Susceptibility Assessment of Rockfall in Karst Regions based on Information Entropy and Multi-Model Coupling

doi:10.21203/rs.3.rs-3753153/v1

Download PDF

Research Article

Susceptibility Assessment of Rockfall in Karst Regions based on Information Entropy and Multi-Model Coupling

https://doi.org/10.21203/rs.3.rs-3753153/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Rockfall is one of the primary geological hazards in karst regions. In order to study the susceptibility distribution patterns of rockfall disasters in karst areas, the research areain Xincheng County is selected in this study and data are collected at 172 historical rockfall points under different geological environments. Various factors, including aspect, slope, elevation, terrain relief, plan curvature, profile curvature, landform type, roughness, coefficient of variation, lithology, fault distance, rainfall, distance to rivers, NDVI (Normalized Difference Vegetation Index), and distance to roads, are employed to construct four coupling models, e.g. IV-RF, IV-CHAID, IV-MLP and IV-SVM. Through comparative analysis of the accuracy and reliability of these models, the optimal evaluation model is determined. The results indicate the corresponding AUC (Area Under the Curve) values for the four models, IV-MLP, IV-CHAID, IV-RF, and IV-SVM, are 0.854, 0.86, 0.862, and 0.888, respectively. For prediction of rockfall in karst areas, rainfall, profile curvature, and coefficient of variation are identified as the most significant factors, accounting for 21%, 18%, and 11%, respectively. These factors indirectly promote water movement in karst areas, consequently influencing rockfall occurrences.

Rockfall

Susceptibility Assessment

IV-SVM

Karst

With the rapid economic development of mountainous urban areas, human engineering activities have increased, leading to geological environmental degradation. Uneven spatial and temporal distribution of precipitation due to climate change, along with intense short-duration rainfall, has resulted in frequent occurrences of rockfall, posing severe threats to the social and economic development of mountainous towns(Zhao et al. 2022; C. Liu et al. 2013). Despite limited disaster prevention resources, comprehensive risk investigation and precise monitoring and warning are challenging tasks. Nonetheless, critical facilities such as residential areas, schools, tourist attractions, and roads in mountainous regions face significant unknown rockfall hazards (Jinhuang et al. 2021). Consequently, achieving the utmost protection of local residents' lives and properties within cost constraints becomes a vital objective of disaster prevention and mitigation efforts. Therefore, conducting zoning and evaluation of rockfall susceptibility based on the local geological environment holds significant implications for disaster prevention and mitigation work.

Rockfall susceptibility assessment refers to the likelihood of rockfall occurrence within a regional scope(Yuan et al. 2022). As a type of geological hazard, the formation of rockfall is governed by geological environmental conditions. Currently, rockfall hazard assessment mainly falls into two categories: qualitative analysis and quantitative evaluation. Qualitative analysis methods include Analytic Hierarchy Process (AHP) and evidence weight method (Castellanos Abella and Van Westen 2008; Youssef, Pradhan, and Hassan 2011). For instance, Adnan et al.(Ozdemir 2015) predicted and evaluated the spatial distribution of sinkholes near Konya, Turkey, using evidence weight method and GIS. Chanu et al. (Chanu and Bakimchandra 2022)assessed landslide susceptibility along the Manipur State Highway in India using the AHP model and multi-resolution DEM. However, qualitative analysis approaches suffer from considerable subjectivity, where the accuracy of the evaluation model depends heavily on the evaluator's expertise, experience, and the accuracy of data collected. In contrast, quantitative analysis employs large datasets and mathematical methods or machine learning to identify data patterns and explore the relationships between various factors and rockfall, reducing the influence of subjective factors (Falaschi et al. 2009; Hungr et al. 2008; Tian et al. 2019; K. Zhang et al. 2017). For example, Li et al. (W.B. Li et al. 2021) evaluated rockfall susceptibility in Anyuan County using machine learning methods such as Logistic Regression (LR), Radial Basis Function Neural Network (RBF), Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), CHAID decision tree, and Random Forest (RF), comparing the accuracy of each model. Li et al. (S. Li et al. 2022) conducted a sensitivity analysis of geological hazards in the Longmen Mountain area of Wenchuan, China, using the Random Forest model after an earthquake. Numerous researchers have conducted studies on rockfall susceptibility assessment. However, due to the complexity and abruptness of rockfall and geological environmental conditions, there is no unified and widely accepted application method yet. Furthermore, the demand for higher accuracy in rockfall susceptibility assessment has increased in recent years, revealing the limitations of single evaluation models. As a result, coupling evaluation models, through modifying and adjusting single evaluation results, have shown improved capabilities to meet assessment requirements. For instance, Wang et al. (Wang et al. 2022)optimized the radio frequency model using Bayesian algorithms and combined it with the Random Forest model to achieve better rockfall susceptibility assessment results. Ling et al. (Ling et al. 2022)utilized the Random Forest model in conjunction with Classification and Regression Trees (CART), Logistic Regression (LR), and Support Vector Classification (SVC) to conduct precision analysis of Geological Hazard Susceptibility Mapping (LSM) in Tianshui, Gansu Province. The above studies comprehensively adjusted their results through corresponding coupling methods, enhancing the accuracy of the respective models. Overall, most existing research focuses on using specific methods to associate models for susceptibility assessment in designated regions(Jia, Dai, and Yang 2019; Sun et al. 2023). Fewer studies provide a sound rationale and thorough analysis and delve into the comprehensive impact of evaluation factors on rockfall. Additionally, there is a limited in-depth investigation of susceptibility assessment and analysis of rockfall disasters in karst regions, with more emphasis placed on surface subsidence. Currently, the majority of research involves utilizing ArcGIS to extract, analyze, and quantify relevant data, employing various modeling techniques to delineate and assess rockfall susceptibility in the study area. This research primarily focuses on the comparison and selection of analysis methods and the acquisition of rockfall formation conditions, while incorporating the specific geological and environmental characteristics of the study area to refine the evaluation criteria. The ultimate goal is to explore the optimal susceptibility zoning and evaluation model.

