EN中文

Journal of Geographical Sciences >

2017 , Vol. 27 >Issue 4: 439 - 462

DOI: https://doi.org/10.1007/s11442-017-1386-4

Research Articles

Comparative evaluation of geological disaster susceptibility using multi-regression methods and spatial accuracy validation

  • JIANG Weiguo , 1, 2 ,
  • RAO Pingzeng 1, 2 ,
  • CAO Ran 3 ,
  • TANG Zhenghong 4 ,
  • CHEN Kun 5
Expand
  • 1. Academy of Disaster Reduction and Emergency Management, Beijing Normal University, Beijing 100875, China
  • 2. Key Laboratory of Environmental Change and Natural Disaster, Beijing Normal University, Beijing 100875, China
  • 3. China Property & Casualty Reinsurance Company Ltd., Beijing 100033, China
  • 4. College of Architecture, University of Nebraska-Lincoln, NE 68588, USA
  • 5. School of Geography and Planning, Ludong University, Yantai 264025, Shandong, China

Author: Jiang Weiguo, PhD, E-mail: ;

Received date: 2016-09-08

  Accepted date: 2016-10-20

  Online published: 2017-04-20

Supported by

National Natural Science Foundation of China, No.41571077, No.41171318

The Fundamental Research Funds for the Central Universities

Copyright

Journal of Geographical Sciences, All Rights Reserved
Fold

Abstract

Geological disasters not only cause economic losses and ecological destruction, but also seriously threaten human survival. Selecting an appropriate method to evaluate susceptibility to geological disasters is an important part of geological disaster research. The aims of this study are to explore the accuracy and reliability of multi-regression methods for geological disaster susceptibility evaluation, including Logistic Regression (LR), Spatial Autoregression (SAR), Geographical Weighted Regression (GWR), and Support Vector Regression (SVR), all of which have been widely discussed in the literature. In this study, we selected Yunnan Province of China as the research site and collected data on typical geological disaster events and the associated hazards that occurred within the study area to construct a corresponding index system for geological disaster assessment. Four methods were used to model and evaluate geological disaster susceptibility. The predictive capabilities of the methods were verified using the receiver operating characteristic (ROC) curve and the success rate curve. Lastly, spatial accuracy validation was introduced to improve the results of the evaluation, which was demonstrated by the spatial receiver operating characteristic (SROC) curve and the spatial success rate (SSR) curve. The results suggest that: 1) these methods are all valid with respect to the SROC and SSR curves, and the spatial accuracy validation method improved their modelling results and accuracy, such that the area under the curve (AUC) values of the ROC curves increased by about 3%-13% and the AUC of the success rate curve values increased by 15%-20%; 2) the evaluation accuracies of LR, SAR, GWR, and SVR were 0.8325, 0.8393, 0.8370 and 0.8539, which proved the four statistical regression methods all have good evaluation capability for geological disaster susceptibility evaluation and the evaluation results of SVR are more reasonable than others; 3) according to the evaluation results of SVR, the central-southern Yunnan Province are the highest susceptibility areas and the lowest susceptibility is mainly located in the central and northern parts of the study area.

Key words： geological disaster susceptibility; multi-regression methods; geographical weighted regression; support vector regression; spatial accuracy validation; Yunnan Province

Cite this article

JIANG Weiguo , RAO Pingzeng , CAO Ran , TANG Zhenghong , CHEN Kun . Comparative evaluation of geological disaster susceptibility using multi-regression methods and spatial accuracy validation[J]. Journal of Geographical Sciences, 2017 , 27(4) : 439 -462 . DOI: 10.1007/s11442-017-1386-4

1 Introduction

Typical geological disasters include collapses, landslides, debris flows, flash floods, earthquakes and mixed type geological disasters, etc. (Yi et al., 2012; Samarasundera et al., 2014). In recent decades, geological disasters occurred more frequently because of immoderate use of natural resources by humans. Geological disasters are quite common all over the world, especially in developing countries (Guzzetti et al., 1999; Alcántara-Ayala, 2002; Huang and Cheng, 2013). In fact, nearly every country or region has experienced geological disaster events during the past decades (Metternicht et al., 2005; Samarasundera et al., 2014). Geological disasters threaten the life and property of local people and cause huge damages to the ecological environment, which severely restricts the sustainable development of human society (Uitto and Shaw, 2016). The frequency and intensity of geological disasters intensified rapidly over the most recent decades (Guzzetti et al., 1999). Therefore, it is very meaningful to evaluate geological disaster susceptibility and identify high susceptibility zones for prevention and control of geological disasters.
Geological disaster susceptibility evaluation is the first step to assessing the associated hazards and risks (Varnes, 1984; Trigila et al., 2015). Qualitative and quantitative methods are often used for geological disaster susceptibility evaluation (Wang and Sassa, 2005; Corominas et al., 2014). Because of technical limitations and inaccurate understanding of geological disasters at early stages, researchers often use qualitative assessment methods to evaluate geological disaster susceptibility (Degg, 1992; Zhou et al., 2002). In recent decades, with the progress of science and technology, especially the continuous development of computer technology and geographic information system technology, the quantitative analysis method has gradually become the primary research direction (Bai et al., 2010). There are some commonly used methods, such as the Analytic Hierarchy Process (Nie et al., 2001; Komac, 2006), Fuzzy Comprehensive Evaluation (Jiang et al., 2009), and Logistic Regression (Lee and Pradhan, 2007). Among them, the Analytic Hierarchy Process and Fuzzy Comprehensive Evaluation are heuristic methods that are subjective and empirical with low accuracy. Logistic Regression is a statistical method that collects and regresses historical data to obtain a susceptibility degree index of geological disasters (Ayalew and Yamagishi 2005; Lee and Pradhan, 2007; Li et al., 2015). At present, statistical methods are widely used to evaluate geological disaster susceptibility (Erener and Düzgün, 2010). In addition, with the development of computer technology in recent years, deterministic methods based on the physical mechanisms of objects have been gradually applied to the physical modelling of geological disasters (Pradhan et al., 2016). Two common methods include stability analysis of individual slope formation (Frattini et al., 2004; Frattini et al., 2008) and simulation and analysis of the evolution of single small watershed (Bregoli et al., 2015). Because the parameters are difficult to obtain and the applicability of deterministic methods are limited, so far they have been rarely used in geological disaster susceptibility evaluation (Carrara et al., 2008; Bregoli et al., 2015).
Statistical methods such as Logistic Regression (LR), Spatial Autoregression (SAR), Geographical Weighted Regression (GWR), and Support Vector Regression (SVR) are usually used to evaluate geological disaster susceptibility, and they are convenient for determining the contribution of various influencing factors to instability (Neuhäuser and Terhorst, 2007; Pourghasemi et al., 2013; Pederson et al., 2015). LR is the most widely used statistical method because of its simplicity and good function, and its effectiveness has been well proven (Ayalew and Yamagishi, 2005; Lee and Sambath, 2006; Lee and Pradhan, 2007; Bai et al., 2010; Ramani et al., 2011; Jiang et al., 2015; Wang et al., 2015). However, there are many factors that LR does not consider that are indispensable, which sometimes causes large errors. Therefore, more and more researchers have adopted an improved statistical regression method or a more comprehensive statistical approach to evaluate geological disaster susceptibility. Currently, the common methods include SAR (Erener and Düzgün, 2011), GWR (Sabokbar et al., 2014), and SVR (Yao et al., 2008; Xu et al., 2012; Pradhan, 2013). SAR is an improved ordinary linear regression technique that recognizes the spatial autocorrelation of the dependent variable by introducing a spatial lag variable and spatial contiguity matrix (Erener and Düzgün, 2010). The traditional logistic regression method does not consider spatial nonstationarity, which is inevitable (Brunsdon et al., 2002; Wu and Zhang, 2013). When spatial nonstationarity is large enough, the error cannot be ignored. In contrast, GWR considers spatial nonstationarity (Fotheringham et al., 1997), and it can be applied well to geological disaster susceptibility evaluation (Sabokbar et al., 2014). SVR applies to small samples and high-dimensional space, thus avoiding over-fitting problems and enabling a strong ability for generalization (Basak et al., 2007). In order to demonstrate the practicality and reliability of SVR, many researchers utilize it for comparison with other traditional methods. Pradhan et al. (2013) contrasted SVR with Decision Trees and neuro-fuzzy models for landslide susceptibility evaluation, and the final result showed that SVR was the best. Besides, there are other commonly used statistical methods, such as frequency ratio model (Lee and Sambath, 2006; Lee and Pradhan, 2007), information quantity model (Tan et al., 2015), weight of evidence (Neuhäuser and Terhorst, 2007), artificial neural network (Xu, 2001; Qiu et al., 2014) etc.
It is difficult to clearly and accurately determine a geological disaster susceptibility value. Davis and Goodrich (1990) considered the accuracy evaluation of the models to be a multi-standard problem. The key is to characterize the model sensitivity, specificity, and accuracy (Melchiorre et al., 2006). At present, commonly used methods include the kappa coefficient, ROC curve (Pourghasemi et al., 2012; Wang et al., 2015), success rate curve (Pradhan, 2013), etc. The kappa coefficient method is simple but not specific that it determines accuracy of the models by calculating kappa values of the modelling results. The ROC curve and success rate curve are simple and intuitive, and both can accurately reflect the specificity and sensitivity of the modelling results with good accuracy validation and wide application to geological disaster susceptibility evaluation (Kavzoglu et al., 2014). In fact, the validation results are usually poor and researchers used to consider the modeling approach as the key factor of them. However, the modelling results are also usually affected by the discrete expression of spatial data, which often causes great deviations in evaluation results (Tang et al., 2013). In this paper, a method called spatial accuracy validation, which can make the modelling values closer to the true values, was introduced to improve the situation because of the discrete expression of spatial data (Shekhar et al., 2002). The improved results are then used to verify model accuracy.
Geological disasters are very frequent, and the types of disasters in China are various (Yi et al., 2012; Liu et al., 2012; Li et al., 2013; Cui, 2014), especially southwest region of China (Tang and Wu, 1990). Yunnan Province is mountainous and rugged, coupled with frequent heavy rainfall and earthquakes. It is susceptible to erupt geological disasters, such as landslides, flash floods, and debris flows (Liu et al., 1992; Lan et al., 2004; Zhang et al., 2011; Jiang et al., 2016). There are few studies regarding geological disaster susceptibility of the entire Yunnan Province. Mostly, studies targeted partial areas of Yunnan Province and utilized traditional empirical assessment methods (Liu and Lei, 2003; Wu, 2015; Zhuang et al., 2015). In this paper, 500 major geological disasters (including landslides, flash floods, debris flows, etc.) that occurred in Yunnan Province from 2000 to 2014 were selected, and related predisposing factors and potential formation conditions were collected.
Therefore, the objective of this paper is to use multi-regression methods (LR, SAR, GWR, SVR) to evaluate the geological disaster susceptibility of the study area. The spatial receiver operating characteristic (SROC) curve and the spatial success rate (SSR) curve are applied to verify and compare the accuracy of these methods. The susceptibility results could provide evidence-based guidance towards preventing and responding to geological disasters in the study area.

2 Study area

Yunnan Province, which is located in southwest China (Figure 1), consists of 16 prefecture- level cities. The study area lies between 21.14°N and 29.25°N and 97.53°E and 106.20°E, encompassing about 394,000 km2. Northwest of the study area is mountainous area where the altitude is very high; the highest elevation is 6471 m. The central part of the study area belongs to the Yunnan-Guizhou Plateau, with an average elevation of about 2,000 m. Yunnan’s mountainous area, which is a typical mountain environment, accounts for 94% of the province (Liu et al., 2002). Climate change is extremely complex and diverse, and dry and rainy seasons are distinctively obvious. The annual precipitation is more than 1000 mm inmost of the study area, and the precipitation mainly from June to August with frequent heavy rains. The spatial distribution of precipitation is very uneven, such that it gradually decreases from the northwest to the southeast (Yu et al., 2013).
Figure 1 The location of Yunnan Province
Earthquakes occur frequently, with approximately 60 instances greater than magnitude 4 per year that lead to great social and economic losses in Yunnan Province (Wen et al., 2008; Yang et al., 2015). The geological structure is extremely complex in the study area. Poor rock slope stability and frequent rain lead to geological disasters such as landslides, flash floods, and debris flows (Zhang et al., 2011; Yang et al., 2015). The annual direct economic loss due to debris flows is more than 8 billion yuan (about 1.3 billion US dollars); furthermore, these disasters resulted in 100 fatalities and a large number of injured people (Liu et al., 2002).

