Evaluation of Geological Hazards Susceptibility along the Hefei-Fuzhou High- Speed Railway Based on Machine Learning Algorithms

Evaluation of Geological Hazards Susceptibility along the Hefei-Fuzhou High- Speed Railway Based on Machine Learning Algorithms | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Evaluation of Geological Hazards Susceptibility along the Hefei-Fuzhou High- Speed Railway Based on Machine Learning Algorithms Jiarong Liang, Wenwen Qi, Chong Xu, Peng Wang, Jingjing Sun, Xuewei Zhang, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6501951/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 27 Nov, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Geological hazards pose significant risks during the construction and operation of railways, demanding effective prevention and control measures to ensure operational safety. The Hefei-Fuzhou High-Speed Railway, a critical transportation artery, traverses complex geological terrain and diverse landforms, leading to prominent geological hazards risk along its route. This study focuses on the Huangshan-Fuzhou section of this railway, evaluating landslide susceptibility within the railway corridor using the Random Forest (RF) algorithm. Furthermore, we analyze spatial variations in the primary influencing factors by comparing two distinct sub-regions. The main findings are as follows: ( 1 ) A landslide susceptibility model developed using a historical landslide inventory and the RF algorithm demonstrated strong predictive performance, achieving an Area Under the Curve (AUC) value of 0.86. ( 2 ) Application of this model produced a landslide susceptibility zonation map, which revealed that approximately 30% of the study area is classified as high (16.92%) and very high (13.22%) susceptibility zones. ( 3 ) Analysis of feature importance identified Slope (0.28), Relief (0.18), and Topographic Wetness Index (TWI, 0.11) as the most influential factors governing landslide susceptibility across the entire study area. ( 4 ) Comparative analysis of factor importance between the northern and southern sub-regions revealed distinct patterns: the primary factors in the north were Slope (0.115), TWI (0.051), and Relief (0.050), whereas in the south, they were Slope (0.149), Rainfall (0.043), and Curvature (0.041). These results highlight the dominant influence of topography on landslide susceptibility at a regional scale, while underscoring the enhanced role of precipitation as a key contributing factor to landslide in the southern region. This research provides a scientific basis for targeted geological hazards prevention along the Hefei-Fuzhou High-Speed Railway and offers a valuable methodological approach for regional infrastructure planning and geological risk management in similar environments. Earth and environmental sciences/Natural hazards Earth and environmental sciences/Environmental social sciences/Sustainability Earth and environmental sciences/Environmental sciences/Environmental impact Landslide traces inventory Landslide susceptibility Geological hazards Random Forest Hefei-Fuzhou High-Speed Railway Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Introduction Economic and social development and technological progress for the construction of key infrastructure needs continue to increase, and the construction of infrastructure, especially railway construction and geologic hazards have a close relationship between the construction of railways, the use of the process will increase the potential risk of geologic hazards, geologic hazards on the safe operation of infrastructure also poses a serious threat. They may not only lead to transportation disruption and property damage, but also cause casualties and adversely affect socio-economic stability and development. So, it is particularly important to evaluate the risk of geologic hazards in the area of road construction and operation. With the development of artificial intelligence, machine learning methods have been continuously applied in landslide susceptibility assessment. Compared with traditional statistical methods, machine learning methods have higher nonlinear feature extraction capabilities and deeper data mining capabilities [ 1 ] . For example, some scholars have used remote sensing, GIS and machine learning models to map landslide susceptibility in Nigeria [ 2 ] . ANN techniques to assess landslide susceptibility have also been well applied [ 3 ] , and the results show that ANN techniques have a high degree of accuracy in landslide susceptibility assessment. Some scholars have proposed a new interpretable deep learning model Deep-Attention-LSF to improve the accuracy and interpretability of landslide susceptibility mapping in the Three Gorges Reservoir area [ 4 ] . In addition, Logistic Regression methods [ 5 , 6 ] , K-means clustering algorithm [ 7 ] , and SVM [ 8 ] are widely used in this task. Many scholars have also applied a variety of machine learning methods to compare the performance of the model in the same research area. For example, using Logistic Regression model and Random Forest model to evaluate the landslide susceptibility in the southeast mountainous area of Sanming City, Fujian Province. The results show that RF has higher accuracy [ 5 ] . Based on Logistic Regression and Random Forest model, landslide sensitivity analysis was carried out in Wuyuan, Jiangxi Province, and the accuracy of Random Forest was higher than Logistic Regression model by comparing the results of nonparametric and linear regression analysis [ 8 ] . By comparing the performance of machine learning methods in the generation of landslide susceptibility map of Yanchuan County concluded that the Random Forest (RF) and Logistic Model Tree (LMT) methods are more accurate than the Categorical Regression Trees (CART) and Bayesian Network (BN) two models are more stable [ 9 ] . Under the multi-model comparison of machine learning susceptibility evaluation based on many scholars in different research areas, we can see that Random Forest is a highly flexible machine learning method with simple model structure and strong generalization ability. In landslide susceptibility evaluation, the Random Forest model performs best among individual machine learning models [ 10 ] , and the accuracy of the results is also significantly higher than other methods [ 5 , 8 ] . The use of Random Forest-related ensemble learning algorithms can couple the advantages of multiple models [ 11 ] , leading to further improvements in the accuracy of multi-hazard susceptibility assessments. For example, the accuracy of multi-hazard susceptibility evaluation is improved based on Random Forest-backpropagation neural network coupled model [ 12 ] . Through coupling optimization sampling and heterogeneous integrated machine learning, the accuracy of landslide susceptibility pre-evaluation in Chenxi County, Hunan Province is improved [ 13 ] . In the evaluation of landslide susceptibility, the selection of factor selection method is very important to the prediction accuracy and reliability of the model. Methods such as Information Gain Ratio (IGR) [ 14 ] , GeoDetector [ 15 ] , Pearson's Correlation Coefficient (PCC) [ 16 , 17 ] , and Multicollinearity analysis [ 18 , 19 ] have been widely used in landslide susceptibility assessment, which can effectively optimize the combination of factors and improve the prediction ability of the model. In this study, the method of calculating PCC was selected to calculate the linear correlation between factors to achieve the purpose of removing redundant factors. In addition, the evaluation of feature importance for landslide assessment factors based on Random Forest is a commonly employed machine learning approach. [ 20 ] . Through this method, the importance of each factor feature is evaluated, and the contribution of various environmental factors can be determined, which is of great significance for the evaluation of landslide hazards. In both the landslide susceptibility assessment of Mang City, Yunnan Province [ 21 ] , and the development of the landslide susceptibility map for the Pyeongchang region, South Korea [ 22 ] , feature importance evaluation was conducted. Additionally, in the process of deformation slope extraction in Chongqing and its surrounding areas, the importance of factors is also sorted to evaluate the importance of each factor for deformation slope [ 23 ] . Due to the significant variations in geological and tectonic settings, topography, climatic conditions, hydrological regimes, and land utilization practices across different regions, the primary factors contributing to landslides exhibit considerable regional discrepancies. For example, the elevation, water system and distance from the road in Luding area are the main environmental factors [ 24 ] . In the prediction of road landslide in Vietnam, NDVI and slope were found to be the main factors affecting the susceptibility of landslide [ 25 ] . In the study of landslide susceptibility evaluation in Yiyuan County, Shandong Province, topographical features and human activities were found to have a significant effect on the occurrence of landslide in this region [ 26 ] . The Hefei-Fuzhou High-Speed Railway passes through Anhui, Jiangxi, and Fujian provinces. There are significant differences in topographic conditions, geological structures, precipitation and other environmental factors between southern Anhui and northern Fujian at the north and south ends of the railway. Fujian Province experiences frequent typhoon landings each year, and these typhoons often bring heavy rainfall, which typically triggers clusters of landslides. Therefore, in the different environments of the north and south regions of the railway, this study constructs a landslide susceptibility evaluation model by applying the Random Forest method and selecting 13 environmental factors such as DEM, slope, slope direction (Aspect), terrain position index (TPI), relief, and terrain wetness index (TWI) in the same framework system. Through the method of ranking importance, the assessment of the actual importance of landslide factors in different regions along the north and south of the line to the region is carried out, and the main environmental impact factors of landslide in each sub-region are derived, and the impacts of different factors on the susceptibility to geologic hazards are analyzed, which provides a new idea and research example for the organic combination of the regional infrastructure planning and the risk management of geologic hazards. Study Area The Hefei-Fuzhou High-Speed Railway traverses the central and southern parts of Anhui Province, the eastern part of Jiangxi Province, and the northeastern part of Fujian Province. Starting from Hefei, it passes through five cities in Anhui Province, including Chaohu, Tongling, Wuhu, Xuancheng, and Huangshan, enters Shangrao City in Jiangxi Province, extends to Nanping and Ningde cities in Fujian Province, and finally reaches Fuzhou City (Fig. 1 ). It serves as a crucial passenger transport corridor connecting the central and eastern inland regions with the southeastern coastal areas of China [ 27 ] .In this paper, the Huangshan-Fuzhou section of Hefei-Fuzhou Railway is selected as the research object. It passes through a variety of geomorphological areas, including plains, hills and mountains. The terrain fluctuates greatly, the geological conditions along the line are complex, and the stratum lithology and geological structure are complex and changeable. The probability of debris flow, collapse, landslide and flash flood is large [ 28 ] . In addition, the study area has a long rainy season and frequent typhoons. Under such climatic conditions, it may lead to soil saturation and decrease of slope stability, which will also increase the risk of landslide. Through the analysis of existing data on the distribution of large-scale landslide remnants, it can be observed that the distribution of landslides along the railway line exhibits distinct spatial distribution patterns, influenced by various factors such as topography and geomorphology, geological structures, and climatic conditions. As illustrated in Fig. 1 , within the study area, the northeast and southwest regions experience a high number and density of landslides. Mountainous areas like Wuyi Mountain and Huangshan Mountain, characterized by significant topographic relief and complex geological conditions, emerge as high-incidence landslide zones, primarily concentrated in mid-altitude regions ranging from 1000 to 2200 meters above sea level. Data and methods Landslide data sample collection Relying on multi-source high-resolution satellite imagery data provided by the Google Earth platform, visual interpretation methods were employed, supplemented by systematic field investigation verification, literature review, and verification, to achieve precise spatial positioning, identification, and delineation of the boundaries of Landslide traces for the compilation of a landslide inventory. This process provides foundational data support for analyzing the distribution characteristics of landslide and for subsequent related research [ 29–31 ] . We interpreted large-scale landslides along the Hefei-Fuzhou Railway and its surrounding areas using the aforementioned methods, delineated landslide polygons, and completed the inventory compilation of large-scale Landslide traces in the study area (landslide distribution shown in Fig. 1 ). The existing surface data of large landslide remnants were converted into a list of landslide point data, thus collecting 2,299 landslide points as positive samples. In order to construct non-landslide samples, or negative samples, we used the strategy of generating random points. Specifically, firstly, the landslide polygon data is spatially intersected with the study area, and its complement is taken to identify the non-landslide area. Subsequently, we use GIS software to generate random points in the non-landslide area to ensure that the minimum distance between any two points is at least 1km to enhance the spatial representation of the sample. Through this method, we obtained 2299 non-landslide points equal to the number of positive samples, that is, negative sample points. After screening out invalid values, a total of 4556 positive and negative sample data were selected as the training sample data for the machine learning model. Environmental factor data The occurrence of landslide is influenced by various environmental factors. Based on existing research cases, so in this study, we chose digital elevation model (DEM), slope (Slope), slope direction (Aspect), terrain position index (TPI), topographic relief (Relief), terrain moisture index (TWI), distance from fault (Fault), slope curvature (Curvature). A total of 13 landslide impact factors, distance from river (River), Geology, land cover type (Landcover), average annual precipitation (Rain), and fractional vegetation cover (FVC), were used to build the model (Table 1 ). Table 1 Environmental impact factors and data sources Factors Data sources DEM Shuttle Radar Topography Mission (STRM) Slope From Elevation Aspect From Elevation TPI From Elevation Relief From Elevation TWI From Elevation Fault National Seismic Active Fault Data Curvature From Elevation River National Geographic Information Resource Catalogue Geology China Geological Survey and United States Geological Survey (USGS) Landcover GLCNMO Rain Global Climate data FVC GLCNMO DEM were obtained from STRM (Shuttle Radar Topography Mission) with a resolution of 3 arc-second meters [ 32 ] . All the terrain indices, including Slope, Aspect, Slope Curvature (m − 1 ), Topographic Position Index (TPI), and topographic relief, were generated based on the