Xincheng County, was selected as the research area due to its typical karst topography and frequent occurrences of rockfall disasters. This study investigates the uncertainty between rockfall and its environmental factors during the process of susceptibility assessment, and further explore its impact on the evaluation of rockfall susceptibility.

The process of susceptibility assessment model and comparative analysis is illustrated in Fig. 1 as follows:

(a)Obtain rockfall and relevant environmental factor data for the study area and construct a spatial dataset.

(b)Utilize Information Value (IV) and the original factor data as input variables for the four coupling models.

(c)Conduct susceptibility modeling for each of the four coupling models, and subsequently create susceptibility maps of rockfall in GIS, including the delineation of susceptibility levels.

(d)Perform an analysis of the susceptibility evaluation results using metrics such as ROC accuracy, mean, standard deviation, and rockfall point density.

Having outlined the processes used to assess rockfall susceptibility in the study area, we now present the key models underpinning our analysis. These models are complex mathematical formulas and provide a quantitative framework for understanding the factors that influence rockfall occurrence

Information Value (IV) model

The Information Value (IV) model is an evaluation model that comprehensively assesses the environmental factors contributing to rockfall occurrences in the geological environment(M.H. Li, Qiu, et al. 2023). To simplify the calculation process, the overall probabilities involved are replaced with sample frequencies. The formula for IV is expressed as follows:

$$IVln\frac{{N}_{j}/N}{{S}_{j}/S}$$

In the above equation, IV represents the likelihood of rockfall occurrences under the condition of environmental factor j, where Nj denotes the number of grid cells with rockfall events, N refers to the total number of grid cells with rockfall, Sj represents the number of grid units with environmental factor j, and S is the total number of grid cells.IV > 0 indicates that the environmental factor favors the occurrence of rockfall, and the probability of its occurrence is directly proportional to its IV value. IV < 0 implies that the environmental factor is unfavorable for rockfall events. When IV = 0, it signifies that the environmental factor does not provide conditions for rockfall development.

Random Forest

Random Forest (RF) is a comprehensive data science classification model composed of multiple decision trees. The optimal classification result is determined through a voting process. The fundamental principle involves using bootstrap to collect samples from the original training set, with the collection size denoted as K, and based on this, obtaining the number of features. Subsequently, the decision tree model is constructed according to the sample classification. The process of RF has been described by Zhang et al. (Z.F. Zhang et al. 2022).

Multi-layer Perceptron

The Multi-layer Perceptron (MLP) is a feedforward artificial neural network (ANN)(Zare et al. 2013). It can be viewed as a directed graph consisting of multiple layers of nodes, where each layer is fully connected to the next one. Apart from the input layer, each node is a neuron with a nonlinear activation function. The MLP network learns the relationship between input (factors) and output (responses) vectors by adjusting the weights and biases. The communication between input and output is determined by changing the weight values of neuron connections, resulting in a well-structured, stable, and decision-capable network.In this study, each environmental factor serves as an input layer, the hidden layers process the environmental factors, and the output layer corresponds to the rockfall susceptibility evaluation results.

Support Vector Machine

Support Vector Machine (SVM) is capable of accurately fitting high-dimensional nonlinear systems based on small samples. It demonstrates several unique advantages in addressing problems of small sample size, nonlinearity, and high-dimensional pattern recognition, making it applicable to various machine learning tasks, including function approximation(Nian-qin et al. 2019). For linear support vector machines, the equation representing the hyperplane is given as:

$$wx+b0$$

In Eq. (2), w represents the weight vector, x denotes the feature vector of the sample point, and b is the intercept term. The solution to the above equation is a convex quadratic programming problem, which can be transformed into the optimization problem of solving the Lagrange function inequality. Details on the Lagrange function and linear equations can be found in the work by Liu et al. (Y.R. Liu et al. 2023).