3 Materials

3.1 Disaster samples and non-disaster samples

The key to susceptibility evaluation is obtaining geological disaster-related spatial attribute data. In this study, data for 500 major disaster events that occurred in Yunnan from 2000 to 2014 were collected, which were mainly downloaded from the National Science & Technology Infrastructure of China (NSTIC), National Earth System Science Data Sharing Infrastructure (NESSDSI, http://www.geodata.cn) and Chinese Academy of Geological Sciences (CAGS, http://www.geoscience.cn/), including 126 mountain flood events, 91 debris flow events, 271 landslide events, and 12 mixed events evens (Figure 2a).
Figure 2 Distribution and types for geological disaster and non-disaster samples
In addition, non-disaster samples are also important for geological disaster susceptibility evaluation, but usually non-disaster samples are randomly selected within the disaster sample buffer (Ramani et al., 2011). In this study, we argue that geological disaster susceptibility is extremely low in large and medium urban areas. Based on this assumption, we randomly generated non-disaster samples by combining the landslide and debris flow risk zoning maps of Yunnan Province. The interval between non-disaster sampling points is 3 km. Finally, a total of 597 non-point disaster samples were determined according to the above conditions (Figure 2b).

3.2 Predisposing factors and formation conditions

It is essential to determine influencing factors that are included as the input variables in the models, including predisposing factors and formation conditions (Figure 3).
Figure 3 Predisposing factors and formation conditions of geological disaster susceptibility: (a) One-hour precipitation; (b) Annual precipitation; (c) Distance to epicenter; (d) Distance to faults; (e) Fault density; (f) Distance to rivers; (g) Slope; (h) Elevation; (i) NDVI; (j) Lithology; (k) Slope position; (l) Aspect
Rainfall and earthquakes are direct triggers of geological disasters such as flash floods, debris flows, and landslides in the study area. This paper selected One-hour precipitation, Annual precipitation, and Distance to epicentre as predisposing factors. The precipitation data was obtained from the China Meteorological Administration (CMA), which provided the data from 1692 meteorological stations in and around Yunnan Province. The Inverse Distance Weighted (IDW) spatial interpolation method in ArcGIS was used to interpolate the spatial data of these sites; then One-hour precipitation and Annual precipitation could be calculated using the annual average values of raster data. Seismic data was downloaded from the China Earthquake Data Center (CEDC, http://data.earthquake.cn), with a total of 941 earthquake events in the study area. Distance to epicentre is the Euclidean distance between each pixel and the epicentre that is calculated by the ArcGIS analysis module.
Topography, geological structure, vegetation cover, and river system are the main formation conditions associated with geological disasters. Distance to faults, Fault density, Elevation, Slope, Aspect, Slope position, Lithology, Distance to rivers, and Normalized Difference Vegetation Index (NDVI) were selected as potential indicator factors. Distance to faults and Fault density, which were extracted from 1:2,500,000 geological fault map of China Geological Survey (CGS), reflect features of the geological structure. Distance to faults was obtained using the ArcGIS spatial distance analysis module to calculate the spatial Euclidean distance, and Fault density was calculated using the ArcGIS kernel density analysis method within 10 km units. Elevation, Slope, Aspect, and Slope position reflect the characteristics of the terrain and were extracted using Shuttle Radar Topography Mission (SRTM) digital elevation data from the National Aeronautics and Space Administration (NASA, http://www. cgiar-csi.org/). River system data was obtained from a 1:250,000 topographic map data of the National Geomatics Center of China (NGCC). Vegetation Index reflects the growing conditions and vegetation cover, and NDVI is the MODIS NDVI product from NASA.

4 Methods

4.1 Evaluation methods of geological disaster susceptibility

Geological disaster susceptibility evaluation is based on three assumptions: (1) outbreak of geological disasters is mainly related to their influencing factors; (2) historical geological disaster zones can represent very high susceptibility areas; (3) geological disaster susceptibility can be predicted. That is to say, geological disasters susceptibility can be predicted by influencing factors (Lee and Pradhan, 2007; Ramani et al., 2011). Therefore, this study adopts the Geological Disasters Susceptibility Index (GDSI) to quantify the degree of geological disaster susceptibility, which can be expressed as a mathematical model. The range of GDSI is [0, 1]. The specific form of the model is as follows:
$GDSI=T(f({{x}_{1}},{{x}_{2}},\cdots {{x}_{n}}))$ (1)
where f is a spatial math function, x1, x2,…, xn are the susceptibility influencing factors, and T represents the results of linear or non-linear transformation. f is set as LR, SAR, GWR, or SVR.
Importantly, this paper considered 500 geological disasters as positive samples, namely very high susceptibility samples, and their GDSI was set to 1. Similarly, 597 non-geological disasters were considered to be negative samples whose GDSI was set as 0. The two types of samples were mixed, and 60% of them were selected as training samples for building and training the evaluation models; the remaining 40% were used as validating samples for accuracy validation of the modelling results.
4.1.1 Logistic regression (LR)
Logistic Regression (LR) is a generalized linear regression method that is widely used to explore the probability of certain events (Wu and Zhang, 2013). Different from ordinary least squares regression, the dependent variables of LR can be categorical variables as well as continuous variables, even if they do not follow the normal distribution. Therefore, it is suitable for geological disaster susceptibility evaluation (Ramani et al., 2011). In the susceptibility evaluation, the dependent variable of LR is GDSI, and the independent variables are influencing factors of geological disasters. GDSI of observed geological disasters samples is either 0 or 1, representing high susceptibility and low susceptibility, and obey the binomial distribution.
where Y=1 represents geological disasters, x1, x2,…, xn are known as n influencing factors, and $P(Y=1|X)$ is the probability of a combination of influencing factors, which is also GDSI. ε is the GDSI linear function, which is assumed to obey a logistic distribution. When ε approaches negative infinity,$P(Y=1|X)=0;$in contrast, when ε approaches positive infinity,$P(Y=1|X)=1.\ {{\beta }_{0}},{{\beta }_{1}},{{\beta }_{2}},\cdots {{\beta }_{n}}$are the regression coefficients of GDSI, which can be estimated from the observed samples.
4.1.2 Spatial Autoregression (SAR)
Spatial Autoregression is a global regression model that introduces spatial lag term and a spatial contiguity matrix (Erener and Düzgün, 2010). The spatial autoregressive linear model expressions are as follows:
where y is the independent variable; ρ is the explanatory variable; W1y is the spatial delay term, where W1 is the spatial weights matrix of y and ρ is the regression coefficient of W1y; W2μ is spatial interference term, where W2 is the spatial weights matrix of μ and λ is the regression coefficient of W2μ; and ε is the residual, which obeys the normal distribution.
4.1.3 Geographically Weighted Regression (GWR)
Geographically Weighted Regression (GWR) is a partial regression model that introduces a spatial non-stationary location and spatial regression coefficients (Fotheringham, 2003). The spatial position is embedded into the regression coefficients, which do not only describe the relationship between explained and explanatory variables but also reflect the spatial variation of their relationship. It is an effective modelling technique used to treat spatial nonstationarity (Su et al., 2012).
${{y}_{i}}={{\beta }_{0}}({{\mu }_{i}},{{v}_{i}})+\sum\limits_{j=1}^{k}{{{\beta }_{j}}({{\mu }_{i}},{{v}_{i}}){{x}_{ij}}+{{\varepsilon }_{i}}}$ (4)
where xij is the independent variable and yi is the dependent variable, εi is the residual and εi~N(0, σ2), (μi, vi) is the location of i, and βj(μi, vi) is the regression coefficient of explanatory variables.
GWR uses a locally weighted least squares method to estimate the parameters, and the resulting regression coefficients usually are not constant but variable with respect to changes in spatial position, which reflects the nonstationarity of the spatial relationship (Gao and Li, 2011). The regression parameters of any point can be estimated by the following formula (Su et al., 2012):
$\hat{\beta }({{\mu }_{i}},{{v}_{i}})={{({{X}^{T}}W({{\mu }_{i}},{{v}_{i}})X)}^{-1}}{{X}^{T}}W({{\mu }_{i}},{{v}_{i}})Y$ (5)
where W(μi, vi) is a diagonal matrix, and diagonal elements that the distance between point i and point j meet monotonically decreasing (Gao and Li, 2011).
4.1.4 Support Vector Regression (SVR)
Support Vector Regression (SVR) based on the statistical learning theory is a nonlinear model for small samples and high-dimensional space (Basak et al., 2007). It converts the inputted space by nonlinear transformation to a high-dimensional feature space and then finds the optimal linear interface in the new space. It avoids the over-fitting problem and has excellent generalization ability.
(xi, yi) is the training sample set, where$i=1,2,\cdots ,l,x\in R,Y\in \{\pm 1\}.$The regression problem can be converted to the optimal problem, such that the expression is: $\left\langle W,X \right\rangle +b=0$
where${{\xi }_{i}}$and$\xi _{i}^{*}$are slack variables, and $\varepsilon $is a positive constant. When$\left\langle w,\phi ({{x}_{i}}) \right\rangle +b-$ ${{y}_{i}}\ge \varepsilon ,$ the value of error is $\left| \left\langle w,\phi ({{x}_{i}}) \right\rangle +b-{{y}_{i}} \right|-\varepsilon ;$ otherwise the error will be discarded.

4.2 Validation methods of geological disaster susceptibility

In general, the results of geological disaster susceptibility evaluation need to meet two assumptions: (1) most disaster samples should be located in high susceptibility zones, and most non-disaster samples should be located in low susceptibility zones; (2) the high susceptibility zones should occupy a relatively small area and the low susceptibility zones should account for a relatively large area (Can, 2005; Ramani et al., 2011). The receiver operating characteristic (ROC) curve and success rate curve are common methods used for accuracy evaluation of the models. The ROC curve can examine the predictive ability and severability for a model through positive and negative class samples. The success rate curve can examine the ability to predict only disaster samples for a model. Based on the two methods, this study introduced spatial accuracy validation to improve the model evaluation results, and the improved ROC curve and success rate curve should be more reliable.
4.2.1 ROC curve and success rate curve
The ROC curve is a method for validating the explanatory capability of the regression models (Pontius, 2001).The contingency table of predicted values and actual values is established by calculating true positive rate (TPR) and false positive rate (FPR) under different susceptibility thresholds of categorized situations. The ROC curve is generated such that TPR is on the ordinate axis and FPR is on the horizontal axis. The area under the curve (AUC) value is the area between the ROC curve and the horizontal axis. The AUC value varies from 0.5 (diagonal line) to 1, with a higher value indicating a better predictive capability of the model. TPR and FPR are calculated as follows:
$TPR=\frac{TP}{TP+FN}\ \ \ \ \ FPR=\frac{FP}{FP+TN}$ (7)
where TP (true positive) is the number of true predictive class samples out of the geological disaster samples; FP (false positive) is the number of false predictive class samples out of the geological disaster samples. TN (true negative) is the number of true predictive class samples out of the non-geological disaster samples; FN (false negative) is the number of false predictive class samples out of the non-geological disaster samples.
Unlike the ROC curve, the success rate curve only considers the predictive capability of the models for the geological disaster samples but disregards the non-geological disaster samples (Pradhan, 2013). The success rate curve is determined by the predictive susceptibility rank (SR) and cumulative percentage (CP) of the corresponding disaster samples, where SR is the horizontal axis and CP is the vertical axis. SR is often divided into 100 ranks with a descending order of susceptibility.
4.2.2 Spatial ROC (SROC) curve and spatial success rate (SSR) curve
Because of the discrete representation of spatial data, the evaluation process often offsets the spatial position. In addition, the spatial data collection and production process also results in accumulation of errors, which yields great uncertainty in the modelling results. Shekhar et al. (2002) considered that the classification for a sample point is correct if the corresponding class appears in or around the pixel of the sample point. That is to say, the predictive values of the observed samples are reasonable (Figure 4). Figure 4a shows the location of the sample points. Figure 4b shows the corresponding pixel location of the actual sample points which is A. P is the corresponding pixel location of the predicted sample points in Figure 4c. Figure 4d shows that A and P differ in spatial location, which is considered to be inaccurate for the traditional classification methods, and the accuracy value is 0. Figure 4e shows that P and A are consistent for a certain space range (3 × 3 pixel areas) because of the existence of a spatial position offset, so the accuracy value is 1.
Figure 4 Classification accuracy and spatial accuracy (Shekhar, 2002)
Considering that the image resolution is 30 m, this study recounts GDSI of the disaster samples and the non-disaster samples in 3 × 3 pixel areas. Finally, we can obtain the spatial ROC (SROC) curve and spatial success rate (SSR) curve.

