Landslide Samples Research Articles

Landslide Detection Using Densely Connected Convolutional Networks and Environmental Conditions

A complete and accurate landslide map is necessary for landslide susceptibility and risk assessment. Currently, deep learning faces the dilemma of insufficient application, scarce samples, and poor efficiency in landslide recognition. This article utilizes the advantages of dense convolutional networks (DenseNets) and their modified technique to solve the three proposed problems. For this purpose, we created a new landslide sample library. On the original remote sensing image, 12 geological, topographic, hydrological and land cover factors that can directly or indirectly reflect the landslide are superimposed. Then, landslide detection was carried out in the three Gorges reservoir area in China to test the performance of the improved method. The quantitative evaluation of the landslide detection map shows that the combination of environmental factors and DenseNet can improve the accuracy of the detection model. Compared with the optical image, kappa and F1 increased by 9.7% and 9.1% respectively. Compared with other traditional neural networks and machine learning algorithms, DenseNet has the highest kappa and F1 values. Based on the base Densenet, through data augmentation and fine-tuning optimization technology, the kappa and F1 values reach the highest values of 0.9474 and 0.9505, respectively. The proposed method has promising applicability in large area landslide identification scenarios.

Landslide Susceptibility Mapping Using Statistical Methods along the Asian Highway, Bhutan

In areas prone to frequent landslides, the use of landslide susceptibility maps can greatly aid in the decision-making process of the socio-economic development plans of the area. Landslide susceptibility maps are generally developed using statistical methods and geographic information systems. In the present study, landslide susceptibility along road corridors was considered, since the anthropogenic impacts along a road in a mountainous country remain uniform and are mainly due to road construction. Therefore, we generated landslide susceptibility maps along 80.9 km of the Asian Highway (AH48) in Bhutan using the information value, weight of evidence, and logistic regression methods. These methods have been used independently by some researchers to produce landslide susceptibility maps, but no comparative analysis of these methods with a focus on road corridors is available. The factors contributing to landslides considered in the study are land cover, lithology, elevation, proximity to roads, drainage, and fault lines, aspect, and slope angle. The validation of the method performance was carried out by using the area under the curve of the receiver operating characteristic on training and control samples. The area under the curve values of the control samples were 0.883, 0.882, and 0.88 for the information value, weight of evidence, and logistic regression models, respectively, which indicates that all models were capable of producing reliable landslide susceptibility maps. In addition, when overlaid on the generated landslide susceptibility maps, 89.3%, 85.6%, and 72.2% of the control landslide samples were found to be in higher-susceptibility areas for the information value, weight of evidence, and logistic regression methods, respectively. From these findings, we conclude that the information value method has a better predictive performance than the other methods used in the present study. The landslide susceptibility maps produced in the study could be useful to road engineers in planning landslide prevention and mitigation works along the highway.

Comparative landslide spatial research based on various sample sizes and ratios in Penang Island, Malaysia

This paper aims to compare and develop the influence on different sample sizes and sample ratios when using machine learning (ML) models, i.e., support vector machine (SVM) and artificial neural network (ANN), to produce landslide susceptibility maps (LSMs) in Penang Island, Malaysia. At the same time, traditional statistical (TS) models are also considered to produce LSMs in this comparative research. The receiver operating characteristic (ROC) curve and recall metric are applied to evaluate the model’s performance. Based on the evaluation criteria, the ML model outperforms the TS models and the ML models trained using the datasets with larger sample size give a better performance. ML models, especially SVM models, have better performance when training with balanced datasets as well as the datasets of more landslide sample data. Kruskal-Wallis test and Mann-Whitney U test are applied to test the significance. The results indicate that sample size and sample ratio are essential factors when considering ML models to produce LSMs. The LSMs produced in this research can provide valid and useful information to the local authorities for landslide mitigation and prediction.

