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.     

Machine learning techniques applied to prediction of residual strength of clay

Stability with first time or reactivated landslides depends upon the residual shear strength of soil. This paper describes prediction of the residual strength of soil based on index properties using two machine learning techniques. Different Artificial Neural Network (ANN) models and Support Vector Machine (SVM) techniques have been used. SVM aims at minimizing a bound on the generalization error of a model rather than at minimizing the error on the training data only. The ANN models along with their generalizations capabilities are presented here for comparisons. This study also highlights the capability of SVM model over ANN models for the prediction of the residual strength of soil. Based on different statistical parameters, the SVM model is found to be better than the developed ANN models. A model equation has been developed for prediction of the residual strength based on the SVM for practicing geotechnical engineers. Sensitivity analyses have been also performed to investigate the effects of different index properties on the residual strength of soil.     

基于支持向量机的水库诱发地震分析

将诱发水库地震的主要因素（岩性、岩体完整性、断层性质、库区区域应力状态、库区地震活动背景）划分为11个因子,并进行定量化;再根据每个样本到所属类内超平面的距离计算每个样本点的模糊因子,确定其对分类超平面影响大小;然后建立水库地震的支持向量机（SVM）和模糊支持向量机（FSVM）模型,并应用于水库诱发地震等级预测。实例分析表明,两种模型均可用于水库诱发地震等级预测,具有预测精度较高、考虑因素全面的特点,相比之下SVM模型预测结果略优于FSVM模型。另外,在应用SVM和FSVM进行分类时,如果样本离散性较高,则SVM模型优于FSVM模型;相反,如果样本离散性较低,则FSVM模型优于SVM模型。     

Is multi-hazard mapping effective in assessing natural hazards and integrated watershed management?

Natural hazards are often studied in isolation.However,there is a great need to examine hazards holistically to better manage the complex of threats found in any region.Many regions of the world have complex hazard landscapes wherein risk from individual and/or multiple extreme events is omnipresent.Extensive parts of Iran experience a complex array of natural hazards-floods,earthquakes,landslides,forest fires,subsidence,and drought.The effectiveness of risk mitigation is in part a function of whether the complex of hazards can be collectively considered,visualized,and evaluated.This study develops and tests individual and collective multihazard risk maps for floods,landslides,and forest fires to visualize the spatial distribution of risk in Fars Province,southern Iran.To do this,two well-known machine-learning algorithms-SVM and MARS-are used to predict the distribution of these events.Past floods,landslides,and forest fires were surveyed and mapped.The locations of occurrence of these events(individually and collectively) were randomly separated into training(70%) and testing(30%) data sets.The conditioning factors(for floods,landslides,and forest fires) employed to model the risk distributions are aspect,elevation,drainage density,distance from faults,geology,LULC,profile curvature,annual mean rainfall,plan curvature,distance from man-made residential structures,distance from nearest river,distance from nearest road,slope gradient,soil types,mean annual temperature,and TWI.The outputs of the two models were assessed using receiver-operating-characteristic(ROC) curves,true-skill statistics(TSS),and the correlation and deviance values from each models for each hazard.The areas-under-the-curves(AUC) for the MARS model prediction were 76.0%,91.2%,and 90.1% for floods,landslides,and forest fires,respectively.Similarly,the AUCs for the SVM model were 75.5%,89.0%,and 91.5%.The TSS reveals that the MARS model was better able to predict landslide risk,but was less able to predict flood-risk patterns and forest-fire risk.Finally,the combination of flood,forest fire,and landslide risk maps yielded a multi-hazard susceptibility map for the province.The better predictive model indicated that 52.3% of the province was at-risk for at least one of these hazards.This multi-hazard map may yield valuable insight for land-use planning,sustainable development of infrastructure,and also integrated watershed management in Fars Province.     