4.3 Flowchart

This study randomly divided disaster samples and non-disaster samples into training samples and validating samples. Training samples were used for modelling, and validating samples were used for validation. The evaluation results of the multi-regression models were evaluated using SROC and SSR. Finally, the results of geological disaster susceptibility were analysed and compared. A flowchart of this process is shown in Figure 5.
Figure 5 The flowchart of geological disasters susceptibility evaluation

5 Results

In this study, four statistical regression methods were utilized to evaluate geological disaster susceptibility of the study area, and the SROC curve and SSR curve were used to verify and compare the evaluation results of the four methods.

5.1 Susceptibility index of geological disaster

The evaluation rsults of the methods are shown in Figure 6. There is a significant difference in the quantity and spatial distribution of the evaluation results for each model. The quantitative feature values of the GDSI, such as the maximum, minimum, mean, and standard deviation are different, as shown in Table 1.
Figure 6 Geological disaster susceptibility index map produced by the multi-regression methods
Table 1 Characteristics of GDSI for multi-regression methods
LR SAR GWR SVR
Minimum 0 0.077 -0.71 -0.45
Maximum 1 0.73 2.05 1.69
Mean 0.52 0.31 0.79 0.8
Standard deviation 0.1 0.06 0.28 0.3
For LR, the minimum value is 0 and the maximum value is 1, with a mean of 0.52 and a standard deviation of 0.1, indicating that the susceptibility level of the study area is relatively uniform. Importantly, for the evaluation results of different models, a maximum value greater than 1 indicates that the predictive capability of the model in high susceptibility areas
is strong; otherwise, it is weak. Similarly, a minimum value less than 0 indicates that the predictive capability of the model in low susceptibility areas is strong; otherwise, it is weak. Furthermore, the mean value reflects the overall assessment bias of the model. If the mean value is greater than 0.5, this suggests that the overall evaluation results of the model are relatively large; conversely, they are relatively small.
For the other three models, the predictive capability of SAR in high susceptibility and low susceptibility areas are both weak, and the mean predictive value is relatively small. GWR and SVR are exactly opposite compared to SAR, which indicates that GWR and SVR are better than SAR for geological disaster susceptibility evaluation in the study area.

5.2 Reclassification of geological disaster susceptibility

In order to understand geological disaster susceptibility of the study area more completely and clearly, appropriate breakpoints were selected for the cumulative frequency curve to reclassify the susceptibility results. The breakpoints were determined at cumulative frequencies of 30%, 50%, 70%, and 85% (Hong et al., 2015). There are 5 levels (Very low, Low, Medium, High and Very high levels) for the classification results shown in Figure 7.
Figure 7 Reclassification of geological disaster susceptibility by the multi-regression methods
Figure 7 shows that the modelling results of LR, SR, and GWR are almost identical, which indicates their principles are relatively similar. The highest susceptibility level is mainly distributed in the northwestern part of the study area, followed by the central and southern regions. Disaster susceptibility is relatively low in the eastern and western parts of the study area. The evaluation results of SVR are clearly different, showing that the northwest of the study area has a low disaster susceptibility but the south has a high disaster susceptibility. It is distinctly different in the northwest of the study area (mainly Diqing city and Lijiang city), as shown in the elliptical area (Figure 7).

5.3 Classification accuracy and spatial accuracy

According to the classification criteria, validating samples were selected to count the classification results of the different susceptibility levels (Figure 8), where the proportions of disaster samples reflect the classification levels of different susceptibility classes and the frequency ratio reflects the susceptibility size of different susceptibility classes. For different regression models, the proportions of disaster samples from low to high susceptibility levels were consistent with the selected grade range, namely 30%, 20%, 20%, 15% and 15% (section 4.2), which proved the selected breakpoints are reasonable to some extent.
Figure 8 The proportion of disasters and frequency ratios used for validating samples
Unfortunately, the results of the frequency ratio are irregular. Theoretically, a higher level of geological disaster susceptibility should correspond to a higher frequency ratio value, but the evaluation results of the models obviously do not follow such regularity. For example, the frequency ratio values of LR, ranging from low to high susceptibility levels, are 0.91, 1.08, 0.94, 1.13, and 1.01, which are all approximately equal to 1. The other three models are similar to the evaluation results of LR. Such a situation most likely occurs for the following two reasons: (1) the 12 influencing factors that this study selected do not reflect the susceptibility well; or (2) there is not a large error for the assessment results of the models, but discrete expression of the spatial data resulted in offset of the spatial position for disaster sample points, which will affect the susceptibility values of the geological disaster sample points.
Considering that the 12 predisposing factors and formation conditions were selected by combining characteristics of geological disasters in the study area, there is a very low probability that the evaluation results of the four regression methods do not reflect the geological disaster susceptibility of the study area. Therefore, in order to get a better disaster susceptibility distribution map of the study area, this study introduced spatial accuracy validation to improve the evaluation results with respect to the second reason mentioned above.
That is, the pixel of a disaster point and the surrounding pixels are all likely to be the true position of the corresponding disaster point, so the highest susceptibility value among them should be considered the susceptibility value of the disaster point. Similarly, for non-disaster sample points, the lowest susceptibility value that appears around them should be considered the susceptibility value of the non-disaster point. In this paper, spatial accuracy is selected as the neighbourhood of 3 × 3 pixels.
Figure 9 shows the improved evaluation results of geological disaster susceptibility. It shows that the proportions of disaster samples and the frequency ratio of the four methods are larger significantly in high levels (including ‘High’ and ‘Very high’) than before. If disaster sample points distributed in ‘Very high’ are considered to be reasonable, the simulation results of SVR whose proportions of disasters of ‘Very high’ is 38.6% are more reasonable than others. If disaster sample points distributed in ‘High’ and ‘Very high’ are considered to be reasonable, the proportions of disasters of LR, SAR, GWR and SVR are 57.6%, 55.6%, 57.2% and 58.0%, respectively, which the simulation results of SVR are also more reasonable than others.
Figure 9 The proportion of disaster samples and frequency ratios introducing spatial a ccuracy validation
The frequency ratio reflects the degree of the disaster susceptibility, which is closely related to the susceptibility levels. As shown in Figure 9, the evaluation results of LR, with frequency ratios from low to high susceptibility levels being 0.36, 0.59, 0.99, 1.54, and 2.25, respectively; for SAR, with frequency ratio values of 0.25, 0.63, 1.24, 1.80 and 1.85, respectively; for GWR, similarly to the results of LR with frequency ratio values of 0.39, 0.60, 0.97, 1.59 and 2.16, respectively; SVR, with frequency ratio values of 0.41, 0.43, 1.07, 1.24 and 2.60, respectively. Obviously, the results of the four methods are all good to meet the regularity that the higher susceptibility levels (including ‘High’ and ‘Very high’) are corresponding to higher frequency ratio values. Of all the four methods, with the results of proportions of disasters and frequency ratio, the evaluation results of SVR are both more reasonable than others.
Additionally, this study also analysed the distribution of the non-disaster samples at different susceptibility levels and found that most of the non-disaster samples are located in the ‘Very low’ level, which is consistent with the actual situation. Comparing Figures 8 and 9, after introducing spatial accuracy validation methods, the results obtained by the models better reflect the geological disaster susceptibility features of this study area, especially the frequency ratio indicators. In summary, the evaluation results of LR, SAR, GWR and SVR all can reflect geological disaster susceptibility in the study area, and the simulation results of SVR are more reasonable than others.

5.4 ROC curves and SROC curves

The ROC curve is used to verify and compare the evaluation results of the models, and the results are shown in Figure 10a. The AUC values of LR, GWR, and SVR are greater than 0.9, indicating that the evaluation effect of each of the three methods is excellent. SAR, with an AUC value of 0.86, is poorer than other methods, but it is good overall. After introducing the method of spatial accuracy validation, the results are shown in Figure 10b. This shows that the accuracy of the models was improved, and AUC values were all close to 1 after introducing spatial accuracy validation. Compared with the classification accuracy, the effects of the spatial accuracy validation increased by 3%-12%, showing that it can better reflect the actual susceptibility results of this study area. It also shows that the four regression methods are all excellent for the geological disaster susceptibility evaluation of this study area.
Figure 10 ROC curves of the multi-regression methods: (a) Classification accuracy; (b) Spatial accuracy

5.5 Success rate curves and SSR curves

Similarly, the success rate curve is also utilized to verify and compare the evaluation results of the four methods; the results are shown in Figure 11a. The AUC values of the four models are in the vicinity of 0.5. Generally, when the AUC value is in the range [0.5, 1], the simulation results of the model can be considered reliable. For the four models, only the AUC value of SAR does not reach 0.5, indicating that the SAR’s simulation result is less reliable. After introducing space accuracy validation, the results are shown in Figure 11b. This figure shows that the accuracy of the models is significantly improved after introducing the method, the AUC values of the models were all close to 0.7 (>0.5, indicating that four models have all passed the validation of success rate curve). The method with the highest accuracy was SVR with an AUC value of 0.7183; the lowest was for SAR with an AUC value of 0.6813. After introducing spatial accuracy validation, AUC increased by 15%-20%. Compared to the results of the success rate curve, obviously, the results of the SSR curve can better reflect geological disaster susceptibility of this study area.
Figure 11 Success rate curves of the multi-regression methods: (a) Classification accuracy; (b) Spatial accuracy
If we think the calculated results of SROC and SSR are reasonable, the AUC values of them are averaged and the AUC values of the four models (LR, SAR, GWR and SVR) are 0.8393, 0.8325, 0.8370 and 0.8539, respectively. Therefore, we can consider the SVR’s evaluation effect is the most reasonable, followed by LR, GWR and SAR.

6 Discussion

The application of statistical methods for geological disaster susceptibility evaluation is one of the focal points of current geological disaster research. There is huge uncertainty regarding the different geological disaster susceptibility evaluation methods. In order to determine better evaluation methods, this study adopted multi-regression models (including LR, SAR, GWR and SVR) to evaluate the geological disaster susceptibility of the study area. Compared to previous studies, this paper aims to explore the application of statistical regression methods for geological disaster susceptibility evaluation. By contrast, we can determine which methods are superior or not and know the sensitivity and uncertainty of each method. In addition, aiming to address the problem of discrete expressions of spatial data causing space positional displacement, this study introduced spatial accuracy validation to improve the modelling results of these four models. Compared to the previous ROC curve and success rate curve, the effectiveness of improved SROC curve and SSR curve were obviously better. The advantages and disadvantages of the regression methods and validation methods are summarized in below section.