Three oversampling methods applied in a comparative landslide spatial research in Penang Island, Malaysia

Two main problems in landslide spatial prediction research are the lack of landslide samples (minority) to train the models and the misunderstanding of assigning equal costs to different misclassifications. In order to handle the problems properly, the research is conducted based on two main objectives, which are to augment the landslide sample data in an efficient way and to assign proper unequal costs to the two types of error when training and evaluating models. Resampling techniques, including random oversampling technique, synthetic minority oversampling technique and self-creating oversampling technique (SCOTE), are used to augment the minority class samples. Logistic regression (LR) and support vector machine (SVM) are used for landslide spatial classification. Receiver operating characteristic and cost curves are used to evaluate the models. The results show that the SVM models trained using the dataset generated by SCOTE with sample size of 10,000 have the best prediction performance. The nonparametric test, Kruskal–Wallis test, is used to test the difference of sample size between different groups, which shows that LR models are more sensitive to the change of sample size. Two landslide susceptibility maps are produced based on the models with the best prediction performance. The verification results show that the maps both successfully predict more than 86% of the susceptible area, which can provide valid information on landslide mitigation and prediction to the local authorities.

Susceptibility mapping of damming landslide based on slope unit using frequency ratio model

In the mountainous area of southwest China, there is a large number of large hydropower stations that have already been built or under construction. The construction and the safety of the operation of hydropower stations directly affect the residents’ safety of life and property on both sides of the downstream river. Landslide hazards are one of the main problems affecting the construction and operation of hydropower stations. Therefore, it is currently more urgent to evaluate landslide susceptibility within the basin. The goal of this paper is to carry out landslide susceptibility mapping that affects the construction and operation of hydropower stations. In the case of a large research area and lack of landslide hazard data, this paper selects the sample of blocking river landslide in a larger area with similar topography and geological conditions and applies the quantitative method to carry out landslide sensitivity mapping. In this paper, the slope unit works as the evaluation unit, 191 landslide sample data were obtained, and nine landslide conditioning factors (slope, aspect, curvature, lithology, distance to fault, distance to the river, land use, soil, and the topographic wetness index (TWI)) were obtained. Combined with the quantitative frequency model (FR model), the landslide susceptibility map was obtained. The research process and results show that (1) when the study area is large, the landslide data is scarce, and it is difficult to get the landslide inventory map; the landslide samples can be selected in a larger area with similar topography and geological conditions. It is feasible to use the analysis data of landslide samples to evaluate the landslide susceptibility; (2) for specific industry applications such as hydropower station operation safety, the landslide evaluation range can be reduced by the spatial correlation between landslide aspect and river flow direction, which is beneficial to the management department to reduce the scope of attention; and (3) the verification results show that the slope unit can be used for quantitative evaluation of landslide sensitivity based on the FR model and can obtain the highest evaluation accuracy. The main advantage of this method is its capability to carry out landslide susceptibility mapping with quantitative methods and to ensure that higher mapping accuracy is obtained in areas where landslide hazard data is lacking.

Landslide susceptibility prediction based on a semi-supervised multiple-layer perceptron model

Conventional supervised and unsupervised machine learning models used for landslide susceptibility prediction (LSP) have many drawbacks, such as an insufficient number of recorded landslide samples, and the subjective and random selection of non-landslide samples. To overcome these drawbacks, a semi-supervised multiple-layer perceptron (SSMLP) is innovatively proposed with several processes: (1) an initial landslide susceptibility map (LSM) is produced using the multiple-layer perceptron (MLP) based on the original recorded landslide samples and related environmental factors; (2) the initial LSM is respectively classified into five areas with very high, high, moderate, low and very low susceptible levels; (3) some reasonable grid units from the areas with very high susceptible level are selected as new landslide samples to expand the original landslide samples; (4) reasonable non-landslide samples are selected from the areas with very low susceptible level; and (5) the expanded landslide samples, reasonable selected non-landslide samples and related environmental factors are put into the MLP once again to predict the final LSM. The Xunwu County of Jiangxi Province in China is selected as the study area. Conventional supervised machine learning (i.e. MLP) and unsupervised machine learning (i.e. K-means clustering model) are selected for comparisons. The comparative results indicate that the SSMLP model has a considerably higher LSP performance than the MLP and K-means clustering in Xunwu County. The SSMLP model successfully addresses the drawbacks existed in the conventional machine learning for LSP.

