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。     

Weights-of-evidence and logistic regression modeling of magmatic nickel sulfide prospectivity in the Yilgarn Craton,Western Australia

Bayesian weight-of-evidence and logistic regression models are implemented in a GIS environment for regional-scale prospectivity modeling of greenstone belts in the Yilgarn Craton, Western Australia, for magmatic nickel sulfide deposits. The input variables for the models consisted of derivative GIS layers that were used as proxies for mappable exploration criteria for magmatic nickel sulfide deposits in the Yilgarn. About 70% of the 165 known deposits of the craton were used to train the models; the remaining 30% was used to validate the models and, therefore, had to be treated as if they had not been discovered. The weights-of-evidence and logistic regression models, respectively, classify 71.4% and 81.6% validation deposits in prospective zones that occupy about 9% of the total area occupied by the greenstone belts in the craton. The superior performance of the logistic regression model is attributed to its capability to accommodate conditional dependencies amongst the input predictor maps, and provide less biased estimates of prospectivity.     

Landslide hazard mapping using logistic regression model in Mackenzie Valley,Canada

A logistic regression model is developed within the framework of a Geographic Information System (GIS) to map landslide hazards in a mountainous environment. A case study is conducted in the mountainous southern Mackenzie Valley, Northwest Territories, Canada. To determine the factors influencing landslides, data layers of geology, surface materials, land cover, and topography were analyzed by logistic regression analysis, and the results are used for landslide hazard mapping. In this study, bedrock, surface materials, slope, and difference between surface aspect and dip direction of the sedimentary rock were found to be the most important factors affecting landslide occurrence. The influence on landslides by interactions among geologic and geomorphic conditions is also analyzed, and used to develop a logistic regression model for landslide hazard mapping. The comparison of the results from the model including the interaction terms and the model not including the interaction terms indicate that interactions among the variables were found to be significant for predicting future landslide probability and locating high hazard areas. The results from this study demonstrate that the use of a logistic regression model within a GIS framework is useful and suitable for landslide hazard mapping in large mountainous geographic areas such as the southern Mackenzie Valley.     

Doline probability map using logistic regression and GIS technology in the central Ebro Basin (Spain)

In the surroundings of Zaragoza, karstification processes are especially intense in covered karst areas where fluvial terraces lie directly on Tertiary evaporites. Since the beginning of Quaternary, these processes have lead to the development of collapse and subsidence dolines with a wide range of sizes, which have significant economic impacts. To reduce economic impact and increase safety, a regional analysis of this phenomenon is needed for spatial management. Therefore, a probability map of dolines was developed using logistic regression and geographic information system (GIS) techniques. This paper covers the selection of input data, manipulation of data using the GIS technology, and the use of logistic regression to generate a doline probability map. The primary variable in the doline development in this area is geomorphology, represented by the location of endorheic areas and different terrace levels. Secondary variables are the presence of irrigation and the water table gradient.     

Delineation of landslide hazard areas on Penang Island, Malaysia, by using frequency ratio, logistic regression, and artificial neural network models

This paper summarizes findings of landslide hazard analysis on Penang Island, Malaysia, using frequency ratio, logistic regression, and artificial neural network models with the aid of GIS tools and remote sensing data. Landslide locations were identified and an inventory map was constructed by trained geomorphologists using photo-interpretation from archived aerial photographs supported by field surveys. A SPOT 5 satellite pan sharpened image acquired in January 2005 was used for land-cover classification supported by a topographic map. The above digitally processed images were subsequently combined in a GIS with ancillary data, for example topographical (slope, aspect, curvature, drainage), geological (litho types and lineaments), soil types, and normalized difference vegetation index (NDVI) data, and used to construct a spatial database using GIS and image processing. Three landslide hazard maps were constructed on the basis of landslide inventories and thematic layers, using frequency ratio, logistic regression, and artificial neural network models. Further, each thematic layer’s weight was determined by the back-propagation training method and landslide hazard indices were calculated using the trained back-propagation weights. The results of the analysis were verified and compared using the landslide location data and the accuracy observed was 86.41, 89.59, and 83.55% for frequency ratio, logistic regression, and artificial neural network models, respectively. On the basis of the higher percentages of landslide bodies predicted in very highly hazardous and highly hazardous zones, the results obtained by use of the logistic regression model were slightly more accurate than those from the other models used for landslide hazard analysis. The results from the neural network model suggest the effect of topographic slope is the highest and most important factor with weightage value (1.0), which is more than twice that of the other factors, followed by the NDVI (0.52), and then precipitation (0.42). Further, the results revealed that distance from lineament has the lowest weightage, with a value of 0. This shows that in the study area, fault lines and structural features do not contribute much to landslide triggering.     