6.1 The advantages and disadvantages of multi-regression methods

Contrasting the four regression methods showed that they have their own distinctive strengths but also many shortcomings. LR is one of the simplest statistical regression methods, whose principle is based on the least squares method. SAR considers spatial autocorrelation, and GWR considers spatial nonstationarity. SVR, unlike LR, SAR and GWR, introduces the vapnik-chervonenkis dimension. Before introducing space accuracy validation, the displayed evaluation results of the four regression methods were poor; afterwards, the evaluation results of all four regression methods reflected the geological disaster susceptibility situation of the study area well, which is demonstrated by the SROC curve (Figure 10) and SSR curve (Figure 11).
For reclassification of geological disaster susceptibility, SAR is lacking in the evaluation of high susceptibility zones, LR, GWR and SVR all yield good simulation results. In terms of the AUC values of the SROC curve and SSR curve, the evaluation results of SVR are the best, followed by LR, GWR and SAR. In addition, compared the evaluation results of the four models, it should be considered why the evaluation results of SVR differed greatly from the evaluation results of the other three methods in northwest of the study area (see Figure 7). In fact, there are little disaster sample points in northwest of the study area. There may be two main reasons which both can cause some deviations for different methods: (1) Predisposing factors and formation conditions selected in this paper are insufficient for geological disaster susceptibility evaluation; or (2) The collection of disaster sample points is very incomplete in northwest of the study area. There are some similar studies on Yunnan Province, and they almost all show that the northwest region witnessed less geological disasters and is not high susceptibility areas (Jiang, 1990; Hu, 2014; Wu, 2015; Jiang 2016). Wu (2015) obtained a similar result with SVR, which shows central and southern parts of the study area are the most susceptibility areas of geological disasters. Based on the analysis of the above, with reference to Figures 2, 6, 7, 8 and 9 and in combination with the disaster sample points and comparison of the evaluation results of the models, the evaluation results of SVR can be considered to be better effect on treating disaster sample points.
The AUC values of LR and GWR are both 0.84 by combining the results of the SROC curves and SSR curves and the evaluation results of LR and GWR are similar, which indicates spatial nonstationarity is weak in the study area. The predictive capability of GWR in high susceptibility areas and low susceptibility areas are both stronger than LR (see Section 4.1), and GWR itself is an improved LR method. Therefore, we can also consider that GWR is better than LR. Unfortunately, compared to the evaluation results of LR, the evaluation results of SAR are relatively poor. It is possible that some regression coefficients deviated or variable factors were selected incompletely.
In this paper, four statistical regression methods are used for geological disaster susceptibility evaluation. In fact, many new similar methods have been widely used in this field, such as multiple logistic regression (Ahmed, 2015), Naïve Bayes classifier (Tsangaratos and Ilia, 2016). In addition, GIS and RS technology is developing rapidly, especially the resolution of image continuously improved. Deterministic methods will be widely applied range in the future, hence the next study is needed to explore the appropriate excitation deterministic method to monitor the susceptibility of various geological disasters.

6.2 Advantages and disadvantages of the validation methods

This paper selected the ROC curve and success rate curve to evaluate the modelling accuracy of the methods. The AUC values of the ROC curves for the models are all approximately 0.9 and nearly 1 for the SROC curves, showing the evaluation accuracy of the models to be excellent. By contrast, the AUC values of success rate curves and SSR curves are about 0.5 and 0.7, respectively, which are significantly lower than the AUC values of the ROC curves and the SROC curves. Compared to the success rate curve method, the ROC curve considers not only the role of disaster sample points, but also the impact of non-disaster sample points. Thus, the reason that the AUC values of the ROC values and the SROC values are always higher than the AUC values of the success rate curves and the SSR curves is mainly because the non-disaster samples are more reasonable as the assumption than the disaster samples (see Section 4.3). Obviously, the ROC curve method is more persuasive than the success rate curve method.
More importantly, the AUC values of the SROC values and the SSR values are obviously higher than the ROC values and the success rate values. Therefore, we had to pay attention to the spatial accuracy validation method, which was considered in this paper by comparing Figures 10 and 11. It is showed that the spatial accuracy validation method can greatly improve the modelling results, which differs from previous research, and is innovative and valuable for the study of spatially discrete problems in the future, especially for study related to geological disasters.
The disadvantages of the validation methods are summarized in here. Typically, potential geological disasters and formation conditions considered in the study are very uncertain.For example, predisposing factors of geological disasters only considered precipitation and earthquakes in the study. In fact, human activity is also a very important predisposing factor. Negative human activity often intensifies the occurrence of landslides and debris flows, such as unreasonable land use, excessive deforestation, mineral extraction, and even some water conservancy projects. Of course, it is still very difficult to calculate the specific impact of human activity. Besides, this study is mainly based on topographic features of the study area and the degree of difficulty of data acquisition to select influencing factor indicators. Some researchers also consider distance to roads, land cover, and Topographic Wetness Index (TWI), etc. (Regmi et al., 2010; Hong et al., 2015; Trigila et al., 2015). Incompleteness of the selected indicators will cause a certain impact on the results, which should also be considered in future research.

7 Conclusion

This paper used different statistical regression methods to evaluate the susceptibility of Yunnan Province, China to typical geological disasters (including landslides, flash floods, debris flows and mixed types). This study introduced the method of spatial accuracy validation to improve the evaluation results which are shown by the SROC curve and SSR curve. Before the improvements, the proportion of disasters and the frequency ratio of each regression model did not correspond to reality. After the improvements, the improved results are consistent with the actual situation, and the proportion of disaster and the frequency ratio increased because the susceptibility levels were increasing. Fortunately, the AUC values of the SROC curves increased by 3%-13%, and the AUC values of the SSR curves increased by 15%-20%, indicating that the space discrete expression causes substantial error for the evaluation of the models.
At last, the evaluation accuracies of LR, SAR, GWR, and SVR were 0.8325, 0.8393, 0.8370 and 0.8539, respectively, showing that the four statistical regression methods have good evaluation capability for geological disaster susceptibility evaluation. Among them, the evaluation results of SVR are proved to be more reasonable than others. Therefore, the evaluation results of SVR can be considered to be the geological disasters susceptibility mapping of Yunnan. According to the evaluation results of SVR, the central- southern Yunnan Province, including Pu’er, Lincang, Wenshan, Xishuangbanna, southwest of Yuxi and south of Honghe, are the highest susceptibility areas in the province, followed by Nujiang, Zhaotong, Baoshan and Dehong. The lowest susceptibility is mainly located in the central and northern parts of the study area, including Kunming, Qujing, Chuxiong, Dali, Diqing and north of Honghe. The disaster susceptibility research framework in this study can be applied to other study areas.

The authors have declared that no competing interests exist.

References

Publishing order | Descend order by publishing year | Descend order by cited within
[1]
Ahmed B, 2015. Landslide susceptibility modelling applying user-defined weighting and data-driven statistical techniques in Cox’s Bazar Municipality, Bangladesh.Natural Hazards, 79(3): 1707-1737.Landslides are common geophysical hazards in the highly urbanized hilly areas of Cox’s Bazar Municipality, Bangladesh. Every year, during the monsoon, landslides cause human casualties, property damag

DOI

[2]
Alcántara-Ayala I, 2002. Geomorphology, natural hazards, vulnerability and prevention of natural disasters in developing countries.Geomorphology, 47(2-4): 107-124.The significance of the prevention of natural disasters is made evident by the commemoration of the I nternational D ecade for N atural D isaster R eduction (IDNDR). This paper focuses on the role of geomorphology in the prevention of natural disasters in developing countries, where their impact has devastating consequences. Concepts such as natural hazards, natural disasters and vulnerability have a broad range of definitions; however, the most significant elements are associated with the vulnerability concept. The latter is further explored and considered as a key factor in understanding the occurrence of natural disasters, and consequently, in developing and applying adequate strategies for prevention. Terms such as natural and human vulnerabilities are introduce and explained as target aspects to be taken into account in the reduction of vulnerability and for prevention and mitigation of natural disasters. The importance of the incorporation not only of geomorphological research, but also of geomorphologists in risk assessment and management programs in the poorest countries is emphasized.

DOI

[3]
Ayalew L, Yamagishi H, 2005. The application of GIS-based logistic regression for landslide susceptibility mapping in the Kakuda-Yahiko Mountains, Central Japan.Geomorphology, 65(1/2): 15-31.As a first step forward in regional hazard management, multivariate statistical analysis in the form of logistic regression was used to produce a landslide susceptibility map in the Kakuda-Yahiko Mountains of Central Japan. There are different methods to prepare landslide susceptibility maps. The use of logistic regression in this study stemmed not only from the fact that this approach relaxes the strict assumptions required by other multivariate statistical methods, but also to demonstrate that it can be combined with bivariate statistical analyses (BSA) to simplify the interpretation of the model obtained at the end. In susceptibility mapping, the use of logistic regression is to find the best fitting function to describe the relationship between the presence or absence of landslides (dependent variable) and a set of independent parameters such as slope angle and lithology. Here, an inventory map of 87 landslides was used to produce a dependent variable, which takes a value of 0 for the absence and 1 for the presence of slope failures. Lithology, bed rock-slope relationship, lineaments, slope gradient, aspect, elevation and road network were taken as independent parameters. The effect of each parameter on landslide occurrence was assessed from the corresponding coefficient that appears in the logistic regression function. The interpretations of the coefficients showed that road network plays a major role in determining landslide occurrence and distribution. Among the geomorphological parameters, aspect and slope gradient have a more significant contribution than elevation, although field observations showed that the latter is a good estimator of the approximate location of slope cuts. Using a predicted map of probability, the study area was classified into five categories of landslide susceptibility: extremely low, very low, low, medium and high. The medium and high susceptibility zones make up 8.87% of the total study area and involve mid-altitude slopes in the eastern part of Kakuda Mountain and the central and southern parts of Yahiko Mountain.

DOI

[4]
Bai Shibiao, Wang Jian, Lü Guonian et al., 2010. GIS-based logistic regression for landslide susceptibility mapping of the Zhongxian segment in the Three Gorges area, China.Geomorphology, 115(1/2): 23-31.A detailed landslide susceptibility map was produced using a logistic regression method with datasets developed for a geographic information system (GIS). Known as one of the most landslide-prone areas in China, the Zhongxian–Shizhu segment in the Three Gorges Reservoir region of China was selected as a suitable case to evaluate the frequency and distribution of landslides. The site covered an area of 260.902km 2 with a landslide area of 5.302km 2 . Four data domains were used in this study: remote sensing products, thematic maps, geological maps, and topographical maps, all with 2502×022502m 2 pixels or cells. Statistical relationships for landslide susceptibility were developed using landslide and landslide causative factor databases. We extended the application of logistic regression approaches to use all continuous variables as they are, and the landslide density is used to transform these nominal variables to numeric variable. According to the map, 2.8% of the study area was identified as an area with very high-susceptibility, whereas very low-, low-, medium- and high-susceptibility zones covered 18.2%, 36.2%, 26.7%, and 16.1% of the area, respectively. The quality of susceptibility mapping was validated, and the correct classification percentage and root mean square error (RMSE) values for the validation data were 81.4% and 0.392, respectively.

DOI

[5]
Basak D, Pal S, Patranabis D C, 2007. Support vector regression. Neural Information Processing-Letters and Reviews, 11(10): 203-224.

[6]
Bregoli F, Medina V, Chevalier G et al., 2014. Debris-flow susceptibility assessment at regional scale: Validation on an alpine environment.Landslides, 12(3): 437-454.A debris flow is an erosional process that frequently occurs in areas with steep slopes. Although the initiation can be triggered in higher and inhabited altitudes, the propagation and then the deposition can endanger urbanized areas. Frequently, the zones of generation of debris flows are placed in unattainable territories complicating field observations. Moreover, the rapid vegetation recovery can hide the scars and tracks created by the occurrence of the flow, unlikely to be recognizable in aerial photos. As a consequence, the assessment of debris-flow susceptibility at regional scale is increasing in interest. Two main techniques can be applied for the regional susceptibility assessment: physical models and statistical models. Here, two physical models are introduced simulating the initiation of debris flow by means of shallow landslides: one based on the lateral groundwater flow and the other on vertical groundwater flow. The presented models algorithms are implemented in a toolbox based on GIS environment and declared as open source code. Being applied in a basin of the Spanish Pyrenees where soil properties and dataset of observations have been gathered, the two models have been evaluated. Successively, they have been compared with a statistical model based on a logistic regression. The results have been aggregated in terrain units, built as fluvio-morphological sub-catchments. Each model has been tested based on the gathered observations. It has been observed that the vertical flow-based model enhances by 18 % the performance of the lateral flow-based model in matching debris-flow susceptible areas. Furthermore, the vertical flow-based model performance is found to be similar to that of the logistic regression, having an averaged quality index of about 70 %.

DOI

[7]
Brunsdon C, Fotheringham A S, Charlton M, 2002. Geographically weighted summary statistics: A framework for localised exploratory data analysis. Computers, Environment and Urban Systems, 26(6): 501-524.Geographical kernel weighting is proposed as a method for deriving local summary statistics from geographically weighted point data. These local statistics are then used to visualise geographical variation in the statistical distribution of variables of interest. Univariate and bivariate summary statistics are considered, for both moment-based and order-based approaches. Several aspects of visualisation are considered. Finally, an example based on house price data is presented.