DEM using GIS software. Additionally, the Aspect was classified into nine categories: Flat, North, Northeast, East, Southeast, South, Southwest, West, and Northwest. The terrain moisture index (TWI) was produced based on DEM, and was produced under GRASS GIS. The fractional vegetation cover (FVC) of land in the study area was set according to 0-100%, and was assigned a value of -1 in the watershed. Fault data [ 33 , 34 ] were adopted from the National Seismic Active Faults Data, which were categorized into the distance from the active faults of 0-1km、1-2km、2-3km、3-4km、4-5km、5-6km、6-7km、7-8km、8-9km、9-10km、>10km for a total of 11 categories. Mean annual rainfall data (Rain) were obtained from the Global Climate data [ 35 ] with a resolution of about 1km, and a map of the mean annual rainfall in the study area was produced using linear interpolation. The distance to rivers was calculated based on the national five-level river data, sourced from the National Geographic Information Resource Catalogue ( https://www.webmap.cn/ ), using GIS software to compute the Euclidean distance. The stratigraphic data were obtained from the 1:2.5 million geologic map of China from the China Geological Survey ( https://www.cgs.gov.cn/ ), and geological maps of some Asian countries provided by the United States Geological Survey (USGS) ( https://www.usgs.gov/ ), and the study area was classified into 12 categories according to the geologic age, from newest to oldest, namely, Quaternary (Q), Tertiary (Te), Cretaceous (K), Jurassic ( J), Triassic (Tr), Permian (P), Carboniferous (C), Devonian (D), Silurian (S), Ordovician (O), Cambrian (∈), and Pre-Cambrian (Pre ∈). The Fractional Vegetation Cover (FVC) and land cover type data are sourced from the Global Land Cover by National Mapping Organizations (GLCNMO) [ 36 ] . The land cover types are divided into 20 categories, which include: Broad-leaved Evergreen Forests, Coniferous Evergreen Forests, Coniferous Deciduous Forests, Mixed Forests, Sparsely Wooded Forests, Broad-leaved Deciduous Forests, Shrublands, Grassy Areas, Grassy Areas with Sparsely Wooded Forests or Shrubs, Sparse Vegetation, Farmlands, Paddy Fields, Farmlands with Other Crops, Mangrove Forests, Wetlands, Consolidated Rock Areas, Exposed Areas of Loose Sandy Soil, Urban Land, Water Bodies, Snow and Ice. In this study, elevation (DEM), Slope, Relief, terrain moisture index (TWI), distance to fault (Fault), slope curvature (Curvature), distance to river (River), average annual precipitation (Rain), and fractional vegetation cover (FVC) were the continuous factors. The slope aspect (Aspect), Geology, land cover type (Landcover), and terrain position index (TPI) are discontinuous factors. Continuous factors were reclassified in the evaluation process using the natural discontinuity method (Fig. 2 ). Random Forests and Feature Importance Assessment Random Forests is an ensemble algorithm based on decision trees, which uses the bootstrap method (sampling with replacement) to perform classification and prediction for each decision tree [ 37 ] . In this study, the Random Forest Machine Learning Model was chosen as the base learner. The generated positive and negative sample data are used as the objects involved in subsequent model training. We further randomly divide these samples into a training set and a validation set, where 70% of the samples are assigned to the training set for model learning and optimization, while the remaining 30% of the samples constitute the validation set for validating the model's prediction accuracy and generalization ability. This division strategy aims to ensure the reliability of the model and provide a scientific basis for landslide susceptibility analysis. Gini Impurity is a key indicator used to measure the purity of data sets in decision tree and Random Forest algorithms. It is based on the Gini Index calculation, which is used to quantify the probability that samples in the data set are misclassified. The smaller the Gini impurity, the higher the purity of the dataset, the more centralized the distribution of categories of the samples. In this study, the built-in feature importance evaluation method of the Random Forest model adopts a strategy based on the reduction of Gini impurity. Specifically, the importance value of the feature is determined by measuring the contribution of the feature to the reduction of Gini impurity during the model training process, thus reflecting its contribution to the classification performance in the entire model. Receiver operating characteristic The Receiver Operating Characteristic (ROC) curve is a tool used to evaluate the performance of classification models. The performance of the classifier is comprehensively evaluated by calculating the True Positive Rate (TPR) and False Positive Rate (FPR) of the model at different thresholds. The closer the ROC curve is to the upper left corner, the better the performance of the model. The area under the curve (AUC) is calculated based on the ROC curve [ 38 , 39 ] , which can be used to quantify the overall performance of the model, and its value ranges from 0–1. The larger the area under the curve, the higher the accuracy of the model, and we generally believe that an AUC value of 0.7 or above indicates a better discriminatory ability and a more accurate model. Result and analysis Correlation Analysis In the process of establishing the evaluation index system, if there is a strong correlation between the influencing factors, it will make the accuracy of the model affected, therefore, based on the 13 influencing factors selected, the values of the correlation factors are extracted at the selected positive and negative sample points, the Pearson correlation coefficients are calculated, and the results of the correlation matrix are obtained as shown in Fig. 3 . The results of several studies show that when the correlation coefficient > 0.7, there is a high degree of covariance between the variables [ 40–42 ] . On the contrary, it means that various variables can meet the requirements of correlation test. And from Fig. 3 , the correlation coefficients between various factors are not greater than 0.7, therefore, the above mentioned 13 influencing factors can be used for this landslide susceptibility evaluation. Model performance analysis In this study, we used the generated positive and negative sample data as the objects involved in model training and randomly divided these samples into training and validation sets, where 70% of the samples are the training set, which is used for model learning and optimization, and 30% of the samples constitute the validation set. Through model training and validation, we plotted the ROC curve of the model and calculated the AUC value. Figure 4 shows an AUC value of 0.86, indicating that the landslide susceptibility evaluation model constructed based on the Random Forest algorithm has good predictive ability, can effectively distinguish landslide susceptible areas from non-susceptible areas, and provides strong support for landslide prediction and risk assessment. Evaluation results and analysis In this study, we used the Random Forest algorithm to obtain a high-precision landslide susceptibility evaluation model, the performance of which is shown in the previous section, and applied this model to the entire study area. The results were obtained and categorized by the Jenks Natural Breaks Classification Method, resulting in five categories: very low, low, medium, high, and very high susceptibility (Fig. 5 ). The results show that the very high susceptibility area accounts for 13.22% of the whole area, and the high susceptibility area also reaches about 16.92%, and these two categories account for a very significant proportion. Combined with the map of the distribution of environmental factors, it can be seen that the very high susceptibility and high susceptibility are mainly distributed in mountainous areas with higher elevation, and are concentrated in the area of high average annual precipitation and high slope, and the vegetation cover is at a low level in the vast majority of the area. It is very consistent with the spatial distribution of the existing landslide remnants, while the low and very low susceptibility areas together account for 51.99% of the study area, and the main land cover types are farmland, paddy land, and urban areas, with a relatively flat terrain. In order to identify the key factors that have a significant impact on landslide susceptibility, the accuracy of landslide susceptibility evaluation as well as the prediction accuracy can be improved. Random Forest-based evaluation of the importance eigenvalues of landslide evaluation factors is a very commonly used machine learning eigen importance. The principle of this method has been described in detail in 3.3, and this study uses this method to evaluate the feature importance to determine the contribution size of the factors, and screen out the environmental factors that are more important for this study area. The importance of landslide evaluation factors is obtained as shown in following figure (Fig. 6 ). We can see that three factors, Slope, Relief and Terrain Wetness Index (TWI), contribute more to landslide susceptibility, reaching 0.28, 0.18 and 0.11, respectively. According to the significant influence of the above three topographic factors on landslide, the degree of landslide susceptibility is closely related to the topographic conditions, and therefore, topographic features also occupy a central position in the evaluation model of the study area. In this paper, based on the different regional environments along the north and south ends of the route, we take a high susceptibility as well as a very high susceptibility more concentrated area at each of the north and south ends of the study area (red rectangular box in Fig. 5 ). The two sub-areas are mainly hilly and mountainous, and combined with the distribution of large landslides, more than 85% of landslides in the two sub-areas are concentrated in the slope range of 10°-30°and the distribution of landslides is very dense. In order to determine the feature importance of each factor within the two typical regions of North and South, the Permutation Importance method has become an important method for assessing the extent to which each feature affects model performance. Permutation Importance is a method that determines the importance of features by randomly disrupting the values of individual features and observing changes in model performance. In this study, we do this by calling a Random Forest model that has been trained over the entire study area and ensuring that the characteristics of the data in the two sub-areas are not fundamentally different from the total study area, and that the influence mechanisms of the factors taken for landslide susceptibility are also consistent with the total area, only narrowed in scope. Therefore, the ranked importance method can be used to reflect the actual importance of the features on the sub-regions. In the northern typical subregion, the top three factors are Slope, TWI and Relief, which are 0.115, 0.051 and 0.050, respectively. Figures 7 a, 7 b and 7 c show the local zoomed-in maps of these three factors. The northern typical area is close to Huangshan City, Anhui Province, the urban area is flat, low slope, and the railway is built in the process of construction as much as possible in the area of relatively low slope, taking into account the difficulty of slope for the construction of the railway and its impact on the possible geological hazards induced by the railway. In Fig. 7 c, we can see that the spatial distribution of topographic relief in the region is uneven, with smaller topographic relief in the north and larger topographic relief in the center. From the TWI, which is also shown in Fig. 7 b, the higher the TWI, the more likely the soil in the area is to be saturated, so it is easier for landslides to develop, and the landslide traces in the northern subregion are also concentrated in the areas with high TWI values. In the typical sub-region in the south, most of the area is located in the north of Fujian Province, and the terrain is mainly mountainous and hilly, and the slope is still in the first place with the value of 0.149 (Fig. 8 a), unlike the northern region of the Hefei-Fuzhou High-Speed Railway, the precipitation is in the second place with the value of 0.043 see precipitation factor has become one of the main environmental impact factors of landslide in Fujian (Fig. 8 b), which may be related to the strong precipitation brought by the typhoon season and the complex topographic and geological conditions in the northern region of Fujian (Fig. 8 b), which may be attributed to Fujian's proximity to the coast, where typhoon seasons bring heavy rainfall, coupled with the complex topography and geological conditions in the northern part of Fujian, which further exacerbate landslide susceptibility. and Curvature is a quantitative indicator of the degree of curvature of the surface of the terrain, the importance of the value of the southern typical region of the railway is located in the third (Fig. 8 c), the curvature of the slopes can be a response to the complexity of terrain and potential geologic instability, and high curvature areas are more prone to landslide and other geologic hazards. Discussion Regional variability in landslide impact factors In this study, the importance eigenvalue assessment of landslide evaluation factors based on Random Forests, it is concluded that the three factors of Slope, Relief, and TWI have a relatively large contribution to landslide susceptibility in the whole study area, which reaches 0.28, 0.18, and 0.11, respectively. And then, this paper selects two typical sub-regions in the north and south, and through the ranking importance method to reflect the actual importance of each factor feature on the sub-region. We get the following results in Section 4.3 in the typical northern subregion, the top three factors are Slope (0.115), TWI (0.051) and Relief (0.050), while in the typical southern subregion, Rain and Curvature are in the second place (0.043) and the third place (0.040). In the northern typical area along the Hefei-Fuzhou High-Speed Railway, most of the area is located in the southern part of Anhui Province and the border with Jiangxi Province, and the terrain is mainly mountainous and hilly. In this paper, after calculating the importance of the characteristics of the region, it is concluded that the slope is one of the main factors affecting landslide [ 43 ] . This conclusion has also been verified in some studies by previous scholars, for example, in Yangbi County, Dali Prefecture, Yunnan Province, the distribution of seismic landslides in the area of 10–30° has the highest number [ 44 ] , and from the distribution of landslides in Jiyuan City, the number of development is highest when the angle of slopes is from 0°to 30°and there is a certain relationship between the slope gradient and the susceptibility to landslides. However, the steepness of slopes is not strongly positively correlated with landslide susceptibility. Regions with a high number of landslides are not entirely those with steep slopes. Instead, in some areas with relatively gentle slopes, if affected by rainfall, rainwater infiltration can easily occur, which may further trigger landslide [ 45 ] . This is also basically consistent with the research results in this paper, which further demonstrates that slope is one of the main factors affecting landslide development [ 46 ] . A number of studies have shown that relief is one of the key environmental factors in landslide susceptibility assessment, especially after