Optimized neural network using beetle antennae search for predicting the unconfined compressive strength of jet grouting coalcretes

This investigation studied the coalcrete, a new supporting material produced by jet grouting (JG) for supporting surrounding coal seams. For support design, the unconfined compressive strength (UCS) of the coalcrete is an essential parameter to evaluate the jet grouting effect in coal mines. In this study, an intelligent technique was proposed for predicting the UCS of the coalcrete by combining back propagation neural network (BPNN) and beetle antennae search (BAS). The architecture of BPNN was first tuned by BAS, and then, the optimized BPNN-BAS model was subsequently used for nonlinear relationship modeling. Several crucial influencing variables including water-cement ratio, coal-grout ratio, and curing time were selected as the inputs. By combining these variables, 360 coalcrete samples were prepared in a controlled laboratory environment and tested for establishing the dataset. The results demonstrate that BAS can tune the BPNN architecture more efficiently compared with random selection. Moreover, in comparison with multiple regression (MLR) and logistic regression (LR), and support vector machine (SVM), the optimized BPNN-BAS model is more reliable and accurate for predicting coalcrete strength. Sensitivity analysis (SA) was used to obtain the variable importance, and the results demonstrate that curing time affects the UCS of the coalcrete most strongly, followed by water-cement ratio and coal-grout ratio. The success of this study provides an accurate and brief approach to coalcrete strength prediction.     

Analysis of soil erosion characteristics in small watersheds with particle swarm optimization, support vector machine, and artificial neuronal networks

Sand production by soil erosion in small watershed is a complex physical process. There are few physical models suitable to describe the characteristics of the intense erosion in domestic loess plateau. Introducing support vector machine (SVM) oriented to small sample data and possessing good extension property can be an effective approach to predict soil erosion because SVM has been applied in hydrological prediction to some extent. But there are no effective methods to select the rational parameters for SVM, which seriously limited the practical application of SVM. This paper explored the application of intelligence-based particle swarm optimization (PSO) algorithm in automatic selection of parameters for SVM, and proposed a prediction model by linking PSO and SVM for small sample data analysis. This method utilized the high efficiency optimization property and swarm paralleling property of PSO algorithm and the relatively strong learning and extending capacity of SVM. For an example of Huangfuchuan small watershed, its intensive fragmentation and intense erosion earn itself the name of “worst erosion in the world”. Using four characteristics selection algorithms of correlation feature selection, the primary affecting factors for soil erosion in this small watershed were determined to be the channel density, ravine area, sand rock proportion, and the total vegetation coverage. Based on the proposed PSO–SVM algorithm, the soil erosion modulus in the small watershed was predicted. The accuracy of the simulation and prediction was good, and the average error was 3.85%. The SVM predicting model was based on the monitoring data of sand production. The construction of the SVM erosion modulus prediction model for the small watershed comprehensively reflected the complex mechanism of soil erosion and sand production. It had certain advantage and relatively high practical value in small sample prediction in the discipline of soil erosion.     

