for landslide susceptibility assessment

In this study, the geological hazard sites in Tongguan District were modeled separately by the information value and CF models based on the seven selected influencing factors (In this study, we selected the information value method and the certainty factor model (CF) as the primary methods for landslide susceptibility zonation (LSZ). The reasons for selecting these two methods are as follows:

1. Theoretical basis and wide application

The information value method was proposed by Shannon, the founder of information theory, and has proven effective in the prediction and evaluation of geological hazards⁴². The CF model was first proposed by Shortliffe and Buchanan⁴³, and then improved by Heckerman⁴⁴. Both methods, due to their solid theoretical foundation and clear calculation process, have been widely used in landslide susceptibility evaluations.

• (2)Applicability of methods

In the study area, the occurrence of landslides is influenced by multiple factors. The information value method effectively integrates the influence of these factors by calculating the event probability for each influencing factor. The CF model quantitatively assesses the impact of each factor using probability functions, providing a relatively clear risk assessment. The combination of these two methods can effectively improve the accuracy and reliability of landslide susceptibility zonation.

The calculation processes of these two methods are transparent and highly operable, making them suitable for modeling and calculation in the ArcGIS software environment. Using these methods, we can obtain the information value and CF value for each pixel, and through raster calculation, ultimately obtain the landslide susceptibility zonation map. These processes are clear and straightforward, facilitating replication and verification.

By calculating the area, proportion, and landslide density sites within the study area, we can comprehensively evaluate the applicability of the information value method and the CF model. The results indicate that both methods demonstrate high accuracy and reliability in the current study area, making them effective for landslide susceptibility evaluation.

Through the above analysis, we believe that the information value method and the CF model have high applicability and effectiveness in the study area, meeting the requirements for landslide susceptibility zonation.

Furthermore, while other advanced models like Support Vector Machines (SVM) and Random Forests (RF) are popular in landslide studies due to their ability to handle complex, non-linear data, these models have several limitations in the context of this study: (1) SVM and RF often require larger datasets and extensive feature engineering to achieve optimal results. In Tongguan District, historical landslide data is limited, making these models less practical. (2) SVM and RF demand significant computational resources, which can pose challenges in regions with limited technical infrastructure. In contrast, the Information Value and CF models are computationally efficient and can be implemented in standard GIS software like ArcGIS. (3) While machine learning models often act as "black boxes," the Information Value and CF models provide clear and interpretable relationships between influencing factors and landslide susceptibility. This transparency is crucial for local stakeholders and decision-makers in hazard mitigation.

Information value method (I)

The formation of landslides is not caused by a single factor, but often by the joint influence of multiple factors. That is to say, the information value can be used to determine whether a landslide occurs or not⁴⁵. The information value is calculated by the event probability, and the main formulas are as follows:

1

where represents the probability of factor A realizing landslide B in state i. In the actual calculation process, the overall probability is converted to sample frequency for estimation to facilitate the calculation, which can be converted from formula (1) to formula (2):

2

where denotes the information value of factor A leading to landslide B in state i, N_i is the number of geologic hazard sites distributed by factor A in state i, N is the total number of geologic hazard sites that have occurred, S_i is the area of factor A in state i, and S is the total area of the region. Under the condition of factor combination, the total information value can be determined by formula (3), and the larger the value of I, the higher the possibility of landslide.

3

Certainty factor model (CF)

The CF model is a quantitative assessment of the factors influencing landslide susceptibility, utilizing probability functions for evaluation. The calculation formula (4) is as follows:

4

where PP_b is the conditional probability of occurrence of landslides events in b, expressed as the ratio of the number of geological hazard sites to their area in the factor classification level b; PP_S is the a priori probability of occurrence of landslides events in the whole area S, expressed as the total number of landslides in the whole area and its total area. CF is usually taken in the range of − 1 to 1. CF > 0 indicates that the certainty of the occurrence of landslides is high, and the negative value indicates low certainty. When the value of CF is close to 0, it means that the conditional probability is very close to the a priori probability, and it is impossible to judge whether the evaluation unit is a landslides prone area or not.

