Landslide susceptibility mapping using machine learning algorithms and comparison of their performance at Abha Basin,Asir Region,Saudi Arabia

The current study aimed at evaluating the capabilities of seven advanced machine learning techniques(MLTs),including,Support Vector Machine(SVM),Random Forest(RF),Multivariate Adaptive Regression Spline(MARS),Artificial Neural Network(ANN),Quadratic Discriminant Analysis(QDA),Linear Discriminant Analysis(LDA),and Naive Bayes(NB),for landslide susceptibility modeling and comparison of their performances.Coupling machine learning algorithms with spatial data types for landslide susceptibility mapping is a vitally important issue.This study was carried out using GIS and R open source software at Abha Basin,Asir Region,Saudi Arabia.First,a total of 243 landslide locations were identified at Abha Basin to prepare the landslide inventory map using different data sources.All the landslide areas were randomly separated into two groups with a ratio of 70%for training and 30%for validating purposes.Twelve landslide-variables were generated for landslide susceptibility modeling,which include altitude,lithology,distance to faults,normalized difference vegetation index(NDVI),landuse/landcover(LULC),distance to roads,slope angle,distance to streams,profile curvature,plan curvature,slope length(LS),and slope-aspect.The area under curve(AUC-ROC)approach has been applied to evaluate,validate,and compare the MLTs performance.The results indicated that AUC values for seven MLTs range from 89.0%for QDA to 95.1%for RF.Our findings showed that the RF(AUC=95.1%)and LDA(AUC=941.7%)have produced the best performances in comparison to other MLTs.The outcome of this study and the landslide susceptibility maps would be useful for environmental protection.     

Creating an advanced backpropagation neural network toolbox within GIS software

An artificial neural network (ANN) toolbox is created within GIS software for spatial interpolation, which will help GIS users to train and test ANNs, perform spatial analysis, and display results as a single process. The performance is compared to that of the open source Fast Artificial Neural Network library and conventional interpolation methods by creating digital elevation models (DEMs) given that nearly exact solutions exist. Simulation results show that the advanced backpropagations such as iRprop speed up the learning, while they can get stuck in a local minimum depending on initial weight sets. Besides, the division of input–output examples into training and test data affects the accuracy, particularly when the distribution of the examples is skewed and peaked, and the number of data is small. ANNs, however, show the similar performance to inversed distance weighted or kriging and outperform polynomial interpolations as a global interpolation method in high-dimensional data. In addition, the neural network residual kriging (NNRK) model, which combines the ANN toolbox and kriging within GIS software, is performed. The NNRK outperforms conventional methods and well captures global trends and local variations. A key outcome of this work is that the ANN toolbox created within the de facto standard GIS software is applicable to various spatial analysis including hazard risk assessment over a large area, in particular when there are multiple potential causes, the relationship between risk factors and hazard events is not clear, and the number of available data is small given its performance for DEM generation.     

Landslide susceptibility mapping at Vaz Watershed (Iran) using an artificial neural network model: a comparison between multilayer perceptron (MLP) and radial basic function (RBF) algorithms

Landslide susceptibility and hazard assessments are the most important steps in landslide risk mapping. The main objective of this study was to investigate and compare the results of two artificial neural network (ANN) algorithms, i.e., multilayer perceptron (MLP) and radial basic function (RBF) for spatial prediction of landslide susceptibility in Vaz Watershed, Iran. At first, landslide locations were identified by aerial photographs and field surveys, and a total of 136 landside locations were constructed from various sources. Then the landslide inventory map was randomly split into a training dataset 70 % (95 landslide locations) for training the ANN model and the remaining 30 % (41 landslides locations) was used for validation purpose. Nine landslide conditioning factors such as slope, slope aspect, altitude, land use, lithology, distance from rivers, distance from roads, distance from faults, and rainfall were constructed in geographical information system. In this study, both MLP and RBF algorithms were used in artificial neural network model. The results showed that MLP with Broyden–Fletcher–Goldfarb–Shanno learning algorithm is more efficient than RBF in landslide susceptibility mapping for the study area. Finally the landslide susceptibility maps were validated using the validation data (i.e., 30 % landslide location data that was not used during the model construction) using area under the curve (AUC) method. The success rate curve showed that the area under the curve for RBF and MLP was 0.9085 (90.85 %) and 0.9193 (91.93 %) accuracy, respectively. Similarly, the validation result showed that the area under the curve for MLP and RBF models were 0.881 (88.1 %) and 0.8724 (87.24 %), respectively. The results of this study showed that landslide susceptibility mapping in the Vaz Watershed of Iran using the ANN approach is viable and can be used for land use planning.     