interacting with elevation, the explanatory power of the relief factor is significantly enhanced [ 47 ] . Relief refers to the difference between the maximum elevation and the minimum elevation in the study area, which can be used to quantify the changes in surface elevation and reveal the complexity of the terrain, and the greater the relief, the greater the surface runoff and erosion in the area will be intensified, so the topographic relief also has a certain influence on the landslide, and it is an important factor that should not be ignored. The topographic relief of the study area has a certain control effect on the landslide size, and there is a positive correlation between topographic relief and landslide area, elevation range and landslide area in the local area. In this study, Relief had a stronger effect on landslide susceptibility in the northern subregion compared to other factors. TWI, on the other hand, is a quantitative indicator for assessing the influence of topography on surface hydrological processes, especially in analyzing surface runoff, soil moisture and waterlogged areas, and is one of the most critical factors in assessing landslide susceptibility models [ 48 ] . In the landslide susceptibility assessment of this study, a higher topographic moisture index also means a relatively high soil moisture, which can also exacerbate the risk of landslide occurrence. Different from the northern sub-region, most of the southern sub-region is located in the northern part of Fujian Province, involving parts of four cities including Nanping, Sanming, Fuzhou and Ningde. Fujian Province is one of the most serious areas in China suffering from such landslide hazards [ 5 ] . The region has a warm and humid climate, abundant rainfall, frequent typhoons, frequent landslides and other geological hazards in mountainous and hilly areas [ 49 , 50 ] . In this sub-region, the slope is still the primary factor affecting the occurrence of landslide, including some previous studies have also proved that the slope has the greatest weight in the risk assessment of geological hazards in Fu 'an area, Fujian Province [ 51 ] . In this sub-region, in addition to Slope, the precipitation (Rain) factor ranks second, which is also closely related to the local climate characteristics. Some scholars have also confirmed in their studies on landslide in Fujian that heavy rainfall-induced landslides are the main cause of casualties [ 5 ] . Furthermore, in the large-scale landslide in Sanming, Fujian in 2016, in addition to topographic factors, heavy rainfall in a short period of time became the main influencing factor [ 52 ] . In addition to rainfall, in the prediction of geological hazards, slope curvature can reflect the complexity and potential geological instability of the topography in the study area. It is an important topographic factor affecting landslide susceptibility [ 53 ] . High-curvature areas usually indicate abrupt topographic changes and may be more prone to geological hazards such as landslides and debris flows. In the southern sub-region taken in this paper, the high curvature region is very consistent with the distribution of identified landslide sites. Coupled with the synergistic effect of rainfall factors in this region, the high curvature region is also more prone to landslide under heavy rainfall conditions [ 54 ] . Limitations and Future Directions of the study This paper only selects the Random Forest algorithm as the main research method, and does not compare it with other machine learning algorithms. It is based on the fact that Random Forest is a flexible machine learning method. Its model structure is simple and has strong generalization ability [ 55 ] , and it is not easy to overfit [ 56 ] .The built-in importance eigenvalue evaluation method of the algorithm can also be used to calculate the contribution of each factor to the landslide susceptibility results in this study, which can well meet the research objectives of this paper. In addition, the Random Forest algorithm has achieved good results in the assessment of landslide hazards at different regional scales. Based on these studies, this study also further expands the application scope of the Random Forest algorithm at different regional scales. In this study, the selection of evaluation factors included a total of 13 factors: DEM, Slope, Aspect, TWI, Relief, TPI, Faults, Curvature, Rivers, Geology, Landcover, Rain, and FVC. These factors were chosen to fully consider the impacts of topography, climate, vegetation, and water systems on landslide, and they have been shown to play significant roles in landslide hazards in this region. However, there are still some limitations. For instance, we can further analyze the specific relationships between landslide distribution characteristics and each factor within the study area [ 57 ] to better understand the relationships between landslide spatial distribution and these factors. Secondly, the study area in this research is relatively large, and there are significant environmental differences between the northern and southern sections of the railway, especially in the Fujian region which is prone to typhoons. Studies have also shown that strong winds can significantly induce landslide, and the risk increases dramatically when they are coupled with heavy rainfall [ 58 ] . Therefore, in future research, appropriate factors can be incorporated into the evaluation model based on regional differences. Additionally, many landslides originate from the evolution of unstable slopes [ 59 ] . Extracting deformation slopes within the region can provide new references for geological hazards warning and management [ 23 ] . Therefore, in future research, factors such as slope types and deformation results can be incorporated for comprehensive analysis to provide more accurate susceptibility assessment results and offer more effective support for landslide hazards risk management and emergency response. Conclusion This study evaluates the landslide susceptibility along the Hefei-Fuzhou High-Speed Railway and the surrounding areas based on the constructed Random Forest model and obtains the following conclusions: This paper comprehensively considered a variety of factors for landslide occurrence, including climate factors, vegetation factors, topographic and geologic factors, totaling 13 kinds. The importance of the characteristics of each factor was evaluated, and the top three factors were: Slope, Relief, and TWI, which reached 0.28, 0.18, and 0.11, respectively. A landslide susceptibility evaluation model was constructed based on existing large landslide remnant data and the Random Forest method, and a landslide susceptibility class zoning map was drawn. The result map was divided into five categories: very low, low, medium, high and very high susceptibility, and the results showed that the very high and high susceptibility zones accounted for a total of about 30.14%. The AUC value of the evaluation model was 0.86, and the accuracy of the model prediction results was high. Based on the landslide susceptibility class zoning map, this study selected two typical areas in the north and south that have a high percentage of high and very high susceptibility classes and a large difference in topography, rainfall and other conditions, and utilized the ranked importance method to assess the degree of influence of each feature on their respective sub-areas. The factors that were identified to be in the top three contributing factors for the northern typical region of the study area were Slope, TWI, and Relief, respectively. In contrast, Rain as well as Curvature factors were ranked second and third for the southern typical region. In summary, this paper uses the Random Forest model to evaluate the landslide susceptibility along the Hefei-Fuzhou High-Speed Railway and the surrounding areas, and obtains the landslide susceptibility zonation map. These results highlight the dominant influence of topography on landslide susceptibility at a regional scale, while underscoring the enhanced role of precipitation as a key contributing factor to landslide in the southern region. This research provides a scientific basis for targeted geological hazards prevention along the Hefei-Fuzhou High-Speed Railway and offers a valuable methodological approach for regional infrastructure planning and geological risk management in similar environments. Declarations Competing interests The authors declare no conflicts of interest. Ethical approval This study did not involve any human participants, animal subjects, or sensitive data requiring ethical approval. No experiments, interventions, or interactions were conducted that necessitate formal review or approval by an ethics committee. Author Contribution J.L.: Writing—original draft preparation, prepared figures. J. L., W. Q., C. X. and Z. X. : Writing—review and editing. W. Q.and C. X. : Project funding support. P.W., J. S., X. Z., J. C., Y. C., J.P., J. W. and Q. S.: Data acquisition. All authors reviewed the manuscript. Acknowledgement This work was supported by the National Key Research and Development Program of China (Grant No. 2024YFC3012603 and 2024YFC3012604), Chongqing Water Resources Bureau, China (Project No. CQS2400836) and Key Project of China Railway Design Corporation (Project No. 2023A0226409). Special thanks also go to the editors for their invaluable assistance in refining this work. Data Availability The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request. References Ishwaran H, andMalley J D. Synthetic learning machines [J]. BioData mining, 2014, 7(1) : 1-12. Nnaji E K, Ubana G O, Enabulele E C, et al. Landslide Susceptibility Mapping in Nigeria Using Remote Sensing, GIS, and Machine Learning Models [J]. The Asian Review of Civil Engineering, 2024. Caniani D, Pascale S D, Sdao F, et al. Neural networks and landslide susceptibility: a case study of the urban area of Potenza [J]. Natural Hazards, 2008, 45: 55-72. Cheng C, andFan L. An Attribution Deep Learning Interpretation Model for Landslide Susceptibility Mapping in the Three Gorges Reservoir Area [J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 1-15. Zhang K, Cui Y, Xu C, et al. Evaluation of Landslide Susceptibility to Heavy Rainfall in Sanming City,Fujian Province Based on Machine Learning Modeling [J]. Journal of Institute of Disaster Prevention, 2024, (3). Hong H, Tsangaratos P, Ilia I, et al. Comparing the Performance of a Logistic Regression and a Random Forest Model in Landslide Susceptibility Assessments. the Case of Wuyaun Area, China [C]. Fourth World Landslide Forum. 2017. Guo Z, Shi Y, Huang F, et al. Landslide susceptibility zonation method based on C5.0 decision tree and K-means cluster algorithms to improve the efficiency of risk management [J]. Geoscience Frontiers, 2021. Hong H, Xu C, andChen W. Providing a Landslide Susceptibility Map in Nancheng County, China, by Implementing Support Vector Machines [J]. American Journal of Geographic Information System, 2017. Chen W, Guo C, Lin F, et al. Exploring advanced machine learning techniques for landslide susceptibility mapping in Yanchuan County, China [J]. Earth Science Informatics, 2024, 17(6) : 5385-5402. Qi W, Xu C, andXu X. AutoGluon: A revolutionary framework for landslide hazard analysis [J]. Natural Hazards Research, 2021, 1(3) : 103-108. Jebur M N, Pradhan B, andTehrany M S. Optimization of landslide conditioning factors using very high-resolution airborne laser scanning (LiDAR) data at catchment scale [J]. Remote Sensing of Environment, 2014, 152: 150-165. Yu B, Xing H, andYan J. Susceptibility assessment of multi-hazards using random forest—back propagation neural network coupling model: a Hangzhou city case study [J]. Scientific Reports, 2024, 14(1). Lu Y, Xu H, Wang C, et al. A Novel Strategy Coupling Optimised Sampling with Heterogeneous Ensemble Machine-Learning to Predict Landslide Susceptibility [J]. Remote Sensing, 2024, 16(19). Yao W, Zhao J, andLi K. Assessment of Debris Flow Susceptibility Based on Information Gain Ratio-Support Vector Machine [J]. Urban Geotechnical Investigation & Surveying, 2023, (05) : 181-186. Chen B, Li C, Song C, et al. Automatic detection of active geohazards with millimeter-to-meter-scale deformation and quantitative analysis of factors influencing spatial distribution: A case study in the Hexi corridor, China [J]. International Journal of Applied Earth Observation and Geoinformation, 2024, 131: 103995-103995. Yang C, Liu L, Huang F, et al. Machine learning-based landslide susceptibility assessment with optimized ratio of landslide to non-landslide samples [J]. Gondwana Research, 2023, 123: 198-216. Huang Y, Xu C, He X, et al. Landslides induced by the 2023 Jishishan Ms6.2 earthquake (NW China): spatial distribution characteristics and implication for the seismogenic fault [J]. npj Natural Hazards, 2025, 2(1) : 14-14. Chen C, Shen Z, Weng Y, et al. Modeling Landslide Susceptibility in Forest-Covered Areas in Lin'an, China, Using Logistical Regression, a Decision Tree, and Random Forests [J]. Remote Sensing, 2023, 15(18) : 19. Liu X, Shao S, Zhang C, et al. Landslide susceptibility prediction in the loess tableland considering geomorphic evolution [J]. Catena, 2025, 249: 108668-108668. Reichenbach P. A review of statistically-based landslide susceptibility models [J]. Science Letter, 2018. Zheng Y, Chen J, Wang C, et al. Application of certainty factor and random forests model in landslide susceptibility evaluation in Mangshi City, Yunnan Province [J]. Bulletin of Geological Science and Technology, 2020, 39(6) : 131-144. Kim J-C, Lee S, Jung H-S, et al. Landslide susceptibility mapping using random forest and boosted tree models in Pyeong-Chang, Korea [J]. Geocarto International, 2018, 33(9) : 1000-1015. Liu J, Xu C, Zhao B, et al. Deformation Slope Extraction and Influencing Factor Analysis Using LT-1 Satellite Data: A Case Study of Chongqing and Surrounding Areas, China [J]. Remote Sensing, 2025, 17(1) : 156-156. Wang J, Zang M, Xu C, et al. Landslide Susceptibility Assessment Following the 2022 Luding Earthquake: a coupled analytic hierarchy Process and area under the receiver operating characteristic curve algorithm [J]. Jouranl of Engineering Geology, 2024, 32(5) : 1696-1711. Viet T N, Nguyen D D, Duc M N, et al. Exploring deep learning models for roadside landslide prediction: Insights and implications from comparative analysis [J]. Physics and Chemistry of the Earth, 2024, 136: 103741-103741. Li S, Yin C, Li J, et al. Landslide susceptibility assessment based on remote sensing interpretation and DBN-MLP model: a case study of Yiyuan County, China [J]. Stochastic Environmental Research Risk Assessment, 2025, 39(2). Li L, Lu L, Sun X, et al. Study on the network structure and interactive relationship of tourism flow along High-Speed Railway: an example of Hefei-Fuzhou High-Speed Railway [J]. Human Geography, 2020, 35(1) : 9. Dai L, Yu S, andCai K. Analysis of Flood Control Measures for Fujian-Jiangxi Section of HefeiFuzhou High-speed Railway [J]. High Speed Railway Technology, 2023, 14(01) : 91-96. Gao H, Xu C, Xie C, et al. Landslides triggered by the July 2023 extreme rainstorm in the Haihe River Basin, China [J]. Landslides, 2024, (prepublish) : 1-6. Wang W, Huang Y-D, Xu C, et al. Identification and distribution of 13003 landslides in the northwest margin of Qinghai-Tibet Plateau based on human-computer interaction remote sensing interpretation [J]. China Geology, 2024, 7(2) : 171-187. Wang Q, Xu C, andXu J. Exploration of Landslide Geomorphology and Inventory Construction in Minhe County, Qinghai, China, Based on Google Earth Remote Sensing Imagery [J]. 2025, 29(1) : e13281. Jarvis A, Reuter H, Nelson A, et al. Hole-filled SRTM for the globe Version 4 [J]. 