Comparison of landslide susceptibility mapping methodologies for Koyulhisar,Turkey: conditional probability,logistic regression,artificial neural networks,and support vector machine

This case study presented herein compares the GIS-based landslide susceptibility mapping methods such as conditional probability (CP), logistic regression (LR), artificial neural networks (ANNs) and support vector machine (SVM) applied in Koyulhisar (Sivas, Turkey). Digital elevation model was first constructed using GIS software. Landslide-related factors such as geology, faults, drainage system, topographical elevation, slope angle, slope aspect, topographic wetness index, stream power index, normalized difference vegetation index, distance from settlements and roads were used in the landslide susceptibility analyses. In the last stage of the analyses, landslide susceptibility maps were produced from ANN, CP, LR, SVM models, and they were then compared by means of their validations. However, area under curve values obtained from all four methodologies showed that the map obtained from ANN model looks like more accurate than the other models, accuracies of all models can be evaluated relatively similar. The results also showed that the CP is a simple method in landslide susceptibility mapping and highly compatible with GIS operating features. Susceptibility maps can be easily produced using CP, because input process, calculation and output processes are very simple in CP model when compared with the other methods considered in this study.     

Integrating an artificial intelligence approach with k-means clustering to model groundwater salinity: the case of Gaza coastal aquifer (Palestine)

Artificial intelligence (AI) techniques have increasingly become efficient alternative modeling tools in the water resources field, particularly when the modeled process is influenced by complex and interrelated variables. In this study, two AI techniques—artificial neural networks (ANNs) and support vector machine (SVM)—were employed to achieve deeper understanding of the salinization process (represented by chloride concentration) in complex coastal aquifers influenced by various salinity sources. Both models were trained using 11 years of groundwater quality data from 22 municipal wells in Khan Younis Governorate, Gaza, Palestine. Both techniques showed satisfactory prediction performance, where the mean absolute percentage error (MAPE) and correlation coefficient (R) for the test data set were, respectively, about 4.5 and 99.8% for the ANNs model, and 4.6 and 99.7% for SVM model. The performances of the developed models were further noticeably improved through preprocessing the wells data set using a k-means clustering method, then conducting AI techniques separately for each cluster. The developed models with clustered data were associated with higher performance, easiness and simplicity. They can be employed as an analytical tool to investigate the influence of input variables on coastal aquifer salinity, which is of great importance for understanding salinization processes, leading to more effective water-resources-related planning and decision making.     

Support Vector Machines for Landslide Susceptibility Mapping: The Staffora River Basin Case Study, Italy

The aim of this study is the application of support vector machines (SVM) to landslide susceptibility mapping. SVM are a set of machine learning methods in which model capacity matches data complexity. The research is based on a conceptual framework targeted to apply and test all the procedural steps for landslide susceptibility modeling from model selection, to investigation of predictive variables, from empirical cross-validation of results, to analysis of predicted patterns. SVM were successfully applied and the final susceptibility map was interpreted via success and prediction rate curves and receiver operating characteristic (ROC) curves, to support the modeling results and assess the robustness of the model. SVM appeared to be very specific learners, able to discriminate between the informative input and random noise. About 78% of occurrences was identified within the 20% of the most susceptible study area for the cross-validation set. Then the final susceptibility map was compared with other maps, addressed by different statistical approaches, commonly used in susceptibility mapping, such as logistic regression, linear discriminant analysis, and naive Bayes classifier. The SVM procedure was found feasible and able to outperform other techniques in terms of accuracy and generalization capacity. The over-performance of SVM against the other techniques was around 18% for the cross-validation set, considering the 20% of the most susceptible area. Moreover, by analyzing receiver operating characteristic (ROC) curves, SVM appeared to be less prone to false positives than the other models. The study was applied in the Staffora river basin (Lombardy, Northern Italy), an area of about 275 km² characterized by a very high density of landslides, mainly superficial slope failures triggered by intense rainfall events.     