Landslide susceptibility assessment based on an incomplete landslide inventory in the Jilong Valley, Tibet, Chinese Himalayas

The Himalayan range is one of the most tectonically active mountain ranges on Earth. The Jilong Valley is a deeply-dissected canyon in the Chinese Himalayas in Tibet, and in this valley, landslide hazard and risk have increased significantly in response to active crustal movements, intense rainfall and ever-increasing human interference. Landslide susceptibility assessment in the valley is fundamental for risk mitigation and to inform land use and planning but remains a challenge owing to the inaccessible high altitude and incomplete landslide inventory. In addition, because of sustained glacial erosion, weathering, and denudation, there is considerable uncertainty in the use of remote sensing interpretation of landslides in high-altitude terrain. In this paper, therefore, a systematic approach for landslide susceptibility assessment is proposed, combining the interpretations from remotely sensed dataset and both heuristic and statistical susceptibility models to overcome the problem of the limited spatial coverage of landslide data and uncertainty in landslide interpretation. The main steps include: 1) landslide field investigation and interpretation and the compilation of a preliminary landslide inventory, including certain and probable landslides, 2) knowledge-driven identification of landslide-prone areas based on the heuristic model, 3) classification and quantification of the uncertainty of probable landslides through heuristic landslide-prone areas mapping and the production of a revised landslide inventory, and 4) data-driven susceptibility assessment using a statistical model. In the statistical model, the landslide samples have multivalent dependent variables between 0 and 1. Therefore, a multinomial statistical classifier and multiclass Receiver Operation Characteristic curves are needed for model calibration and validation. The statistical susceptibility mapping showed good performance in the study area, with an average AUC of 0.867, which is a significant improvement over the heuristic model. The methodology presented can be used for landslide susceptibility mapping in data-scarce environments, not only in the Himalayan region but also in other mountainous areas where there is a lack of data on past landslide occurrences.

Landslides Information Extraction Using Object-Oriented Image Analysis Paradigm Based on Deep Learning and Transfer Learning

How to acquire landslide disaster information quickly and accurately has become the focus and difficulty of disaster prevention and relief by remote sensing. Landslide disasters are generally featured by sudden occurrence, proposing high demand for emergency data acquisition. The low-altitude Unmanned Aerial Vehicle (UAV) remote sensing technology is widely applied to acquire landslide disaster data, due to its convenience, high efficiency, and ability to fly at low altitude under cloud. However, the spectrum information of UAV images is generally deficient and manual interpretation is difficult for meeting the need of quick acquisition of emergency data. Based on this, UAV images of high-occurrence areas of landslide disaster in Wenchuan County and Baoxing County in Sichuan Province, China were selected for research in the paper. Firstly, the acquired UAV images were pre-processed to generate orthoimages. Subsequently, multi-resolution segmentation was carried out to obtain image objects, and the barycenter of each object was calculated to generate a landslide sample database (including positive and negative samples) for deep learning. Next, four landslide feature models of deep learning and transfer learning, namely Histograms of Oriented Gradients (HOG), Bag of Visual Word (BOVW), Convolutional Neural Network (CNN), and Transfer Learning (TL) were compared, and it was found that the TL model possesses the best feature extraction effect, so a landslide extraction method based on the TL model and object-oriented image analysis (TLOEL) was proposed; finally, the TLOEL method was compared with the object-oriented nearest neighbor classification (NNC) method. The research results show that the accuracy of the TLOEL method is higher than the NNC method, which can not only achieve the edge extraction of large landslides, but also detect and extract middle and small landslides accurately that are scatteredly distributed.

A Holistic Analysis for Landslide Susceptibility Mapping Applying Geographic Object-Based Random Forest: A Comparison between Protected and Non-Protected Forests

Despite recent progress in landslide susceptibility mapping, a holistic method is still needed to integrate and customize influential factors with the focus on forest regions. This study was accomplished to test the performance of geographic object-based random forest in modeling the susceptibility of protected and non-protected forests to landslides in northeast Iran. Moreover, it investigated the influential conditioning and triggering factors that control the susceptibility of these two forest areas to landslides. After surveying the landslide events, segment objects were generated from the Landsat 8 multispectral images and digital elevation model (DEM) data. The features of conditioning factors were derived from the DEM and available thematic layers. Natural triggering factors were derived from the historical events of rainfall, floods, and earthquake. The object-based image analysis was used for deriving anthropogenic-induced forest loss and fragmentation. The layers of logging and mining were obtained from available historical data. Landslide samples were extracted from field observations, satellite images, and available database. A single database was generated including all conditioning and triggering object features, and landslide samples for modeling the susceptibility of two forest areas to landslides using the random forest algorithm. The optimal performance of random forest was obtained after building 500 trees with the area under the receiver operating characteristics (AUROC) values of 86.3 and 81.8% for the protected and non-protected forests, respectively. The top influential factors were the topographic and hydrologic features for mapping landslide susceptibility in the protected forest. However, the scores were loaded evenly among the topographic, hydrologic, natural, and anthropogenic triggers in the non-protected forest. The topographic features obtained about 60% of the importance values with the domination of the topographic ruggedness index and slope in the protected forest. Although the importance of topographic features was reduced to 36% in the non-protected forest, anthropogenic and natural triggering factors remarkably gained 33.4% of the importance values in this area. This study confirms that some anthropogenic activities such as forest fragmentation and logging significantly intensified the susceptibility of the non-protected forest to landslides.