Susceptibility assessments of the Wenchuan earthquake-triggered landslides in Longnan using logistic regression

In this study a Wenchuan earthquake-induced landslide susceptibility assessment was carried out in the Longnan area in northwestern China using a GIS-based logistic regression model. This region has frequently been affected by landslides in the past, and was intensively affected by the 5.12 Wenchuan earthquake which received considerable international attention. The data used for this study consist of the landslides triggered by the Wenchuan earthquake and a landslide pre-disposing factor database. Information regarding the landslide causative factors came from additional data sources, such as a digital elevation model (DEM) with a 30 × 30 m² resolution, orthophotos, geological and land-use maps, precipitation records, and information on peak ground acceleration data from the 2008 earthquake. The statistical analysis of the relationship between the Wenchuan earthquake-triggered landslides and pre-disposing factors showed the great influence of lithological and topographical conditions on slope failures. The quality of susceptibility mapping was validated by splitting the study area into training and validation sections. The prediction capability analysis demonstrated that the landslide susceptibility map could be used for land planning as well as emergency planning by local authorities.     

Regional-scale landslide activity and landslide susceptibility zonation in the Nepal Himalaya

Landslide susceptibility zonation mapping is a fundamental procedure for geo-disaster management in tropical and sub-tropical regions. Recently, various landslide susceptibility zonation models have been introduced in Nepal with diverse approaches of assessment. However, validation is still a problem. Additionally, the role of various predisposing causative parameters for landslide activity is still not well understood in the Nepal Himalaya. To address these issues of susceptibility zonation and landslide activity, about 4,000 km² area of central Nepal was selected for regional-scale assessment of landslide activity and susceptibility zonation mapping. In total, 655 new landslides and 9,229 old landslides were identified with the study area with the help of satellite images, aerial photographs, field data and available reports. The old landslide inventory was “blind landslide database” and could not explain the particular rainfall event responsible for the particular landslide. But considering size of the landslide, blind landslide inventory was reclassified into two databases: short-duration high-intensity rainfall-induced landslide inventory and long-duration low-intensity rainfall-induced landslide inventory. These landslide inventory maps were considered as proxy maps of multiple rainfall event-based landslide inventories. Similarly, all 9,884 landslides were considered for the activity assessment of predisposing causative parameters. For the Nepal Himalaya, slope, slope aspect, geology and road construction activity (anthropogenic cause) were identified as most affective predisposing causative parameters for landslide activity. For susceptibility zonation, multivariate approach was considered and two proxy rainfall event-based landslide databases were used for the logistic regression modelling, while a relatively recent landslide database was used in validation. Two event-based susceptibility zonation maps were merged and rectified to prepare the final susceptibility zonation map and its prediction rate was found to be more than 82 %. From this work, it is concluded that rectification of susceptibility zonation map is very appropriate and reliable. The results of this research contribute to a significant improvement in landslide inventory preparation procedure, susceptibility zonation mapping approaches as well as role of various predisposing causative parameters for the landslide activity.     