Explorative data analysis of heavy metal contaminated soil using multidimensional spatial regression

To obtain data on heavy metal contaminated soil requires laborious and time-consuming data sampling and analysis. Not only has the contamination to be measured, but also additional data characterizing the soil and the boundary conditions of the site, such as pH, land use, and soil fertility. For an integrative approach, combining the analysis of spatial distribution, and of factors influencing the contamination, and its treatment, the Mollifier interpolation was used, which is a non-parametric kernel density regression. The Mollifier was capable of including additional independent variables (beyond the spatial dimensions x and y) in the spatial interpolation and hence explored the combined influence of spatial and other variables, such as land use, on the heavy metal distribution. The Mollifier could also represent the interdependence between different heavy metal concentrations and additional site characteristics. Although the uncertainty measure supplied by the Mollifier at first seems somewhat unusual, it is a valuable feature and supplements the geostatistical uncertainty assessment.     

Isolation of wells contaminated by tannery industries using principal component analysis

The groundwater in the upper Kodaganar basin is contaminated due to the discharge of effluents from tannery industries. The water in the wells, whose physico-chemical characteristics are altered due to the influence of the effluents, is statistically analyzed. The physico-chemical variables such as EC, Na⁺, K⁺, Ca²⁺, Mg²⁺, F^?, Cl^?, HCO₃ ^?,CO₃ ^2?, NO₃ ^?, SO₄ ^2?, pH, and Cr_total were used for this study. An attempt was made to identify the contaminated wells based on suitability for drinking, suitability for industrial requirements, and through principal component analysis (PCA). Classification based on suitability helped in identifying the contaminated wells. However, this resulted in failure when identifying the wells that are contaminated by tanneries. PCA has proved to be effective in the segregation of contaminated wells influenced by tannery industries. The physico-chemical variables that are 13 in number are transformed into two orthogonal components and Eigen values based on the variance. The Eigen values are used to select the first two principal components PC1 (7.26) and PC2 (2.24) that accounted for 73.04% variance in the data. The components of the variables and the wells are plotted in a biplot to isolate the contaminated samples. The contaminated samples are analyzed in the spatial domain in geographic information system and found to be clustered around the tannery belt. The study reveals that 35% of the samples are contaminated due to discharge from tannery industries.     

A GIS-based landslide susceptibility evaluation using bivariate and multivariate statistical analyses

Bivariate and multivariate statistical analyses were used to predict the spatial distribution of landslides in the Cuyahoga River watershed, northeastern Ohio, U.S.A. The relationship between landslides and various instability factors contributing to their occurrence was evaluated using a Geographic Information System (GIS) based investigation. A landslide inventory map was prepared using landslide locations identified from aerial photographs, field checks, and existing literature. Instability factors such as slope angle, soil type, soil erodibility, soil liquidity index, landcover pattern, precipitation, and proximity to stream, responsible for the occurrence of landslides, were imported as raster data layers in ArcGIS, and ranked using a numerical scale corresponding to the physical conditions of the region. In order to investigate the role of each instability factor in controlling the spatial distribution of landslides, both bivariate and multivariate models were used to analyze the digital dataset. The logistic regression approach was used in the multivariate model analysis. Both models helped produce landslide susceptibility maps and the suitability of each model was evaluated by the area under the curve method, and by comparing the maps with the known landslide locations. The multivariate logistic regression model was found to be the better model in predicting landslide susceptibility of this area. The logistic regression model produced a landslide susceptibility map at a scale of 1:24,000 that classified susceptibility into four categories: low, moderate, high, and very high. The results also indicated that slope angle, proximity to stream, soil erodibility, and soil type were statistically significant in controlling the slope movement.     

Modeling of tensile strength of rocks materials based on support vector machines approaches

In the predicting of geological variables, artificial neural networks (ANNs) have some drawbacks including possibility of getting trapped in local minima, over training, subjectivity in the determining of model parameters and the components of its complex structure. Recently, support vector machines (SVM) has been found to be popular in prediction studies due to its some advantages over ANNs. Because the least squares SVM (LS‐SVM) provides a computational advantage over SVM by converting quadratic optimization problem into a system of linear equations, LS‐SVM method is also tried in study. The main purpose of this study is to examine the capability of these two SVM algorithms for the prediction of tensile strength of rock materials and to compare its performance with ANN and linear regression (MLR) models. Total porosity, sonic velocity, slake durability index and aggregate impact value were used as input in modeling applications. Favorite performance evaluation measures were employed to assess developed models. The results determined in study indicate that the SVM, LS‐SVM and ANN methods are successful tools for prediction of tensile strength variable and can give good prediction performances than MLR model. Although these three methods are powerful artificial intelligence techniques, LS‐SVM makes the running time considerably faster with the higher accuracy. In terms of accuracy, the LS‐SVM model resulted in error reductions relative to that of the other models. Copyright © 2012 John Wiley & Sons, Ltd.     