A hybrid model considering spatial heterogeneity for landslide susceptibility mapping in Zhejiang Province, China

Landslides are a type of serious geologic disaster causing great damage to the human environment. Landslide susceptibility mapping is an effective means to reduce landslide risk. However, previous studies have not considered spatial heterogeneity. In this study, a hybrid model considering spatial heterogeneity is designed by integrating GeoSOM and Stacking ensemble methods and is applied to map the landslide susceptibility of Zhejiang Province, China. The GeoSOM method was used to cluster the study area into several homogeneous regions to solve the heterogeneity problem, and each region was assigned a cluster attribute as one of the landslide model inputs. The Stacking ensemble technique was utilized to design a high-performance landslide model by combining three traditional machine learning methods (support vector machine (SVM), artificial neural network (ANN), and gradient-boosting decision tree (GBDT)). We collected 1051 landslide samples and fourteen affecting factors after feature selection. For landslide modelling, the landslides were randomly split into two subsets: 70% samples for training and the rest for validation. Landslide models were assessed by the receiver operating characteristic (ROC) curve and statistical measures. The results indicated that the hybrid model was 0.11–0.135 higher than those of traditional machine learning methods in term of the area under the ROC curve (AUC). In general, this hybrid model can generate high-quality landslide susceptibility maps and help to develop policies that reduce the burden of landslides.

Using the rotation and random forest models of ensemble learning to predict landslide susceptibility

Ensemble learning methods can be used to evaluate landslide susceptibility when combined with remote sensing (RS) and geographic information systems (GIS). In this study, the rotation forest (ROF) and random forest (RF) ensemble learning models were applied to evaluate landslide susceptibility. The experiments selected the factors by analysing the linear relationship between the factors, explored the optimal proportions of non-landslide samples and landslide samples based on an unbalanced sample dataset, and used the factors before and after the selection to generate landslide susceptibility maps (LSMs) in the Zigui-Badong area. The results show that a suitable ratio between the sample types in the training set can achieve good results for both sensitivity and specificity. The RF models of the study area with 21 factors and 16 factors had sensitivities of 94.22% and 93.59%, respectively. The ROF models with 21 factors and 16 factors had sensitivities of 90.63% and 88.84%, respectively. Although both the RF and ROF models exhibited high accuracy, the RF model achieved a more reasonable and accurate LSM.

Comparison of GA-BP and PSO-BP neural network models with initial BP model for rainfall-induced landslides risk assessment in regional scale: a case study in Sichuan, China

With the increase in inclement weather conditions, many countries would experience more and more landslide hazards in the process of planning, designing and construction for engineering projects, especially in the mountainous regions. How to quickly and accurately assess potential landslide risk in a large region (> 10,000 km2) is facing challenge due to its complex geological conditions and large amount of landslides in the region. To optimize the accuracy of the existing models for a large region, in this study, the genetic algorithm (GA) and particle swarm optimization (PSO) are, respectively, coupled with the backpropagation (BP) neural network to determine the initial weights and thresholds in the BP neural network, which can be called GA-BP model and PSO-BP model. To show the reliability and accuracy of the new models in large region, the BP, GA-BP and PSO-BP models are evaluated based on root mean square error (RMSE), coefficient of determination (R2), Kappa coefficient (k), receiver operating characteristic (ROC), training time and condition factor weights by using 100 landslide samples from Sichuan Province, China. Results show that the RMSE values of the GA-BP model and the PSO model are, respectively, 22.6% and 5.1% lower than those of the BP model; the R2 values of the GA-BP model and the PSO model are, respectively, 24.9% and 6.2% higher than those of the BP model; the k values of the GA-BP model and the PSO model are, respectively, 44.3% and 15.4% higher than those of the BP model, and the areas under ROC of the GA-BP model and the PSO model are, respectively, 32.4% and 9.6% larger than those of the BP model. The GA-BP model and the PSO-BP model have better accuracy in the assessment of the overall risk value and the risk-level classification. The difference of the training time is small, and the sequences of condition factor weights given by the three models are consistent. In general, the GA-BP model is more effective for landslide risk assessment in large region. At last, this study gives proposed models under different engineering conditions, which can increase efficiency of the risk assessment for landslides.