2008. Xu X, Han Z, Yang X, et al. Seismotectonic map in China and its adjacent regions. In: Beijing: Seismological Press. [Z]. In: Beijing: Seismological Press. 2016 Wu X, Xu X, Yu G, et al. China Active Faults Database and its web system [J]. Earth System Science Data Discussions, 2023. Fick S E, andHijmans R J. WorldClim 2: new 1‐km spatial resolution climate surfaces for global land areas [J]. International Journal of Climatology, 2017, 37(12) : 4302-4315. Tateishi R, Uriyangqai B, Al-Bilbisi H, et al. Production of global land cover data - GLCNMO [J]. International journal of digital Earth, 2011, (1) : 4. Ma M, Wang J, Chen Q, et al. Regional Landslide Susceptibility Evaluation Based on Random Forest Weighting Information: a Case Study of Zigui to Badong Section in the Three Gorges Reservoir Area [J]. South China Geology, 2024, 40(04) : 749-763. Nhu V-H, Hoang N-D, Nguyen H, et al. Effectiveness assessment of Keras based deep learning with different robust optimization algorithms for shallow landslide susceptibility mapping at tropical area [J]. Catena, 2020, 188: 104458. Fawcett T. An introduction to ROC analysis [J]. Pattern Recognition Letters, 2006, 27(8) : 861-874. Bui D T, Tuan T A, Klempe H, et al. Spatial prediction models for shallow landslide hazards: a comparative assessment of the efficacy of support vector machines, artificial neural networks, kernel logistic regression, and logistic model tree [J]. Landslides, 2016, 13(2) : 361-378. Meng Q. LightGBM: A Highly Efficient Gradient Boosting Decision Tree [C]. Neural Information Processing Systems. 2017. Yao J, Qin S, Qiao S, et al. Assessment of Landslide Susceptibility Combining Deep Learning with Semi-Supervised Learning in Jiaohe County, Jilin Province, China [J]. Applied Sciences, 2020, 10(16). Sun H, Li W, Scaioni M, et al. Influence of spatial heterogeneity on landslide susceptibility in the transboundary area of the Himalayas [J]. Geomorphology, 2023, 433. He X, andXu C. Spatial distribution and tectonic significance of the landslides triggered by the 2021 Ms6.4 Yangbi earthquake, Yunnan, China [J]. Frontiers in Earth Science, 2022, 10. Deng J, Deng Y, Qiao S, et al. Spatial Distribution and Susceptibility Evaluation of Landslide Disasters in Loess Plateau Based on Information-CNN Coupling Model: A Case Study of Jiyuan City in Henan Province [J]. Geology and Resources, 2024, 33(06) : 836-845+835. Cui Y, Chen C, Zhu L, et al. Spatial distribution law of landslides and landslide susceptibility assessment in the eastern Himalayan syntaxis region [J]. Quarterly Journal of Engineering Geology and Hydrogeology, 2024, 57(3). Chen B, Zhang C, Li Z, et al. Developmental Characteristics and Controlling Factors of Landslides Triggered by Extreme Rainfalls on 16 June 2024 in Longyan, Fujian Province [J]. Geomatics and Information Science of Wuhan University, 2024, 49(11) : 2145-2155. Achu A L, Gopinath G, andH. T J a C D V. Redefining landslide susceptibility under extreme rainfall events using deep learning [J]. Geomorphology, 2024, 448(Mar.1) : 1.1-1.18. Huang C. Stability Evaluation of Rainfall-induced Landslide in Wuqu Village of Shouning County, Fujian Province [J]. Geology of Fujian, 2024, 43(04) : 297-305. Wang L, Deng Z, Xiao C, et al. Analysis of landslide temporal characteristics based on 50 years of landslide disaster data in Fujian Province [J]. Shanxi Architecture, 2025, 51(02) : 93-96+103. Zhu Q, Tan Q, Zhou J, et al. Risk Assessment of Regional Geological Landslide Hazards Based on GIS and AHP in Fu'an City [J]. Geomatics & Spatial Information Technology, 2021, 44(05) : 25-30. Ma S, Shao X, andXu C. Landslides Triggered by the 2016 Heavy Rainfall Event in Sanming, Fujian Province: Distribution Pattern Analysis and Spatio-Temporal Susceptibility Assessment [J]. Remote Sensing, 2023, 15(11) : 24. Li Y, Mei H, Ren X, et al. Geological Disaster Susceptibility Evaluation Based on Certainty Factor and Support Vector Machine [J]. Journal of Geo-information Science, 2018, 20(12) : 1699-1709. Luo X, Li C, andZhou J. GIS-based prediction method of shallow landslides induced by heavy rainfall in large mountainous areas [J]. Journal of Mountain Science, 2024, 21(5) : 1534-1548. Torres-Vázquez M Á, Herrera S, Gincheva A, et al. Enhancing seasonal fire predictions with hybrid dynamical and random forest models [J]. npj Natural Hazards, 2025, 2(1) : 20. Dey H, Haque M M, Shao W, et al. Simulating flood risk in Tampa Bay using a machine learning driven approach [J]. npj Natural Hazards, 2024, 1(1) : 40. Huang Y, Xu C, He X, et al. Distribution characteristics and cumulative effects of landslides triggered by multiple moderate-magnitude earthquakes: a case study of the comprehensive seismic impact area in Yibin, Sichuan, China [J]. Landslides, 2024, (prepublish) : 1-17. Lin Y-C, Hsieh J-Y, Shih H-S, et al. Strong wind is one of the important factors that trigger landslides [J]. npj Natural Hazards, 2025, 2(1) : 12. Dai Y, Li S, Zhang Y, et al. Continous–discontinous analysis of an unstable slope: evolution of damage zones and potential influencing areas [J]. npj Natural Hazards, 2025, 2(1) : 23. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 27 Nov, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 27 May, 2025 Reviews received at journal 26 May, 2025 Reviews received at journal 12 May, 2025 Reviewers agreed at journal 04 May, 2025 Reviewers agreed at journal 01 May, 2025 Reviewers invited by journal 30 Apr, 2025 Editor assigned by journal 30 Apr, 2025 Editor invited by journal 30 Apr, 2025 Submission checks completed at journal 29 Apr, 2025 First submitted to journal 22 Apr, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6501951","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":451766054,"identity":"915c8aba-39f4-45f2-98ca-56c711d822f4","order_by":0,"name":"Jiarong Liang","email":"","orcid":"","institution":"National Institute of Natural Hazards, Ministry of Emergency Management of China","correspondingAuthor":false,"prefix":"","firstName":"Jiarong","middleName":"","lastName":"Liang","suffix":""},{"id":451766055,"identity":"df2353da-f63a-4ec4-b6ff-18311ccb3208","order_by":1,"name":"Wenwen Qi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIiWNgGAWjYDACZiD+UMDAwEaSFsYZBiRpAeniMSBFOd9xHjNpG4N7cnz8Bxg//GCwyyOoRfIwUEuOQbExm0QCs2QPQ3IxQS0Gh3m33c4xSEhsk2BgkGZgOJDYQJQWC4OE+jb+A8y/idfCYJCQwMaQwEacLZKH+b//7DFIMGyTSGyz7DFIJqyF7/yxZIMfFQny8v2HD9/4UWFHWAvDATiLEaiYqAg6QFDFKBgFo2AUjHgAAPVuNJKOP8wWAAAAAElFTkSuQmCC","orcid":"","institution":"National Institute of Natural Hazards, Ministry of Emergency Management of China","correspondingAuthor":true,"prefix":"","firstName":"Wenwen","middleName":"","lastName":"Qi","suffix":""},{"id":451766056,"identity":"12d9c566-fbc5-4740-a1a4-cd762f031656","order_by":2,"name":"Chong Xu","email":"","orcid":"","institution":"National Institute of Natural Hazards, Ministry of Emergency Management of China","correspondingAuthor":false,"prefix":"","firstName":"Chong","middleName":"","lastName":"Xu","suffix":""},{"id":451766057,"identity":"c14f1956-9964-4781-821c-9e469b7d3dcc","order_by":3,"name":"Peng Wang","email":"","orcid":"","institution":"Beijing Engineering Corporation Limited","correspondingAuthor":false,"prefix":"","firstName":"Peng","middleName":"","lastName":"Wang","suffix":""},{"id":451766058,"identity":"0003fbb9-ff93-4134-b004-3914954b2e25","order_by":4,"name":"Jingjing Sun","email":"","orcid":"","institution":"National Institute of Natural Hazards, Ministry of Emergency Management of China","correspondingAuthor":false,"prefix":"","firstName":"Jingjing","middleName":"","lastName":"Sun","suffix":""},{"id":451766059,"identity":"33e38f3a-06e8-4ab0-bd72-b75f058186bb","order_by":5,"name":"Xuewei Zhang","email":"","orcid":"","institution":"National Institute of Natural Hazards, Ministry of Emergency Management of China","correspondingAuthor":false,"prefix":"","firstName":"Xuewei","middleName":"","lastName":"Zhang","suffix":""},{"id":451766060,"identity":"63667540-e3f8-472e-8c18-0a0b4c910a0d","order_by":6,"name":"Zhiwen Xue","email":"","orcid":"","institution":"National Institute of Natural Hazards, Ministry of Emergency Management of China","correspondingAuthor":false,"prefix":"","firstName":"Zhiwen","middleName":"","lastName":"Xue","suffix":""},{"id":451766061,"identity":"287aa3d4-ff10-49d2-8edf-45eb3db04805","order_by":7,"name":"Jingyu Chen","email":"","orcid":"","institution":"National Institute of Natural Hazards, Ministry of Emergency Management of China","correspondingAuthor":false,"prefix":"","firstName":"Jingyu","middleName":"","lastName":"Chen","suffix":""},{"id":451766062,"identity":"1557e912-6a1d-4709-b2ad-84dc0a9eb674","order_by":8,"name":"Yulong Cui","email":"","orcid":"","institution":"Anhui University of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Yulong","middleName":"","lastName":"Cui","suffix":""},{"id":451766063,"identity":"b1c81a64-40a7-4fdf-b7d2-6af523d6497d","order_by":9,"name":"Junwei Pan","email":"","orcid":"","institution":"China MCC5 Group Co., Ltd","correspondingAuthor":false,"prefix":"","firstName":"Junwei","middleName":"","lastName":"Pan","suffix":""},{"id":451766064,"identity":"d357af59-5f3b-4350-a9f9-42b5c4a017f8","order_by":10,"name":"Juanling Wang","email":"","orcid":"","institution":"China Railway Xi'an Group Company Limited","correspondingAuthor":false,"prefix":"","firstName":"Juanling","middleName":"","lastName":"Wang","suffix":""},{"id":451766065,"identity":"d53696e3-5563-4818-ad47-ff781c18be01","order_by":11,"name":"Qihao Sun","email":"","orcid":"","institution":"China Railway Design Corporation, Railway Design Corporation","correspondingAuthor":false,"prefix":"","firstName":"Qihao","middleName":"","lastName":"Sun","suffix":""}],"badges":[],"createdAt":"2025-04-22 08:23:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6501951/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6501951/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-26496-x","type":"published","date":"2025-11-27T15:57:35+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":82040476,"identity":"2b510d8c-ad81-4949-9c3a-bae9af86ec27","added_by":"auto","created_at":"2025-05-06 08:59:40","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":807840,"visible":true,"origin":"","legend":"\u003cp\u003eLocation of the study area\u003c/p\u003e","description":"","filename":"1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6501951/v1/33a90bb4578558980ffdd873.jpeg"},{"id":82040477,"identity":"a2983081-b99c-4c72-b4f9-1d90e433f9e1","added_by":"auto","created_at":"2025-05-06 08:59:40","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2436075,"visible":true,"origin":"","legend":"\u003cp\u003eInfluence factors in the study area: (a) DEM (b) River (c) Rain (d) Relief (e) Faults (f) Geology (g) Landcover (h) Aspect (i) Curvature (j) Slope (k) TWI (l) FVC (m) TPI\u003c/p\u003e","description":"","filename":"2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6501951/v1/9ab30bcb02f82a3ebce9189b.jpeg"},{"id":82040474,"identity":"f10e315b-7687-4207-8ccc-a50f9ab643cc","added_by":"auto","created_at":"2025-05-06 08:59:40","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":111129,"visible":true,"origin":"","legend":"\u003cp\u003ePlot of correlation coefficients between factors\u003c/p\u003e","description":"","filename":"3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6501951/v1/ceb2acf667363834d316a586.jpeg"},{"id":82040475,"identity":"c7447680-a6b5-41ab-99ec-f6535da89be0","added_by":"auto","created_at":"2025-05-06 08:59:40","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":49922,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves and AUC values\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6501951/v1/cc1ff0085ca07f87da0d477e.png"},{"id":82042111,"identity":"00a50ce5-4d0e-45e0-b35b-e3a3d380777d","added_by":"auto","created_at":"2025-05-06 09:15:40","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1123456,"visible":true,"origin":"","legend":"\u003cp\u003eSpatial distribution of landslide susceptibility evaluation results (The northern and southern rectangular boxes in the figure represent the two typical sub-regions selected for this study)\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6501951/v1/c38ddd5a1d0c3b54ec1ca2c5.jpg"},{"id":82040480,"identity":"b777c01a-4438-4b77-b61d-5480523aa70e","added_by":"auto","created_at":"2025-05-06 08:59:40","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":217661,"visible":true,"origin":"","legend":"\u003cp\u003eAssessment of the importance of each factor's characteristics\u003c/p\u003e","description":"","filename":"6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6501951/v1/3f556db118e714722aa68289.jpeg"},{"id":82042112,"identity":"74a89116-00fb-4574-924e-95f9c30f33ec","added_by":"auto","created_at":"2025-05-06 09:15:40","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":1068822,"visible":true,"origin":"","legend":"\u003cp\u003eThe figure on the left shows the spatial distribution of landslide susceptibility evaluation results in a typical sub-region in the northern part of the study area. The right figure shows the local zoom-in map of the factors with the top three eigenvalues in the northern typical region: (a) Slope, (b) Relief and (c) TWI are the zoom-in maps of each factor in the northern typical region, respectively.\u003c/p\u003e","description":"","filename":"7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6501951/v1/255b33ca17081d03950f8fd1.jpeg"},{"id":82040487,"identity":"2cafe270-2b69-412d-b264-901def4f8419","added_by":"auto","created_at":"2025-05-06 08:59:40","extension":"jpeg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":1113322,"visible":true,"origin":"","legend":"\u003cp\u003eThe figure on the left shows the spatial distribution of landslide susceptibility evaluation results in a typical sub-region in the southern part of the study area. The right figure shows the local zoom-in map of the factors with the top three eigenvalues in the southern typical region: (a) Slope, (b) Rain, and (c) Curvature are the local zoom-in maps of each factor in the southern typical region.\u003c/p\u003e","description":"","filename":"8.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6501951/v1/2e85c442b60c5f663882c5a9.jpeg"},{"id":97178392,"identity":"fd2149cc-5657-47e4-a081-c7361634988c","added_by":"auto","created_at":"2025-12-01 16:09:32","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":7661519,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6501951/v1/5482466b-2be2-41ab-b4b0-a73dfd6d8110.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Evaluation of Geological Hazards Susceptibility along the Hefei-Fuzhou High- Speed Railway Based on Machine Learning Algorithms","fulltext":[{"header":"Introduction","content":"\u003cp\u003eEconomic and social development and technological progress for the construction of key infrastructure needs continue to increase, and the construction of infrastructure, especially railway construction and geologic hazards have a close relationship between the construction of railways, the use of the process will increase the potential risk of geologic hazards, geologic hazards on the safe operation of infrastructure also poses a serious threat. They may not only lead to transportation disruption and property damage, but also cause casualties and adversely affect socio-economic stability and development. So, it is particularly important to evaluate the risk of geologic hazards in the area of road construction and operation.