A comparative study of landslide susceptibility mapping using weight of evidence,logistic regression and support vector machine and evaluated by SBAS-InSAR monitoring: Zhouqu to Wudu segment in Bailong River Basin,China

The determining of landslide-prone areas in mountainous terrain is essential for land planning and hazard mitigation. In this paper, a comparative study using three statistical models including weight of evidence model (WoE), logistic regression model (LR) and support vector machine method (SVM) was undertaken in the Zhouqu to Wudu segment in the Bailong River Basin, Southern Gansu, China. Six conditionally independent environmental factors, elevation, slope, aspect, distance from fault, lithology and settlement density, were selected as the explanatory variables that may contribute to landslide occurrence based on principal component analysis (PCA) and Chi-square test. The relation between landslide distributions and these variables was analyzed using the three models and the results then used to calculate the landslide susceptibility (LS). The performance of the models was then evaluated using both the highly accurate deformation signals produced by using the Small Baseline Subset Interferometric Synthetic Aperture Radar technique and Receiver Operating Characteristic (ROC) curve. Results show more deformation points in areas with high and very high LS levels, and also more stable points in areas with low and very low LS levels for the SVM model. In addition, the SVM has larger area under the ROC curve. It indicates that the SVM has better prediction accuracy and classified ability. For the interpretability, the WoE derives the class of factors that most contributed to landsliding in the study area, and the LR reveals that factors including elevation, settlement density and distance from fault played major roles in landslide occurrence and distribution, whereas the SVM cannot provide relative weights for the variables. The outperformed SVM could be employed to determine potential landslide zones in the study area. Outcome of this research would provide preliminary basis for general land planning such as choosing new urban areas and infrastructure construction in the future, as well as for landslide hazard mitigation in Bailong River Basin.     

Predicting natural arsenic contamination of bedrock groundwater for a local region in Korea and its application

A logistic regression model for the probability of arsenic exceeding the drinking water guidelines (10 μg/L) in bedrock groundwater was developed for a selected county in Korea, where arsenic occurrence and release reactions have been investigated. Arsenic was enriched naturally by the oxidation of sulfide minerals in metasedimentary rocks and mineralized zones, and due to high mobility in alkaline pH conditions, concentrations were high in groundwater of the county. When considering these reactions of arsenic release and water quality characteristics, several geological and geochemical factors were selected as influencing variables in the model. In the final logistic regression model, geological units of limestone and metasedimentary rocks, the concentrations of nitrate and sulfate, and distances to closed mines and adjacent granite were retained as statistically significant variables. Predicted areas of high probability agreed well with known spatial contamination patterns in the county. The model was also applied to an adjacent county, where the groundwater has not previously been tested for the presence of arsenic, and a probability map for arsenic contamination was then produced. Through the analysis of arsenic concentrations at the wells of high probability, it was determined that the applied model accurately indicated the arsenic contamination of groundwater. The logistic regression approach of this study can be applied to predict arsenic contamination in areas of similar geological and geochemical conditions to the county used in this model.     