Chi-squared Automatic Interaction Detection

CHAID (Chi-squared Automatic Interaction Detection) utilizes the chi-squared statistic to construct an optimal decision tree, which automatically discriminates the dependent variable based on the explanatory relationship between the independent and dependent variables(Faming et al. 2022). The CHAID model sets criteria such as the number of tree growth layers, splitting, and aggregation thresholds to build accurate and efficient prediction or classification models. To prevent overfitting, the model is constructed using randomly divided training samples. Finally, CHAID is stepwise tested with randomly divided test samples to adjust model parameters. For a more comprehensive CHAID computational process, please refer to the work by Park et al. (Park et al. 2018).

Study Area

Xincheng County is situated in the western part of Laibin City, Guangxi, with geographical coordinates ranging from 108°24′ to 109°07′ east longitude and 23°40′ to 24°23′ north latitude. The total area of the county is 2541 square kilometers. The topography of the entire county is characterized by typical karst mountainous terrain, exhibiting significant variations in landforms, mainly comprised of karst peaks and depressions. The elevation difference between mountains and valleys can reach 200 to 500 meters, with more mountains and fewer flat areas.The climate of Xincheng County falls under the category of South Subtropical Monsoon Climate, with ample annual precipitation. The average annual temperature is 20.7°C, and extreme maximum temperature can reach 39°C, while the extreme minimum temperature can drop to -3°C. The average annual rainfall ranges from 1304.0 to 1693.0 millimeters over multiple years.

Data source and processing

The main data sources for this study include:

(a) Basic data on geological hazard points, primarily obtained from 1:50,000 and 1:100,000 geological hazard investigation materials for Xincheng County, provided by the Geological Environment Monitoring Station of Guangxi Zhuang Autonomous Region;

(b) Derived data such as slope, aspect, and elevation extracted from the Digital Elevation Model (DEM), sourced from the Geographic Spatial Data Cloud, GDEMV3 version, with a spatial resolution of 30 m × 30 m;

(d) Data on major rivers, road networks, and administrative boundaries acquired from the 1:50,000 geographical base map of the study area;

(e) NDVI data from the 2020 China 30-meter annual maximum NDVI dataset (J.L. Yang et al. 2019);

(f) Spatial distribution map of rainfall data obtained from the National Science and Technology Resource Sharing Service Platform - National Earth System Science Data Center(Xu 2021).

In this study, a grid with a resolution of 30m × 30m was used as the basic unit for rockfall susceptibility assessment, resulting in the division of the entire study area into 2,874,376 grid cells using ArcGIS software. Firstly, various evaluation factors were extracted using ArcGIS, and each factor was classified and graded based on its categories and impact on rockfall. Then, the evaluation factor data was integrated with the rockfall disaster data to calculate the information value of each graded factor. To address issues of multicollinearity and potential overfitting, collinearity diagnostics were conducted.Subsequently, the information value data for each evaluation factor was obtained by transforming the graded factor data, resulting in the creation of information value layers for each factor. ArcGIS was used to process and analyze the information value data of the whole research area, leading to the establishment of a rockfall susceptibility database. For modeling, 70% of the samples were randomly selected from 172 rockfall disaster points and an equal number of non-disaster points as training data, while the remaining 30% of the samples were used to test the classification performance of the model.

Data set construction

Topographic and geomorphic factors

Elevation, slope, planar curvature, profile curvature, surface roughness, surface roughness, coefficient of variation, and landform type data can reflect the complex and diverse terrain conditions and directly or indirectly impact the stability of slopes (W.A. Zhang et al. 2023; X.M. Li, Cao, et al. 2023). In karst regions, planar curvature and profile curvature influence the erosion rate of karst areas(X.W. Yang 2019). Terrain roughness, surface roughness, and coefficient of variation affect surface runoff velocity, the transport and deposition of slope materials, and consequently influence soil infiltration and runoff generation time(Amen et al. 2023; Bi et al. 2023).

Normalized Difference Vegetation Index (NDVI) reflects the surface vegetation cover, and in general, areas with higher vegetation coverage have better soil and water conservation, which enhances the stability of geological formations. However, in karst regions, the relationship between vegetation types and soil-climate factors can affect the erosional effects in the karst areas (Shuai et al. 2023). In the study area, the soil is mainly acidic red soil, and the respiration of plant roots generates a large amount of CO₂, which leads to the fragmentation of the surface rock layers and increases erosion. Additionally, the decay of the litter layer and subsequent microbial activities also produce organic acids and a significant amount of CO₂, providing favorable conditions for continuous karst processes but unfavorable for the integrity of rock structures.

Basic geological factor

The distance of faults reflects the influence of tectonic activity on the development of rock joints and fractures (Y.Y. Li, Chen, and Shang 2017). Rock lithology is an important indicator reflecting the physical properties of rock and soil, including shear strength, among others (Samodra et al. 2016). The study area is predominantly composed of high-purity carbonate rock, accounting for over 90% of the total area of the study region.