\u003c/p\u003e \u003cp\u003eWith the development of artificial intelligence, machine learning methods have been continuously applied in landslide susceptibility assessment. Compared with traditional statistical methods, machine learning methods have higher nonlinear feature extraction capabilities and deeper data mining capabilities\u003csup\u003e[\u003c/sup\u003e1\u003csup\u003e]\u003c/sup\u003e. For example, some scholars have used remote sensing, GIS and machine learning models to map landslide susceptibility in Nigeria \u003csup\u003e[\u003c/sup\u003e2\u003csup\u003e]\u003c/sup\u003e. ANN techniques to assess landslide susceptibility have also been well applied \u003csup\u003e[\u003c/sup\u003e3\u003csup\u003e]\u003c/sup\u003e, and the results show that ANN techniques have a high degree of accuracy in landslide susceptibility assessment. Some scholars have proposed a new interpretable deep learning model Deep-Attention-LSF to improve the accuracy and interpretability of landslide susceptibility mapping in the Three Gorges Reservoir area \u003csup\u003e[\u003c/sup\u003e4\u003csup\u003e]\u003c/sup\u003e. In addition, Logistic Regression methods \u003csup\u003e[\u003c/sup\u003e5\u003csup\u003e,\u003c/sup\u003e6\u003csup\u003e]\u003c/sup\u003e, K-means clustering algorithm \u003csup\u003e[\u003c/sup\u003e7\u003csup\u003e]\u003c/sup\u003e, and SVM \u003csup\u003e[\u003c/sup\u003e8\u003csup\u003e]\u003c/sup\u003eare widely used in this task.\u003c/p\u003e \u003cp\u003eMany scholars have also applied a variety of machine learning methods to compare the performance of the model in the same research area. For example, using Logistic Regression model and Random Forest model to evaluate the landslide susceptibility in the southeast mountainous area of Sanming City, Fujian Province. The results show that RF has higher accuracy \u003csup\u003e[\u003c/sup\u003e5\u003csup\u003e]\u003c/sup\u003e. Based on Logistic Regression and Random Forest model, landslide sensitivity analysis was carried out in Wuyuan, Jiangxi Province, and the accuracy of Random Forest was higher than Logistic Regression model by comparing the results of nonparametric and linear regression analysis\u003csup\u003e[\u003c/sup\u003e8\u003csup\u003e]\u003c/sup\u003e. By comparing the performance of machine learning methods in the generation of landslide susceptibility map of Yanchuan County concluded that the Random Forest (RF) and Logistic Model Tree (LMT) methods are more accurate than the Categorical Regression Trees (CART) and Bayesian Network (BN) two models are more stable\u003csup\u003e[\u003c/sup\u003e9\u003csup\u003e]\u003c/sup\u003e. Under the multi-model comparison of machine learning susceptibility evaluation based on many scholars in different research areas, we can see that Random Forest is a highly flexible machine learning method with simple model structure and strong generalization ability.\u003c/p\u003e \u003cp\u003eIn landslide susceptibility evaluation, the Random Forest model performs best among individual machine learning models \u003csup\u003e[\u003c/sup\u003e10\u003csup\u003e]\u003c/sup\u003e, and the accuracy of the results is also significantly higher than other methods\u003csup\u003e[\u003c/sup\u003e5\u003csup\u003e,\u003c/sup\u003e8\u003csup\u003e]\u003c/sup\u003e. The use of Random Forest-related ensemble learning algorithms can couple the advantages of multiple models \u003csup\u003e[\u003c/sup\u003e11\u003csup\u003e]\u003c/sup\u003e, leading to further improvements in the accuracy of multi-hazard susceptibility assessments.\u003c/p\u003e \u003cp\u003eFor example, the accuracy of multi-hazard susceptibility evaluation is improved based on Random Forest-backpropagation neural network coupled model\u003csup\u003e[\u003c/sup\u003e12\u003csup\u003e]\u003c/sup\u003e. Through coupling optimization sampling and heterogeneous integrated machine learning, the accuracy of landslide susceptibility pre-evaluation in Chenxi County, Hunan Province is improved \u003csup\u003e[\u003c/sup\u003e13\u003csup\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn the evaluation of landslide susceptibility, the selection of factor selection method is very important to the prediction accuracy and reliability of the model. Methods such as Information Gain Ratio (IGR)\u003csup\u003e[\u003c/sup\u003e14\u003csup\u003e]\u003c/sup\u003e, GeoDetector\u003csup\u003e[\u003c/sup\u003e15\u003csup\u003e]\u003c/sup\u003e, Pearson's Correlation Coefficient (PCC) \u003csup\u003e[\u003c/sup\u003e16\u003csup\u003e,\u003c/sup\u003e17\u003csup\u003e]\u003c/sup\u003e, and Multicollinearity analysis\u003csup\u003e[\u003c/sup\u003e18\u003csup\u003e,\u003c/sup\u003e19\u003csup\u003e]\u003c/sup\u003e have been widely used in landslide susceptibility assessment, which can effectively optimize the combination of factors and improve the prediction ability of the model. In this study, the method of calculating PCC was selected to calculate the linear correlation between factors to achieve the purpose of removing redundant factors.\u003c/p\u003e \u003cp\u003eIn addition, the evaluation of feature importance for landslide assessment factors based on Random Forest is a commonly employed machine learning approach. \u003csup\u003e[\u003c/sup\u003e20\u003csup\u003e]\u003c/sup\u003e. Through this method, the importance of each factor feature is evaluated, and the contribution of various environmental factors can be determined, which is of great significance for the evaluation of landslide hazards. In both the landslide susceptibility assessment of Mang City, Yunnan Province\u003csup\u003e[\u003c/sup\u003e21\u003csup\u003e]\u003c/sup\u003e, and the development of the landslide susceptibility map for the Pyeongchang region, South Korea \u003csup\u003e[\u003c/sup\u003e22\u003csup\u003e]\u003c/sup\u003e, feature importance evaluation was conducted. Additionally, in the process of deformation slope extraction in Chongqing and its surrounding areas, the importance of factors is also sorted to evaluate the importance of each factor for deformation slope \u003csup\u003e[\u003c/sup\u003e23\u003csup\u003e]\u003c/sup\u003e. Due to the significant variations in geological and tectonic settings, topography, climatic conditions, hydrological regimes, and land utilization practices across different regions, the primary factors contributing to landslides exhibit considerable regional discrepancies. For example, the elevation, water system and distance from the road in Luding area are the main environmental factors\u003csup\u003e[\u003c/sup\u003e24\u003csup\u003e]\u003c/sup\u003e. In the prediction of road landslide in Vietnam, NDVI and slope were found to be the main factors affecting the susceptibility of landslide\u003csup\u003e[\u003c/sup\u003e25\u003csup\u003e]\u003c/sup\u003e. In the study of landslide susceptibility evaluation in Yiyuan County, Shandong Province, topographical features and human activities were found to have a significant effect on the occurrence of landslide in this region\u003csup\u003e[\u003c/sup\u003e26\u003csup\u003e]\u003c/sup\u003e. The Hefei-Fuzhou High-Speed Railway passes through Anhui, Jiangxi, and Fujian provinces. There are significant differences in topographic conditions, geological structures, precipitation and other environmental factors between southern Anhui and northern Fujian at the north and south ends of the railway. Fujian Province experiences frequent typhoon landings each year, and these typhoons often bring heavy rainfall, which typically triggers clusters of landslides.\u003c/p\u003e \u003cp\u003eTherefore, in the different environments of the north and south regions of the railway, this study constructs a landslide susceptibility evaluation model by applying the Random Forest method and selecting 13 environmental factors such as DEM, slope, slope direction (Aspect), terrain position index (TPI), relief, and terrain wetness index (TWI) in the same framework system. Through the method of ranking importance, the assessment of the actual importance of landslide factors in different regions along the north and south of the line to the region is carried out, and the main environmental impact factors of landslide in each sub-region are derived, and the impacts of different factors on the susceptibility to geologic hazards are analyzed, which provides a new idea and research example for the organic combination of the regional infrastructure planning and the risk management of geologic hazards.\u003c/p\u003e\n\u003ch3\u003eStudy Area\u003c/h3\u003e\n\u003cp\u003eThe Hefei-Fuzhou High-Speed Railway traverses the central and southern parts of Anhui Province, the eastern part of Jiangxi Province, and the northeastern part of Fujian Province. Starting from Hefei, it passes through five cities in Anhui Province, including Chaohu, Tongling, Wuhu, Xuancheng, and Huangshan, enters Shangrao City in Jiangxi Province, extends to Nanping and Ningde cities in Fujian Province, and finally reaches Fuzhou City (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). It serves as a crucial passenger transport corridor connecting the central and eastern inland regions with the southeastern coastal areas of China\u003csup\u003e[\u003c/sup\u003e27\u003csup\u003e]\u003c/sup\u003e.In this paper, the Huangshan-Fuzhou section of Hefei-Fuzhou Railway is selected as the research object. It passes through a variety of geomorphological areas, including plains, hills and mountains. The terrain fluctuates greatly, the geological conditions along the line are complex, and the stratum lithology and geological structure are complex and changeable. The probability of debris flow, collapse, landslide and flash flood is large\u003csup\u003e[\u003c/sup\u003e28\u003csup\u003e]\u003c/sup\u003e. In addition, the study area has a long rainy season and frequent typhoons. Under such climatic conditions, it may lead to soil saturation and decrease of slope stability, which will also increase the risk of landslide.\u003c/p\u003e \u003cp\u003eThrough the analysis of existing data on the distribution of large-scale landslide remnants, it can be observed that the distribution of landslides along the railway line exhibits distinct spatial distribution patterns, influenced by various factors such as topography and geomorphology, geological structures, and climatic conditions. As illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, within the study area, the northeast and southwest regions experience a high number and density of landslides. Mountainous areas like Wuyi Mountain and Huangshan Mountain, characterized by significant topographic relief and complex geological conditions, emerge as high-incidence landslide zones, primarily concentrated in mid-altitude regions ranging from 1000 to 2200 meters above sea level.\u003c/p\u003e "},{"header":"Data and methods","content":"\u003ch2\u003eLandslide data sample collection\u003c/h2\u003e\u003cp\u003eRelying on multi-source high-resolution satellite imagery data provided by the Google Earth platform, visual interpretation methods were employed, supplemented by systematic field investigation verification, literature review, and verification, to achieve precise spatial positioning, identification, and delineation of the boundaries of Landslide traces for the compilation of a landslide inventory. This process provides foundational data support for analyzing the distribution characteristics of landslide and for subsequent related research\u003csup\u003e[\u003c/sup\u003e29–31\u003csup\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eWe interpreted large-scale landslides along the Hefei-Fuzhou Railway and its surrounding areas using the aforementioned methods, delineated landslide polygons, and completed the inventory compilation of large-scale Landslide traces in the study area (landslide distribution shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The existing surface data of large landslide remnants were converted into a list of landslide point data, thus collecting 2,299 landslide points as positive samples. In order to construct non-landslide samples, or negative samples, we used the strategy of generating random points. Specifically, firstly, the landslide polygon data is spatially intersected with the study area, and its complement is taken to identify the non-landslide area. Subsequently, we use GIS software to generate random points in the non-landslide area to ensure that the minimum distance between any two points is at least 1km to enhance the spatial representation of the sample. Through this method, we obtained 2299 non-landslide points equal to the number of positive samples, that is, negative sample points. After screening out invalid values, a total of 4556 positive and negative sample data were selected as the training sample data for the machine learning model.\u003c/p\u003e\u003ch3\u003eEnvironmental factor data\u003c/h3\u003e\u003cp\u003eThe occurrence of landslide is influenced by various environmental factors. Based on existing research cases, so in this study, we chose digital elevation model (DEM), slope (Slope), slope direction (Aspect), terrain position index (TPI), topographic relief (Relief), terrain moisture index (TWI), distance from fault (Fault), slope curvature (Curvature). A total of 13 landslide impact factors, distance from river (River), Geology, land cover type (Landcover), average annual precipitation (Rain), and fractional vegetation cover (FVC), were used to build the model (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eEnvironmental impact factors and data sources\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFactors\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eData sources\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDEM\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eShuttle Radar Topography Mission (STRM)\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSlope\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFrom Elevation\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAspect\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFrom Elevation\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTPI\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFrom Elevation\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRelief\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFrom Elevation\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTWI\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFrom Elevation\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFault\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNational Seismic Active Fault Data\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCurvature\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFrom Elevation\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRiver\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNational Geographic Information Resource Catalogue\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGeology\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina Geological Survey and United States Geological Survey (USGS)\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLandcover\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGLCNMO\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRain\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGlobal Climate data\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFVC\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGLCNMO\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003eDEM were obtained from STRM (Shuttle Radar Topography Mission) with a resolution of 3 arc-second meters \u003csup\u003e[\u003c/sup\u003e32\u003csup\u003e]\u003c/sup\u003e. All the terrain indices, including Slope, Aspect, Slope Curvature (m\u003csup\u003e− 1\u003c/sup\u003e), Topographic Position Index (TPI), and topographic relief, were generated based on the DEM using GIS software. Additionally, the Aspect was classified into nine categories: Flat, North, Northeast, East, Southeast, South, Southwest, West, and Northwest. The terrain moisture index (TWI) was produced based on DEM, and was produced under GRASS GIS. The fractional vegetation cover (FVC) of land in the study area was set according to 0-100%, and was assigned a value of -1 in the watershed. Fault data \u003csup\u003e[\u003c/sup\u003e33\u003csup\u003e,\u003c/sup\u003e34\u003csup\u003e]\u003c/sup\u003ewere adopted from the National Seismic Active Faults Data, which were categorized into the distance from the active faults of 0-1km、1-2km、2-3km、3-4km、4-5km、5-6km、6-7km、7-8km、8-9km、9-10km、\u0026gt;10km for a total of 11 categories. Mean annual rainfall data (Rain) were obtained from the Global Climate data \u003csup\u003e[\u003c/sup\u003e35\u003csup\u003e]\u003c/sup\u003e with a resolution of about 1km, and a map of the mean annual rainfall in the study area was produced using linear interpolation. The distance to rivers was calculated based on the national five-level river data, sourced from the National Geographic Information Resource Catalogue (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.webmap.cn/\u003c/span\u003e\u003cspan address=\"https://www.webmap.