Exploring deep learning for landslide mapping:A comprehensive review

A detailed and accurate inventory map of landslides is crucial for quantitative hazard assessment and land planning.Traditional methods relying on change detection and object-oriented approaches have been criticized for their dependence on expert knowledge and subjective factors.Recent advancements in highresolution satellite imagery,coupled with the rapid development of artificial intelligence,particularly datadriven deep learning algorithms(DL)such as convolutional neural networks(CNN),have provided rich feature indicators for landslide mapping,overcoming previous limitations.In this review paper,77representative DL-based landslide detection methods applied in various environments over the past seven years were examined.This study analyzed the structures of different DL networks,discussed five main application scenarios,and assessed both the advancements and limitations of DL in geological hazard analysis.The results indicated that the increasing number of articles per year reflects growing interest in landslide mapping by artificial intelligence,with U-Net-based structures gaining prominence due to their flexibility in feature extraction and generalization.Finally,we explored the hindrances of DL in landslide hazard research based on the above research content.Challenges such as black-box operations and sample dependence persist,warranting further theoretical research and future application of DL in landslide detection.

Landslide mapping and analysis along the China-Pakistan Karakoram Highway based on SBAS-InSAR detection in 2017

The Karakoram Highway(KKH),a part of the China–Pakistan Economic Corridor(CPEC),is a major highway connecting northern Pakistan to China.The inventorying and analysis of landslides along KKH are challenging because of poor accessibility,vast study area,limited availability of ground-based datasets,and the complexity of landslide processes in the region.In order to preserve life,property,and infrastructure,and to enable the uninterrupted and efficient operation of the KKH,it is essential to strengthen measures for the prevention and control of geological disasters.In the present study,SBASInSAR(Small Baseline Subsets-Interferometric Synthetic Aperture Radar)was used to process 150 scenes of Sentinel 1-A images in the year 2017 along the Karakoram Highway.A total of 762 landslides,including 57 complex landslides,126 rock falls,167 debris slides,and 412 unstable slopes,ranging in size between 0.0017 and 10.63 km2 were identified.Moreover,this study also gains an inventory of 40 active glacier movements in this region.Landslide categorization,displacements characteristics,spatial distribution,and their relationship with various contributing factors have been successfully investigated along the entire KKH using image interpretation and frequency-area statistics.The criteria adopted for landslides categorization is presented in the study.The results showed that the 2-D ground deformation derived in Hunza valley echoes well with the general regional landslides characteristics.The spatial distribution analysis revealed that there are clumped distributions of landslides in the Gaizi,Tashkurgan,and Khunjerab in China,as well as in Hunza valley,and north of Chilas city in Pakistan.Statistical results indicated that these landslides mainly occur on south-facing slopes with a slope angle of 20°–45°and elevation relief of 550–2,100 m.Landslide development is also related to low vegetation cover and weathering effects in mountain gullies.Overall,our study provides scientific data support and theoretical references for prevention,control,and mitigation of geological disasters in the Karakoram region.

Landslide susceptibility mapping of mountain roads based on machine learning combined model

Landslide susceptibility mapping of mountain roads is frequently confronted by insufficient historical landslide sample data,multicollinearity of existing evaluation index factors,and inconsistency of evaluation factors due to regional environmental variations.Then,a single machine learning model can easily become overfitting,thus reducing the accuracy and robustness of the evaluation model.This paper proposes a combined machine-learning model to address the issues.The landslide susceptibility in mountain roads were mapped by using factor analysis to normalize and reduce the dimensionality of the initial condition factor and generating six new combination factors as evaluation indexes.The mountain roads in the Youxi County,Fujian Province,China were used for the landslide susceptibility mapping.Three most frequently used machine learning techniques,support vector machine(SVM),random forest(RF),and artificial neural network(ANN)models,were used to model the landslide susceptibility of the study area and validate the accuracy of this evaluation index system.The global minimum variance portfolio was utilized to construct a machine learning combined model.5-fold cross-validation,statistical indexes,and AUC(Area Under Curve)values were implemented to evaluate the predictive accuracy of the landslide susceptibility model.The mean AUC values for the SVM,RF,and ANN models in the training stage were 89.2%,88.5%,and 87.9%,respectively,and 78.0%,73.7%,and 76.7%,respectively,in the validating stage.In the training and validation stages,the mean AUC values of the combined model were 92.4% and 87.1%,respectively.The combined model provides greater prediction accuracy and model robustness than one single model.