A Synergetic Analysis of Sentinel-1 and -2 for Mapping Historical Landslides Using Object-Oriented Random Forest in the Hyrcanian Forests

Despite increasing efforts in the mapping of landslides using Sentinel-1 and -2, research on their combination for discerning historical landslides in forest areas is still lacking, particularly using object-oriented machine learning approaches. This study was accomplished to test the efficiency of Sentinel-derived features and digital elevation model (DEM) derivatives for mapping old and new landslides, using object-oriented random forest. Two forest subsets were selected including a protected and non-protected forest in northeast Iran. Landslide samples were obtained from CORONA images and aerial photos (old landslides), and also field mensuration and high-resolution images (new landslides). Segment objects were generated from a set combination of Sentinel-1A, Sentinel-2A, and some topographic-derived indices using multiresolution segmentation algorithm. Various object features were derived from the main channels of Sentinel images and DEM derivatives in the seven main groups, including spectral layers, spectral indices, geometric, contextual, textural, topographic, and hydrologic features. A single database was created, including landslide samples and Sentinel- and DEM-derived object features. Roughly 20% of landslide-affected objects and non-landslide-affected objects were randomly selected as an input for training the random forest classifier. Two-thirds of the selected objects were assigned as learning samples for classification, and the remainder were used for testing the accuracy of landslide and non-landslide classification. Results indicated that: (1) The sensitivity of mapping historical landslides was 86.6% and 80.3% in the protected and non-protected forests, respectively; (2) the object features of Sentinel-2A and DEM obtained the highest importance with the total scores of 55.6% and 32%, respectively in the protected forests, and 65.4% and 21% respectively in the non-protected forests; (3) the features derived from the combination of Sentinel-1 and -2A demonstrated a total importance of 10% for mapping new landslides; and (4) textural features were obtained in approximately two-thirds of the total scores for mapping new landslides, however a combination of topographic, spectral, textural, and contextual features were the effective predictors for mapping old landslides. This research proposes applying a synergetic analysis of Sentinel- and DEM-derived features for mapping historical landslides; however, there are no uniformly pre-defined influential variables for mapping historical landslides in different forest areas.

Landslide susceptibility: a statistically-based assessment on a depositional pyroclastic ramp

This study aimed to produce a high-quality landslide susceptibility map for Teziutlan municipality, a landslide-prone region in Mexico, which is characterised by a depositional pyroclastic ramp. The heterogeneous quality of available topographic information (i.e. higher resolution digital elevation model only for a sub-region) encouraged to confront modelling results based on two different study area delineations and two raster resolutions. Input data was based on the larger modelling region L15 (163 km2) and smaller S (70 km2; located inside L15) with an associated raster cell size of 15 m (region L15 and S15) and 5 m (region S5). The resulting three data sets (L15, S15 and S5) were included into three differently flexible modelling techniques (Generalized Linear Model - GLM, General Additive Model - GAM, Support Vector Machine-SVM) to produce nine landslide susceptibility models. Preceding variable selection was performed heuristically and supported by an exploratory data analysis. The final models were based on the explanatory variables slope angle, slope aspect, lithology, relative slope position, elevation, convergence index, distance to streams, distance to springs and topographic wetness index. The ability of the models to classify independent test data was elaborated using a k-fold cross validation procedure and the AUROC (Area Under the Receiver Operating Characteristic) metric. In general, all produced landslide susceptibility maps depicted the hillslopes of the ravines, which cut the pyroclastic ramp, as prone to landsliding. The modelling results showed that predictive performances (i.e. AUROC values) slightly increased with an increasing flexibility of the applied modelling technique. Thus, SVM performed best, while the GAM outperformed the GLM. This tendency was most distinctive when modelling with the largest landslide sample size (i.e. data set L15; n = 662 landslides). Non-linear classifiers (GAMs, SVMs) performed slightly better when trained on the basis of lower raster resolution (data set S15) compared to the 5 m counterparts (data set S5). Highest predictive performance was obtained for the model based on data set L15 and the SVM classifier (median AUROC: 0.82). However, SVMs also indicated the highest degree of model overfitting. This study indicates that the decision to delineate a study area, the selection of a raster resolution as well as the chosen classification technique can affect varying aspects of subsequent modelling results. The results do not support the assumption that a higher raster resolution (i.e. a more detailed digital representation of the terrain) inevitably leads to better performing or geomorphically more plausible landslide susceptibility maps.