Spatial distribution analysis and susceptibility mapping of landslides triggered before and after Mw7.8 Gorkha earthquake along Upper Bhote Koshi,Nepal

The 2015 Mw7.8 Gorkha earthquake triggered thousands of landslides of various types scattered over a large area. In the current study, we utilized pre- and post-earthquake high-resolution satellite imagery to compile two landslide inventories before and after earthquake and prepared three landslide susceptibility maps within 404 km² area using frequency ratio (FR) model. From the study, we could map about 519 landslides including 178 pre-earthquake slides and 341 coseismic slides were identified. This study investigated the relationship between landslide occurrence and landslide causative factors, i.e., slope, aspect, altitude, plan curvature, lithology, land use, distance from streams, distance from road, distance from faults, and peak ground acceleration. The analysis showed that the majority of landslides both pre-earthquake and coseismic occurred at slope >30°, preferably in S, SE, and SW directions and within altitude ranging from 1000 to 1500 m and 1500 to 3500 m. Scatter plots between number of landslides per km^?2 (LN) and percentage of landslide area (LA) and causative factors indicate that slope is the most influencing factor followed by lithology and PGA for the landslide formation. Higher landslide susceptibility before earthquake is observed along the road and rivers, whereas landslides after earthquake are triggered at steeper slopes and at higher altitudes. Combined susceptibility map indicates the effect of topography, geology, and land cover in the triggering of landslides in the entire basin. The resultant landslide susceptibility maps are verified through AUC showing success rates of 78, 81, and 77%, respectively. These susceptibility maps are helpful for engineers and planners for future development work in the landslide prone area.     

Artificial neural network coupled with wavelet transform for estimating snow water equivalent using passive microwave data

Snow Water Equivalent (SWE) is an important parameter in hydrologic engineering involving the streamflow forecasting of high-elevation watersheds. In this paper, the application of classic Artificial Neural Network model (ANN) and a hybrid model combining the wavelet and ANN (WANN) is investigated in estimating the value of SWE in a mountainous basin. In addition, k-fold cross validation method is used in order to achieve a more reliable and robust model. In this regard, microwave images acquired from Spectral Sensor Microwave Imager (SSM/I) are used to estimate the SWE of Tehran sub-basins during 1992–2008 period. Also for obtaining measured SWE within the corresponding Equal-Area Scalable Earth-Grid (EASE-Grid) cell of SSM/I image, approach of Cell-SWE extraction using height–SWE relations is applied in order to reach more precise estimations. The obtained results reveal that the wavelet-ANN model significantly increases the accuracy of estimations, mainly because of using multi-scale time series as the ANN inputs. The Nash–Sutcliffe Index (NSE) for ANN and WANN models are respectively 0.09 and 0.44 which shows a firm improvement of 0.35 in NSE parameter when WANN is applied. Similar trend is observed in other parameters including RMSE where the value is 0.3 for ANN and 0.07 for WANN.     

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

Preparation of landslide susceptibility maps is considered as the first important step in landslide risk assessments, but these maps are accepted as an end product that can be used for land use planning. The main objective of this study is to explore some new state-of-the-art sophisticated machine learning techniques and introduce a framework for training and validation of shallow landslide susceptibility models by using the latest statistical methods. The Son La hydropower basin (Vietnam) was selected as a case study. First, a landslide inventory map was constructed using the historical landslide locations from two national projects in Vietnam. A total of 12 landslide conditioning factors were then constructed from various data sources. Landslide locations were randomly split into a ratio of 70:30 for training and validating the models. To choose the best subset of conditioning factors, predictive ability of the factors were assessed using the Information Gain Ratio with 10-fold cross-validation technique. Factors with null predictive ability were removed to optimize the models. Subsequently, five landslide models were built using support vector machines (SVM), multi-layer perceptron neural networks (MLP Neural Nets), radial basis function neural networks (RBF Neural Nets), kernel logistic regression (KLR), and logistic model trees (LMT). The resulting models were validated and compared using the receive operating characteristic (ROC), Kappa index, and several statistical evaluation measures. Additionally, Friedman and Wilcoxon signed-rank tests were applied to confirm significant statistical differences among the five machine learning models employed in this study. Overall, the MLP Neural Nets model has the highest prediction capability (90.2 %), followed by the SVM model (88.7 %) and the KLR model (87.9 %), the RBF Neural Nets model (87.1 %), and the LMT model (86.1 %). Results revealed that both the KLR and the LMT models showed promising methods for shallow landslide susceptibility mapping. The result from this study demonstrates the benefit of selecting the optimal machine learning techniques with proper conditioning selection method in shallow landslide susceptibility mapping.     