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), using GIS software to compute the Euclidean distance. The stratigraphic data were obtained from the 1:2.5\u0026nbsp;million geologic map of China from the China Geological Survey (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.cgs.gov.cn/\u003c/span\u003e\u003cspan address=\"https://www.cgs.gov.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), and geological maps of some Asian countries provided by the United States Geological Survey (USGS) (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.usgs.gov/\u003c/span\u003e\u003cspan address=\"https://www.usgs.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), and the study area was classified into 12 categories according to the geologic age, from newest to oldest, namely, Quaternary (Q), Tertiary (Te), Cretaceous (K), Jurassic ( J), Triassic (Tr), Permian (P), Carboniferous (C), Devonian (D), Silurian (S), Ordovician (O), Cambrian (∈), and Pre-Cambrian (Pre ∈). The Fractional Vegetation Cover (FVC) and land cover type data are sourced from the Global Land Cover by National Mapping Organizations (GLCNMO)\u003csup\u003e[\u003c/sup\u003e36\u003csup\u003e]\u003c/sup\u003e. The land cover types are divided into 20 categories, which include: Broad-leaved Evergreen Forests, Coniferous Evergreen Forests, Coniferous Deciduous Forests, Mixed Forests, Sparsely Wooded Forests, Broad-leaved Deciduous Forests, Shrublands, Grassy Areas, Grassy Areas with Sparsely Wooded Forests or Shrubs, Sparse Vegetation, Farmlands, Paddy Fields, Farmlands with Other Crops, Mangrove Forests, Wetlands, Consolidated Rock Areas, Exposed Areas of Loose Sandy Soil, Urban Land, Water Bodies, Snow and Ice. In this study, elevation (DEM), Slope, Relief, terrain moisture index (TWI), distance to fault (Fault), slope curvature (Curvature), distance to river (River), average annual precipitation (Rain), and fractional vegetation cover (FVC) were the continuous factors. The slope aspect (Aspect), Geology, land cover type (Landcover), and terrain position index (TPI) are discontinuous factors. Continuous factors were reclassified in the evaluation process using the natural discontinuity method (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\u003ch3\u003eRandom Forests and Feature Importance Assessment\u003c/h3\u003e\u003cp\u003eRandom Forests is an ensemble algorithm based on decision trees, which uses the bootstrap method (sampling with replacement) to perform classification and prediction for each decision tree\u003csup\u003e[\u003c/sup\u003e37\u003csup\u003e]\u003c/sup\u003e. In this study, the Random Forest Machine Learning Model was chosen as the base learner. The generated positive and negative sample data are used as the objects involved in subsequent model training. We further randomly divide these samples into a training set and a validation set, where 70% of the samples are assigned to the training set for model learning and optimization, while the remaining 30% of the samples constitute the validation set for validating the model's prediction accuracy and generalization ability. This division strategy aims to ensure the reliability of the model and provide a scientific basis for landslide susceptibility analysis.\u003c/p\u003e\u003cp\u003eGini Impurity is a key indicator used to measure the purity of data sets in decision tree and Random Forest algorithms. It is based on the Gini Index calculation, which is used to quantify the probability that samples in the data set are misclassified. The smaller the Gini impurity, the higher the purity of the dataset, the more centralized the distribution of categories of the samples. In this study, the built-in feature importance evaluation method of the Random Forest model adopts a strategy based on the reduction of Gini impurity. Specifically, the importance value of the feature is determined by measuring the contribution of the feature to the reduction of Gini impurity during the model training process, thus reflecting its contribution to the classification performance in the entire model.\u003c/p\u003e\u003ch2\u003eReceiver operating characteristic\u003c/h2\u003e\u003cp\u003eThe Receiver Operating Characteristic (ROC) curve is a tool used to evaluate the performance of classification models. The performance of the classifier is comprehensively evaluated by calculating the True Positive Rate (TPR) and False Positive Rate (FPR) of the model at different thresholds. The closer the ROC curve is to the upper left corner, the better the performance of the model.\u003c/p\u003e\u003cp\u003eThe area under the curve (AUC) is calculated based on the ROC curve \u003csup\u003e[\u003c/sup\u003e38\u003csup\u003e,\u003c/sup\u003e39\u003csup\u003e]\u003c/sup\u003e, which can be used to quantify the overall performance of the model, and its value ranges from 0–1. The larger the area under the curve, the higher the accuracy of the model, and we generally believe that an AUC value of 0.7 or above indicates a better discriminatory ability and a more accurate model.\u003c/p\u003e"},{"header":"Result and analysis","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eCorrelation Analysis\u003c/h2\u003e \u003cp\u003eIn the process of establishing the evaluation index system, if there is a strong correlation between the influencing factors, it will make the accuracy of the model affected, therefore, based on the 13 influencing factors selected, the values of the correlation factors are extracted at the selected positive and negative sample points, the Pearson correlation coefficients are calculated, and the results of the correlation matrix are obtained as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. The results of several studies show that when the correlation coefficient\u0026thinsp;\u0026gt;\u0026thinsp;0.7, there is a high degree of covariance between the variables \u003csup\u003e[\u003c/sup\u003e40\u0026ndash;42\u003csup\u003e]\u003c/sup\u003e. On the contrary, it means that various variables can meet the requirements of correlation test. And from Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, the correlation coefficients between various factors are not greater than 0.7, therefore, the above mentioned 13 influencing factors can be used for this landslide susceptibility evaluation.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eModel performance analysis\u003c/h2\u003e \u003cp\u003eIn this study, we used the generated positive and negative sample data as the objects involved in model training and randomly divided these samples into training and validation sets, where 70% of the samples are the training set, which is used for model learning and optimization, and 30% of the samples constitute the validation set.\u003c/p\u003e \u003cp\u003eThrough model training and validation, we plotted the ROC curve of the model and calculated the AUC value. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows an AUC value of 0.86, indicating that the landslide susceptibility evaluation model constructed based on the Random Forest algorithm has good predictive ability, can effectively distinguish landslide susceptible areas from non-susceptible areas, and provides strong support for landslide prediction and risk assessment.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eEvaluation results and analysis\u003c/h2\u003e \u003cp\u003eIn this study, we used the Random Forest algorithm to obtain a high-precision landslide susceptibility evaluation model, the performance of which is shown in the previous section, and applied this model to the entire study area. The results were obtained and categorized by the Jenks Natural Breaks Classification Method, resulting in five categories: very low, low, medium, high, and very high susceptibility (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). The results show that the very high susceptibility area accounts for 13.22% of the whole area, and the high susceptibility area also reaches about 16.92%, and these two categories account for a very significant proportion. Combined with the map of the distribution of environmental factors, it can be seen that the very high susceptibility and high susceptibility are mainly distributed in mountainous areas with higher elevation, and are concentrated in the area of high average annual precipitation and high slope, and the vegetation cover is at a low level in the vast majority of the area. It is very consistent with the spatial distribution of the existing landslide remnants, while the low and very low susceptibility areas together account for 51.99% of the study area, and the main land cover types are farmland, paddy land, and urban areas, with a relatively flat terrain.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn order to identify the key factors that have a significant impact on landslide susceptibility, the accuracy of landslide susceptibility evaluation as well as the prediction accuracy can be improved. Random Forest-based evaluation of the importance eigenvalues of landslide evaluation factors is a very commonly used machine learning eigen importance. The principle of this method has been described in detail in 3.3, and this study uses this method to evaluate the feature importance to determine the contribution size of the factors, and screen out the environmental factors that are more important for this study area. The importance of landslide evaluation factors is obtained as shown in following figure (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). We can see that three factors, Slope, Relief and Terrain Wetness Index (TWI), contribute more to landslide susceptibility, reaching 0.28, 0.18 and 0.11, respectively.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAccording to the significant influence of the above three topographic factors on landslide, the degree of landslide susceptibility is closely related to the topographic conditions, and therefore, topographic features also occupy a central position in the evaluation model of the study area. In this paper, based on the different regional environments along the north and south ends of the route, we take a high susceptibility as well as a very high susceptibility more concentrated area at each of the north and south ends of the study area (red rectangular box in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). The two sub-areas are mainly hilly and mountainous, and combined with the distribution of large landslides, more than 85% of landslides in the two sub-areas are concentrated in the slope range of 10\u0026deg;-30\u0026deg;and the distribution of landslides is very dense.\u003c/p\u003e \u003cp\u003eIn order to determine the feature importance of each factor within the two typical regions of North and South, the Permutation Importance method has become an important method for assessing the extent to which each feature affects model performance. Permutation Importance is a method that determines the importance of features by randomly disrupting the values of individual features and observing changes in model performance. In this study, we do this by calling a Random Forest model that has been trained over the entire study area and ensuring that the characteristics of the data in the two sub-areas are not fundamentally different from the total study area, and that the influence mechanisms of the factors taken for landslide susceptibility are also consistent with the total area, only narrowed in scope. Therefore, the ranked importance method can be used to reflect the actual importance of the features on the sub-regions.\u003c/p\u003e \u003cp\u003eIn the northern typical subregion, the top three factors are Slope, TWI and Relief, which are 0.115, 0.051 and 0.050, respectively. Figures\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea, \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb and \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ec show the local zoomed-in maps of these three factors. The northern typical area is close to Huangshan City, Anhui Province, the urban area is flat, low slope, and the railway is built in the process of construction as much as possible in the area of relatively low slope, taking into account the difficulty of slope for the construction of the railway and its impact on the possible geological hazards induced by the railway. In Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ec, we can see that the spatial distribution of topographic relief in the region is uneven, with smaller topographic relief in the north and larger topographic relief in the center. From the TWI, which is also shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb, the higher the TWI, the more likely the soil in the area is to be saturated, so it is easier for landslides to develop, and the landslide traces in the northern subregion are also concentrated in the areas with high TWI values.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn the typical sub-region in the south, most of the area is located in the north of Fujian Province, and the terrain is mainly mountainous and hilly, and the slope is still in the first place with the value of 0.149 (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003ea), unlike the northern region of the Hefei-Fuzhou High-Speed Railway, the precipitation is in the second place with the value of 0.043 see precipitation factor has become one of the main environmental impact factors of landslide in Fujian (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eb), which may be related to the strong precipitation brought by the typhoon season and the complex topographic and geological conditions in the northern region of Fujian (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eb), which may be attributed to Fujian's proximity to the coast, where typhoon seasons bring heavy rainfall, coupled with the complex topography and geological conditions in the northern part of Fujian, which further exacerbate landslide susceptibility. and Curvature is a quantitative indicator of the degree of curvature of the surface of the terrain, the importance of the value of the southern typical region of the railway is located in the third (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003ec), the curvature of the slopes can be a response to the complexity of terrain and potential geologic instability, and high curvature areas are more prone to landslide and other geologic hazards.