Back-propagation neural network and support vector machines for gold mineral prospectivity mapping in the Hatu region,Xinjiang, China

Machine Learning technologies have the potential to deliver new nonlinear mineral prospectivity mapping (MPM) models. In this study, Back Propagation (BP) neural network Support Vector Machine (SVM) methods were applied to MPM in the Hatu region of Xinjiang, northwestern China. First, a conceptual model of mineral prospectivity for Au deposits was constructed by analysis of geological background. Evidential layers were selected and transformed into a binary data format. Then, the processes of selecting samples and parameters were described. For the BP model, the parameters of the network were 9–10???1; for the SVM model, a radial basis function was selected as the kernel function with best C?=?1 and γ = 0.25. MPM models using these parameters were constructed, and threshold values of prediction results were determined by the concentration-area (C-A) method. Finally, prediction results from the BP neural network and SVM model were compared with that of a conventional method that is the weight- of- evidence (W- of- E). The prospectivity efficacy was evaluated by traditional statistical analysis, prediction-area (P-A) plots, and the receiver operating characteristic (ROC) technique. Given the higher intersection position (74% of the known deposits were within 26% of the total area) and the larger AUC values (0.825), the result shows that the model built by the BP neural network algorithm has a relatively better prediction capability for MPM. The BP neural network algorithm applied in MPM can elucidate the next investigative steps in the study area.     

Assessment of climate change impacts on rainfall using large scale climate variables and downscaling models – A case study

Many of the applied techniques in water resources management can be directly or indirectly influenced by hydro-climatology predictions. In recent decades, utilizing the large scale climate variables as predictors of hydrological phenomena and downscaling numerical weather ensemble forecasts has revolutionized the long-lead predictions. In this study, two types of rainfall prediction models are developed to predict the rainfall of the Zayandehrood dam basin located in the central part of Iran. The first seasonal model is based on large scale climate signals data around the world. In order to determine the inputs of the seasonal rainfall prediction model, the correlation coefficient analysis and the new Gamma Test (GT) method are utilized. Comparison of modelling results shows that the Gamma test method improves the Nash–Sutcliffe efficiency coefficient of modelling performance as 8% and 10% for dry and wet seasons, respectively. In this study, Support Vector Machine (SVM) model for predicting rainfall in the region has been used and its results are compared with the benchmark models such as K-nearest neighbours (KNN) and Artificial Neural Network (ANN). The results show better performance of the SVM model at testing stage. In the second model, statistical downscaling model (SDSM) as a popular downscaling tool has been used. In this model, using the outputs from GCM, the rainfall of Zayandehrood dam is projected under two climate change scenarios. Most effective variables have been identified among 26 predictor variables. Comparison of the results of the two models shows that the developed SVM model has lesser errors in monthly rainfall estimation. The results show that the rainfall in the future wet periods are more than historical values and it is lower than historical values in the dry periods. The highest monthly uncertainty of future rainfall occurs in March and the lowest in July.     

Improved flood susceptibility mapping using a best first decision tree integrated with ensemble learning techniques

Improving the accuracy of flood prediction and mapping is crucial for reducing damage resulting from flood events. In this study, we proposed and validated three ensemble models based on the Best First Decision Tree (BFT) and the Bagging (Bagging-BFT), Decorate (Bagging-BFT), and Random Subspace (RSS-BFT) ensemble learning techniques for an improved prediction of flood susceptibility in a spatially-explicit manner. A total number of 126 historical flood events from the Nghe An Province (Vietnam) were connected to a set of 10 flood influencing factors (slope, elevation, aspect, curvature, river density, distance from rivers, flow direction, geology, soil, and land use) for generating the training and validation datasets. The models were validated via several performance metrics that demonstrated the capability of all three ensemble models in elucidating the underlying pattern of flood occurrences within the research area and predicting the probability of future flood events. Based on the Area Under the receiver operating characteristic Curve (AUC), the ensemble Decorate-BFT model that achieved an AUC value of 0.989 was identified as the superior model over the RSS-BFT (AUC = 0.982) and Bagging-BFT (AUC = 0.967) models. A comparison between the performance of the models and the models previously reported in the literature confirmed that our ensemble models provided a reliable estimate of flood susceptibilities and their resulting susceptibility maps are trustful for flood early warning systems as well as development of mitigation plans.     

Predictive Models for Permeability of Cracked Rock Masses Based on Support Vector Machine Techniques

In this study, a database developed from existing literature about permeability of cracked rock was established. The performance of Support Vector Machine (SVM) combined with optimisation algorithms: Genetic Algorithm (GA) and Particle Swarm Optimisation Algorithm (PSO) in predicting the permeability of cracked rock masses (CRM) is evaluated. Also, the sensitivity analysis of the influence factors to the permeability of CRM is conducted. The results indicate that the hybrid GA–SVM and hybrid PSO–SVM models can accurately predict the permeability of CRM in terms of the statistical performance criteria: Coefficient of Determination R², Regression Coefficient R and Mean Residual Error (MSE); Additionally, optimisation algorithms: PSO and GA can improve significantly the predictive performance of the SVM model. Based on the sensitivity analysis, crack angle is the most important factor to change the permeability of CRM, followed by confining pressure.