DOI

[8]
Can T, Nefeslioglu H A, Gokceoglu C et al., 2005. Susceptibility assessments of shallow earth flows triggered by heavy rainfall at three catchments by logistic regression analyses.Geomorphology, 72(1-4): 250-271.Sometimes regional meteorological anomalies trigger different types of mass movements. In May 1998, the western Black Sea region of Turkey experienced such a meteorological anomaly. Numerous residential and agricultural areas and engineering lifelines were buried under the flood waters. Besides the reactivation of many previously delineated landslides, thousands of small-scale landslides (mostly the earthflow type) occurred all over the region. The earthflows were mainly developed in flysch-type units, which have already presented high landslide concentrations. In this study, three different catchments namely Agustu, Egerci, and Kelemen were selected because they have the most landslide-prone geological units of the region. The purposes of the present study are to put forward the spatial distributions of the shallow earthflows triggered, to describe the possible factors conditioning the earthflows, and to produce the shallow earthflow susceptibility maps of the three catchments. The unique condition units (UCU) were employed during the production of susceptibility maps and during statistical analyses. The unique condition units numbered 4052 for the Agustu catchment, 13,241 for the Egerci catchment and 12,314 for the Kelemen catchment. The earthflow intensity is the highest in the Agustu catchment (0.038 flow/UCU) and lowest in the Egerci catchment (0.0035 flow/UCU). Logistic regression analyses were also employed. However, during the analyses, some difficulties were encountered. To overcome the difficulties, a series of sensitivity analyses were performed based on some decision rules introduced in the present study. Considering the decision rules, the proper ratios of UCU free from earthflow (0) / UCU including the earthflow (1) for the Agustu, Egerci and Kelemen catchments were obtained as 3, 6, and 5, respectively. Also, a chart for the proper ratio selection was developed. The regression equations from the selected ratios were then applied to the entire catchment and the earthflow susceptibility maps were produced. The landslide susceptibility maps revealed that 15% of the Agustu catchment, 8% of the Egerci catchment, and 7% of the Kelemen catchment have very high earthflow susceptibility; and most of the earthflows triggered by the May 1998 meteorological event were found in the very high susceptibility zones.

DOI

[9]
Carrara A, Crosta G, Frattini P, 2008. Comparing models of debris-flow susceptibility in the alpine environment.Geomorphology, 94(3/4): 353-378.Model performance was evaluated from the percentages of terrain units that each model correctly classifies, the number of debris-flow falling within the area classified as unstable by each model, and through the metric of ROC curves. Although all techniques implemented yielded results essentially comparable; the discriminant model based on the partition of the study area into small slope units may constitute the most suitable approach to regional debris-flow assessment in the Alpine environment.

DOI

[10]
Corominas J, van W C, Frattini P et al., 2013. Recommendations for the quantitative analysis of landslide risk.Bulletin of Engineering Geology and the Environment.This paper presents recommended methodologies for the quantitative analysis of landslide hazard, vulnerability and risk at different spatial scales (site-specific, local, regional and national), as well as for the verification and validation of the results. The methodologies described focus on the evaluation of the probabilities of occurrence of different landslide types with certain characteristics. Methods used to determine the spatial distribution of landslide intensity, the characterisation of the elements at risk, the assessment of the potential degree of damage and the quantification of the vulnerability of the elements at risk, and those used to perform the quantitative risk analysis are also described. The paper is intended for use by scientists and practising engineers, geologists and other landslide experts.

DOI

[11]
Cui Peng, 2014. Progress and prospects in research on mountain hazards in China.Progress in Geography, 33(2): 145-152. (in Chinese)In this paper, first, the research progress on fundamental theories on mountain hazards and prevention techniques are briefly reviewed. In the current research, the distribution of mountain hazards has been recognized and the hazards assessment methods for mountain torrents, debris flow and landslides have been established, and the principles and calculations of landslide stability analysis have been developed. In addition, the stress-strain relation of debris flow slurry and the formulae of velocity, discharge and impact of debris flow have been established, and a model of viscous debris flow initiation has also been proposed. We also presented the magnitude amplification effects of mountain torrents and debris flows. Meanwhile, the methods to forecast mountain hazards have been developed by analyzing both rainfall and ground factors. Based on the understanding of the physical characteristics of mountain hazards, a number of monitoring instruments have been invented, and especially the early warning methods have been used in the underdeveloped regions. Furthermore, a systematic technology for coping with mountain hazards has been developed. The topics on the formation, development, forecasting and prevention, protection strategies and risk management of mountain hazards have been proposed for future research. Finally, aiming at the requirements of the reduction of disasters for the nation and the goals of the academic progresses, we propose the topics that are worthy of more attention in the future research, such as the response of mountain hazards to ecology, the effects of climate change on mountain hazards, the prediction of catastrophe, the mesomechanic behaviors of the soil affected by water, the theories and methods of hazard risk assessment, the hazard forecasting model based on the formation mechanisms and the improvement of prediction and recovery technology.

DOI

[12]
Davis P A, Goodrich M T, 1990. A proposed strategy for the validation of ground-water flow and solute transport models. Sandia National Labs., Albuquerque, NM (USA).Ground-water flow and transport models can be thought of as a combination of conceptual and mathematical models and the data that characterize a given system. The judgment of the validity or invalidity of a model depends both on the adequacy of the data and the model structure. This report proposes a validation strategy for testing both components independently. The strategy is based on the philosophy that a model cannot be proven valid, only invalid or not invalid. In addition, the authors believe that a model should not be judged in absence of its intended purpose. Hence, a flow and transport model may be invalid for one purpose but not invalid for another. 9 refs.

[13]
Degg M, 1992. Natural disasters: Recent trends and future prospects.Geography, 198-209.The 1990s have been declared the "International Decade for Natural Disaster Reduction" (IDNDR) by the United Nations. This paper presents a global analysis of the nature and distribution of natural hazard losses since the Second World War. Regional disparities in loss experience are highlighted, in particular the contrasts between developed and developing regions. The paper concludes with a preliminary evaluation of future disaster potential, through analysis of exposure of the world's 100 largest urban agglomerations (accomodating 10 per cent of the global population) to earthquake and windstorm hazards. We are currently witnessing an increase in exposure to these hazards, coupled with escalating potential for catastrophic loss.

DOI

[14]
Erener A, Düzgün H SB, 2010. Improvement of statistical landslide susceptibility mapping by using spatial and global regression methods in the case of More and Romsdal (Norway).Landslides, 7(1): 55-68.Statistical models are one of the most preferred methods among many landslide susceptibility assessment methods. As landslide occurrences and influencing factors have spatial variations, global models like neural network or logistic regression (LR) ignore spatial dependence or autocorrelation characteristics of data between the observations in susceptibility assessment. However, to assess the probability of landslide within a specified period of time and within a given area, it is important to understand the spatial correlation between landslide occurrences and influencing factors. By including these relations, the predictive ability of the developed model increases. In this respect, spatial regression (SR) and geographically weighted regression (GWR) techniques, which consider spatial variability in the parameters, are proposed in this study for landslide hazard assessment to provide better realistic representations of landslide susceptibility. The proposed model was implemented to a case study area from More and Romsdal region of Norway. Topographic (morphometric) parameters (slope angle, slope aspect, curvature, plan, and profile curvatures), geological parameters (geological formations, tectonic uplift, and lineaments), land cover parameter (vegetation coverage), and triggering factor (precipitation) were considered as landslide influencing factors. These influencing factors together with past rock avalanche inventory in the study region were considered to obtain landslide susceptibility maps by using SR and LR models. The comparisons of susceptibility maps obtained from SR and LR show that SR models have higher predictive performance. In addition, the performances of SR and LR models at the local scale were investigated by finding the differences between GWR and SR and GWR and LR maps. These maps which can be named as comparison maps help to understand how the models estimate the coefficients at local scale. In this way, the regions where SR and LR models over or under estimate the landslide hazard potential were identified.

DOI

[15]
Erener A, Düzgün H S B, 2011. Landslide susceptibility assessment: What are the effects of mapping unit and mapping method?Environmental Earth Sciences, 66(3): 859-877.Landslide susceptibility assessment forms the basis of any hazard mapping, which is one of the essential parts of quantitative risk mapping. For the same study area, different susceptibility maps can be achieved depending on the type of susceptibility mapping methods, mapping unit, and scale. Although there are various methods of obtaining susceptibility maps, the efficiency and performance of each method should be evaluated. In this study the effect of mapping unit and susceptibility mapping method on landslide susceptibility assessment is investigated. When analyzing the effect of susceptibility mapping method, logistic regression (LR) which is widely used in landslide susceptibility mapping and, spatial regression (SR), which have not been used for landslide susceptibility mapping, are selected. The susceptibility maps with logistic and spatial regression models are obtained using two different mapping units namely slope unit-based and grid-based mapping units. The procedure for investigation of effect of mapping unit on different susceptibility mapping methods is applied to Kumluca watershed, in Bartin Province of Western Black Sea Region, Turkey. 18 factor maps are prepared for landslide susceptibility assessment in the study region. Geographic information systems and remote sensing techniques are used to create the landslide factor maps, to obtain susceptibility maps and to compare the results. The relative operating characteristics (ROC) curve is used to compare the predictive abilities of each model and mapping unit and also the accuracy is evaluated depending on the observations made during field surveys. By analyzing the area under the ROC curve for grid-based and slope unit-based mapping units, it can be concluded that SR model provide better predictive performance (0.774 in grids and 0.898 in slope units) as compared to the LR model (0.744 in grids and 0.820 in slope units). This result is also supported by the accuracy analysis. For both mapping units, the SR model provides more accurate result (0.55 for grids and 0.57 for slope units) than the LR model (0.50 for grids and 0.48 for slopes). The main reason for this better performance is that the spatial correlations between the mapping units are incorporated into the model in SR while this fact is not considered in LR model.

DOI

[16]
FotheringhamA S, Brunsdon C, Charlton M, 2003. Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. John Wiley & Sons.ABSTRACT of the original paper because they are different from those reported in the CSD le, on the Internet site and in Table 1 of the original paper.

DOI

[17]
Fotheringham A S, Charlton M, Brunsdon C, 1997. Measuring spatial variations in relationships with geographically weighted regression. Recent Developments in Spatial Analysis. Springer Berlin Heidelberg, 60-82.

[18]
Frattini P, Crosta G, Carrara A et al., 2008. Assessment of rockfall susceptibility by integrating statistical and physically-based approaches.Geomorphology, 94(3/4): 419-437.In Val di Fassa (Dolomites, Eastern Italian Alps) rockfalls constitute the most significant gravity-induced natural disaster that threatens both the inhabitants of the valley, who are few, and the thousands of tourists who populate the area in summer and winter. To assess rockfall susceptibility, we developed an integrated statistical and physically-based approach that aimed to predict both the susceptibility to onset and the probability that rockfalls will attain specific reaches. Through field checks and multi-temporal aerial photo-interpretation, we prepared a detailed inventory of both rockfall source areas and associated scree-slope deposits. Using an innovative technique based on GIS tools and a 3D rockfall simulation code, grid cells pertaining to the rockfall source-area polygons were classified as active or inactive, based on the state of activity of the associated scree-slope deposits. The simulation code allows one to link each source grid cell with scree deposit polygons by calculating the trajectory of each simulated launch of blocks. By means of discriminant analysis, we then identified the mix of environmental variables that best identifies grid cells with low or high susceptibility to rockfalls. Among these variables, structural setting, land use, and morphology were the most important factors that led to the initiation of rockfalls. We developed 3D simulation models of the runout distance, intensity and frequency of rockfalls, whose source grid cells corresponded either to the geomorphologically-defined source polygons ( geomorphological scenario) or to study area grid cells with slope angle greater than an empirically-defined value of 37 ( empirical scenario). For each scenario, we assigned to the source grid cells an either fixed or variable onset susceptibility; the latter was derived from the discriminant model group (active/inactive) membership probabilities. Comparison of these four models indicates that the geomorphological scenario with variable onset susceptibility appears to be the most realistic model. Nevertheless, political and legal issues seem to guide local administrators, who tend to select the more conservative empirically-based scenario as a land-planning tool.