Machine learning solution for regional landslide susceptibility based on fault zone division strategy

Landslide susceptibility assessment is an essential tool for disaster prevention and management. In areas with multiple fault zones, the impact of fault zone on slope stability cannot be disregarded. This study performed qualitative analysis of fault zones and proposed a zoning method to assess the landslide susceptibility in Chengkou County, Chongqing Municipality, China. The region within a distance of 1 km from the faults was designated as sub-zone A, while the remaining area was labeled as sub-zone B. To accomplish the assessment, a dataset comprising 388 historical landslides and 388 non-landslide points was used to train the random forest model. 10-fold cross-validation was utilized to select the training and testing datasets for the model. The results of the models were analyzed and discussed, with a focus on model performance and prediction uncertainty. By implementing the proposed division strategy based on fault zone, the accuracy, precision, recall, F-score, and AUC of both two sub-zones surpassed those of the whole region. In comparison to the results obtained for the whole region, sub-zone B exhibited an increase in AUC by 6.15%, while sub-zone A demonstrated a corresponding increase of 1.66%. Moreover, the results of 100 random realizations indicated that the division strategy has little effect on the prediction uncertainty. This study introduces a novel approach to enhance the prediction accuracy of the landslide susceptibility mapping model in areas with multiple fault zones.

Landslide susceptibility mapping using hybrid random forest with GeoDetector and RFE for factor optimization

The present study aims to develop two hybrid models to optimize the factors and enhance the predictive ability of the landslide susceptibility models.For this,a landslide inventory map was created with 406 historical landslides and 2030 non-landslide points,which was randomly divided into two datasets for model training(70%)and model testing(30%).22 factors were initially selected to establish a landslide factor database.We applied the GeoDetector and recursive feature elimination method(RFE)to address factor optimization to reduce information redundancy and collinearity in the data.Thereafter,the frequency ratio method,multicollinearity test,and interactive detector were used to analyze and evaluate the optimized factors.Subsequently,the random forest(RF)model was used to create a landslide susceptibility map with original and optimized factors.The resultant hybrid models GeoDetector-RF and RFE-RF were evaluated and compared by the area under the receiver operating characteristic curve(AUC)and accuracy.The accuracy of the two hybrid models(0.868 for GeoDetector-RF and 0.869 for RFE-RF)were higher than that of the RF model(0.860),indicating that the hybrid models with factor optimization have high reliability and predictability.Both RFE-RF GeoDetector-RF had higher AUC values,respectively 0.863 and 0.860,than RF(0.853).These results confirm the ability of factor optimization methods to improve the performance of landslide susceptibility models.

Landslide susceptibility mapping using Genetic Algorithm for the Rule Set Production(GARP) model