     

Modelling of shallow landslides with machine learning algorithms

This paper introduces three machine learning(ML)algorithms,the‘ensemble'Random Forest(RF),the‘ensemble'Gradient Boosted Regression Tree(GBRT)and the Multi Layer Perceptron neural network(MLP)and applies them to the spatial modelling of shallow landslides near Kvam in Norway.In the development of the ML models,a total of 11 significant landslide controlling factors were selected.The controlling factors relate to the geomorphology,geology,geo-environment and anthropogenic effects:slope angle,aspect,plan curvature,profile curvature,flow accumulation,flow direction,distance to rivers,water content,saturation,rainfall and distance to roads.It is observed that slope angle was the most significant controlling factor in the ML analyses.The performance of the three ML models was evaluated quantitatively based on the Receiver Operating Characteristic(ROC)analysis.The results show that the‘ensemble'GBRT machine learning model yielded the most promising results for the spatial prediction of shallow landslides,with a 95%probability of landslide detection and 87%prediction efficiency.     

A comparative study on the landslide susceptibility mapping using evidential belief function and weights of evidence models

The purpose of this study is to produce landslide susceptibility map of a landslide-prone area (Daguan County, China) by evidential belief function (EBF) model and weights of evidence (WoE) model to compare the results obtained. For this purpose, a landslide inventory map was constructed mainly based on earlier reports and aerial photographs, as well as, by carrying out field surveys. A total of 194 landslides were mapped. Then, the landslide inventory was randomly split into a training dataset; 70% (136 landslides) for training the models and the remaining 30% (58 landslides) was used for validation purpose. Then, a total number of 14 conditioning factors, such as slope angle, slope aspect, general curvature, plan curvature, profile curvature, altitude, distance from rivers, distance from roads, distance from faults, lithology, normalized difference vegetation index (NDVI), sediment transport index (STI), stream power index (SPI), and topographic wetness index (TWI) were used in the analysis. Subsequently, landslide susceptibility maps were produced using the EBF and WoE models. Finally, the validation of landslide susceptibility map was accomplished with the area under the curve (AUC) method. The success rate curve showed that the area under the curve for EBF and WoE models were of 80.19% and 80.75% accuracy, respectively. Similarly, the validation result showed that the susceptibility map using EBF model has the prediction accuracy of 80.09%, while for WoE model, it was 79.79%. The results of this study showed that both landslide susceptibility maps obtained were successful and would be useful for regional spatial planning as well as for land cover planning.     

Application of statistical learning algorithms to ultimate bearing capacity of shallow foundation on cohesionless soil

This study employs two statistical learning algorithms (Support Vector Machine (SVM) and Relevance Vector Machine (RVM)) for the determination of ultimate bearing capacity (q_u) of shallow foundation on cohesionless soil. SVM is firmly based on the theory of statistical learning, uses regression technique by introducing varepsilon‐insensitive loss function. RVM is based on a Bayesian formulation of a linear model with an appropriate prior that results in a sparse representation. It also gives variance of predicted data. The inputs of models are width of footing (B), depth of footing (D), footing geometry (L/B), unit weight of sand (γ) and angle of shearing resistance (?). Equations have been developed for the determination of q_u of shallow foundation on cohesionless soil based on the SVM and RVM models. Sensitivity analysis has also been carried out to determine the effect of each input parameter. This study shows that the developed SVM and RVM are robust models for the prediction of q_u of shallow foundation on cohesionless soil. Copyright © 2010 John Wiley & Sons, Ltd.     