Optimizing the Predictive Ability of Machine Learning Methods for Landslide Susceptibility Mapping Using SMOTE for Lishui City in Zhejiang Province, China

The main goal of this study was to use the synthetic minority oversampling technique (SMOTE) to expand the quantity of landslide samples for machine learning methods (i.e., support vector machine (SVM), logistic regression (LR), artificial neural network (ANN), and random forest (RF)) to produce high-quality landslide susceptibility maps for Lishui City in Zhejiang Province, China. Landslide-related factors were extracted from topographic maps, geological maps, and satellite images. Twelve factors were selected as independent variables using correlation coefficient analysis and the neighborhood rough set (NRS) method. In total, 288 soil landslides were mapped using field surveys, historical records, and satellite images. The landslides were randomly divided into two datasets: 70% of all landslides were selected as the original training dataset and 30% were used for validation. Then, SMOTE was employed to generate datasets with sizes ranging from two to thirty times that of the training dataset to establish and compare the four machine learning methods for landslide susceptibility mapping. In addition, we used slope units to subdivide the terrain to determine the landslide susceptibility. Finally, the landslide susceptibility maps were validated using statistical indexes and the area under the curve (AUC). The results indicated that the performances of the four machine learning methods showed different levels of improvement as the sample sizes increased. The RF model exhibited a more substantial improvement (AUC improved by 24.12%) than did the ANN (18.94%), SVM (17.77%), and LR (3.00%) models. Furthermore, the ANN model achieved the highest predictive ability (AUC = 0.98), followed by the RF (AUC = 0.96), SVM (AUC = 0.94), and LR (AUC = 0.79) models. This approach significantly improves the performance of machine learning techniques for landslide susceptibility mapping, thereby providing a better tool for reducing the impacts of landslide disasters.

Landslide Susceptibility Mapping Based on Weighted Gradient Boosting Decision Tree in Wanzhou Section of the Three Gorges Reservoir Area (China)

The main goal of this study is to produce a landslide susceptibility map in the Wanzhou section of the Three Gorges reservoir area (China) with a weighted gradient boosting decision tree (weighted GBDT) model. According to the current research on landslide susceptibility mapping (LSM), the GBDT method is rarely used in LSM. Furthermore, previous studies have rarely considered the imbalance of landslide samples and simply regarded the LSM problem as a binary classification problem. In this paper, we considered LSM as an imbalanced learning problem and obtained a better predictive model using the weighted GBDT method. The innovations of the article mainly include the following two points: introducing the GBDT model into the evaluation of landslide susceptibility; using the weighted GBDT method to deal with the problem of landslide sample imbalance. The logistic regression (LR) model and gradient boosting decision tree (GBDT) model were also used in the study to compare with the weighted GBDT model. Five kinds of data from different data source were used in the study: geology, topography, hydrology, land cover, and triggered factors (rainfall, earthquake, land use, etc.). Twenty nine environmental parameters and 233 landslides were used as input data. The receiver operating characteristic (ROC) curve, the area under the ROC curve (AUC) value, and the recall value were used to estimate the quality of the weighted GBDT model, the GBDT model, and the LR model. The results showed that the GBDT model and the weighted GBDT model had a higher AUC value (0.977, 0.976) than the LR model (0.845); the weighted GBDT model had a little higher AUC value (0.977) than the GBDT model (0.976); and the weighted GBDT model had a higher recall value (0.823) than the GBDT model (0.426) and the LR model (0.004). The weighted GBDT method could be considered to have the best performance considering the AUC value and the recall value in landslide susceptibility mapping dealing with imbalanced landslide data.

Modeling landslide susceptibility in data-scarce environments using optimized data mining and statistical methods

This study evaluated the generalizability of five models to select a suitable approach for landslide susceptibility modeling in data-scarce environments. In total, 418 landslide inventories and 18 landslide conditioning factors were analyzed. Multicollinearity and factor optimization were investigated before data modeling, and two experiments were then conducted. In each experiment, five susceptibility maps were produced based on support vector machine (SVM), random forest (RF), weight-of-evidence (WoE), ridge regression (Rid_R), and robust regression (RR) models. The highest accuracy (AUC=0.85) was achieved with the SVM model when either the full or limited landslide inventories were used. Furthermore, the RF and WoE models were severely affected when less landslide samples were used for training. The other models were affected slightly when the training samples were limited.