Evaluation of landslide susceptibility

Selection and correlation analysis of influencing factors

Influencing factors of landslides

Landslides are caused by a variety of factors such as natural and man-made factors, and the importance of each influencing factor to the occurrence of landslides is different. Therefore, when evaluating the degree of susceptibility to landslide, the reasonable selection of evaluation indexes (influencing factors) is an important prerequisite for evaluating the landslide susceptible areas. Among the influencing factors selected in this research, there are both qualitatively described levels and quantitatively described indicators:

1. Elevation: Elevation can have a significant impact on slopes. As the elevation increases, the potential energy of geologic hazards such as landslides increases, which has a greater impact on the susceptibility to landslides. In addition, there are differences in human activities, vegetation cover, and temperature at different elevation ranges⁴⁶.

2. Slope: The scale of slope is closely related to the degree of susceptibility to landslide⁴⁷. In addition to being able to represent the undulation pattern of the ground surface, slope is also one of the important factors affecting the occurrence of landslide, which directly influences the stability of slopes.

3. Aspect: The influence of aspect on landslides lies mainly in the degree of solar radiation received by the slopes. Slopes of different aspects receive different intensities of solar radiation, which, under equal conditions, can lead to differences in the internal water content of the rock mass⁴⁸. Thus, aspect influences the possibilities of landslides.

4. Profile curvature: Profile curvature measures the rate of change in ground elevation along the direction of the steepest slope⁴⁹. It indirectly represents slope morphology, which reflects the stress distribution and physical properties of rock and soil masses. This, in turn, affects the occurrence and development of landslides.

5. Geological rock group: The geological rock group is the material basis for the generation of landslides, reflecting the physical and chemical properties of the rock mass. The type and hardness of the rock, along with interlayer structures, determine the physical and mechanical strength, weathering resistance, stress distribution, and deformation or failure characteristics of rock and soil masses. These factors, in turn, affect slope stability and the difficulty or ease of surface erosion. Therefore, the geological rock group is one of the critical influencing factors and inherent conditions for the occurrence of landslides⁵⁰.

6. Distance to fault: Faults affect the stability of slopes, as tectonic activity near faults is typically more intense⁵¹. Consequently, there is a close relationship between landslides and faults. Generally, the closer the distance to a fault, the more fractured the rock is, and the greater the possibilities of landslides.

7. Land cover type: The relationship between land cover type and landslides is mainly reflected in the impact of different land cover methods on the geological environment⁵². Different land cover types directly affect the surface and underground hydrological conditions, soil stability and geological structure. Also, some land cover types represent the impact of human construction activities on rock and soil masses. Therefore, land cover type is an important indicator for evaluating the susceptibility to landslides⁵³.

In summary, this paper combines the characteristics of Tongguan District’s topography, geological structures, and land cover types, selecting seven influencing factors as indicators for evaluating the susceptibility to landslides. The classification standards for each influencing factor are shown in Table 2.

Table 2.

Classification of landslide susceptibility influencing factors in Tongguan District.

Influencing Factor	Classification
Elevation (m)	< 25, 25 ~ 75, 75 ~ 150, 150 ~ 250, > 250
Slope (°)	< 5, 5 ~ 10, 10 ~ 20, 20 ~ 30, > 30
Aspect (°)	− 1 (Flat), 0 ~ 22.5 (N), 22.5 ~ 67.5 (NE), 67.5 ~ 112.5 (E), 112.5 ~ 157.5 (SE), 157.5 ~ 202.5 (S), 202.5 ~ 247.5 (SW), 247.5 ~ 292.5 (W), 292.5 ~ 337.5 (NE), 337.5 ~ 360 (N)
Profile curvature	< − 0.5, − 0.5 ~ 0.5, > 0.5
Geological rock group	Slightly dense to medium dense gravelly soil, dense sand and gravel rock group (Q2 + Q1), Medium dense silt, highly compressible clayey soil dominated rock group (Q4), Hard massive diorite & granodiorite rock group (RM1), Hard thick-bedded sandstone & fine sandstone rock group (RM2-1), Moderately hard to soft siltstone & mudstone rock group (RM2-2), Moderately hard to soft thick-bedded conglomerate & gravel rock group (RM3), Moderately hard to soft siliceous shale, sandy shale rock group (RM4) Hard thick-bedded strong to moderately karstified limestone rock group (RM5) Hard thick-bedded weakly karstified limestone, dolomite rock group (RM6)
Distance to fault (m)	0 ~ 50, 50 ~ 100, 100 ~ 300, 300 ~ 500, > 500
Land cover type	Road, Tree cover, Grassland, Cropland, Building, Barren and sparse veg., Water, Wetland

Among them, the geological rock group and land cover type are qualitative descriptive factors with clearer classification criteria. Other influencing factors are continuous numerical influencing factors, and the grading standards are subjective except for the fixed grading of aspects. The reclassification mapping of each influencing factor using ArcGIS 10.2 software is shown in Fig. 3.