基于BMA多模型组合的疏勒河径流预测研究

准确可靠的径流预测对于水资源的科学管理与规划具有重要意义,特别是在水资源紧缺的干旱半干旱地区,径流预测对流域内水资源高效利用与水利工程经济运行具有重要现实意义。针对径流预测通常采用单一方法进行建模与预测,难以利用各预测模型优势的问题,本文利用极限学习机（ELM）模型、支持向量机（SVM）模型、多元自适应回归样条（MARS）等机器学习方法建立了疏勒河上游未来1~7日的径流预测模型。在此基础上,运用贝叶斯模型平均（BMA）方法对ELM、SVM、MARS模型的预测结果进行组合,构建了径流组合预测模型,以获取更可靠的预测结果,并采用蒙特卡洛抽样方法获取BMA的95%置信区间,对预测结果进行了不确定性分析。结果表明：ELM、SVM、MARS模型以及BMA组合模型均适用于干旱半干旱地区的中长期日径流预测;BMA的预测精度较单一模型更高,能够提供更准确的预测值;BMA的95%置信区间对实测值覆盖率高,同时能够提供较好的确定性预测和概率预测结果。表明BMA在资料有限的条件下,表现出较单一模型更高的预测性能,可以成为干旱半干旱地区中长期日径流预测的有效方法。     

Disaster prediction model based on support vector machine for regression and improved differential evolution

The kernel parameters setting of SVM influences prediction precision. The hybrid model based on SVM for regression and improved differential evolution is proposed to enhance the prediction precision. The improved differential evolution is used to optimize the kernel parameters. The improved differential evolution algorithm employs two trial vector generation strategies and two control parameter settings. The first-generation strategy is with best solution, and the second strategy is without best solution. Three categories of disasters time series including flood, drought and storm from Ministry of agriculture of China are used to verify the validity of the proposed model. Compared with the grid SVM and other models, the proposed hybrid model improves the prediction precision of SVM.     

A Comparative Study of Least Square Support Vector Machines and Multiclass Alternating Decision Trees for Spatial Prediction of Rainfall-Induced Landslides in a Tropical Cyclones Area

The objective of this study is to explore and compare the least square support vector machine (LSSVM) and multiclass alternating decision tree (MADT) techniques for the spatial prediction of landslides. The Luc Yen district in Yen Bai province (Vietnam) has been selected as a case study. LSSVM and MADT are effective machine learning techniques of classification applied in other fields but not in the field of landslide hazard assessment. For this, Landslide inventory map was first constructed with 95 landslide locations identified from aerial photos and verified from field investigations. These landslide locations were then divided randomly into two parts for training (70 % locations) and validation (30 % locations) processes. Secondly, landslide affecting factors such as slope, aspect, elevation, curvature, lithology, land use, distance to roads, distance to faults, distance to rivers, and rainfall were selected and applied for landslide susceptibility assessment. Subsequently, the LSSVM and MADT models were built to assess the landslide susceptibility in the study area using training dataset. Finally, receiver operating characteristic curve and statistical index-based evaluations techniques were employed to validate the predictive capability of these models. As a result, both the LSSVM and MADT models have high performance for spatial prediction of landslides in the study area. Out of these, the MADT model (AUC = 0.853) outperforms the LSSVM model (AUC = 0.803). From the landslide study of Luc Yen district in Yen Bai province (Vietnam), it can be conclude that the LSSVM and MADT models can be applied in other areas of world also for and spatial prediction. Landslide susceptibility maps obtained from this study may be helpful in planning, decision making for natural hazard management of the areas susceptible to landslide hazards.     

基于PCA-PSO-SVM的凡口铅锌矿地区岩溶地表塌陷危险性预测

岩溶地表塌陷是由多个影响因素共同作用导致地面形成塌陷坑（洞）的一种动力地质现象,具有隐蔽性和突发性的特点,常规简单数学模型难以对地表塌陷危险性准确预测。文章先通过主成分分析法（PCA）对选取的地下水位、地下水位波动幅度、给水度等11个影响因素提取5个主成分,对导致地表塌陷危险性的主成分进行全新的解释,同时引入粒子群算法（PSO）优化的支持向量机（SVM）方法,建立PCA-PSO-SVM岩溶地表塌陷危险性预测模型,并结合凡口铅锌矿地区工程实例,将预测结果与单一的SVM模型预测结果进行对比,表明PCA-PSO-SVM危险性预测模型精度更高,可以更好地为岩溶地表塌陷防治工作提供依据。      