Meteorological and hydrological factors

Rainfall affects the deep infiltration of rock and soil, thereby altering the strength of the rock and soil and subsequently impacting the stability of slopes. In karst regions, rainfall is a crucial water supply source for the karst process, and studies have shown a positive correlation between rainfall and karst erosion(Qian, Wu, and Li 2013; Wu et al. 2020; Zhou et al. 2023). Under favorable rain and heat conditions, the rock is more susceptible to dissolution by water, resulting in the formation of dissolution voids that reduce the integrity of the rock and promote rockfall. The distance to water systems reflects the spatial distribution of water systems in the karst region and indirectly indicates the extent of water system erosion in the study area.

Human activity factor

The construction of roads, especially the slope excavation during the road construction process, alters the slope's original condition, changes the stress distribution of the slope, and increases the probability of disasters. The blasting activities during road construction and the dynamic loads generated by the movement of large vehicles after road completion have adverse effects on the surrounding unstable geological formations.

Data set analysis

The majority of this study adopts the natural breakpoint classification method, which minimizes the differences within each category and maximizes the differences between categories, ensuring category consistency. Due to limitations in data sources, there are instances where certain factors have no disaster points within their interval ranges. To reduce the error caused by handling such data, a discretization approach is employed in the treatment of disaster points to avoid invalid information.

Through the analysis of information values, it is found that the development pattern of rockfalls in the study area is similar to the patterns discovered in most studies on rockfall susceptibility evaluation concerning factors such as slope, rainfall, fault distance, and road distance. However, there are differences in the consideration of certain factors:

(a) The primary lithology in the study area is carbonate rock, which is relatively hard. The rock mass exhibits varied karstification and weathering degrees, leading to different overall integrity of the rock mass. Therefore, considering the rock mass combination types rather than solely focusing on the shear strength and compressive strength of the rock mass is more crucial for karst areas. In this study, the rock mass combinations in the research area are categorized into moderately thick bedded hard to semi-hard strongly karstified carbonate rock groups, moderately thick bedded hard to weak to moderately karstified carbonate rock with interbedded clastic rock groups, and thin to thick bedded hard to weak clastic rock groups.

(b) In the study area, rockfalls occur within the NDVI range of 0.65 to 0.75. The development of rockfalls is unfavorable when NDVI is lower than 0.65 and higher than 0.83, indicating that vegetation conditions have varied effects on the development of rockfalls in the research area.

(c) Regarding landform types, peak clusters and depressions are favorable for the development of rockfalls in karst areas.

(d) Factors such as coefficient of variation, planar curvature, profile curvature, and surface roughness all have promoting effects on the development of rockfalls, suggesting that the role of water in the occurrence and development of rockfalls in the study area cannot be ignored.

Results of regional evaluation

First, based on the results of the proportions of each factor's classification intervals among the 15 conditioning factors and the occurred rockfall disasters' classification intervals, information values for each interval of the 15 conditioning factors are calculated, and the original data is revalued and tabulated accordingly. Then, the integrated information values of rockfall points and an equal number of non-collapse points are taken as the sample data, with 70% randomly selected as training data and the remaining 30% as testing data, which are then loaded into SPSS software for standardization.

In the SPSS software, the processed datasets are imported, and the data-driven models including CHAID, MLP, RF, and SVM are selected in the SPSS Modeler software for training. The revalued information of the original data is loaded into the trained models, and the four models generate predictive values for the occurrence of rockfalls in the 2,874,376 grid cells within the study area, which are compiled into tables.

Finally, the predictive result tables from the four data-driven models are imported into ArcGIS software. By using the data linking function, the DEM point layer is connected with the predictive results, and the point-to-raster function is applied to generate four models: IV-RF, IV-CHAID, IV-MLP, and IV-SVM. The entire grid cells are then classified into different levels based on proportions of 25%, 50%, and 75% (Zhou et al. 2023). Consequently, the final rockfall susceptibility evaluation is obtained by combining information values with the various coupled models (Fig. 5,Table 1).

Table 1. Results of sensitivity partitioning in the study area

	IV-SVM		IV-RF		IV-CHAID		IV-MLP
	Proportion(%)	Number of grids	Proportion(%)	Number of grids	Proportion(%)	Number of grids	Proportion(%)	Number of grids
Very Low	19.19	551597	15.71	451564	7.52	216129	14.53	417711
Low	21.85	627950	24.81	713193	56.11	1612842	20.81	598236
Moderate	24.85	714287	27.98	804232	14.89	427853	23.30	669611
High	20.10	577734	20.11	577959	11.15	320474	23.12	664541
Very High	14.01	402808	11.38	327134	10.34	297079	18.24	524277

According to Fig. 5, it can be observed that the distribution of high susceptibility and very high susceptibility zones for rockfalls in the study area is relatively scattered, which is closely related to the terrain and geomorphological factors of the karst region, with numerous peak clusters and depressions in the karst area. The common feature among the evaluation results of the four coupled models is that the high susceptibility and very high susceptibility zones are mainly distributed in the northeast and central-southwest sides of the study area. The main difference lies in the enrichment of the very high susceptibility zone: IV-RF, IV-MLP, and IV-SVM show significant enrichment of high susceptibility and very high susceptibility zones, while the distribution of various susceptibility zones in IV-CHAID is relatively scattered and uniform. Some susceptibility zoning patterns in certain models appear as block-like contours, which are closely related to the importance of influencing factors in their respective evaluation models (Fig. 6). Among them, the importance of rainfall is all greater than 0.1. Therefore, distinct boundaries are evident in the susceptibility assessment maps of each model, and such boundaries are highly correlated with the distribution pattern of rainfall.