DOI

[19]
Frattini P, Crosta G B, Fusi N et al., 2004. Shallow landslides in pyroclastic soils: A distributed modelling approach for hazard assessment.Engineering Geology, 73(3/4): 277-295.An effective assessment of shallow landslide hazard requires spatially distributed modelling of triggering processes. This is possible by using physically based models that allow us to simulate the transient hydrological and geotechnical processes responsible for slope instability. Some simplifications are needed to address the lack of data and the difficulty of calibration over complex terrain at the catchment's scale. We applied two simple hydrological models, coupled with the infinite slope stability analysis, to the May 1998 landslide event in Sarno, Southern Italy. A quasi-dynamic model (Barling et al., 1994) was used to model the contribution to instability of lateral flow by simulating the time-dependent formation of a groundwater table in response to rainfall. A diffusion model [Water Resour. Res. 36 (2000) 1897] was used to model the role of vertical flux by simulating groundwater pressures that develop in response to heavy rainstorms. The quasi-dynamic model overestimated the slope instability over the whole area (more than 16%) but was able to predict correctly slope instability within zero order basins where landslides occurred and developed into large debris flows. The diffusion model simulated correctly the triggering time of more than 70% of landslides within an unstable area amounting to 7.3% of the study area. These results support the hypothesis that both vertical and lateral fluxes were responsible for landslide triggering during the Sarno event, and confirm the utility of such models as tools for hazard planning and land management.

DOI

[20]
Gao Jiangbo, Li Shuangcheng, 2011. Detecting spatially non-stationary and scale-dependent relationships between urban landscape fragmentation and related factors using geographically weighted regression. Applied Geography, 31(1): 292-302.Landscape fragmentation is usually caused by many different anthropogenic influences and landscape elements. Scientifically revealing the spatial relationships between landscape fragmentation and related factors is highly significant for land management and urban planning. The former studies on statistical relationships between landscape fragmentation and related factors were almost global and single-scaled. In fact, landscape fragmentations and their causal factors are usually location-dependent and scale-dependent. Therefore, we used geographically Weighted Regression (GWR), with a case study in Shenzhen City, Guangdong Province, China, to examine spatially varying and scale-dependent relationships between effective mesh size , an indicator of landscape fragmentation, and related factors. We employed the distance to main roads as a direct influencing factor, and slope and the distance to district centers as indirect influencing factors, which affect landscape fragmentation through their impacts on land use and urbanization, respectively. The results show that these relationships are spatially non-stationary and scale-dependent, indicated by clear spatial patterns of parameter estimates obtained from GWR models, and the curves with a characteristic scale of 12km for three explanatory variables, respectively. Moreover, GWR models have better model performance than OLS models with the same independent variable, as is indicated by lower AICc values, higher Adjusted R 2 values from GWR and the reduction of the spatial autocorrelation of residuals. GWR models can reveal detailed site information on the different roles of related factors in different parts of the study area. Therefore, this finding can provide a scientific basis for policy-making to mitigate the negative effects of landscape fragmentation.

DOI

[21]
Guzzetti F, Carrara A, Cardinali M et al., 1999. Landslide hazard evaluation: A review of current techniques and their application in a multi-scale study, central Italy.Geomorphology, 31(1): 181-216.In recent years, growing population and expansion of settlements and life-lines over hazardous areas have largely increased the impact of natural disasters both in industrialized and developing countries. Third world countries have difficulty meeting the high costs of controlling natural hazards through major engineering works and rational land-use planning. Industrialized societies are increasingly reluctant to invest money in structural measures that can reduce natural risks. Hence, the new issue is to implement warning systems and land utilization regulations aimed at minimizing the loss of lives and property without investing in long-term, costly projects of ground stabilization. Government and research institutions worldwide have long attempted to assess landslide hazard and risks and to portray its spatial distribution in maps. Several different methods for assessing landslide hazard were proposed or implemented. The reliability of these maps and the criteria behind these hazard evaluations are ill-formalized or poorly documented. Geomorphological information remains largely descriptive and subjective. It is, hence, somewhat unsuitable to engineers, policy-makers or developers when planning land resources and mitigating the effects of geological hazards. In the Umbria and Marche Regions of Central Italy, attempts at testing the proficiency and limitations of multivariate statistical techniques and of different methodologies for dividing the territory into suitable areas for landslide hazard assessment have been completed, or are in progress, at various scales. These experiments showed that, despite the operational and conceptual limitations, landslide hazard assessment may indeed constitute a suitable, cost-effective aid to land-use planning. Within this framework, engineering geomorphology may play a renewed role in assessing areas at high landslide hazard, and helping mitigate the associated risk.

DOI

[22]
Hong H, Pradhan B, Xu C et al., 2015. Spatial prediction of landslide hazard at the Yihuang area (China) using two-class kernel logistic regression, alternating decision tree and support vector machines.Catena, 133: 266-281.Preparation of landslide susceptibility map is the first step for landslide hazard mitigation and risk assessment. The main aim of this study is to explore potential applications of two new models such as two-class Kernel Logistic Regression (KLR) and Alternating Decision Tree (ADT) for landslide susceptibility mapping at the Yihuang area (China). The ADT has not been used in landslide susceptibility modeling and this paper attempts a novel application of this technique. For the purpose of comparison, a conventional method of Support Vector Machines (SVM) which has been widely used in the literature was included and their results were assessed. At first, a landslide inventory map with 187 landslide locations for the study area was constructed from various sources. Landslide locations were then spatially randomly split in a ratio of 70/30 for building landslide models and for the model validation. Then a spatial database with a total of fourteen landslide conditioning factors was prepared, including slope, aspect, altitude, topographic wetness index (TWI), stream power index (SPI), sediment transport index (STI), plan curvature, landuse, normalized difference vegetation index (NDVI), lithology, distance to faults, distance to rivers, distance to roads, and rainfall. Using the KLR, the SVM, and the ADT, three landslide susceptibility models were constructed using the training dataset. The three resulting models were validated and compared using the receive operating characteristic (ROC), Kappa index, and five statistical evaluation measures. In addition, pairwise comparisons of the area under the ROC curve were carried out to assess if there are significant differences on the overall performance of the three models. The goodness-of-fits are 92.5% (the KLR model), 88.8% (the SVM model), and 95.7% (the ADT model). The prediction capabilities are 81.1%, 84.2%, and 93.3% for the KLR, the SVM, and the ADT models, respectively. The result shows that the ADT model yielded better overall performance and accurate results than the KLR and SVM models. The KLR model considered slightly better than SVM model in terms of the positive prediction values. The ADT and KLR are the two promising data mining techniques which might be considered to use in landslide susceptibility mapping. The results from this study may be useful for landuse planning and decision making in landslide prone areas.

DOI

[23]
Hu Juan, Min Ying, Li Huahong et al., 2014. Meteorological early-warning research of mountain torrent and geologic hazard in Yunnan Province.Journal of Catastrophology, 29(1): 62-66. (in Chinese)By analyzing 1101 cases of mountain torrent and geologic hazard and data of rainfall between 2000 to 2011 of 116 meteorological stations in Yunnan Province,the relationships of mountain torrent and geologic hazard occurred by precipitation are discussed,and based on which the meteorological early-warning method for hazards is proposed. It indicated that the temporal variation of mountain torrent and geologic hazard is consistent with which of precipitation. The rainy season( May. to Oct.) is the fastigium of mountain torrent and geologic hazard,with the peak value in major flood season. The frequency of mountain torrent and geologic hazard is more in west of Yunnan than in eastern and more in northwestern than in southeastern. The space distribution of mountain torrent and geologic hazard is basically consistent with which of heavy rainfall days in south and southwest of Yunnan Province. According to the 1101 daily synthesized rainfall data composed of real-time precipitation and effective antecedent precipitation,the critical rainfall is confirmed,on which five grades for early-warning of mountain torrent and geologic hazard are based. There had good effect of the application of the model in 2012 events.

[24]
Huang Yu, Cheng Hualin, 2013. The impact of climate change on coastal geological disasters in southeastern China. Natural Hazards, 65(1): 377-390.Climate change is presently a major global challenge. As the world’s largest developing country, China is particularly vulnerable to global warming, especially in the rapidly developing coastal regions in the southeast of the country. This paper provides an overview of the impacts of climate change on the nature of geological disasters in the coastal regions of southeastern China. In the context of climate change, processes with the potential for causing geological disasters in this region, including sea-level rise, land subsidence, storm surges, and slope failures, which already have a substantial occurrence history, are all aggravated. All these processes have their own characteristics and relevance to climate change. Sea-level rise together with land subsidence reduces the function of dikes and flood prevention infrastructure in the study areas and makes the region more vulnerable to typhoons, storm surges, floods, and astronomical tidal effects. Storm surges have caused great losses in the study areas and also have contributed to increases in rainstorms. As a result, numerous rainfall-induced slope failures, characterized by focused time concentration, high frequencies, strong “burstiness,” and substantial damage, occur in the study areas. To prevent and mitigate such disasters that are accelerated by climate change, and to reduce losses, a series of measures is proposed that may help to achieve sustainable development in coastal southeastern China.

DOI

[25]
Jiang Chaosong, 1990. Geological hazards in Yunnan Province.Journal of Catastrophology, (4): 42-48. (in Chinese)Yunnan is one of the provinces in China where geological hazards, such as muflow, landslide and mountain creep are most serious, ln this paper, the general conditions. distribution, damages and basic characteristics of major geological hazards are discussed so as to draw attention to geological hazards and enhance public awareness of disaster prevention and reduction.

[26]
Jiang Weiguo, Chen Zheng, Lei Xuan et al., 2015. Simulating urban land use change by incorporating an autologistic regression model into a CLUE-S model.Journal of Geographical Sciences, 25(7): 836-850.The Conversion of Land Use and its Effects at Small regional extent (CLUE-S) model is a widely used method to simulate land use change. An ordinary logistic regression model was integrated into the CLUE-S model to identify explanatory variables without considering the spatial autocorrelation effect. Using image-derived maps of the Changsha-Zhuzhou-Xiangtan urban agglomeration, the CLUE-S model was integrated with the ordinary logistic regression and autologistic regression models in this paper to simulate land use change in 2000, 2005 and 2009 based on an observation map from 1995. Significant positive spatial autocorrelation was detected in residuals of ordinary logistic models. Some variables that were much more significant than they should be were selected. Autologistic regression models, which used autocovariate incorporation, were better able to identify driving factors. The Receiver Operating Characteristic Curve (ROC) values of autologistic regression models were larger than 0.8 and the pseudo R 2 values were improved, compared with results of logistic regression model. By overlapping the observation maps, the Kappa values of the ordinary logistic regression model (OL)-CLUE-S and autologistic regression model (AL)-CLUE-S models were larger than 0.75. The results showed that the simulation results were indeed accurate. The Kappa fuzzy (Kfuzzy) values of the AL-CLUE-S models (0.780, 0.773, 0.606) were larger than the values of the OL-CLUE-S models (0.759, 0.760, 0.599) during the three periods. The AL-CLUE-S models performed better than the OL-CLUE-S models in the simulation of land use change. The results showed that it is reasonable to integrate autocovariates into CLUE-S models. However, the Kfuzzy values decreased with prolonged duration of simulation and the maximum range of time was not discussed in this paper.

DOI

[27]
Jiang Weiguo, Deng Lei, Chen Luyao et al., 2009. Risk assessment and validation of flood disaster based on fuzzy mathematics.Progress in Natural Science, 19(10): 1419-1425.Floods often take place around rivers and plains, which indicates a higher risk of flooding in these areas. This paper adopts fuzzy comprehensive assessment (FCA), simple fuzzy classification (SFC), and the fuzzy similarity method (FSM) to assess flood disaster risk in Kelantan, Malaysia. Validation data, such as the flooded area, paddy area, urban area, residential area, and refuges, were overlaid to validate and analyze the accuracy of flood disaster risk. The results show that (1) 70-75% of flooded areas lie within the higher and high-est risk zones, which shows an effective assessment accuracy; (2) paddy, built-up, and residential areas concentrated in the higher and highest risk zones are more likely to be destroyed by flood disasters; (3) 200-225 refuges in the higher and highest risk zones account for around 50% of all refuges, which means that more refuges should be built in the higher and highest risk zones to meet the accom-modation requirement; (4) three methods proved to be feasible and effective in evaluating flood disaster risk, among which FCA is more suitable for the study area than the two other methods.