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eRegional variability in landslide impact factors\u003c/h2\u003e \u003cp\u003eIn this study, the importance eigenvalue assessment of landslide evaluation factors based on Random Forests, it is concluded that the three factors of Slope, Relief, and TWI have a relatively large contribution to landslide susceptibility in the whole study area, which reaches 0.28, 0.18, and 0.11, respectively.\u003c/p\u003e \u003cp\u003eAnd then, this paper selects two typical sub-regions in the north and south, and through the ranking importance method to reflect the actual importance of each factor feature on the sub-region. We get the following results in Section 4.3 in the typical northern subregion, the top three factors are Slope (0.115), TWI (0.051) and Relief (0.050), while in the typical southern subregion, Rain and Curvature are in the second place (0.043) and the third place (0.040).\u003c/p\u003e \u003cp\u003eIn the northern typical area along the Hefei-Fuzhou High-Speed Railway, most of the area is located in the southern part of Anhui Province and the border with Jiangxi Province, and the terrain is mainly mountainous and hilly. In this paper, after calculating the importance of the characteristics of the region, it is concluded that the slope is one of the main factors affecting landslide \u003csup\u003e[\u003c/sup\u003e43\u003csup\u003e]\u003c/sup\u003e. This conclusion has also been verified in some studies by previous scholars, for example, in Yangbi County, Dali Prefecture, Yunnan Province, the distribution of seismic landslides in the area of 10\u0026ndash;30\u0026deg; has the highest number \u003csup\u003e[\u003c/sup\u003e44\u003csup\u003e]\u003c/sup\u003e, and from the distribution of landslides in Jiyuan City, the number of development is highest when the angle of slopes is from 0\u0026deg;to 30\u0026deg;and there is a certain relationship between the slope gradient and the susceptibility to landslides. However, the steepness of slopes is not strongly positively correlated with landslide susceptibility. Regions with a high number of landslides are not entirely those with steep slopes. Instead, in some areas with relatively gentle slopes, if affected by rainfall, rainwater infiltration can easily occur, which may further trigger landslide \u003csup\u003e[\u003c/sup\u003e45\u003csup\u003e]\u003c/sup\u003e. This is also basically consistent with the research results in this paper, which further demonstrates that slope is one of the main factors affecting landslide development \u003csup\u003e[\u003c/sup\u003e46\u003csup\u003e]\u003c/sup\u003e. A number of studies have shown that relief is one of the key environmental factors in landslide susceptibility assessment, especially after interacting with elevation, the explanatory power of the relief factor is significantly enhanced\u003csup\u003e[\u003c/sup\u003e47\u003csup\u003e]\u003c/sup\u003e. Relief refers to the difference between the maximum elevation and the minimum elevation in the study area, which can be used to quantify the changes in surface elevation and reveal the complexity of the terrain, and the greater the relief, the greater the surface runoff and erosion in the area will be intensified, so the topographic relief also has a certain influence on the landslide, and it is an important factor that should not be ignored. The topographic relief of the study area has a certain control effect on the landslide size, and there is a positive correlation between topographic relief and landslide area, elevation range and landslide area in the local area. In this study, Relief had a stronger effect on landslide susceptibility in the northern subregion compared to other factors. TWI, on the other hand, is a quantitative indicator for assessing the influence of topography on surface hydrological processes, especially in analyzing surface runoff, soil moisture and waterlogged areas, and is one of the most critical factors in assessing landslide susceptibility models\u003csup\u003e[\u003c/sup\u003e48\u003csup\u003e]\u003c/sup\u003e. In the landslide susceptibility assessment of this study, a higher topographic moisture index also means a relatively high soil moisture, which can also exacerbate the risk of landslide occurrence.\u003c/p\u003e \u003cp\u003eDifferent from the northern sub-region, most of the southern sub-region is located in the northern part of Fujian Province, involving parts of four cities including Nanping, Sanming, Fuzhou and Ningde. Fujian Province is one of the most serious areas in China suffering from such landslide hazards \u003csup\u003e[\u003c/sup\u003e5\u003csup\u003e]\u003c/sup\u003e. The region has a warm and humid climate, abundant rainfall, frequent typhoons, frequent landslides and other geological hazards in mountainous and hilly areas\u003csup\u003e[\u003c/sup\u003e49\u003csup\u003e,\u003c/sup\u003e50\u003csup\u003e]\u003c/sup\u003e. In this sub-region, the slope is still the primary factor affecting the occurrence of landslide, including some previous studies have also proved that the slope has the greatest weight in the risk assessment of geological hazards in Fu 'an area, Fujian Province\u003csup\u003e[\u003c/sup\u003e51\u003csup\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn this sub-region, in addition to Slope, the precipitation (Rain) factor ranks second, which is also closely related to the local climate characteristics. Some scholars have also confirmed in their studies on landslide in Fujian that heavy rainfall-induced landslides are the main cause of casualties \u003csup\u003e[\u003c/sup\u003e5\u003csup\u003e]\u003c/sup\u003e. Furthermore, in the large-scale landslide in Sanming, Fujian in 2016, in addition to topographic factors, heavy rainfall in a short period of time became the main influencing factor \u003csup\u003e[\u003c/sup\u003e52\u003csup\u003e]\u003c/sup\u003e. In addition to rainfall, in the prediction of geological hazards, slope curvature can reflect the complexity and potential geological instability of the topography in the study area. It is an important topographic factor affecting landslide susceptibility \u003csup\u003e[\u003c/sup\u003e53\u003csup\u003e]\u003c/sup\u003e. High-curvature areas usually indicate abrupt topographic changes and may be more prone to geological hazards such as landslides and debris flows. In the southern sub-region taken in this paper, the high curvature region is very consistent with the distribution of identified landslide sites. Coupled with the synergistic effect of rainfall factors in this region, the high curvature region is also more prone to landslide under heavy rainfall conditions \u003csup\u003e[\u003c/sup\u003e54\u003csup\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eLimitations and Future Directions of the study\u003c/h2\u003e \u003cp\u003eThis paper only selects the Random Forest algorithm as the main research method, and does not compare it with other machine learning algorithms. It is based on the fact that Random Forest is a flexible machine learning method. Its model structure is simple and has strong generalization ability\u003csup\u003e[\u003c/sup\u003e55\u003csup\u003e]\u003c/sup\u003e, and it is not easy to overfit\u003csup\u003e[\u003c/sup\u003e56\u003csup\u003e]\u003c/sup\u003e.The built-in importance eigenvalue evaluation method of the algorithm can also be used to calculate the contribution of each factor to the landslide susceptibility results in this study, which can well meet the research objectives of this paper. In addition, the Random Forest algorithm has achieved good results in the assessment of landslide hazards at different regional scales. Based on these studies, this study also further expands the application scope of the Random Forest algorithm at different regional scales.\u003c/p\u003e \u003cp\u003eIn this study, the selection of evaluation factors included a total of 13 factors: DEM, Slope, Aspect, TWI, Relief, TPI, Faults, Curvature, Rivers, Geology, Landcover, Rain, and FVC. These factors were chosen to fully consider the impacts of topography, climate, vegetation, and water systems on landslide, and they have been shown to play significant roles in landslide hazards in this region. However, there are still some limitations. For instance, we can further analyze the specific relationships between landslide distribution characteristics and each factor within the study area \u003csup\u003e[\u003c/sup\u003e57\u003csup\u003e]\u003c/sup\u003e to better understand the relationships between landslide spatial distribution and these factors. Secondly, the study area in this research is relatively large, and there are significant environmental differences between the northern and southern sections of the railway, especially in the Fujian region which is prone to typhoons. Studies have also shown that strong winds can significantly induce landslide, and the risk increases dramatically when they are coupled with heavy rainfall\u003csup\u003e[\u003c/sup\u003e58\u003csup\u003e]\u003c/sup\u003e. Therefore, in future research, appropriate factors can be incorporated into the evaluation model based on regional differences. Additionally, many landslides originate from the evolution of unstable slopes\u003csup\u003e[\u003c/sup\u003e59\u003csup\u003e]\u003c/sup\u003e. Extracting deformation slopes within the region can provide new references for geological hazards warning and management\u003csup\u003e[\u003c/sup\u003e23\u003csup\u003e]\u003c/sup\u003e. Therefore, in future research, factors such as slope types and deformation results can be incorporated for comprehensive analysis to provide more accurate susceptibility assessment results and offer more effective support for landslide hazards risk management and emergency response.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study evaluates the landslide susceptibility along the Hefei-Fuzhou High-Speed Railway and the surrounding areas based on the constructed Random Forest model and obtains the following conclusions:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThis paper comprehensively considered a variety of factors for landslide occurrence, including climate factors, vegetation factors, topographic and geologic factors, totaling 13 kinds. The importance of the characteristics of each factor was evaluated, and the top three factors were: Slope, Relief, and TWI, which reached 0.28, 0.18, and 0.11, respectively.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eA landslide susceptibility evaluation model was constructed based on existing large landslide remnant data and the Random Forest method, and a landslide susceptibility class zoning map was drawn. The result map was divided into five categories: very low, low, medium, high and very high susceptibility, and the results showed that the very high and high susceptibility zones accounted for a total of about 30.14%. The AUC value of the evaluation model was 0.86, and the accuracy of the model prediction results was high.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eBased on the landslide susceptibility class zoning map, this study selected two typical areas in the north and south that have a high percentage of high and very high susceptibility classes and a large difference in topography, rainfall and other conditions, and utilized the ranked importance method to assess the degree of influence of each feature on their respective sub-areas. The factors that were identified to be in the top three contributing factors for the northern typical region of the study area were Slope, TWI, and Relief, respectively. In contrast, Rain as well as Curvature factors were ranked second and third for the southern typical region.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eIn summary, this paper uses the Random Forest model to evaluate the landslide susceptibility along the Hefei-Fuzhou High-Speed Railway and the surrounding areas, and obtains the landslide susceptibility zonation map. These results highlight the dominant influence of topography on landslide susceptibility at a regional scale, while underscoring the enhanced role of precipitation as a key contributing factor to landslide in the southern region. This research provides a scientific basis for targeted geological hazards prevention along the Hefei-Fuzhou High-Speed Railway and offers a valuable methodological approach for regional infrastructure planning and geological risk management in similar environments.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003cstrong\u003eCompeting interests\u003c/strong\u003e \u003cp\u003eThe authors declare no conflicts of interest.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eEthical approval\u003c/strong\u003e \u003cp\u003eThis study did not involve any human participants, animal subjects, or sensitive data requiring ethical approval. No experiments, interventions, or interactions were conducted that necessitate formal review or approval by an ethics committee.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eJ.L.: Writing\u0026mdash;original draft preparation, prepared figures. J. L., W. Q., C. X. and Z. X. : Writing\u0026mdash;review and editing. W. Q.and C. X. : Project funding support. P.W., J. S., X. Z., J. C., Y. C., J.P., J. W. and Q. S.: Data acquisition. All authors reviewed the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThis work was supported by the National Key Research and Development Program of China (Grant No. 2024YFC3012603 and 2024YFC3012604), Chongqing Water Resources Bureau, China (Project No. CQS2400836) and Key Project of China Railway Design Corporation (Project No. 2023A0226409). Special thanks also go to the editors for their invaluable assistance in refining this work.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eIshwaran H, andMalley J D. Synthetic learning machines [J]. BioData mining, 2014, 7(1) : 1-12.\u003c/li\u003e\n\u003cli\u003eNnaji E K, Ubana G O, Enabulele E C, et al. Landslide Susceptibility Mapping in Nigeria Using Remote Sensing, GIS, and Machine Learning Models [J]. The Asian Review of Civil Engineering, 2024.\u003c/li\u003e\n\u003cli\u003eCaniani D, Pascale S D, Sdao F, et al. Neural networks and landslide susceptibility: a case study of the urban area of Potenza [J]. Natural Hazards, 2008, 45: 55-72.\u003c/li\u003e\n\u003cli\u003eCheng C, andFan L. An Attribution Deep Learning Interpretation Model for Landslide Susceptibility Mapping in the Three Gorges Reservoir Area [J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 1-15.\u003c/li\u003e\n\u003cli\u003eZhang K, Cui Y, Xu C, et al. Evaluation of Landslide Susceptibility to Heavy Rainfall in Sanming City,Fujian Province Based on Machine Learning Modeling [J]. Journal of Institute of Disaster Prevention, 2024, (3).\u003c/li\u003e\n\u003cli\u003eHong H, Tsangaratos P, Ilia I, et al. Comparing the Performance of a Logistic Regression and a Random Forest Model in Landslide Susceptibility Assessments. the Case of Wuyaun Area, China [C]. Fourth World Landslide Forum. 2017.\u003c/li\u003e\n\u003cli\u003eGuo Z, Shi Y, Huang F, et al. Landslide susceptibility zonation method based on C5.0 decision tree and K-means cluster algorithms to improve the efficiency of risk management [J]. Geoscience Frontiers, 2021.\u003c/li\u003e\n\u003cli\u003eHong H, Xu C, andChen W. Providing a Landslide Susceptibility Map in Nancheng County, China, by Implementing Support Vector Machines [J]. American Journal of Geographic Information System, 2017.\u003c/li\u003e\n\u003cli\u003eChen W, Guo C, Lin F, et al. Exploring advanced machine learning techniques for landslide susceptibility mapping in Yanchuan County, China [J]. Earth Science Informatics, 2024, 17(6) : 5385-5402.\u003c/li\u003e\n\u003cli\u003eQi W, Xu C, andXu X. AutoGluon: A revolutionary framework for landslide hazard analysis [J]. Natural Hazards Research, 2021, 1(3) : 103-108.