Relative effect method of landslide susceptibility zonation in weathered granite soil: a case study in Deokjeok-ri Creek,South Korea

The objective of this study was to produce and evaluate a landslide susceptibility map for weathered granite soils in Deokjeok-ri Creek, South Korea. The relative effect (RE) method was used to determine the relationship between landslide causative factors (CFs) and landslide occurrence. To determine the effect of CFs on landslides, data layers of aspect, elevation, slope, internal relief, curvature, distance to drainage, drainage density, stream power index, sediment transport index, topographic wetness index, soil drainage character, soil type, soil depth, forest type, timber age, and geology were analyzed in a geographical information system (GIS) environment. A GIS-based landslide inventory map of 748 landslide locations was prepared using data from previous reports, aerial photographic interpretation, and extensive field work. A RE model was generated from a training set consisting of 673 randomly selected landslides in the inventory map, with the remaining 75 landslides used for validation of the susceptibility map. The results of the analysis were verified using the landslide location data. According to the analysis, the RE model had a success rate of 86.3 % and a predictive accuracy of 88.6 %. The validation results showed satisfactory agreement between the susceptibility map and the existing data on landslide locations. The results of this study can therefore be used to mitigate landslide-induced hazards and to plan land use.     

A test of transferability for landslides susceptibility models under extreme climatic events: application to the Messina 2009 disaster

A model building strategy is tested to assess the susceptibility for extreme climatic events driven shallow landslides. In fact, extreme climatic inputs such as storms typically are very local phenomena in the Mediterranean areas, so that with the exception of recently stricken areas, the landslide inventories which are required to train any stochastic model are actually unavailable. A solution is here proposed, consisting in training a susceptibility model in a source catchment, which was implemented by applying the binary logistic regression technique, and exporting its predicting function (selected predictors regressed coefficients) in a target catchment to predict its landslide distribution. To test the method, we exploit the disaster that occurred in the Messina area (southern Italy) on 1 October 2009 where, following a 250-mm/8-h storm, approximately two thousand debris flow/debris avalanches landslides in an area of 21 km² triggered, killing 37 people and injuring more than 100, and causing 0.5 M € worth of structural damage. The debris flows and debris avalanches phenomena involved the thin weathered mantle of the Varisican low to high-grade metamorphic rocks that outcrop in the eastern slopes of the Peloritani Mounts. Two 10-km²-wide stream catchments, which are located inside the storm core area, were exploited: susceptibility models trained in the Briga catchment were tested when exported to predict the landslides distribution in the Giampilieri catchment. The prediction performance (based on goodness of fit, prediction skill, accuracy and precision assessment) of the exported model was then compared with that of a model prepared in the Giampilieri catchment exploiting its landslide inventory. The results demonstrate that the landslide scenario observed in the Giampilieri catchment can be predicted with the same high performance without knowing its landslide distribution: we obtained, in fact, a very poor decrease in predictive performance when comparing the exported model to the native random partition-based model.     