Fig. 3.

Distribution of landslide evaluation factors. (A) elevation; (B) slope; (C) aspect; (D) profile curvature; (E) geological rock group; (F) distance to fault and (G) land cover type.

Influencing factors correlation analysis

The multitude of factors influencing landslides may exhibit interrelationships. This paper utilizes the band statistics tool in ArcGIS 10.2 software to conduct band value statistics on the classified influencing factors. The Pearson correlation coefficient (r) is calculated to assess the correlation between each influencing factor. If |r|> 0.8, it indicates a strong correlation between factors; 0.5 <|r|≤ 0.8 suggests a moderate correlation, 0.3 <|r|≤ 0.5 indicates a weak correlation, and |r|≤ 0.3 indicates no correlation.

The calculation results of the correlation coefficients for the landslide susceptibility factors are shown in Table 3. It can be observed that the correlation coefficient between elevation and slope is relatively high, while other factors show no correlation. However, since elevation and slope are essential indicators for conditions that trigger landslides, they are not excluded from the analysis.

Table 3.

Correlation coefficient of factors influencing the susceptibility to landslide.

Influencing factor	Elevation	Slope	Aspect	Profile curvature	Geological rock group	Distance to fault	Land cover type
Elevation	1	0.736	0.087	0.082	− 0.114	− 0.539	− 0.335
Slope	0.736	1	0.126	0.021	− 0.126	− 0.446	− 0.340
Aspect	0.087	0.126	1	0.001	− 0.027	− 0.315	− 0.065
Profile curvature	0.082	0.021	0.001	1	− 0.001	0.004	− 0.030
Geological rock group	− 0.114	− 0.126	− 0.027	− 0.001	1	− 0.148	0.068
Distance to fault	− 0.539	− 0.446	− 0.315	0.004	− 0.148	1	− 0.204
Land cover type	− 0.335	− 0.340	− 0.065	− 0.030	0.068	− 0.204	1

Calculation results of information value and CF value

To facilitate the area under the grading conditions of each influence factor of information value and CF models, the vector data used in the statistics of the occurrence of geological hazard sites was converted into raster data with 10 m × 10 m resolution. The area and the number of geological hazard sites under different grading conditions of the seven influencing factors of landslides were counted separately. And then the information value and CF value of each influencing factor under different grading conditions were calculated separately. The calculation results were shown in Table 4.

Table 4.

Calculation results of information value and CF value.

Influencing factor	Classification	I	CF
Elevation (m)	< 25	− 1.46	− 0.65
	25 ~ 75	2.30	0.90
	75 ~ 150	1.22	0.58
	150 ~ 250	0.00	− 1.00
	> 250	0.00	− 1.00
Slope (°)	< 5	− 1.17	− 0.56
	5 ~ 10	1.30	0.46
	10 ~ 20	2.25	0.90
	20 ~ 30	2.07	0.93
	> 30	2.46	0.87
Aspect	Flat	− 1.88	− 0.81
	N	− 0.69	− 0.29
	NE	− 0.49	− 0.09
	E	− 0.76	− 0.35
	SE	− 2.84	− 0.27
	S	− 0.41	0.02
	WE	− 0.55	− 0.15
	W	− 1.43	0.53
	NW	− 1.22	0.55
	N	− 1.68	0.32
Profile curvature	< − 0.5	1.05	0.69
	− 0.5 ~ 0.5	− 1.34	− 0.54
	> 0.5	0.59	0.41
Geological rock group	Q1 + Q2	− 3.06	0.22
	Q2	0.00	− 1.00
	Q4	− 1.17	− 0.32
	RM1	− 1.64	0.81
	RM2-1	− 2.16	0.51
	RM2-2	0	− 1.00
	RM3	− 2.7	− 0.87
	RM4	− 5.47	− 0.70
	RM5	0.00	− 1.00
	RM6	0.00	− 1.00
Distance to fault (m)	< 50	2.84	0.94
	50 ~ 100	2.96	0.97
	100 ~ 300	1.63	0.63
	300 ~ 500	1.36	0.54
	> 500	− 1.51	− 0.48
Land cover type	Road	2.51	0.99
	Tree cover	0.36	0.50
	Grassland	0.00	− 1.00
	Cropland	− 1.59	− 0.57
	Building	0.90	0.45
	Barren and sparse veg	1.02	0.01
	Water	0.00	− 1.00
	Wetland	0.00	− 1.00