Investigation on landslide phenomenon is necessary for understanding and delineating the landslide prone and safer places for different land use practices. On this basis, a new model known as genetic algorithm for the rule set production was applied in order to assess its efficacy to obtain a better result and a more precise landslide susceptibility map in Klijanerestagh area of Iran. This study considered twelve landslide conditioning factors(LCF) like altitude, slope, aspect, plan curvature, profile curvature, topographic wetness index(TWI), distance from rivers, faults, and roads, land use/cover, and lithology. For modeling purpose, the Genetic Algorithm for the Rule Set Production(GARP) algorithm was applied in order to produce the landslide susceptibility map. Finally, to evaluate the efficacy of the GARP model, receiver operating characteristics curve as well as the Kappa index were employed. Based on these indices, the GARP model predicted the probability of future landslide incidences with the area under the receiver operating characteristics curve(AUC-ROC) values of 0.932, and 0.907 for training and validating datasets, respectively. In addition, Kappa values for the training and validating datasets were computed as 0.775, and 0.716, respectively. Thus, it can be concluded that the GARP algorithm can be a new but effective method for generating landslide susceptibility maps(LSMs). Furthermore, higher contribution of the lithology, distance from roads, and distance from faults was observed, while lower contribution was attributed to soil, profile curvature, and TWI factors. The introduced methodology in this paper can be suggested for other areas with similar topographical and hydrogeological characteristics for land use planning and reducing the landslide damages.

Landslides susceptibility mapping in Guizhou province based on fuzzy theory

The purpose of this study was to assess the susceptibility of landslides around the area of Guizhou province based on fuzzy theory.In first instance, slope, elevation, lithology, proximity to tectonic lines, proximity to drainage and annual precipitation were taken as independent, causal factors in this study.A landslide hazard evaluation factor system was established by classifying these factors into more subclasses according to some rules.Secondly, a trapezoidal fuzzy number weighting(TFNW) approach was used to assess the importance of six causal factors to landslides in an ArcGIS environment.Thirdly, a landslide susceptibility map was created based on a weighted linear combination model.According to this susceptibility map, the study area was classified into four categories of landslide susceptibility:low, moderate, high and very high.Finally, in order to verify the results obtained, the susceptibility map and the landslide inventory map were combined in the GIS.In addition, the weighting procedure showed that TFNW is an efficient method for weighting causal landslide factors.

GIS and ANN model for landslide susceptibility mapping

Landslide hazard is as the probability of occurrence of a potentially damaging landslide phenomenon within specified period of time and within a given area. The susceptibility map provides the relative spatial probability of landslides occurrence. A study is presented of the application of GIS and artificial neural network model to landslide susceptibility mapping, with particular reference to landslides on natural terrain in this paper. The method has been applied to Lantau Island, the largest outlying island within the territory of Hong Kong. A three-level neural network model was constructed and trained by the back-propagate algorithm in the geographical database of the study area. The data in the database includes digital elevation modal and its derivatives, landslides distribution and their attributes, superficial geological maps, vegetation cover, the raingauges distribution and their 14 years 5-minute observation. Based on field inspection and analysis of correlation between terrain variables and landslides frequency, lithology, vegetation cover, slope gradient, slope aspect, slope curvature, elevation, the characteristic value, the rainstorms corresponding to the landslide, and distance to drainage Une are considered to be related to landslide susceptibility in this study. The artificial neural network is then coupled with the ArcView3.2 GIS software to produce the landslide susceptibility map, which classifies the susceptibility into three levels: low, moderate, and high. The results from this study indicate that GIS coupled with artificial neural network model is a flexible and powerful approach to identify the spatial probability of hazards.

Geospatial susceptibility mapping of earthquake-induced landslides in Nuweiba area, Gulf of Aqaba, Egypt

Earthquake induced landslides are one of the most severe geo-environmental hazards that cause enormous damage to infrastructure, property, and loss of life in Nuweiba area. This study developed a model for mapping the earthquake-induced landslide susceptibility in Nuweiba area in Egypt with considerations of geological, geomorphological, topographical, and seismological factors. An integrated approach of remote sensing and GIS technologies were applied for that target. Several data sources including Terra SAR-X and SPOT 5 satellite imagery, topographic maps, field data, and other geospatial resources were used to model landslide susceptibility. These data were used specifically to produce important thematic layers contributing to landslide occurrences in the region. A rating scheme was developed to assign ranks for the thematic layers and weights for their classes based on their contribution in landslide susceptibility. The ranks and weights were defined based on the knowledge from field survey and authors experiences related to the study area. The landslide susceptibility map delineates the hazard zones to three relative classes of susceptibility: high, moderate, and low. Therefore, the current approach provides a way to assess landslide hazards and serves for geo-hazard planning and prediction in Nuweiba area.