Modelling shallow landslide susceptibility: a new approach in logistic regression by using favourability assessment

A new method for estimating shallow landslide susceptibility by combining Geographical Information System (GIS), nonparametric kernel density estimation and logistic regression is described. Specifically, a logistic regression is applied to predict the spatial distribution by estimating the probability of occurrence of a landslide in a 16 km² area. For this purpose, a GIS is employed to gather the relevant sample information connected with the landslides. The advantages of pre-processing the explanatory variables by nonparametric density estimation (for continuous variables) and a reclassification (for categorical/discrete ones) are discussed. The pre-processing leads to new explanatory variables, namely, some functions which measure the favourability of occurrence of a landslide. The resulting model correctly classifies 98.55% of the inventaried landslides and 89.80% of the landscape surface without instabilities. New data about recent shallow landslides were collected in order to validate the model, and 92.20% of them are also correctly classified. The results support the methodology and the extrapolation of the model to the whole study area (278 km²) in order to obtain susceptibility maps.     

Modelling pile capacity using Gaussian process regression

This paper investigates the potential of a Gaussian process (GP) regression approach to predict the load-bearing capacity of piles. Support vector machines (SVM) and empirical relations were used to compare the performance of the GP regression approach. The first dataset used in this study was derived from actual pile-driving records in cohesion-less soil. Out of a total of 94 pieces of data, 59 were used to train and the remaining 35 data were used to test the created models. A radial basis function and Pearson VII function kernels were used with both GP and SVM. The results from this dataset indicate improved performance by GP regression in comparison to SVM and empirical relations. To validate the performance of the GP regression approach, another dataset consisting of 38 pieces of data was considered. The results from this dataset also suggest improved performance by the Pearson VII function kernel-based GP regression modelling approach in comparison to SVM.     

Evaluation of generalized linear models for soil liquefaction probability prediction

A soil deposit subjected to seismic loading can be viewed as a binary system: it will either liquefy or not liquefy. Generalized linear models are versatile tools for predicting the response of a binary system and hence potentially applicable to liquefaction prediction. In this study, the applicability of four generalized linear models (i.e., logistic, probit, log–log, and c-log–log) for liquefaction potential evaluation is assessed and compared. Eight liquefaction models based on the four generalized linear models and two sets of explanatory variables are evaluated. These models are first calibrated with past liquefaction performance data. A weighted-likelihood function method is used to consider the sampling bias in the calibration database. The predicted liquefaction probabilities from various models are then compared. When liquefaction probability is small, the predicted liquefaction probability is sensitive to the regression models used. The effect of sampling bias is more marked in the high cyclic stress ratio region. The eight models are finally ranked using a Bayesian model comparison method. For the generalized linear models examined, the logistic and c-log–log regression models are most supported by the past performance data. On the other hand, the probit and c-log–log regression models are much less applicable to liquefaction prediction.     

Ensemble-based landslide susceptibility maps in Jinbu area, Korea

Ensemble techniques were developed, applied and validated for the analysis of landslide susceptibility in Jinbu area, Korea using the geographic information system (GIS). Landslide-occurrence areas were detected in the study by interpreting aerial photographs and field survey data. Landslide locations were randomly selected in a 70/30 ratio for training and validation of the models, respectively. Topography, geology, soil and forest databases were also constructed. Maps relevant to landslide occurrence were assembled in a spatial database. Using the constructed spatial database, 17 landslide-related factors were extracted. The relationships between the detected landslide locations and the factors were identified and quantified by frequency ratio, weight of evidence, logistic regression and artificial neural network models and their ensemble models. The relationships were used as factor ratings in the overlay analysis to create landslide susceptibility indexes and maps. Then, the four landslide susceptibility maps were used as new input factors and integrated using the frequency ratio, weight of evidence, logistic regression and artificial neural network models as ensemble methods to make better susceptibility maps. All of the susceptibility maps were validated by comparison with known landslide locations that were not used directly in the analysis. As the result, the ensemble-based landslide susceptibility map that used the new landslide-related input factor maps showed better accuracy (87.11% in frequency ratio, 83.14% in weight of evidence, 87.79% in logistic regression and 84.54% in artificial neural network) than the individual landslide susceptibility maps (84.94% in frequency ratio, 82.82% in weight of evidence, 87.72% in logistic regression and 81.44% in artificial neural network). All accuracy assessments showed overall satisfactory agreement of more than 80%. The ensemble model was found to be more effective in terms of prediction accuracy than the individual model.