Precision comparative analysis

The Receiver Operating Characteristic (ROC) curve was employed to accurately evaluate various models used. In the ROC curve, points closer to the upper-left corner indicate higher accuracy of the corresponding results. The area under the ROC curve (AUC) is used to quantify the performance of the predictive classification results. A larger AUC value indicates better classification performance, stronger algorithm generalization ability, and better responsiveness to new data. The y-axis represents the cumulative percentage of landslide occurrences within each susceptibility level in the study area, which can indicate the detection probability or sensitivity. The x-axis represents the cumulative percentage of area within each susceptibility level in the study area, which can represent the false alarm rate or specificity.

From the ROC curves shown in Fig. 6, the corresponding AUC values for the IV-MLP, IV-CHAID, IV-RF, and IV-SVM models are 0.854, 0.86, 0.862, and 0.888, respectively. The results indicate significant differences among the models in obtaining relevant results. It is evident that the IV-SVM coupled model outperforms the others in terms of accuracy and predictive ability.

Table 2
Rockfall density in the study area
	IV-SVM			IV-RF			IV-CHAID			IV-MLP
	Proportion of zone(%)	The number of rockfalls	Density of rockfalls	Proportion of zone(%)	The number of rockfalls	Density of rockfalls	Proportion of zone(%)	The number of rockfalls	Density of rockfalls	Proportion of zone(%)	The number of rockfalls	Density of rockfalls
Very Low	19.19	38	0.003	15.71	32	0.002	7.52	10	0.000	14.53	26	0.001
Low	21.85	38	0.003	24.81	38	0.004	56.11	104	0.023	20.81	40	0.003
Moderate	24.85	39	0.004	27.98	49	0.005	14.89	26	0.002	23.30	41	0.004
High	20.10	28	0.003	0.11	28	0.002	11.15	16	0.001	23.12	35	0.003
Very High	14.01	29	0.002	11.38	25	0.001	10.34	16	0.001	18.24	30	0.002

The mean and standard deviation of the evaluation results reflect the average level and dispersion of the assessment outcomes, respectively. The ranking of means from smallest to largest is as follows: IV-MBL (0.334) < IV-CHAID (0.335) < IV-RF (0.363) < IV-SVM (0.391). The ranking of standard deviations from smallest to largest is: IV-SVM (0.149) < IV-RF (0.154) < IV-CHAID (0.155) < IV-MLB (0.16). Notably, the IV-SVM model exhibits a relatively higher mean and a smaller standard deviation, indicating its enhanced discriminative capacity for rockfall assessment.

Conducting rockfall susceptibility assessment and analysis using a coupled model not only enhances the precision of sensitivity evaluation but also offers robust insights into the mechanisms underlying rockfall formation. Taking Xincheng County as the study area, we selected fifteen factors from four categories to establish the rockfall sensitivity index assessment system. Through the utilization of the IV-SVM model, we investigated the driving mechanisms between rockfall occurrences and geological-environmental factors in karst regions, leading to the identification of three key conclusions:

(a) The results analysis indicates that the following geological and environmental factors significantly contribute to the occurrence of rockfalls in the study area:Topographic relief; Elevation;Rainfall;Profile curvature;Coefficiem of variation;Landform type;Aspect;Distance from road;NDVI.

(b) The distribution of high susceptibility and extremely high susceptibility zones of rockfall disasters in the study area appears highly scattered, without large contiguous areas. This is related to the special geological environmental conditions in karst regions. Rainfall is an important factor affecting the susceptibility evaluation results of rockfalls in karst areas. Rainfall interacts with geological environmental elements directly or indirectly through surface water or groundwater, thereby promoting the occurrence of rockfalls.

(c) Compared to traditional single evaluation models, the coupling model demonstrates higher evaluation accuracy. Through the comparison of ROC curves for multiple coupling model results, it was found that the IV-SVM model has higher accuracy, with an AUC value of 0.888. This indicates that the model is more suitable for predicting and evaluating geological hazards such as rockfalls in karst areas.

The coupling model provides more accurate delineation of rockfall susceptibility zones, and the evaluation results obtained in this study can provide reliable data for urban development planning in karst mountainous areas.

Acknowledgements: The writers would like to greatly acknowledge all these fnancial supports and express their most sincere gratitude.

Availability of data and material: The datasets used or analyzed during the current study are available from the corresponding author on reasonable request.

Code availability: Not applicable.