Application of Chebyshev theorem to data preparation in landslide susceptibility mapping studies:an example from Yenice(Karabük,Turkey)region

Landslide database construction is one of the most crucial stages of the landslide susceptibility mapping studies. Although there are many techniques for preparing landslide database in the literature, representative data selection from huge data sets is a challenging, and, to some extent, a subjective task. Thus, in order to produce reliable landslide susceptibility maps, data-driven, objective and representative database construction is a very important stage for these maps. This study mainly focuses on a landslide database construction task. In this study, it was aimed at building a representative landslide database extraction approach by using Chebyshev theorem to evaluate landslide susceptibility in a landslide prone area in the Western Black Sea region of Turkey. The study area was divided into two different parts such as training （Basin 1） and testing areas （Basin 2）. A total of nine parameters such as topographical elevation, slope, aspect, planar and profile curvatures, stream power index, distance to drainage, normalized difference vegetation index and topographical wetness index were used in the study. Next, frequency distributions of the considered parameters in both landslide and nonlandslide areas were extracted using different sampling strategies, and a total of nine different landslide databases were obtained. Of these, eight databases were gathered by the methodology proposed by this study based on different standard deviations and algebraic multiplication of raster parameter maps. To evaluate landslide susceptibility, Artificial Neural Network method was used in the study area considering the different landslide and nonlandslide data. Finally, to assess the performances of the so-produced landslide susceptibility maps based on nine data sets, Area Under Curve （AUC） approach was implemented both in Basin 1 and Basin 2. The best performances （the greatest AUC values） were gathered by the landslide susceptibility map produced by two standard deviation database extracted by the Chebyshev theorem, as 0.873 and 0.761, respectively. Results revealed that the methodology proposed by this study is a powerful and objective approach in landslide susceptibility mapping.

Enhancing landslide susceptibility mapping incorporating landslide typology via stacking ensemble machine learning in Three Gorges Reservoir,China

Different types of landslides exhibit distinct relationships with environmental conditioning factors.Therefore,in regions where multiple types of landslides coexist,it is required to separate landslide types for landslide susceptibility mapping(LSM).In this paper,a landslideprone area located in Chongqing Province within the middle and upper reaches of the Three Gorges Reservoir area(TGRA),China,was selected as the study area.733 landslides were classified into three types:reservoir-affected landslides,non-reservoir-affected landslides,and rockfalls.Four landslide inventory datasets and 15 landslide conditional factors were trained by three Machine Learning models(logistic regression,random forest,support vector machine),and a Deep Learning(DL)model.After comparing the models using receiver operating characteristics(ROC),the landslide susceptibility indexes of three types landslides were acquired by the best performing model.These indexes were then used as input to generate the final map based on the Stacking method.The results revealed that DL model showed the best performance in LSM without considering landslide types,achieving an area under the curve(AUC)of 0.854 for testing and 0.922 for training.Moreover,when we separated the landslide types for LSM,the AUC improved by 0.026 for testing and 0.044 for training.Thus,this paper demonstrates that considering different landslide types in LSM can significantly improve the quality of landslide susceptibility maps.These maps in turn,can be valuable tools for evaluating and mitigating landslide hazards.