Landslide susceptibility mapping using Bayesian approach in the Sultan Mountains (Akşehir,Turkey)

Landslides cause heavy damage to property and infrastructure, in addition to being responsible for the loss of human lives in many parts of the Turkey. The paper presents GIS-based spatial data analysis for landslide susceptibility mapping in the regions of the Sultan Mountains, West of Akşehir, and central part of Turkey. Landslides occur frequently in the area and seriously affect local living conditions. Therefore, spatial analysis of landslide susceptibility in the Sultan Mountains is important. The relationships between landslide distributions with the 19 landslide affecting parameters were analysed using a Bayesian model. In the study area, 90 landslides were observed. The landslides were randomly subdivided into 80 training landslides and 10 test landslides. A landslide susceptibility map was produced by using the training landslides. The test landslides were used in the accuracy control of the produced landslide susceptibility map. Approximately 9% of the study area was classified as high susceptibility zone. Medium, low and very low susceptibility zones covered 8, 23 and 60% of the study area, respectively. Most of the locations of the observed landslides actually fall into moderate (17.78%) and high (77.78. %) susceptibility zones of the produced landslide susceptibility map. This validates the applicability of proposed methods, approaches and the classification scheme. The high susceptibility zone is along both sides of the Akşehir Fault and at the north-eastern slope of the Sultan Mountains. It was determined that the surface area of the Harlak and Deresenek formations, which have attained lithological characteristics of clayey limestone with a broken and separated base, and where area landslides occur, possesses an elevation of 1,100–1,600 m, a slope gradient of 25°–35° and a slope aspect of 22.5°–157.5° facing slopes.     

Use of artificial neural network for spatial rainfall analysis

In the present study, the precipitation data measured at 23 rain gauge stations over the Achaia County, Greece, were used to estimate the spatial distribution of the mean annual precipitation values over a specific catchment area. The objective of this work was achieved by programming an Artificial Neural Network (ANN) that uses the feed-forward back-propagation algorithm as an alternative interpolating technique. A Geographic Information System (GIS) was utilized to process the data derived by the ANN and to create a continuous surface that represented the spatial mean annual precipitation distribution. The ANN introduced an optimization procedure that was implemented during training, adjusting the hidden number of neurons and the convergence of the ANN in order to select the best network architecture. The performance of the ANN was evaluated using three standard statistical evaluation criteria applied to the study area and showed good performance. The outcomes were also compared with the results obtained from a previous study in the area of research which used a linear regression analysis for the estimation of the mean annual precipitation values giving more accurate results. The information and knowledge gained from the present study could improve the accuracy of analysis concerning hydrology and hydrogeological models, ground water studies, flood related applications and climate analysis studies.     

Spatial landslide susceptibility assessment using machine learning techniques assisted by additional data created with generative adversarial networks

In recent years,landslide susceptibility mapping has substantially improved with advances in machine learning.However,there are still challenges remain in landslide mapping due to the availability of limited inventory data.In this paper,a novel method that improves the performance of machine learning techniques is presented.The proposed method creates synthetic inventory data using Generative Adversarial Networks(GANs)for improving the prediction of landslides.In this research,landslide inventory data of 156 landslide locations were identified in Cameron Highlands,Malaysia,taken from previous projects the authors worked on.Elevation,slope,aspect,plan curvature,profile curvature,total curvature,lithology,land use and land cover(LULC),distance to the road,distance to the river,stream power index(SPI),sediment transport index(STI),terrain roughness index(TRI),topographic wetness index(TWI)and vegetation density are geo-environmental factors considered in this study based on suggestions from previous works on Cameron Highlands.To show the capability of GANs in improving landslide prediction models,this study tests the proposed GAN model with benchmark models namely Artificial Neural Network(ANN),Support Vector Machine(SVM),Decision Trees(DT),Random Forest(RF)and Bagging ensemble models with ANN and SVM models.These models were validated using the area under the receiver operating characteristic curve(AUROC).The DT,RF,SVM,ANN and Bagging ensemble could achieve the AUROC values of(0.90,0.94,0.86,0.69 and 0.82)for the training;and the AUROC of(0.76,0.81,0.85,0.72 and 0.75)for the test,subsequently.When using additional samples,the same models achieved the AUROC values of(0.92,0.94,0.88,0.75 and 0.84)for the training and(0.78,0.82,0.82,0.78 and 0.80)for the test,respectively.Using the additional samples improved the test accuracy of all the models except SVM.As a result,in data-scarce environments,this research showed that utilizing GANs to generate supplementary samples is promising because it can improve the predictive capability of common landslide prediction models.     