The area and the proportion of geological hazard sites under each grading condition of each influencing factor were calculated separately (Fig. 4). It can be seen that Tongguan District has the largest percentage of geological hazard sites and values for both models calculated for elevation of 25–75 m, slope > 10°, profile curvature < -0.5, 50–100 m from geologic faults, and road areas. In addition, the information value and the CF models reflect very different results in some of the grading conditions for the two influencing factors of aspect and geological rock group. The information values calculated under both influencing factors are non-positive. The CF model, on the other hand, shows positive values in areas where the aspects are south, west, northwest, and north and where the geological rock groups are Q1 + Q2, RM1, and RM2-1. This phenomenon coincides with the statistics of the percentage of geological hazard sites in Fig. 3.

Fig. 4.

Area of classification of each influencing factor and the percentage of the geological hazard sites. (A) elevation; (B) slope; (C) aspect; (D) profile curvature; (E) geological rock group; (F) distance to fault and (G) land cover type.

The total information value and CF value of each factor were calculated in ArcGIS 10.2, and the natural breaks method was used to reclassify the total information value and CF value of geological hazard in Tongguan District. The data were resampled to 10 m. The whole study area was divided into lowest susceptibility zone, lowest susceptibility zone, medium susceptibility zone, high susceptibility zone and highest susceptibility zone, and the landslide susceptibility zoning map of Tongguan District was drawn (Fig. 5). These classifications were determined using the Natural Breaks (Jenks) method, which identifies breaks in the data distribution to maximize the variance between classes. This method is particularly effective for our data as it highlights natural groupings inherent in the dataset, thus providing a more accurate reflection of varying landslide susceptibility across Tongguan District. The application of this method allows for a clear differentiation between areas of differing stability, facilitating targeted mitigation strategies.

Fig. 5.

Landslide susceptibility zonation of each model. (A) Information value model and (B) CF model.

The landslides areas are mainly distributed in the south and southeast of Tongguan, accounting for about 10% of the area, which is highly consistent with the distribution of historical geological hazard sites in Tongguan District. However, it is worth noting that there are no geological hazard sites on the faults in the east-central part of Tongguan District, and that this part of the district is highly prone to landslides, so it is necessary for the relevant departments to take precautions to prevent landslides from occurring in this area in the future.

Combined with Table 5, it can be seen that the landslide susceptibility zone of Tongguan District divided according to the results of both models is consistent with the distribution of historical geologic hazards. The sum of the proportions of the number of landslides in the high susceptibility and highest susceptibility zones for the information value and the CF models were 75.0% and 77.8%, respectively. It shows that both models can accurately reflect the zones where landslides occur, while the CF model is more effective.

Table 5.

Landslide susceptibility zonation results and ratios.

Model	Landslide susceptibility zonation	Area (km2)	Number of disasters	Disasters proportion (%)	Landslide density (pcs/km2)
I	Lowest susceptibility area	43.47	1	2.78	0.0230
	Low susceptibility area	39.10	2	5.56	0.0511
	Moderate susceptibility area	26.74	6	16.67	0.2244
	High susceptibility area	22.46	13	36.11	0.5788
	Highest susceptibility area	12.84	14	38.89	1.0905
CF	Lowest susceptibility area	32.61	1	2.78	0.0307
	Low susceptibility area	44.88	3	8.33	0.0668
	Moderate susceptibility area	30.38	4	11.11	0.1317
	High susceptibility area	20.32	12	33.33	0.5906
	Highest susceptibility area	16.41	16	44.44	0.9751

Landslide density refers to the density of landslide occurrences within a given area, expressed as the number of landslide events per square kilometer (pcs/km²). This metric is calculated as:

5

Accuracy evaluation of information value and CF models

In this study, the receiver operating characteristic (ROC) curve is used to evaluate the accuracy of the information value and CF models as shown in Fig. 6. The area under the curve (AUC) is used as the model evaluation index. The ROC curve can indicate the relationship between the data of geologic hazard sites fitted by the two models and the actual data of geologic hazard sites⁵⁴. The vertical axis of the curve is the true positive category rate, i.e., the percentage accumulation of the actual number of geological hazard sites; the horizontal axis is the false positive category rate, i.e., the percentage accumulation of the susceptibility zone. The AUC is a measure of how good the model is at classification. The closer the AUC value is to 1, the higher the model prediction accuracy⁵⁵. The information value and CF models were tested to obtain the ROC curve of landslide susceptibility (Fig. 6). The AUC values of the information value and CF models were 0.879 and 0.921, respectively. Both models were able to evaluate the landslide susceptibility, and the CF model had a higher evaluation accuracy.

Fig. 6.

ROC curve of each model.

Results and discussion

Influencing factor analysis

The analysis of influencing factors (Table 4) shows that slope, distance to faults, and land cover types contribute significantly to landslide susceptibility in Tongguan District. For instance, slopes greater than 20° have the highest information value (2.46) and CF value (0.93), indicating a strong correlation with landslide occurrences. Negative values, such as for areas with elevations above 250 m, suggest reduced landslide susceptibility.

These findings align with the geological characteristics of Tongguan District, where active faults and steep slopes dominate. Faults create fracture zones that reduce rock mass stability, while steep slopes increase gravitational stress, making them prone to failures.

Profile curvature, which measures the rate of elevation change along the steepest slope, is a critical factor influencing landslide susceptibility. Negative profile curvature (concave surfaces) tends to accumulate surface runoff, increasing soil moisture and pore water pressure, which reduces the shear strength of the soil and rock masses. Conversely, positive profile curvature (convex surfaces) facilitates water dispersion, thereby lowering the likelihood of slope failure.

In Tongguan District, landslide occurrences are primarily concentrated in areas with negative profile curvature (< − 0.5), as shown in Table 4. These concave slopes enhance water accumulation, particularly during heavy rainfall events, making them more prone to instability. Similar findings have been reported in other landslide studies, where profile curvature significantly correlates with landslide susceptibility^7,21.

Landslide susceptibility zonation

The landslide susceptibility maps (Fig. 5) show that high and highest susceptibility zones are primarily distributed along faults and in the southeastern part of Tongguan District. These zones account for 38.89% and 44.44% of the total landslide occurrences for the IV and CF models, respectively.

The strong agreement between the zonation results and historical landslide occurrences demonstrates the reliability of the models. High-risk areas near faults and road-cut slopes highlight the need for targeted mitigation measures, particularly in regions with active construction projects.

Model performance comparison

The ROC curve analysis (Fig. 6) indicates that the CF model has a higher AUC value (0.921) compared to the IV model (0.879), suggesting better predictive accuracy. Both models, however, are effective in delineating high-risk zones, as evidenced by their strong correlation with historical landslide data.

The superior performance of the CF model can be attributed to its ability to incorporate conditional probabilities, providing a more nuanced assessment of landslide susceptibility. Previous studies using machine learning models, such as SVM and RF, have also achieved high AUC values, but their reliance on large datasets limits their applicability in data-scarce regions like Tongguan District.

Global applications and methodological relevance

The findings from this study, centered on the Tongguan District, offer valuable insights not only for local hazard mitigation but also for global applications. The use of the information value and CF models demonstrates robust methodologies that could be adapted to other regions with similar geological settings, such as areas characterized by carbonate rock formations and active fault zones. These models have proven effective in assessing landslide susceptibility, particularly in complex terrains where human activities exacerbate natural vulnerabilities.

Internationally, regions prone to landslides—such as the Pacific Northwest in the United States, the Himalayan range, and various parts of Europe—could benefit from the application of these models. The methodologies employed in this study can be integrated into broader global efforts to enhance landslide risk assessments. The similarities in geological conditions across these regions suggest that the information value and CF models could be effectively utilized to predict susceptibility, thus contributing to disaster risk reduction on a global scale.

Furthermore, this study highlights the critical need for a more nuanced understanding of the interplay between natural and anthropogenic factors in landslide-prone areas. As global climate change intensifies, regions with high precipitation and steep terrains are expected to face increasing landslide risks. Therefore, incorporating climate data and land-use changes into these models could improve their predictive power and applicability across diverse environments.