Microstructures in landslides in northwest China – Implications for creeping displacements?

Microstructures, mineralogical composition and texture of selected landslide samples from three landslides in the southern part of the Gansu Province (China) were examined with optical microscopy, transmission electron microscopy (TEM), x-ray diffraction (XRD) and synchrotron x-ray diffraction measurements. Common sheet silicates are chlorite, illite, muscovite, kaolinite, pyrophyllite and dickite. Other minerals are quartz, calcite, dolomite and albite. In one sample, graphite and amorphous carbon were detected by TEM-EDX analyses and TEM high-angle annular dark-field images. The occurrence of graphite and pyrophyllite with very low friction coefficients in the gouge material of the Suoertou and Xieliupo landslides is particularly significant for reducing the frictional strength of the landslides. It is proposed that the landslides underwent comparable deformation processes as fault zones. The low friction coefficients provide strong evidence that slow-moving landsliding is controlled by the presence of weak minerals. In addition, TEM observations document that grain size reduction in clayey slip zone material was produced mainly by mechanical abrasion. For calcite and quartz, grain size reduction was attributed to both pressure solution and cataclasis. Therefore, besides landslide composition, the occurrence of ultrafine-grained slip zone material may also contribute to weakening processes of landslides. TEM images of slip-zone samples show both locally aligned clay particles, as well as kinked and folded sheet silicates, which are widely disseminated in the whole matrix. Small, newly formed clay particles have random orientations. Based on synchrotron x-ray diffraction measurements, the degree of preferred orientation of constituent sheet silicates in local shear zones of the Suoertou and Duang-He-Ba landslide is strong. This work is the first reported observation of well-oriented clay fabrics in landslides.

Improving the accuracy of landslide susceptibility model using a novel region-partitioning approach

Landslide is a natural disaster that threatens human lives and properties worldwide. Numerous have been conducted on landslide susceptibility mapping (LSM), in which each has attempted to improve the accuracy of final outputs. This study presents a novel region-partitioning approach for LSM to understand the effects of partitioning a focused region into smaller areas on the prediction accuracy of common regression models. Results showed that the partitioning of the study area into two regions using the proposed method improved the prediction rate from 0.77 to 0.85 when support vector machine was used, and from 0.87 to 0.88 when logistic regression model was utilized. The spatial agreements of the models were also improved after partitioning the area into two regions based on Shannon entropy equations. Our comparative study indicated that the proposed method outperformed the geographically weighted regression model that considered the spatial variations in landslide samples. Overall, the main advantages of the proposed method are improved accuracy and the reduction of the effects of spatial variations exhibited in landslide-conditioning factors.

Predictive performance of rainfall thresholds for shallow landslides in Switzerland from gridded daily data

AbstractA high‐resolution gridded daily precipitation data set was combined with a landslide inventory containing over 2000 events in the period 1972–2012 to analyze rainfall thresholds which lead to landsliding in Switzerland. We colocated triggering rainfall to landslides, developed distributions of triggering and nontriggering rainfall event properties, and determined rainfall thresholds and intensity‐duration ID curves and validated their performance. The best predictive performance was obtained by the intensity‐duration ID threshold curve, followed by peak daily intensity Imax and mean event intensity Imean. Event duration by itself had very low predictive power. A single country‐wide threshold of Imax = 28 mm/d was extended into space by regionalization based on surface erodibility and local climate (mean daily precipitation). It was found that wetter local climate and lower erodibility led to significantly higher rainfall thresholds required to trigger landslides. However, we showed that the improvement in model performance due to regionalization was marginal and much lower than what can be achieved by having a high‐quality landslide database. Reference cases in which the landslide locations and timing were randomized and the landslide sample size was reduced showed the sensitivity of the Imax rainfall threshold model. Jack‐knife and cross‐validation experiments demonstrated that the model was robust. The results reported here highlight the potential of using rainfall ID threshold curves and rainfall threshold values for predicting the occurrence of landslides on a country or regional scale with possible applications in landslide warning systems, even with daily data.