Landslide susceptibility analysis of Karakoram highway using analytical hierarchy process and scoops 3D

Landslides are prevalent,regular,and expensive hazards in the Karakoram Highway(KKH)region.The KKH connects Pakistan with China in the present China-Pakistan Economic Corridor(CPEC)context.This region has not only immense economic importance but also ecological significance.The purpose of the study was to map the landslide-prone areas along KKH using two different techniquesAnalytical Hierarchy Process(AHP)and Scoops 3 D model.The causative parameters for running AHP include the lithology,presence of thrust,land use land cover,precipitation,and Digital Elevation Model(DEM)derived variables(slope,curvature,aspect,and elevation).The AHP derived final landslide susceptibility map was classified into four zones,i.e.,low,moderate,high,and extremely high.Over 80%of the study area falls under the moderate(43%)and high(40%)landslide susceptible zones.To assess the slope stability of the study area,the Scoops 3 D model was used by integrating with the earthquake loading data.The results of the limit equilibrium analysis categorized the area into four groups(low,moderate,high,and extremely high mass)of slope failure.The areas around Main Mantle Thrust(MMT)including Dubair,Jijal,and Kohistan regions,had high volumes of potential slope failures.The results from AHP and Scoops 3 D techniques were validated with the landslides inventory record of the Geological Survey of Pakistan and Google Earth.The results from both the techniques showed similar output that coincides with the known landslides areas.However,Scoops 3 D provides not only susceptible zones but also the range of volume of the potential slope failures.Further,these techniques could be used in other mountainous areas,which could help in the landslide mitigation measures.

Improving pixel-based regional landslide susceptibility mapping

Regional landslide susceptibility mapping(LSM)is essential for risk mitigation.While deep learning algorithms are increasingly used in LSM,their extensive parameters and scarce labels(limited landslide records)pose training challenges.In contrast,classical statistical algorithms,with typically fewer parameters,are less likely to overfit,easier to train,and offer greater interpretability.Additionally,integrating physics-based and data-driven approaches can potentially improve LSM.This paper makes several contributions to enhance the practicality,interpretability,and cross-regional generalization ability of regional LSM models:(1)Two new hybrid models,composed of data-driven and physics-based modules,are proposed and compared.Hybrid ModelⅠcombines the infinite slope stability analysis(ISSA)with logistic regression,a classical statistical algorithm.Hybrid ModelⅡintegrates ISSA with a convolutional neural network,a representative of deep learning techniques.The physics-based module constructs a new explanatory factor with higher nonlinearity and reduces prediction uncertainty caused by incomplete landslide inventory by pre-selecting non-landslide samples.The data-driven module captures the rela-tion between explanatory factors and landslide inventory.(2)A step-wise deletion process is proposed to assess the importance of explanatory factors and identify the minimum necessary factors required to maintain satisfactory model performance.(3)Single-pixel and local-area samples are compared to understand the effect of pixel spatial neighborhood.(4)The impact of nonlinearity in data-driven algorithms on hybrid model performance is explored.Typical landslide-prone regions in the Three Gorges Reservoir,China,are used as the study area.The results show that,in the testing region,by using local-area samples to account for pixel spatial neighborhoods,Hybrid ModelⅠachieves roughly a 4.2%increase in the AUC.Furthermore,models with 30 m resolution land-cover data surpass those using 1000 m resolution data,showing a 5.5%improvement in AUC.The optimal set of explanatory factors includes elevation,land-cover type,and safety factor.These findings reveal the key elements to enhance regional LSM,offering valuable insights for LSM practices.

Uncertainty pattern in landslide susceptibility prediction modelling:Effects of different landslide boundaries and spatial shape expressions