Landslide susceptibility zonation using GIS and evidential belief function model

The main goal of this paper is to generate a landslide susceptibility map through evidential belief function (EBF) model by using Geographic Information System (GIS) for Qianyang County, Shaanxi Province, China. At first, a detailed landslide inventory map was prepared, and the following ten landslide-conditioning factors were collected: slope angle, slope aspect, curvature, plan curvature, profile curvature, altitude, distance to rivers, geomorphology, lithology, and rainfall. The landslides were detected from the interpretation of aerial photographs and supported by field surveys. A total of 81 landslides were randomly split into the following two parts: the training dataset 70 % (56 landslides) were used for establishing the model and the remaining 30 % (25 landslides) were used for the model validation. The ArcGIS was used to analyze landslide-conditioning factors and evaluate landslide susceptibility; as a result, a landslide susceptibility map was generated by using EBF and ArcGIS 10.0, thus divided into the following five susceptibility classes: very low, low, moderate, high, and very high. Finally, when we validated the accuracy of the landslide susceptibility map, both the success-rate and prediction-rate curve methods were applied. The results reveal that a final susceptibility map has the success rate of 83.31 % and the prediction rate of 79.41 %.     

逻辑回归与支持向量机模型在滑坡敏感性评价中的应用

白龙江流域是我国滑坡泥石流灾害四大高发区之一,进行该区域滑坡敏感性评价,能够为决策者在灾害管理和设施建设规划方面提供帮助,对区域防灾减灾具有重要指导意义。本研究采用边坡单元为基本研究单元,在野外调查及前人研究基础上,选择控制该区域滑坡发育的19个要素作为影响因子; 经过主成分分析和独立性检验得到该区域对滑坡形成贡献最大的6个因子:高程、坡度、坡向、岩性、断裂距离和人口密度; 分别使用二元逻辑回归模型（LR）和支持向量机模型（SVM）对该区域进行滑坡敏感性评价; 最后,采用ROC曲线对模型精度进行验证。研究结果表明,两模型各能将38.76%、14.48%、9.40%、11.28%、26.07%和13.49%、21.61%、8.17%、26.70%、30.04%的边坡单元分别预测为极高危险区、高危险区、中度危险区、低危险区和极低危险区; 精度验证结果表明两种模型均能有效地进行该区域滑坡敏感性评价,并且支持向量机模型具有更好的分类能力、预测精度和稳定性。     

Landslide displacement prediction based on combining method with optimal weight

Predicting the deformation and evolution tendency of landslides is essential to landslide disaster prevention and mitigation. At present, most of the proposed models for landslide displacement prediction belong to single models. It is difficult to accurately describe the deformation and evolution law only by a single model for the complexity of landslides and limitation of the models. In this paper, we presented an application of linear combination model with optimal weight in landslide displacement prediction. We took Huanlongxicun and Saleshan landslides in Gansu province of China as examples, firstly to build GM(1,1) and Verhulst models for displacement prediction of the two landslides; then build two linear combination models of the two landslides, on the basis of the combining theory with optimal weight and the prediction results of the GM(1,1) and Verhulst models. The results show that the prediction accuracies of the combining models are much higher than those of the single models for both Huanglongxicun landslide and Saleshan landslide. Therefore, the combining model with optimal weight is an effective and feasible method to further improve accuracy for landslide displacement prediction.     

滑坡灾害空间预测支持向量机模型及其应用

随着GIS技术在滑坡灾害空间预测研究中的广泛应用,滑坡灾害空间预测模型成为研究的热点问题。在总结滑坡灾害空间预测研究现状的基础上,简要介绍了两类和单类支持向量机的基本原理。以香港自然滑坡空间预测为例,采用两类和单类支持向量机进行滑坡灾害空间预测,并与Logistic回归模型进行了比较。结果表明,两类支持向量机模型优于Logistic回归模型,而Logistic回归模型优于单类支持向量机模型。     

Landslide susceptibility assessment using SVM machine learning algorithm

This paper introduces the current machine learning approach to solving spatial modeling problems in the domain of landslide susceptibility assessment. The latter is introduced as a classification problem, having multiple (geological, morphological, environmental etc.) attributes and one referent landslide inventory map from which to devise the classification rules. Three different machine learning algorithms were compared: Support Vector Machines, Decision Trees and Logistic Regression. A specific area of the Fruška Gora Mountain (Serbia) was selected to perform the entire modeling procedure, from attribute and referent data preparation/processing, through the classifiers' implementation to the evaluation, carried out in terms of the model's performance and agreement with the referent data. The experiments showed that Support Vector Machines outperformed the other proposed methods, and hence this algorithm was selected as the model of choice to be compared with a common knowledge-driven method – the Analytical Hierarchy Process – to create a landslide susceptibility map of the relevant area. The SVM classifier outperformed the AHP approach in all evaluation metrics (κ index, area under ROC curve and false positive rate in stable ground class).