\u003c/li\u003e\n\u003cli\u003eJebur M N, Pradhan B, andTehrany M S. Optimization of landslide conditioning factors using very high-resolution airborne laser scanning (LiDAR) data at catchment scale [J]. Remote Sensing of Environment, 2014, 152: 150-165.\u003c/li\u003e\n\u003cli\u003eYu B, Xing H, andYan J. Susceptibility assessment of multi-hazards using random forest\u0026mdash;back propagation neural network coupling model: a Hangzhou city case study [J]. Scientific Reports, 2024, 14(1).\u003c/li\u003e\n\u003cli\u003eLu Y, Xu H, Wang C, et al. A Novel Strategy Coupling Optimised Sampling with Heterogeneous Ensemble Machine-Learning to Predict Landslide Susceptibility [J]. Remote Sensing, 2024, 16(19).\u003c/li\u003e\n\u003cli\u003eYao W, Zhao J, andLi K. Assessment of Debris Flow Susceptibility Based on Information Gain Ratio-Support Vector Machine [J]. Urban Geotechnical Investigation \u0026amp; Surveying, 2023, (05) : 181-186.\u003c/li\u003e\n\u003cli\u003eChen B, Li C, Song C, et al. Automatic detection of active geohazards with millimeter-to-meter-scale deformation and quantitative analysis of factors influencing spatial distribution: A case study in the Hexi corridor, China [J]. International Journal of Applied Earth Observation and Geoinformation, 2024, 131: 103995-103995.\u003c/li\u003e\n\u003cli\u003eYang C, Liu L, Huang F, et al. Machine learning-based landslide susceptibility assessment with optimized ratio of landslide to non-landslide samples [J]. Gondwana Research, 2023, 123: 198-216.\u003c/li\u003e\n\u003cli\u003eHuang Y, Xu C, He X, et al. Landslides induced by the 2023 Jishishan Ms6.2 earthquake (NW China): spatial distribution characteristics and implication for the seismogenic fault [J]. npj Natural Hazards, 2025, 2(1) : 14-14.\u003c/li\u003e\n\u003cli\u003eChen C, Shen Z, Weng Y, et al. Modeling Landslide Susceptibility in Forest-Covered Areas in Lin\u0026apos;an, China, Using Logistical Regression, a Decision Tree, and Random Forests [J]. Remote Sensing, 2023, 15(18) : 19.\u003c/li\u003e\n\u003cli\u003eLiu X, Shao S, Zhang C, et al. Landslide susceptibility prediction in the loess tableland considering geomorphic evolution [J]. Catena, 2025, 249: 108668-108668.\u003c/li\u003e\n\u003cli\u003eReichenbach P. A review of statistically-based landslide susceptibility models [J]. Science Letter, 2018.\u003c/li\u003e\n\u003cli\u003eZheng Y, Chen J, Wang C, et al. Application of certainty factor and random forests model in landslide susceptibility evaluation in Mangshi City, Yunnan Province [J]. Bulletin of Geological Science and Technology, 2020, 39(6) : 131-144.\u003c/li\u003e\n\u003cli\u003eKim J-C, Lee S, Jung H-S, et al. Landslide susceptibility mapping using random forest and boosted tree models in Pyeong-Chang, Korea [J]. Geocarto International, 2018, 33(9) : 1000-1015.\u003c/li\u003e\n\u003cli\u003eLiu J, Xu C, Zhao B, et al. Deformation Slope Extraction and Influencing Factor Analysis Using LT-1 Satellite Data: A Case Study of Chongqing and Surrounding Areas, China [J]. Remote Sensing, 2025, 17(1) : 156-156.\u003c/li\u003e\n\u003cli\u003eWang J, Zang M, Xu C, et al. Landslide Susceptibility Assessment Following the 2022 Luding Earthquake: a coupled analytic hierarchy Process and area under the receiver operating characteristic curve algorithm [J]. Jouranl of Engineering Geology, 2024, 32(5) : 1696-1711.\u003c/li\u003e\n\u003cli\u003eViet T N, Nguyen D D, Duc M N, et al. Exploring deep learning models for roadside landslide prediction: Insights and implications from comparative analysis [J]. Physics and Chemistry of the Earth, 2024, 136: 103741-103741.\u003c/li\u003e\n\u003cli\u003eLi S, Yin C, Li J, et al. Landslide susceptibility assessment based on remote sensing interpretation and DBN-MLP model: a case study of Yiyuan County, China [J]. Stochastic Environmental Research Risk Assessment, 2025, 39(2).\u003c/li\u003e\n\u003cli\u003eLi L, Lu L, Sun X, et al. Study on the network structure and interactive relationship of tourism flow along High-Speed Railway: an example of Hefei-Fuzhou High-Speed Railway [J]. Human Geography, 2020, 35(1) : 9.\u003c/li\u003e\n\u003cli\u003eDai L, Yu S, andCai K. Analysis of Flood Control Measures for Fujian-Jiangxi Section of HefeiFuzhou High-speed Railway [J]. High Speed Railway Technology, 2023, 14(01) : 91-96.\u003c/li\u003e\n\u003cli\u003eGao H, Xu C, Xie C, et al. Landslides triggered by the July 2023 extreme rainstorm in the Haihe River Basin, China [J]. Landslides, 2024, (prepublish) : 1-6.\u003c/li\u003e\n\u003cli\u003eWang W, Huang Y-D, Xu C, et al. Identification and distribution of 13003 landslides in the northwest margin of Qinghai-Tibet Plateau based on human-computer interaction remote sensing interpretation [J]. China Geology, 2024, 7(2) : 171-187.\u003c/li\u003e\n\u003cli\u003eWang Q, Xu C, andXu J. Exploration of Landslide Geomorphology and Inventory Construction in Minhe County, Qinghai, China, Based on Google Earth Remote Sensing Imagery [J]. 2025, 29(1) : e13281.\u003c/li\u003e\n\u003cli\u003eJarvis A, Reuter H, Nelson A, et al. Hole-filled SRTM for the globe Version 4 [J]. 2008.\u003c/li\u003e\n\u003cli\u003eXu X, Han Z, Yang X, et al. Seismotectonic map in China and its adjacent regions. In: Beijing: Seismological Press. [Z]. In: Beijing: Seismological Press. 2016\u003c/li\u003e\n\u003cli\u003eWu X, Xu X, Yu G, et al. China Active Faults Database and its web system [J]. Earth System Science Data Discussions, 2023.\u003c/li\u003e\n\u003cli\u003eFick S E, andHijmans R J. WorldClim 2: new 1‐km spatial resolution climate surfaces for global land areas [J]. International Journal of Climatology, 2017, 37(12) : 4302-4315.\u003c/li\u003e\n\u003cli\u003eTateishi R, Uriyangqai B, Al-Bilbisi H, et al. Production of global land cover data - GLCNMO [J]. International journal of digital Earth, 2011, (1) : 4.\u003c/li\u003e\n\u003cli\u003eMa M, Wang J, Chen Q, et al. Regional Landslide Susceptibility Evaluation Based on Random Forest Weighting Information: a Case Study of Zigui to Badong Section in the Three Gorges Reservoir Area [J]. South China Geology, 2024, 40(04) : 749-763.\u003c/li\u003e\n\u003cli\u003eNhu V-H, Hoang N-D, Nguyen H, et al. Effectiveness assessment of Keras based deep learning with different robust optimization algorithms for shallow landslide susceptibility mapping at tropical area [J]. Catena, 2020, 188: 104458.\u003c/li\u003e\n\u003cli\u003eFawcett T. An introduction to ROC analysis [J]. Pattern Recognition Letters, 2006, 27(8) : 861-874.\u003c/li\u003e\n\u003cli\u003eBui D T, Tuan T A, Klempe H, et al. Spatial prediction models for shallow landslide hazards: a comparative assessment of the efficacy of support vector machines, artificial neural networks, kernel logistic regression, and logistic model tree [J]. Landslides, 2016, 13(2) : 361-378.\u003c/li\u003e\n\u003cli\u003eMeng Q. LightGBM: A Highly Efficient Gradient Boosting Decision Tree [C]. Neural Information Processing Systems. 2017.\u003c/li\u003e\n\u003cli\u003eYao J, Qin S, Qiao S, et al. Assessment of Landslide Susceptibility Combining Deep Learning with Semi-Supervised Learning in Jiaohe County, Jilin Province, China [J]. Applied Sciences, 2020, 10(16).\u003c/li\u003e\n\u003cli\u003eSun H, Li W, Scaioni M, et al. Influence of spatial heterogeneity on landslide susceptibility in the transboundary area of the Himalayas [J]. Geomorphology, 2023, 433.\u003c/li\u003e\n\u003cli\u003eHe X, andXu C. Spatial distribution and tectonic significance of the landslides triggered by the 2021 Ms6.4 Yangbi earthquake, Yunnan, China [J]. Frontiers in Earth Science, 2022, 10.\u003c/li\u003e\n\u003cli\u003eDeng J, Deng Y, Qiao S, et al. Spatial Distribution and Susceptibility Evaluation of Landslide Disasters in Loess Plateau Based on Information-CNN Coupling Model: A Case Study of Jiyuan City in Henan Province [J]. Geology and Resources, 2024, 33(06) : 836-845+835.\u003c/li\u003e\n\u003cli\u003eCui Y, Chen C, Zhu L, et al. Spatial distribution law of landslides and landslide susceptibility assessment in the eastern Himalayan syntaxis region [J]. Quarterly Journal of Engineering Geology and Hydrogeology, 2024, 57(3).\u003c/li\u003e\n\u003cli\u003eChen B, Zhang C, Li Z, et al. Developmental Characteristics and Controlling Factors of Landslides Triggered by Extreme Rainfalls on 16 June 2024 in Longyan, Fujian Province [J]. Geomatics and Information Science of Wuhan University, 2024, 49(11) : 2145-2155.\u003c/li\u003e\n\u003cli\u003eAchu A L, Gopinath G, andH. T J a C D V. Redefining landslide susceptibility under extreme rainfall events using deep learning [J]. Geomorphology, 2024, 448(Mar.1) : 1.1-1.18.\u003c/li\u003e\n\u003cli\u003eHuang C. Stability Evaluation of Rainfall-induced Landslide in Wuqu Village of Shouning County, Fujian Province [J]. Geology of Fujian, 2024, 43(04) : 297-305.\u003c/li\u003e\n\u003cli\u003eWang L, Deng Z, Xiao C, et al. Analysis of landslide temporal characteristics based on 50 years of landslide disaster data in Fujian Province [J]. Shanxi Architecture, 2025, 51(02) : 93-96+103.\u003c/li\u003e\n\u003cli\u003eZhu Q, Tan Q, Zhou J, et al. Risk Assessment of Regional Geological Landslide Hazards Based on GIS and AHP in Fu\u0026apos;an City [J]. Geomatics \u0026amp; Spatial Information Technology, 2021, 44(05) : 25-30.\u003c/li\u003e\n\u003cli\u003eMa S, Shao X, andXu C. Landslides Triggered by the 2016 Heavy Rainfall Event in Sanming, Fujian Province: Distribution Pattern Analysis and Spatio-Temporal Susceptibility Assessment [J]. Remote Sensing, 2023, 15(11) : 24.\u003c/li\u003e\n\u003cli\u003eLi Y, Mei H, Ren X, et al. Geological Disaster Susceptibility Evaluation Based on Certainty Factor and Support Vector Machine [J]. Journal of Geo-information Science, 2018, 20(12) : 1699-1709.\u003c/li\u003e\n\u003cli\u003eLuo X, Li C, andZhou J. GIS-based prediction method of shallow landslides induced by heavy rainfall in large mountainous areas [J]. Journal of Mountain Science, 2024, 21(5) : 1534-1548.\u003c/li\u003e\n\u003cli\u003eTorres-V\u0026aacute;zquez M \u0026Aacute;, Herrera S, Gincheva A, et al. Enhancing seasonal fire predictions with hybrid dynamical and random forest models [J]. npj Natural Hazards, 2025, 2(1) : 20.\u003c/li\u003e\n\u003cli\u003eDey H, Haque M M, Shao W, et al. Simulating flood risk in Tampa Bay using a machine learning driven approach [J]. npj Natural Hazards, 2024, 1(1) : 40.\u003c/li\u003e\n\u003cli\u003eHuang Y, Xu C, He X, et al. Distribution characteristics and cumulative effects of landslides triggered by multiple moderate-magnitude earthquakes: a case study of the comprehensive seismic impact area in Yibin, Sichuan, China [J]. Landslides, 2024, (prepublish) : 1-17.\u003c/li\u003e\n\u003cli\u003eLin Y-C, Hsieh J-Y, Shih H-S, et al. Strong wind is one of the important factors that trigger landslides [J]. npj Natural Hazards, 2025, 2(1) : 12.\u003c/li\u003e\n\u003cli\u003eDai Y, Li S, Zhang Y, et al. Continous\u0026ndash;discontinous analysis of an unstable slope: evolution of damage zones and potential influencing areas [J]. npj Natural Hazards, 2025, 2(1) : 23.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Landslide traces inventory, Landslide susceptibility, Geological hazards, Random Forest, Hefei-Fuzhou High-Speed Railway","lastPublishedDoi":"10.21203/rs.3.rs-6501951/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6501951/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eGeological hazards pose significant risks during the construction and operation of railways, demanding effective prevention and control measures to ensure operational safety. The Hefei-Fuzhou High-Speed Railway, a critical transportation artery, traverses complex geological terrain and diverse landforms, leading to prominent geological hazards risk along its route. This study focuses on the Huangshan-Fuzhou section of this railway, evaluating landslide susceptibility within the railway corridor using the Random Forest (RF) algorithm. Furthermore, we analyze spatial variations in the primary influencing factors by comparing two distinct sub-regions. The main findings are as follows: (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) A landslide susceptibility model developed using a historical landslide inventory and the RF algorithm demonstrated strong predictive performance, achieving an Area Under the Curve (AUC) value of 0.86. (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) Application of this model produced a landslide susceptibility zonation map, which revealed that approximately 30% of the study area is classified as high (16.92%) and very high (13.22%) susceptibility zones. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) Analysis of feature importance identified Slope (0.28), Relief (0.18), and Topographic Wetness Index (TWI, 0.11) as the most influential factors governing landslide susceptibility across the entire study area. (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e) Comparative analysis of factor importance between the northern and southern sub-regions revealed distinct patterns: the primary factors in the north were Slope (0.115), TWI (0.051), and Relief (0.050), whereas in the south, they were Slope (0.149), Rainfall (0.043), and Curvature (0.041). These results highlight the dominant influence of topography on landslide susceptibility at a regional scale, while underscoring the enhanced role of precipitation as a key contributing factor to landslide in the southern region. This research provides a scientific basis for targeted geological hazards prevention along the Hefei-Fuzhou High-Speed Railway and offers a valuable methodological approach for regional infrastructure planning and geological risk management in similar environments.\u003c/p\u003e","manuscriptTitle":"Evaluation of Geological Hazards Susceptibility along the Hefei-Fuzhou High- Speed Railway Based on Machine Learning Algorithms","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-06 08:59:35","doi":"10.21203/rs.3.rs-6501951/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-05-27T08:53:40+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-26T06:56:46+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-12T12:06:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"36033737308729822155607044277105449489","date":"2025-05-05T00:32:23+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"92115707194496179020650026174324911242","date":"2025-05-01T20:06:09+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-30T15:34:21+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-04-30T15:27:22+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-04-30T12:04:04+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-29T12:21:13+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-04-22T08:10:03+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d7495749-1f43-4131-93c7-7c153e73416d","owner":[],"postedDate":"May 6th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":48049102,"name":"Earth and environmental sciences/Natural hazards"},{"id":48049103,"name":"Earth and environmental sciences/Environmental social sciences/Sustainability"},{"id":48049104,"name":"Earth and environmental sciences/Environmental sciences/Environmental impact"}],"tags":[],"updatedAt":"2025-12-01T16:01:44+00:00","versionOfRecord":{"articleIdentity":"rs-6501951","link":"https://doi.org/10.1038/s41598-025-26496-x","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-11-27 15:57:35","publishedOnDateReadable":"November 27th, 2025"},"versionCreatedAt":"2025-05-06 08:59:35","video":"","vorDoi":"10.1038/s41598-025-26496-x","vorDoiUrl":"https://doi.org/10.1038/s41598-025-26496-x","workflowStages":[]},"version":"v1","identity":"rs-6501951","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6501951","identity":"rs-6501951","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}