基于GIS与ANN模型的地震滑坡易发性区划

基于遥感数据、地理信息系统(GIS)技术和人工神经网络(ANN)模型,开展地震滑坡易发性区划研究.2010年4月14日玉树地震后,基于航片与卫星影像目视解译,并辅以野外调查的方法,在地震区圈定了2036处地震诱发滑坡.选择高程、坡度、坡向、斜坡曲率、坡位、与水系距离、地层岩性、与断裂距离、与公路距离、归一化植被指数(NDVI)、与同震地表破裂距离、地震动峰值加速度(PGA)共12个因子作为地震滑坡易发性评价因子.这些因子均是应用GIS技术与遥感影像处理技术,基于地形数据、地质数据、遥感数据得到.训练样本中的滑动样本有两组,一组是滑坡区整个单滑坡体的质心位置,另一组是滑坡滑源区滑前的坡体高程最高的位置.应用这12个影响因子,分别采用这两组评价样本,基于ANN模型建立地震滑坡易发性索引图,基于GIS工具建立地震滑坡易发性分级图.分别应用训练样本中滑坡分布的点数据去检验各自的结果正确率,正确率分别为81.53％与81.29％,表明ANN模型是一种高效科学的地震滑坡易发性区划模型.     

Landslide susceptibility mapping using analytical hierarchy process (AHP) in Tehri reservoir rim region,Uttarakhand

A comprehensive use of analytical hierarchy process (AHP) method in landslide susceptibility mapping (LSM) has been presented for rim region of Tehri reservoir. Using remote sensing data, various landslide causative factors responsible for inducing instability in the area were derived. Ancillary data such as geological map, soil map, and topographic map were also considered along with remote sensing data. Exhaustive field checks were performed to define the credibility of the random landslide conditioning factors considered in this study. Apart from universally acceptable inherent causative factors used in the susceptibility mapping, others such as impact of reservoir impoundment on terrain, topographic wetness index and stream power index were found to be important causative factors in rim region of the Tehri reservoir. The AHP method was used to acquire weights of factors and their classes respectively. Weights achieved from AHP method matched with the existing field conditions. Acceptable consistency ratio (CR) value was achieved for each AHP matrix. Weights of each factor were integrated with weighted sum technique and a landslide susceptibility index map was generated. Jenk’s natural break classifier was used to classify LSI map into very low, low, moderate, high and very high landslide susceptible classes. Validation of the susceptibility map was performed using cumulative percentage/success rate curve technique. Area under curve value of the success rate curve was converted to percentage validation accuracy and a reasonable 78.7% validation accuracy was achieved.     

Landslide susceptibility assessment by Dempster–Shafer and Index of Entropy models,Sarkhoun basin,Southwestern Iran

Landslides are natural disasters often activated by interaction of different controlling environmental factors, especially in mountainous terrains. In this research, the landslide susceptibility map was developed for the Sarkhoun catchment using Index of Entropy (IoE) and Dempster–Shafer (DS) models. For this purpose, 344 landslides were mapped in GIS environment. 241 (70%) out of the landslides were selected for the modeling and the remaining (30%) were employed for validation of the models. Afterward, 10 landslide conditioning factor layers were prepared including land use, distance to drainage, slope gradient, altitude, lithology, distance to roads, distance to faults, slope aspect, Topography Wetness Index, and Stream Power Index. The relationship between the landslide conditioning factors and landslide inventory maps was determined using the IoE and DS models. In order to verify the models, the results were compared with validation landslide data not employed in training process of the models. Accordingly, Receiver Operating Characteristic (ROC) curves were applied, and Area Under the Curve (AUC) was calculated for the obtained susceptibility maps using the success (training data) and prediction (validation data) rate curves. The land use was found to be the most important factor in the study area. The AUC are 0.82, and 0.81 for success rates of the IoE, and DS models, respectively, while the prediction rates are 0.76 and 0.75. Therefore, the results of the IoE model are more accurate than the DS model. Furthermore, a satisfactory agreement is observed between the generated susceptibility maps by the models and true location of the landslides.     