Comparative analysis of landslide susceptibility studies

In China, numerous studies have focused on landslide susceptibility using various methods and conditioning factors. For instance, Yu et al. (2023) utilized logistic regression and machine learning techniques to map landslide susceptibility in the Sichuan Province, emphasizing the critical role of lithology and slope gradient⁵⁶. Similarly, Yang et al. (2022) highlighted the importance of vegetation cover and soil properties in landslide occurrences in the Yunnan Province²². Our study aligns with these findings by identifying similar conditioning factors but differs in the methodological approach by employing the information value and CF models. This methodological distinction allows for a comparative analysis of model performance in different geological settings.

Internationally, landslide susceptibility mapping has been extensively studied using various statistical and machine learning models. Studies such as those by Guzzetti et al. (2006) in Italy⁷ and van Westen et al. (2008) in the Caribbean highlight the effectiveness of probabilistic methods like the information value and CF models⁸. Our findings corroborate the effectiveness of these models in predicting landslide susceptibility, as demonstrated in other global studies. Moreover, we integrate the impact of hydrological and seismic factors, which have been less emphasized in some international studies but are crucial in our study area due to its specific climatic and tectonic conditions.

Analysis of landslides distribution

The distribution of landslides in the east-central and southwestern parts of the area can be attributed to several key factors:

Local geological conditions

The east-central and southwestern regions have complex geological conditions. The frequent landslides such as landslides and collapses are mainly related to these regions’ geological features. These areas are predominantly characterized by the development of carbonate rocks, which are highly susceptible to erosion, leading to the formation of underground caves and surface subsidence. Additionally, the dense distribution and activity of faults in these regions increase the risk of landslides. Studies have shown that landslides in these areas are primarily concentrated within 300 m of fault lines, with fault-induced geological fracture zones being significant triggers for landslides and collapses. Field investigations confirm that the faults within the study area do not exhibit signs of recent activity and are currently classified as stable. Historical records and seismic data support the conclusion. Although the faults in Tongguan District are currently inactive, their presence creates zones of fractured and weakened rock, which can impact slope stability. These zones may be more susceptible to landslides during heavy rainfall or human-induced activities, despite the lack of tectonic movement⁵⁷.

Impact of human activities

Human engineering and construction activities play a crucial role in the formation of landslides. The economic development and urbanization processes in the east-central and southwestern regions have led to extensive civil engineering and infrastructure construction. These activities, including road cutting and mining, have disrupted the natural stability of slopes, resulting in frequent landslides and collapses. Moreover, the concentrated distribution of buildings and roads has increased the susceptibility to landslides in these areas.

Climatic factors

The east-central and southwestern regions are located in a subtropical monsoon climate zone, characterized by hot and rainy summers, with an average annual precipitation of 1122.5 mm and a maximum monthly precipitation of 224.5 mm. Heavy rainfall saturates the soil, increasing the weight and sliding force of slopes, thereby triggering landslides and collapses. Additionally, intense rainfall exacerbates fault activity and erosion, further raising the likelihood of landslides.

Limitations and future directions

The study indicates that landslides in Tongguan District are primarily influenced by factors such as elevation, slope, aspect, profile curvature, geological rock group, distance to fault and land cover type. Both the information value and the CF models demonstrated the capability to effectively assess the susceptibility to landslides, with the CF model showing slightly higher accuracy compared to the information value model. This suggests that the CF model may provide more precise predictions regarding landslide susceptibility than information value model.

The results of the study have made a significant contribution to the avoidance and emergency response work of the relevant departments, but there are still some limitations. For example, (1) although seven influencing factors were selected in this study, other factors that may affect landslides such as rainfall and changes in vegetation cover were not taken into account⁵⁸; (2) among the multiple influencing factors, the degree of influence of each factor on the susceptibility to landslides is different, which may generate errors when grading the susceptibility to landslides. In the future, consideration should be given to introducing more influencing factors to improve the accuracy of the model, and a certain weight should be considered to be given to each factor for measuring its contribution to landslides. In addition, longitudinal studies should be conducted to collect information on geological hazard sites in different years, track the performance of the two models in different periods, and strengthen the field validation of the modes’ prediction results, to ensure the accuracy in evaluating the susceptibility to landslides.