DOI

[28]
Jiang Weiguo, Deng Yue, Tang Zhenghong et al., 2016. Adaptive capacity of mountainous rural communities under restructuring to geological disasters: The case of Yunnan Province.Journal of Rural Studies, 47: 622-629.Understanding the relationship between adaptive capacity and natural disasters plays a crucial role in mountainous rural development planning in the context of rapid urbanization and the increasing risk of natural disasters in China. Previous studies have examined the adaptive capacity while ignoring the differences between surrounding environments. We analysed the adaptive capacity while incorporating a variety of geological disaster susceptibilities in prefecture-level cities in the mountainous rural Yunnan Province by using socio-economic data from the Yunnan Statistical Yearbook 2014 and geological disaster data from 2000 to 2014. In addition, the major socio-economic determinants and correlation between the adaptive capacity and geological disaster susceptibility were further explored. The results demonstrate the following: (1) 90.2% of geological disasters were primarily distributed in mountainous rural areas, and nearly 70% of prefecture-level cities had moderate, high or very high geological disaster susceptibility; (2) the adaptive capacity was generally higher in south-eastern Yunnan Province, and only 25% of the prefecture-level cities had high or very high adaptive capacity; (3) although the major factors that affected the adaptive capacity varied between cities, the economic conditions were the most important; and (4) although no significant correlation existed between the adaptive capacity and geological disaster susceptibility, cities with high geological disaster susceptibility but low adaptive capacity required more attention to prepare for geological disasters and improve their adaptive capacity. Both the socio-economic background and natural disaster conditions should be considered when evaluating the adaptive capacity. Therefore, determining the local adaptive capacity and then implementing targeted measures to improve the adaptive capacity is more realistic.

DOI

[29]
Kavzoglu T, Sahin E K, Colkesen I, 2013. Landslide susceptibility mapping using GIS-based multi-criteria decision analysis, support vector machines, and logistic regression.Landslides, 11(3): 425-439.Identification of landslides and production of landslide susceptibility maps are crucial steps that can help planners, local administrations, and decision makers in disaster planning. Accuracy of the landslide susceptibility maps is important for reducing the losses of life and property. Models used for landslide susceptibility mapping require a combination of various factors describing features of the terrain and meteorological conditions. Many algorithms have been developed and applied in the literature to increase the accuracy of landslide susceptibility maps. In recent years, geographic information system-based multi-criteria decision analyses (MCDA) and support vector regression (SVR) have been successfully applied in the production of landslide susceptibility maps. In this study, the MCDA and SVR methods were employed to assess the shallow landslide susceptibility of Trabzon province (NE Turkey) using lithology, slope, land cover, aspect, topographic wetness index, drainage density, slope length, elevation, and distance to road as input data. Performances of the methods were compared with that of widely used logistic regression model using ROC and success rate curves. Results showed that the MCDA and SVR outperformed the conventional logistic regression method in the mapping of shallow landslides. Therefore, multi-criteria decision method and support vector regression were employed to determine potential landslide zones in the study area.

DOI

[30]
Komac M, 2006. A landslide susceptibility model using the analytical hierarchy process method and multivariate statistics in perialpine Slovenia. Geomorphology, 74(1): 17-28.Landslides cause damage to property and unfortunately pose a threat even to human lives. Good landslide susceptibility, hazard, and risk models could help mitigate or even avoid the unwanted consequences resulted from such hillslope mass movements. For the purpose of landslide susceptibility assessment the study area in the central Slovenia was divided to 78 365 slope units, for which 24 statistical variables were calculated. For the land-use and vegetation data, multi-spectral high-resolution images were merged using Principal Component Analysis method and classified with an unsupervised classification. Using multivariate statistical analysis (factor analysis), the interactions between factors and landslide distribution were tested, and the importance of individual factors for landslide occurrence was defined. The results show that the slope, the lithology, the terrain roughness, and the cover type play important roles in landslide susceptibility. The importance of other spatial factors varies depending on the landslide type. Based on the statistical results several landslide susceptibility models were developed using the Analytical Hierarchy Process method. These models gave very different results, with a prediction error ranging from 4.3% to 73%. As a final result of the research, the weights of important spatial factors from the best models were derived with the AHP method. Using probability measures, potentially hazardous areas were located in relation to population and road distribution, and hazard classes were assessed.

DOI

[31]
Lan H X, Zhou C H, Wang L J et al., 2004. Landslide hazard spatial analysis and prediction using GIS in the Xiaojiang watershed, Yunnan, China.Engineering Geology, 76(1/2): 109-128.The Xiaojiang watershed in Southwest China has high landslide hazard and has been given the title of he Museum of Geohazards in China . However, the available information on landslides in the Xiaojiang watershed is still limited. We constructed the essential spatial database of landslides using the GIS techniques. The quantitative relationships between landslides and factors affecting landslides are established by the Certainty Factor model (CF). The affecting factors such as lithology, structure, slope angle, slope aspect, elevation and off-fault distance are recognized. By applying CF value integration and landslide zonation, the most significant affecting factors are selected. The widespread landslide activities in the Xiaojiang watershed are caused and triggered by heavy rain. A promising approach to modeling the spatial distribution of rainfall-triggered landslide is combining the mechanistic infinite slope stability model with a hydrological model. A model modified from Stability INdex Mapping (SINMAP) is used to prepare the landslide susceptibility maps for different rainfall conditions. Information on these maps could be useful for explaining the known existing landslide, making emergency decisions and relieving the efforts on the avoidance and mitigation of future landslide hazards.

DOI

[32]
Lee S, Pradhan B, 2007. Landslide hazard mapping at Selangor, Malaysia using frequency ratio and logistic regression models. Landslides, 4(1): 33-41.The aim of this study is to evaluate the landslide hazards at Selangor area, Malaysia, using Geographic Information System (GIS) and Remote Sensing. Landslide locations of the study area were identified from aerial photograph interpretation and field survey. Topographical maps, geological data, and satellite images were collected, processed, and constructed into a spatial database in a GIS platform. The factors chosen that influence landslide occurrence were: slope, aspect, curvature, distance from drainage, lithology, distance from lineaments, land cover, vegetation index, and precipitation distribution. Landslide hazardous areas were analyzed and mapped using the landslide-occurrence factors by frequency ratio and logistic regression models. The results of the analysis were verified using the landslide location data and compared with probability model. The comparison results showed that the frequency ratio model (accuracy is 93.04%) is better in prediction than logistic regression (accuracy is 90.34%) model.

DOI

[33]
Lee S, Sambath T, 2006. Landslide susceptibility mapping in the Damrei Romel area, Cambodia using frequency ratio and logistic regression models.Environmental Geology, 50(6): 847-855.This study applied, tested and compared a probability model, a frequency ratio and statistical model, a logistic regression to Damre Romel area, Cambodia, using a geographic information system. For landslide susceptibility mapping, landslide locations were identified in the study area from interpretation of aerial photographs and field surveys, and a spatial database was constructed from topographic maps, geology and land cover. The factors that influence landslide occurrence, such as slope, aspect, curvature and distance from drainage were calculated from the topographic database. Lithology and distance from lineament were extracted and calculated from the geology database. Land cover was classified from Landsat TM satellite imagery. The relationship between the factors and the landslides was calculated using frequency ratio and logistic regression models. The relationships, frequency ratio and logistic regression coefficient were overlaid to make landslide susceptibility map. Then the landslide susceptibility map was compared with known landslide locations and tested. As the result, the frequency ratio model (86.97%) and the logistic regression (86.37%) had high and similar prediction accuracy. The landslide susceptibility map can be used to reduce hazards associated with landslides and to land cover planning.

DOI

[34]
Li Mingze, Lv Jun, Chen Xin et al., 2015. Provincial evaluation of vulnerability to geological disaster in China and its influencing factors: A three-stage DEA-based analysis.Natural Hazards, 79(3): 1649-1662.China is a country with frequent natural disasters. In order to prevent the losses caused by disaster, this paper plans to make evaluation on vulnerability to geological disaster in 31 provinces in China based on overcoming the disadvantages of traditional data envelopment analysis evaluation methods. The research selected some relevant indexes in China from 2004 to 2010, including the frequency of geological disasters, GDP, population density, personal injury and property loss so as to analyze vulnerability to geological disaster in each province (municipality), and it found that geological vulnerability in China presented an overall pattern of East China > Central China > West China. In addition, it found from the analysis of the influencing factors of vulnerability that industrial development and scientific and technological advancement could reduce vulnerability to geological disasters significantly, while the growth in per-capita GDP and mean sea level could increase vulnerability to geological disasters to a certain extent. Meanwhile, the research indicated that the investment in the prevention and control of geological disasters in China did not have significant effects on the whole vulnerability to geological disasters. Copyright Springer Science+Business Media Dordrecht 2015

DOI

[35]
Li Xue, Liu Xiaoli, Li Jinggang et al., 2013. Factor analysis of earthquake-induced geological disasters of the M7.0 Lushan earthquake in China.Geodesy and Geodynamics, 4(2): 22-29.

DOI

[36]
Liu X, Yue Z Q, Tham L G et al., 2002. Empirical assessment of debris flow risk on a regional scale in Yunnan province, southwestern China.Environ. Manage., 30(2): 249-264.Adopting the definition suggested by the United Nations, a risk model for regional debris flow assessment is presented. Risk is defined as the product of hazard and vulnerability, both of which are necessary for evaluation. A Multiple-Factor Composite Assessment Model is developed for quantifying regional debris flow hazard by taking into account eight variables that contribute to debris flow magnitude and its frequency of occurrence. Vulnerability is a measure of the potential total losses. On a regional scale, it can be measured by the fixed asset, gross domestic product, land resources, population density, as well as the age, education, and wealth of the inhabitants. A nonlinear power-function assessment model that accounts for these indexes is developed. As a case study, the model is applied to compute the hazard, vulnerability and risk for each prefecture of the Yunnan province in southwestern China.

DOI PMID

[37]
Liu Xilin, Lei Junzhong, 2003. A method for assessing regional debris flow risk: An application in Zhaotong of Yunnan province (SW China).Geomorphology, 52(3/4): 181-191.Based on the definitions of the United Nations, the assessment of risk involves the evaluation of both hazard and vulnerability. This forms the basis of a generalized assessment model of debris flow risk. Hazard is a measure of the threatening degree of an extreme event and is expressed theoretically as a function of event magnitude and frequency of occurrence. Mathematically, it is the definite integral area under the magnitude requency curve. Based on the need for a model applicable in regions that lack data, a new method that incorporates theoretical concepts with empirical analysis is presented to calculate the regional hazardousness of debris flows. Debris flow hazard can be estimated from gully density, mean annual rainfall and percentage of cultivated land on steep slope. Vulnerability is defined as the potential total maximum losses due to a potential damaging phenomenon for a specified area and during a reference period. On a regional scale, it is dependent on the fixed assets, gross domestic product, land resources and population density, as well as age, education and wealth of the inhabitants. A nonlinear, power-function model to compute the vulnerability degree is presented. An application of the proposed method to Zhaotong prefecture of Yunnan province, SW China, provides high accuracy and reasonable risk estimates. The highest risk of debris flow is in Zhaotong county with a value of 0.48; the lowest risk of debris flow is in Yanjin county with a value of 0.16. The other counties have debris flow risks ranging from 0.22 to 0.46. This provides an approach for assessing the regional debris flow risk and a basis for the formulation of a regional risk management policy in Zhaotong prefecture.

DOI

[38]
Liu Yang, Liu Ronggao, Ge Quansheng, 2010. Evaluating the vegetation destruction and recovery of Wenchuan earthquake using MODIS data.Natural Hazards, 54(3): 851-862.The Ms 8.0 Wenchuan earthquake in 2008 has led to huge damage to land surface vegetation in northwest Sichuan, one of the typical ecological fragile regions in China. In this paper, the vegetation degradation by the earthquake and its recovery after the disaster are evaluated from analysis of MODIS Gross Primary Productivity (GPP) time series products and other ancillary GIS data. The results suggest that local vegetation GPP after the earthquake in the heavy afflicted area has decreased by 22%. The local vegetation productivity in the heavy afflicted area had recovered to 84 and 87% after 1 and 2 months later. Since August 2008, the vegetation productivity has increased to a nearly normal level.