In some studies on landslide susceptibility mapping(LSM),landslide boundary and spatial shape characteristics have been expressed in the form of points or circles in the landslide inventory instead of the accurate polygon form.Different expressions of landslide boundaries and spatial shapes may lead to substantial differences in the distribution of predicted landslide susceptibility indexes(LSIs);moreover,the presence of irregular landslide boundaries and spatial shapes introduces uncertainties into the LSM.To address this issue by accurately drawing polygonal boundaries based on LSM,the uncertainty patterns of LSM modelling under two different landslide boundaries and spatial shapes,such as landslide points and circles,are compared.Within the research area of Ruijin City in China,a total of 370 landslides with accurate boundary information are obtained,and 10 environmental factors,such as slope and lithology,are selected.Then,correlation analyses between the landslide boundary shapes and selected environmental factors are performed via the frequency ratio(FR)method.Next,a support vector machine(SVM)and random forest(RF)based on landslide points,circles and accurate landslide polygons are constructed as point-,circle-and polygon-based SVM and RF models,respectively,to address LSM.Finally,the prediction capabilities of the above models are compared by computing their statistical accuracy using receiver operating characteristic analysis,and the uncertainties of the predicted LSIs under the above models are discussed.The results show that using polygonal surfaces with a higher reliability and accuracy to express the landslide boundary and spatial shape can provide a markedly improved LSM accuracy,compared to those based on the points and circles.Moreover,a higher degree of uncertainty of LSM modelling is present in the expression of points because there are too few grid units acting as model input variables.Additionally,the expression of the landslide boundary as circles introduces errors in measurement and is not as accurate as the polygonal boundary in most LSM modelling cases.In addition,the results under different conditions show that the polygon-based models have a higher LSM accuracy,with lower mean values and larger standard deviations compared with the point-and circle-based models.Finally,the overall LSM accuracy of the RF is superior to that of the SVM,and similar patterns of landslide boundary and spatial shape affecting the LSM modelling are reflected in the SVM and RF models.

Applying deep learning and benchmark machine learning algorithms for landslide susceptibility modelling in Rorachu river basin of Sikkim Himalaya, India

Landslide is considered as one of the most severe threats to human life and property in the hilly areas of the world.The number of landslides and the level of damage across the globe has been increasing over time.Therefore,landslide management is essential to maintain the natural and socio-economic dynamics of the hilly region.Rorachu river basin is one of the most landslide-prone areas of the Sikkim selected for the present study.The prime goal of the study is to prepare landslide susceptibility maps(LSMs)using computer-based advanced machine learning techniques and compare the performance of the models.To properly understand the existing spatial relation with the landslide,twenty factors,including triggering and causative factors,were selected.A deep learning algorithm viz.convolutional neural network model(CNN)and three popular machine learning techniques,i.e.,random forest model(RF),artificial neural network model(ANN),and bagging model,were employed to prepare the LSMs.Two separate datasets including training and validation were designed by randomly taken landslide and nonlandslide points.A ratio of 70:30 was considered for the selection of both training and validation points.Multicollinearity was assessed by tolerance and variance inflation factor,and the role of individual conditioning factors was estimated using information gain ratio.The result reveals that there is no severe multicollinearity among the landslide conditioning factors,and the triggering factor rainfall appeared as the leading cause of the landslide.Based on the final prediction values of each model,LSM was constructed and successfully portioned into five distinct classes,like very low,low,moderate,high,and very high susceptibility.The susceptibility class-wise distribution of landslides shows that more than 90%of the landslide area falls under higher landslide susceptibility grades.The precision of models was examined using the area under the curve(AUC)of the receiver operating characteristics(ROC)curve and statistical methods like root mean square error(RMSE)and mean absolute error(MAE).In both datasets(training and validation),the CNN model achieved the maximum AUC value of 0.903 and 0.939,respectively.The lowest value of RMSE and MAE also reveals the better performance of the CNN model.So,it can be concluded that all the models have performed well,but the CNN model has outperformed the other models in terms of precision.