Application of an analytical hierarchical process approach for landslide susceptibility mapping in A Luoi district,Thua Thien Hue Province,Vietnam

A Luoi is a Vietnamese–Laotian border district situated in the western part of Thua Thien Hue province, central Vietnam, where landslides occur frequently and seriously affect local living conditions. This study focuses on the spatial analysis of landslide susceptibility in this 263-km² area. To analyze landslide manifestation in the study area, causative factor maps are derived of slope angle, weathering, land use, geomorphology, fault density, geology, drainage distance, elevation, and precipitation. The analytical hierarchical process approach is used to combine these maps for landslide susceptibility mapping. A landslide susceptibility zonation map with four landslide susceptibility classes, i.e. low, moderate, high, and very high susceptibility for landsliding, is derived based on the correspondence with an inventory of observed landslides. The final map indicates that about 37% of the area is very highly susceptible for landsliding and about 22% is highly susceptible, which means that more than half of the area should be considered prone to landsliding.     

GIS-based landslide susceptibility mapping using bivariate statistical analysis in Devrek (Zonguldak-Turkey)

Devrek town with increasing population is located in a hillslope area where some landslides exist. Therefore, landslide susceptibility map of the area is required. The purpose of this study was to generate a landslide susceptibility map using a bivariate statistical index and evaluate and compare the results of the statistical analysis conducted with three different approaches in seed cell concept resulting in different data sets in Geographical Information Systems (GIS) based landslide susceptibility mapping applied to the Devrek region. The data sets are created from the seed cells of (a) crowns and flanks, (b) only crowns, and (c) only flanks of the landslides by using ten different causative parameters of the study area. To increase the data dependency of the analysis, all parameter maps are classified into equal frequency classes based directly on the percentile divisions of each corresponding seed cell data set. The resultant maps of the landslide susceptibility analysis indicate that all data sets produce fairly acceptable results. In each data set analysis, elevation, lithology, slope, aspect, and drainage density parameters are found to be the most contributing factors in landslide occurrences. The results of the three data sets are compared using Seed Cell Area Indexes (SCAI). This comparison shows that the crown data set produces the most accurate and successful landslide susceptibility map of the study area.     

Landslide susceptibility mapping using rough sets and back-propagation neural networks in the Three Gorges, China

In the Three Gorges of China, there are frequent landslides, and the potential risk of landslides is tremendous. An efficient and accurate method of generating landslide susceptibility maps is very important to mitigate the loss of lives and properties caused by these landslides. This paper presents landslide susceptibility mapping on the Zigui-Badong of the Three Gorges, using rough sets and back-propagation neural networks (BPNNs). Landslide locations were obtained from a landslide inventory map, supported by field surveys. Twenty-two landslide-related factors were extracted from the 1:10,000-scale topographic maps, 1:50,000-scale geological maps, Landsat ETM + satellite images with a spatial resolution of 28.5 m, and HJ-A satellite images with a spatial resolution of 30 m. Twelve key environmental factors were selected as independent variables using the rough set and correlation coefficient analysis, including elevation, slope, profile curvature, catchment aspect, catchment height, distance from drainage, engineering rock group, distance from faults, slope structure, land cover, topographic wetness index, and normalized difference vegetation index. The initial, three-layered, and four-layered BPNN were trained and then used to map landslide susceptibility, respectively. To evaluate the models, the susceptibility maps were validated by comparing with the existing landslide locations according to the area under the curve. The four-layered BPNN outperforms the other two models with the best accuracy of 91.53 %. Approximately 91.37 % of landslides were classified as high and very high landslide-prone areas. The validation results show sufficient agreement between the obtained susceptibility maps and the existing landslide locations.