Author contribution: Conceptualization: Wei-an Xie; Methodology: Yi-wen Li, Wang-jun Fei and Xing-duo Xiang; Formal analysis and investigation: Shi-fei Gu, Yi-wen Li, Wang-jun Fei and Xing-duo Xiang; Writing - original draft preparation: Wei-an Xie; Writing - review and editing: Wei-an Xie; Funding acquisition: San-xi Peng; Resources: Shi-fei Gu; Supervision: San-xi Peng

Funding: This work was supported by the National Natural Science Foundation of China under Grant Nos. 42062015.

Competing interests: The authors declare no competing interests.

Amen, A. R. M., A. Mustafa, D. A. Kareem, H. M. Hameed, A. A. Mirza, M. Szydlowski, and B. K. M. Saleem. 2023. "Mapping of Flood-Prone Areas Utilizing GIS Techniques and Remote Sensing: A Case Study of Duhok, Kurdistan Region of Iraq." Remote Sensing 15 (4). https://doi.org/10.3390/rs15041102. <Go to ISI>://WOS:000941772100001.
Bi, X. J., Q. Fan, L. He, C. J. Zhang, Y. F. Diao, and Y. L. Han. 2023. "Analysis and Evaluation of Extreme Rainfall Trends and Geological Hazards Risk in the Lower Jinshajiang River." Applied Sciences-Basel 13 (6). https://doi.org/10.3390/app13064021. <Go to ISI>://WOS:000957711100001.
Castellanos Abella, Enrique A., and Cees J. Van Westen. 2008. "Qualitative landslide susceptibility assessment by multicriteria analysis: A case study from San Antonio del sur, Guantanamo, Cuba." Geomorphology 94 (3-4): 453-466. https://doi.org/10.1016/j.geomorph.2006.10.038. <Go to ISI>://WOS:000253302700013.
Chanu, M. L., and O. Bakimchandra. 2022. "Landslide susceptibility assessment using AHP model and multi resolution DEMs along a highway in Manipur, India." Environmental Earth Sciences 81 (5). https://doi.org/10.1007/s12665-022-10281-4. <Go to ISI>://WOS:000761950900002.
Falaschi, F., F. Giacomelli, P. R. Federici, A. Puccinelli, G. D'Amato Avanzi, A. Pochini, and A. Ribolini. 2009. "Logistic regression versus artificial neural networks: landslide susceptibility evaluation in a sample area of the Serchio River valley, Italy." Natural Hazards 50 (3): 551-569. https://doi.org/10.1007/s11069-009-9356-5. <Go to ISI>://WOS:000268982900009.
Faming, HUANG, SHI Yu, OUYANG Weiping, HONG Anyu, ZENG Ziqiang, and XU Fugang. 2022. Journal of Civil and Environmental Engineering 44 (05): 16-28. https://kns.cnki.net/kcms/detail/50.1218.TU.20220106.1407.002.html.
Hungr, Oldrich, Scott McDougall, Mike Wise, and Michael Cullen. 2008. "Magnitude-frequency relationships of debris flows and debris avalanches in relation to slope relief." Geomorphology 96 (3-4): 355-365. https://doi.org/10.1016/j.geomorph.2007.03.020. <Go to ISI>://WOS:000255676400008.
Jia, X. L., Q. M. Dai, and H. Z. Yang. 2019. "Susceptibility zoning of karst geological hazards using machine learning and cloud model." Cluster Computing-the Journal of Networks Software Tools and Applications 22: S8051-S8058. https://doi.org/10.1007/s10586-017-1590-0. <Go to ISI>://WOS:000502007000036.
Jinhuang, Lin, Chen Wenhui, Qi Xinhua, and Hou Haoran. 2021. "Risk assessment and its influencing factors analysis of geological hazards in typical mountain environment." Journal of Cleaner Production 309.
Li, M. H., Y. Qiu, H. X. Xiong, and Z. C. Zhang. 2023. "Evaluation of landslide susceptibility based on VW-AHP-IV model: a case of Pengyang County, Ningxia, China." Environmental Earth Sciences 82 (4). https://doi.org/10.1007/s12665-023-10787-5. <Go to ISI>://WOS:000932078900001.
Li, S., Z. Y. Ni, Y. B. Zhao, W. Hu, Z. R. Long, H. Y. Ma, G. L. Zhou, Y. H. Luo, and C. T. Geng. 2022. "Susceptibility Analysis of Geohazards in the Longmen Mountain Region after the Wenchuan Earthquake." International Journal of Environmental Research and Public Health 19 (6). https://doi.org/10.3390/ijerph19063229. <Go to ISI>://WOS:000775336200001.
Li, W. B., Y. Shi, F. M. Huang, H. Y. Hong, and G. Q. Song. 2021. "Uncertainties of Collapse Susceptibility Prediction Based on Remote Sensing and GIS: Effects of Different Machine Learning Models." Frontiers in Earth Science 9. https://doi.org/10.3389/feart.2021.731058. <Go to ISI>://WOS:000724264600001.
Li, X. M., K. M. Cao, S. P. Huang, R. M. Xing, and Y. H. Yang. 2023. "Development Pattern of Toppling Deformation Slopes in Western China." Applied Sciences-Basel 13 (11). https://doi.org/10.3390/app13116596. <Go to ISI>://WOS:001004661200001.