Comparative performance assessment of landslide susceptibility models with presence-only,presenceabsence,and pseudo-absence data

The quality of the data for statistical methods plays an important role in landslide susceptibility mapping.How different data types influence the performance of landslide susceptibility maps is worth studying.The goal of this study was to explore the effects of different data types namely,presence-only(PO),presence-absence(PA),and pseudo-absence(PAs) data,on the predictive capability of landslide susceptibility mapping.This was completed by conducting a case study in the landslide-prone Honghe County in the Yunnan Province of China.A total of 428 landslide PO data points were selected.An equivalent number of nonlandslide locations were generated as PA data by random sampling,and 10,000 sites were uniformly selected at random from each region as PAs data.Three landslide susceptibility models,namely the information value model(IVM),logistic regression(LR) model,and maximum entropy(MaxEnt) model,corresponding to the three data types were investigated.Additionally,the area under the receiver operating characteristic curves(ROC-AUC),seven statistical indices(i.e.accuracy,sensibility,falsepositive rate,specificity,precision,Kappa,and Fmeasure),and a landslide density analysis were used to evaluate model performance regarding landslide susceptibility mapping.Our results indicated that the MaxEnt model using PAs data performed the best and had the highest fitness with the highest ROC-AUC values and statistical indices,followed by the IVM model with only landslide data(PO),and the LR model using PA data.Using PAs data avoided the inherent over-predictive shortcomings of PO data by limiting the predicted area of high-landslide susceptibility.Additionally,the random sampling design of landslide PA data increased the uncertainty of landslide susceptibility mapping and influenced the performance of the model.Therefore,our results suggested that the PAs data sampling provided a useful data type in the absence of high-quality data.Finally,we summarized the principles,advantages,and disadvantages of the three data types to assist with model optimization and the improvement of predicted performance and fitness.

Ensemble learning framework for landslide susceptibility mapping:Different basic classifier and ensemble strategy

The application of ensemble learning models has been continuously improved in recent landslide susceptibility research,but most studies have no unified ensemble framework.Moreover,few papers have discussed the applicability of the ensemble learning model in landslide susceptibility mapping at the township level.This study aims at defining a robust ensemble framework that can become the benchmark method for future research dealing with the comparison of different ensemble models.For this purpose,the present work focuses on three different basic classifiers:decision tree(DT),support vector machine(SVM),and multi-layer perceptron neural network model(MLPNN)and two homogeneous ensemble models such as random forest(RF)and extreme gradient boosting(XGBoost).The hierarchical construction of deep ensemble relied on two leading ensemble technologies(i.e.,homogeneous/heterogeneous model ensemble and bagging,boosting,stacking ensemble strategy)to provide a more accurate and effective spatial probability of landslide occurrence.The selected study area is Dazhou town,located in the Jurassic red-strata area in the Three Gorges Reservoir Area of China,which is a strategic economic area currently characterized by widespread landslide risk.Based on a long-term field investigation,the inventory counting thirty-three slow-moving landslide polygons was drawn.The results show that the ensemble models do not necessarily perform better;for instance,the Bagging based DT-SVM-MLPNNXGBoost model performed worse than the single XGBoost model.Amongst the eleven tested models,the Stacking based RF-XGBoost model,which is a homogeneous model based on bagging,boosting,and stacking ensemble,showed the highest capability of predicting the landslide-affected areas.Besides,the factor behaviors of DT,SVM,MLPNN,RF and XGBoost models reflected the characteristics of slow-moving landslides in the Three Gorges reservoir area,wherein unfavorable lithological conditions and intense human engineering activities(i.e.,reservoir water level fluctuation,residential area construction,and farmland development)are proven to be the key triggers.The presented approach could be used for landslide spatial occurrence prediction in similar regions and other fields.