Local strategies and global applications

Based on our findings, specific mitigation strategies are recommended for high-prone landslide areas in Tongguan District. These include but are not limited to:

1. Implementing vegetation and soil conservation practices to stabilize slopes.

2. Enhancing early warning systems and monitoring mechanisms for rapid response to landslide triggers.

3. Restricting or regulating construction activities on steep slopes and near fault lines. Promoting community awareness and preparedness through education and training programs.

4. Collaborating with local authorities and stakeholders to enforce land use regulations and zoning laws that minimize landslide risk.

5. These strategies aim to mitigate the impact of landslides and enhance the resilience of communities in Tongguan District

The findings of this study offer valuable insights that can be applied to other regions with similar geological conditions, such as areas characterized by steep slopes, carbonate rock formations, and active fault zones. The methodologies employed, including the information value and certainty factor models, have demonstrated their robustness and adaptability in evaluating landslide susceptibility. These models can be utilized in regions with analogous conditions to identify high-risk zones, prioritize mitigation efforts, and enhance disaster preparedness. Moreover, the integration of remote sensing and GIS techniques provides a scalable framework for hazard assessment, making it practical for large-scale applications. By adopting these approaches, local governments and disaster management agencies in other regions can improve their strategies for disaster prevention, risk reduction, and emergency response.

Conclusions

1. Based on the data of historical geological hazard sites in Tongguan District, the information value model and CF model were used to assess the susceptibility to landslides, and the following conclusions were drawn:

   1. Landslides occur mainly in zones with elevation of 25–75 m, slope of 20°–30°, northwestern orientation, curvature of less than -0.5, and distance of 300 m from faults.
   2. In the zones of highest susceptibility classified by the two models, the disasters proportion of the total number of landslides sites was 38.9% and 44.4%, respectively; the landslide density is 1.1 km⁻² and 1.0 km⁻².
   3. The AUC values of the information value and CF model under the ROC curve are 0.879 and 0.921, respectively, and the CF model improves the identification efficiency compared with the traditional investigation method, which provides a decision-making basis for the prevention of landslide risks in Tongguan District.

2. This research underscores the efficacy of the information value and CF models in evaluating landslides in the Tongguan District, providing a framework that could be replicated in other regions globally. The ability of these models to accurately delineate areas of highest susceptibility demonstrates their potential utility in international contexts, particularly in regions with analogous geological conditions.

3. Despite the effectiveness of the Information Value (I) and Certainty Factor (CF) models in identifying high-susceptibility areas, several limitations must be acknowledged:

   1. The accuracy of the models relies heavily on the quality and completeness of input data. In Tongguan District, the limited availability of historical landslide data and the absence of high-resolution environmental variables, such as rainfall intensity and soil properties, may have impacted the model’s performance.
   2. The models are constrained by the selected influencing factors. While the chosen factors are highly relevant to the study area, the exclusion of other potential variables, such as vegetation dynamics and anthropogenic activities, may have introduced bias into the susceptibility assessment.
   3. The IV and CF models primarily handle linear relationships between influencing factors and landslide occurrences. In areas with complex, non-linear interactions, advanced machine learning models, such as Random Forests or Support Vector Machines, may offer improved performance.

4. Future research should focus on the application of these models in different geological settings to validate their robustness and adaptability. Expanding the model inputs to include global climate data, land-use patterns, and socioeconomic factors could further enhance the accuracy of susceptibility assessments. Additionally, a comparative analysis with other predictive models across various regions could establish a more comprehensive understanding of landslide risks, contributing to more effective global disaster management strategies.

Author contributions

Conceptualization, D.D., Y.W. and C.G.; methodology, D.D. and C.G.; software, Y.W. and T.W.; validation, D.D., Y.W. and T.W.; formal analysis, D.D. and Y.W.; investigation, D.D.; data cura-tion, D.D.; writing—original draft preparation, D.D., Y.W. and T.W.; writing—review and editing, D.D., Y.W., T.W. and C.G.; visualization, Y.W. and T.W.; funding acquisition, C.G.; supervision, C.G. All authors have read and agreed to the published version of the manuscript.

Funding

This study was financially supported by the National Natural Science Foundation of China (Grant No. 52204182).

Data availability

The data presented in this study are available upon request from the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.