Li, Y. Y., J. P. Chen, and Y. J. Shang. 2017. "An RVM-Based Model for Assessing the Failure Probability of Slopes along the Jinsha River, Close to the Wudongde Dam Site, China." Sustainability 9 (1). https://doi.org/10.3390/su9010032. <Go to ISI>://WOS:000394842700032.
Ling, X., Y. Q. Zhu, D. P. Ming, Y. Y. Chen, L. Zhang, and T. Y. Du. 2022. "Feature Engineering of Geohazard Susceptibility Analysis Based on the Random Forest Algorithm: Taking Tianshui City, Gansu Province, as an Example." Remote Sensing 14 (22). https://doi.org/10.3390/rs14225658. <Go to ISI>://WOS:000887598400001.
Liu, C., W. Y. Li, H. B. Wu, P. Lu, K. Sang, W. W. Sun, W. Chen, Y. Hong, and R. X. Li. 2013. "Susceptibility evaluation and mapping of China's landslides based on multi-source data." Natural Hazards 69 (3): 1477-1495. https://doi.org/10.1007/s11069-013-0759-y. <Go to ISI>://WOS:000327875400012.
Liu, Y. R., Z. Q. Meng, L. Zhu, D. Hu, and H. D. He. 2023. "Optimizing the Sample Selection of Machine Learning Models for Landslide Susceptibility Prediction Using Information Value Models in the Dabie Mountain Area of Anhui, China." Sustainability 15 (3). https://doi.org/10.3390/su15031971. <Go to ISI>://WOS:000931105200001.
Nian-qin, Wang, Guo Youjin, Liu Tieming, and Zhu Qinghua. 2019. "Assessment of landslide susceptibility based on SVM-LＲ model: A case study of Lintong district." Science Technology and Engineering 19 (30): 62-69.
Ozdemir, A. 2015. "Investigation of sinkholes spatial distribution using the weights of evidence method and GIS in the vicinity of Karapinar (Konya, Turkey)." Geomorphology 245: 40-50. https://doi.org/10.1016/j.geomorph.2015.04.034. <Go to ISI>://WOS:000360511200004.
Park, S. J., C. W. Lee, S. Lee, and M. J. Lee. 2018. "Landslide Susceptibility Mapping and Comparison Using Decision Tree Models: A Case Study of Jumunjin Area, Korea." Remote Sensing 10 (10). https://doi.org/10.3390/rs10101545. <Go to ISI>://WOS:000448555800045.
Qian, Hui, Jianhua Wu, and Peiyue Li. 2013. "Chemical Characteristics and Formation of Karst Water in Guilin Area, China." Asian Journal of Chemistry 25 (3): 1541-1546. <Go to ISI>://WOS:000312621400082.
Samodra, G., G. Q. Chen, J. Sartohadi, D. S. Hadmoko, K. Kasama, and M. A. Setiawan. 2016. "Rockfall susceptibility zoning based on back analysis of rockfall deposit inventory in Gunung Kelir, Java." Landslides 13 (4): 805-819. https://doi.org/10.1007/s10346-016-0713-7. <Go to ISI>://WOS:000385044500014.
Shuai, ZHU, CAO Jianhua, YANG Hui, and et al. 2023. "A Review of the Interaction Mechanism and Law between Vegetation and Rock Geochemical Background in Karst Areas." Rock and Mineral Analysis 42 (01): 59-71. https://doi.org/10.15898/j.cnki.11-2131/td.202108090095. https://kns.cnki.net/kcms/detail/11.2131.TD.20220727.1613.001.html.
Sun, D. L., D. L. Chen, J. L. Zhang, C. L. Mi, Q. Y. Gu, and H. J. Wen. 2023. "Landslide Susceptibility Mapping Based on Interpretable Machine Learning from the Perspective of Geomorphological Differentiation." Land 12 (5). https://doi.org/10.3390/land12051018. <Go to ISI>://WOS:000997535100001.
Tian, Yingying, Chong Xu, Haoyuan Hong, Qing Zhou, and Duo Wang. 2019. "Mapping earthquake-triggered landslide susceptibility by use of artificial neural network (ANN) models: an example of the 2013 Minxian (China) Mw 5.9 event." Geomatics Natural Hazards & Risk 10 (1): 1-25. https://doi.org/10.1080/19475705.2018.1487471. <Go to ISI>://WOS:000454617900001.
Wang, G. L., J. Y. Hao, H. J. Wen, and C. Cao. 2022. "A random forest model of karst ground collapse susceptibility based on factor and parameter coupling optimization." Geocarto International 37 (27): 15548-15567. https://doi.org/10.1080/10106049.2022.2102216. <Go to ISI>://WOS:000835211800001.
Wu, Zeyan, Cheng Zhang, Zhongcheng Jiang, Weiqun Luo, Qiong Xiao, Zhaoxin Hu, and Huaying Wu. 2020. "Subsoil carbonate dissolution rates and their determining factors in a karst drainage basin, SW China." Environmental Earth Sciences 79 (22). https://doi.org/10.1007/s12665-020-09255-1. <Go to ISI>://WOS:000584571600001.