A robust discretization method of factor screening for landslide susceptibility mapping using convolution neural network,random forest,and logistic regression models

The selection of discretization criteria and interval numbers of landslide-related environmental factors generally fails to quantitatively determine orfilter,resulting in uncertainties and limitations in the performance of machine learning(ML)methods for landslide susceptibility mapping(LSM).The aim of this study is to propose a robust discretization criterion(RDC)to quantify and explore the uncertainty and subjectivity of different discretization methods.The RDC consists of two steps:raw classification dataset generation and optimal dataset extraction.To evaluate the robustness of the proposed RDC method,Lushan County of Sichuan Province in China was chosen as the study area to generate the LSM based on three datasets(optimal dataset,original dataset with continuous values,and statistical dataset)using three popular ML methods,namely,convolution neural network,random forest,and logistic regression.The results show that the areas under the receiver operating characteristic curve(AUCs)of the optimal dataset for the abovementioned ML models are 0.963,0.961,and 0.930 which are higher than those of the original dataset(0.938,0.947,and 0.900)and statistical dataset(0.948,0.954,and 0.897).In conclusion,the RDC method can extract the more representative features from environmental factors and outperform the other conventional discretization methods.

The application of ResU-net and OBIA for landslide detection from multi-temporal Sentinel-2 images

Landslide detection is a hot topic in the remote sensing community,particularly with the current rapid growth in volume(and variety)of Earth observation data and the substantial progress of computer vision.Deep learning algorithms,especially fully convolutional networks(FCNs),and variations like the ResU-Net have been used recently as rapid and automatic landslide detection approaches.Although FCNs have shown cutting-edge results in automatic landslide detection,accuracy can be improved by adding prior knowledge through possible frameworks.This study evaluates a rulebased object-based image analysis(OBIA)approach built on probabilities resulting from the ResU-Net model for landslide detection.We train the ResU-Net model using a landslide dataset comprising landslide inventories from various geographic regions,including our study area and test the testing area not used for training.In the OBIA stage,we frst calculate land cover and image difference indices for pre-and post-landslide multi-temporal images.Next,we use the generated indices and the resulting ResU-Net probabilities for image segmentation;the extracted landslide object candidates are then optimized using rule-based classification.In the result validation section,the landslide detection of the proposed integration of the ResU-Net with a rule-based classification of OBIA is compared with that of the ResU-Net alone.Our proposed approach improves the mean intersection-over-union of the resulting map from the ResU-Net by more than 22%.

Landslide Hazard Mapping Using GIS and Weight of Evidence Model in Qingshui River Watershed of 2008 Wenchuan Earthquake Struck Region

Tens of thousands of landslides were triggered by May 12, 2008 earthquake over a broad area. The main purpose of this article is to apply and verify earthquake-triggered landslide hazard analysis techniques by using weight of evidence modeling in Qingshui （清水） River watershed, Deyang （德阳） City, Sichuan （四川） Province, China. Two thousand three hundred and twenty-one landslides were interpreted in the study area from aerial photographs and multi-source remote sensing imageries post-earthquake, verified by field surveys. The landslide inventory in the study area was established. A spatial database, including landslides and associated controlling parameters that may have influence on the occurrence of landslides, was constructed from topographic maps, geological maps, and enhanced thematic mapper （ETM＋） remote sensing imageries. The factors that influence landslide occurrence,such as slope angle, aspect, curvature, elevation, flow accumulation, distance from drainages, and distance from roads were calculated from the topographic maps. Lithology, distance from seismogenic fault, distance from all faults, and distance from stratigraphic boundaries were derived from the geological maps. Normalized difference vegetation index （NDV1） was extracted from ETM＋ images. Seismic intensity zoning was collected from Wenchuan （汶川） Ms8.0 Earthquake Intensity Distribution Map published by the China Earthquake Administration.Landslide hazard indices were calculated using the weight of evidence model, and landslide hazard maps were calculated from using different controlling parameters cases. The hazard map was compared with known landslide locations and verified. The success accuracy percentage of using all 13 controlling parameters was 71.82%. The resulting landslide hazard map showed five classes of landslide hazard, i.e., very high, high, moderate, low, and very low. The validation results showed satisfactory agreement between the hazard map and the existing landslides distribution data. The landslide hazard map can be used to identify and delineate unstable hazard-prone areas. It can also help planners to choose favorable locations for development schemes, such as infrastructural, buildings, road constructions, and environmental protection.