Xu, Yang. 2021. "Annual precipitation data of 1km resolution in China（2001-2020)." National Science and Technology Resource sharing Service platform - National Earth System Science Data Center(China). Last Modified 2021. http://www.geodata.cn/.
Yang, J. L., J. W. Dong, X. M. Xiao, J. H. Dai, C. Y. Wu, J. Y. Xia, G. S. Zhao, M. M. Zhao, Z. L. Li, Y. Zhang, and Q. S. Ge. 2019. "Divergent shifts in peak photosynthesis timing of temperate and alpine grasslands in China." Remote Sensing of Environment 233. https://doi.org/10.1016/j.rse.2019.111395. <Go to ISI>://WOS:000497601000029.
Yang, X.W. 2019. "DEM Based Research on the Topographic Characteristics and Spatial Varition of Fenglin and Fengcong Karst landforms." Doctoral, Nanjing Normal University, Cnki.
Youssef, Ahmed M., Biswajeet Pradhan, and Abdallah Mohamed Hassan. 2011. "Flash flood risk estimation along the St. Katherine road, southern Sinai, Egypt using GIS based morphometry and satellite imagery." Environmental Earth Sciences 62 (3): 611-623. https://doi.org/10.1007/s12665-010-0551-1. <Go to ISI>://WOS:000286193200013.
Yuan, X. Y., C. Liu, R. H. Nie, Z. L. Yang, W. L. Li, X. A. Dai, J. Y. Cheng, J. M. Zhang, L. Ma, X. Fu, M. Tang, Y. N. Xu, and H. Lu. 2022. "A Comparative Analysis of Certainty Factor-Based Machine Learning Methods for Collapse and Landslide Susceptibility Mapping in Wenchuan County, China." Remote Sensing 14 (14). https://doi.org/10.3390/rs14143259. <Go to ISI>://WOS:000831707000001.
Zare, M., H. R. Pourghasemi, M. Vafakhah, and B. Pradhan. 2013. "Landslide susceptibility mapping at Vaz Watershed (Iran) using an artificial neural network model: a comparison between multilayer perceptron (MLP) and radial basic function (RBF) algorithms." Arabian Journal of Geosciences 6 (8): 2873-2888. https://doi.org/10.1007/s12517-012-0610-x. <Go to ISI>://WOS:000322028400012.
Zhang, Kaixiang, Xueling Wu, Ruiqing Niu, Ke Yang, and Lingran Zhao. 2017. "The assessment of landslide susceptibility mapping using random forest and decision tree methods in the Three Gorges Reservoir area, China." Environmental Earth Sciences 76 (11). https://doi.org/10.1007/s12665-017-6731-5. <Go to ISI>://WOS:000403434600024.
Zhang, W. A., Y. W. He, L. Q. Wang, S. L. Liu, and X. Y. Meng. 2023. "Landslide Susceptibility mapping using random forest and extreme gradient boosting: A case study of Fengjie, Chongqing." Geological Journal 58 (6): 2372-2387. https://doi.org/10.1002/gj.4683. <Go to ISI>://WOS:000930701900001.
Zhang, Z. F., Z. K. Qian, Y. Wei, X. Zhu, and L. J. Wang. 2022. "Evaluation of Geological Disaster Sensitivity in Shuicheng District Based on the WOE-RF Model." Sustainability 14 (23). https://doi.org/10.3390/su142316247. <Go to ISI>://WOS:000896275600001.
Zhao, J. Q., Q. Zhang, D. Z. Wang, W. H. Wu, and R. Y. Yuan. 2022. "Machine Learning-Based Evaluation of Susceptibility to Geological Hazards in the Hengduan Mountains Region, China." International Journal of Disaster Risk Science 13 (2): 305-316. https://doi.org/10.1007/s13753-022-00401-w. <Go to ISI>://WOS:000782549000001.
Zhou, J. X., S. C. Tan, J. Li, J. Xu, C. Wang, and H. Ye. 2023. "Landslide Susceptibility Assessment Using the Analytic Hierarchy Process (AHP): A Case Study of a Construction Site for Photovoltaic Power Generation in Yunxian County, Southwest China." Sustainability 15 (6). https://doi.org/10.3390/su15065281. <Go to ISI>://WOS:000967832400001.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Susceptibility Assessment of Rockfall in Karst Regions based on Information Entropy and Multi-Model Coupling

Status:

Version 1

Abstract

Figures

Introduction

Research Methods

Information Value (IV) model

Random Forest

Multi-layer Perceptron

Support Vector Machine

Chi-squared Automatic Interaction Detection

Study Area

Data source and processing

Data set construction and analysis

Data set construction

Topographic and geomorphic factors

Basic geological factor

Meteorological and hydrological factors

Human activity factor

Data set analysis

Evaluation results of rockfall susceptibility

Results of regional evaluation

Precision comparative analysis

Conclusion

Declarations

References

Additional Declarations

Status:

Version 1