DOI

[39]
Melchiorre C, Matteucci M, Azzoni A et al., 2008. Artificial neural networks and cluster analysis in landslide susceptibility zonation.Geomorphology, 94(3): 379-400.A landslide susceptibility analysis is performed by means of Artificial Neural Network (ANN) and Cluster Analysis (CA). This kind of analysis is aimed at using ANNs to model the complex non linear relationships between mass movements and conditioning factors for susceptibility zonation, in order to identify unstable areas. The proposed method adopts CA to improve the selection of training, validation, and test records from data, managed within a Geographic Information System (GIS). In particular, we introduce a domain-specific distance measure in cluster formation. Clustering is used in data pre-processing to select non landslide records and is performed on the whole dataset, excluding the test set landslides. Susceptibility analysis is carried out by means of ANNs on the so-generated data and compared with the common strategy to select random non-landslide samples from pixels without landslides. The proposed method has been applied in the Brembilla Municipality, a landslide-prone area in the Southern Alps, Italy. The results show significant differences between the two sampling methods: the classification of the test set, previously separated and excluded from the training data, is always better when the non-landslide patterns are obtained using the proposed cluster sampling. The case study validates that, by means of a domain-specific distance measure in cluster formation, it is possible to introduce expert knowledge into the black-box modelling method, implemented by ANNs, to improve the predictive capability and the robustness of the models obtained.

DOI

[40]
Metternicht G, Hurni L, Gogu R, 2005. Remote sensing of landslides: An analysis of the potential contribution to geo-spatial systems for hazard assessment in mountainous environments.Remote Sensing of Environment, 98(2): 284-303.Natural hazards like landslides, avalanches, floods and debris flows can result in enormous property damage and human casualties in mountainous regions. Switzerland has always been exposed to a wide variety of natural hazards mostly located in its alpine valleys. Recent natural disasters comprising avalanches, floods, debris flows and slope instabilities led to substantial loss of life and damage to property, infrastructure, cultural heritage and environment. In order to offer a solid technical infrastructure, a new concept and expert-tool based on an integrated web-based database/GIS structure is being developed under HazNETH. Given the HazNETH database design contemplates the detection and mapping of diagnostic features from remote sensors (e.g., ground, air and space borne) this paper analyses the use of remote sensing data in landslides studies during the 1980s, 1990s and 2000s, including a discussion of its potential and research challenges as result of new operational and forthcoming technologies such as the very high spatial resolution optical and infrared imagery of Ikonos, Quickbird, IRS CartoSat-1, ALOS, the satellite based interferometric SAR (InSAR and DInSAR of Radarsat, ERS, Envisat, TerraSAR-X, Cosmo/SkyMed, ALOS), micro-satellites like the Pl iades, DMC, RapidEye, airborne LASER altimetry or ground-based differential interferometric SAR. The use of remote sensing data, whether air-, satellite- or ground-based varies according to three main stages of a landslide related study, namely a) detection and identification; b) monitoring; c) spatial analysis and hazard prediction. Accordingly, this paper presents and discusses previous applications of remote sensing tools as related to these three main phases, proposing a conceptual framework for the contribution of remote sensing to the design of databases for natural hazards like debris flows, and identifying areas for further research.

DOI

[41]
Neuhäuser B, Terhorst B, 2007. Landslide susceptibility assessment using “weights-of-evidence” applied to a study area at the Jurassic escarpment (SW-Germany).Geomorphology, 86(1/2): 12-24.A GIS-based model for the assessment of the landslide susceptibility in a selected area of the Jurassic escarpment in the Swabian Alb (SW-Germany) is described, using the weights-of-evidence method. A quantitative model applied to landslides and their causative factors was created and illustrated in susceptibility maps. While previous research work in this area concentrated on large-scale investigations, the present study was carried out at a regional level with a target scale of 1:150,000. The method is based on the assumption that future landslides will occur under the conditions similar or equal to those of past comparable landslides of the same type. Therefore the analysis was limited to one single type of landslides where the causative factors can be assumed as stable over a period of time. Due to uncertainties in the model, mainly because of variances of the weights assigned to the causative factors, the derived probability values, representing the susceptibility for future landslides, have to be considered relative. However, potential susceptible areas can be delineated and landslide indicators can be identified from the available data set. Slopes with angles from 11 to 26 , composed of the Oxford limestone/marls as well as strongly argillaceous and silty colluvial material such as solifluction layers and colluvial cones, are susceptible. The main soil type of the escarpment and the other steep slopes of the Swabian Alb valleys are Rendzinas formed in solifluction layers. Rendzina profiles including rock debris and clay, which are superimposed on marl debris, were also identified as landslide indicators. These findings are in agreement with previous geomorphological studies in the same area. The methodology seems to have widespread applicability beyond this local research area, with the limitation that the knowledge of past landslides input to the model affects the absolute value of the final probability.

DOI

[42]
Nie H, Diao S, Liu J et al., 2001. The application of remote sensing technique and AHP-fuzzy method in comprehensive analysis and assessment for regional stability of Chongqing City, China. Paper presented at the 22nd Asian Conference on Remote Sensing, 5: 9.Regional stability is an important factor to engineering construction. The paper, taking the TM image as a main information source, combining with the AHP-fuzzy method, comprehensively analyses the regional geological environmental conditions of Chongqing municipality. The results show that the stable region accounts for 72.24% of total area, the less stable region and the unstable region 27.76%. This means that Chongqing city is basically in a good condition of regional stability. This reasonable result can provide references for regional planning of mid and long range planning, selecting sites for important projects and immigrants in Chongqing city

DOI

[43]
Pederson C A, Santi P M, Pyles D R, 2015. Relating the compensational stacking of debris-flow fans to characteristics of their underlying stratigraphy: Implications for geologic hazard assessment and mitigation.Geomorphology, 248: 47-56.61A method to predict compensational (avulsion) tendencies of debris fans is proposed.61The compensation index of 3 debris fans in Colorado was analyzed.61Indices were statistically compared to other readily observable fan characteristics.61Relationships can be used to advise debris-flow hazard recognition and mitigation.

DOI

[44]
Pontius R G, Schneider L C, 2001. Land-cover change model validation by an ROC method for the Ipswich watershed, Massachusetts, USA. Agriculture, Ecosystems & Environment, 85(1): 239-248.Scientists need a better and larger set of tools to validate land-use change models, because it is essential to know a model prediction accuracy. This paper describes how to use the relative operating characteristic (ROC) as a quantitative measurement to validate a land-cover change model. Typically, a crucial component of a spatially explicit simulation model of land-cover change is a map of suitability for land-cover change, for example a map of probability of deforestation. The model usually selects locations for new land-cover change at locations that have relatively high suitability. The ROC can compare a map of actual change to maps of modeled suitability for land-cover change. ROC is a summary statistic derived from several two-by-two contingency tables, where each contingency table corresponds to a different simulated scenario of future land-cover change. The categories in each contingency table are actual change and actual non-change versus simulated change and simulated non-change. This paper applies the theoretical concepts to a model of deforestation in the Ipswich watershed, USA.

DOI

[45]
Pourghasemi H R, Moradi H R, Fatemi A S M, 2013. Landslide susceptibility mapping by binary logistic regression, analytical hierarchy process, and statistical index models and assessment of their performances.Natural Hazards, 69(1): 749-779.The current research presents a detailed landslide susceptibility mapping study by binary logistic regression, analytical hierarchy process, and statistical index models and an assessment of their performances. The study area covers the north of Tehran metropolitan, Iran. When conducting the study, in the first stage, a landslide inventory map with a total of 528 landslide locations was compiled from various sources such as aerial photographs, satellite images, and field surveys. Then, the landslide inventory was randomly split into a testing dataset 70 % (370 landslide locations) for training the models, and the remaining 30 % (158 landslides locations) was used for validation purpose. Twelve landslide conditioning factors such as slope degree, slope aspect, altitude, plan curvature, normalized difference vegetation index, land use, lithology, distance from rivers, distance from roads, distance from faults, stream power index, and slope-length were considered during the present study. Subsequently, landslide susceptibility maps were produced using binary logistic regression (BLR), analytical hierarchy process (AHP), and statistical index (SI) models in ArcGIS. The validation dataset, which was not used in the modeling process, was considered to validate the landslide susceptibility maps using the receiver operating characteristic curves and frequency ratio plot. The validation results showed that the area under the curve (AUC) for three mentioned models vary from 0.7570 to 0.8520 \( ({\text{AUC}}_{\text{AHP}} = 75.70\;\% ,\;{\text{AUC}}_{\text{SI}} = 80.37\;\% ,\;{\text{and}}\;{\text{AUC}}_{\text{BLR}} = 85.20\;\% ) \) . Also, plot of the frequency ratio for the four landslide susceptibility classes of the three landslide susceptibility models was validated our results. Hence, it is concluded that the binary logistic regression model employed in this study showed reasonably good accuracy in predicting the landslide susceptibility of study area. Meanwhile, the results obtained in this study also showed that the statistical index model can be used as a simple tool in the assessment of landslide susceptibility when a sufficient number of data are obtained.

DOI

[46]
Pradhan A M S, Kim Y T, 2016. Evaluation of a combined spatial multi-criteria evaluation model and deterministic model for landslide susceptibility mapping.Catena, 140: 125-139.61We compared predictive capability of two different models (SMCE, SHALSTAB).61A total mismatch of 53.27% resulted for three susceptibility classes.61Two results were combined to improve reliability of susceptibility map using NFR.61Combination of SMCE with SHALSTAB can add rainfall triggering factor in semi-quantitative model.61Combined model was more accurate than either individual model at delineating landslide-prone areas.

DOI

[47]
Pradhan B, 2013. A comparative study on the predictive ability of the decision tree, support vector machine and neuro-fuzzy models in landslide susceptibility mapping using GIS.Computers & Geosciences, 51: 350-365.The purpose of the present study is to compare the prediction performances of three different approaches such as decision tree (DT), support vector machine (SVM) and adaptive neuro-fuzzy inference system (ANFIS) for landslide susceptibility mapping at Penang Hill area, Malaysia. The necessary input parameters for the landslide susceptibility assessments were obtained from various sources. At first, landslide locations were identified by aerial photographs and field surveys and a total of 113 landslide locations were constructed. The study area contains 340,608 pixels while total 8403 pixels include landslides. The landslide inventory was randomly partitioned into two subsets: (1) part 1 that contains 50% (4000 landslide grid cells) was used in the training phase of the models; (2) part 2 is a validation dataset 50% (4000 landslide grid cells) for validation of three models and to confirm its accuracy. The digitally processed images of input parameters were combined in GIS. Finally, landslide susceptibility maps were produced, and the performances were assessed and discussed. Total fifteen landslide susceptibility maps were produced using DT, SVM and ANFIS based models, and the resultant maps were validated using the landslide locations. Prediction performances of these maps were checked by receiver operating characteristics (ROC) by using both success rate curve and prediction rate curve. The validation results showed that, area under the ROC curve for the fifteen models produced using DT, SVM and ANFIS varied from 0.8204 to 0.9421 for success rate curve and 0.7580 to 0.8307 for prediction rate curves, respectively. Moreover, the prediction curves revealed that model 5 of DT has slightly higher prediction performance (83.07), whereas the success rate showed that model 5 of ANFIS has better prediction (94.21) capability among all models. The results of this study showed that landslide susceptibility mapping in the Penang Hill area using the three approaches (e.g., DT, SVM and ANFIS) is viable. As far as the performance of the models are concerned, the results appeared to be quite satisfactory, i.e., the zones determined on the map being zones of relative susceptibility.

DOI

[48]
Qiu Haijun, Cao Mingming, Liu Wen et al., 2014. The susceptibility assessment of landslide and its calibration of the models based on three different models. Scientia Geographica Sinica, 34(1): 110-115. (in Chinese)Landslide disaster restricts the sustainable development of human beings which would cause deaths and injuries,property damage and living environment ruins seriously. The regions should be divided into deferent types on the base of disaster risk when making macroeconomic policy of regional geological disaster. Thus,it is very necessary to make susceptibility assessment on zoning prone and risk of geological disasters in these regions firstly. When different assessment models are employed,the results are different. Furthermore,land types according to result of the susceptibility would results in difference in economy. Thus,it was more important to employ suitable model whose susceptibility assessment results were objective and realistic to the fact; however,there were few reports in this field in China yet. This study made assessments on the susceptibility of landslide disaster and evaluated the results. The employed susceptibility assessment models were information value,logistic regression and artificial neural network model. The relative relief,slope,aspect,river system,distance to fault,vegetation cover,formation lithology and road were chosen as factors. The results were showed as following. Firstly,the accuracy of classification influenced the social economic cost. Cohen's Kappa factor method,precision evaluation method proposed by Sridevie Jadi and ROC curve method as the evaluation methods were used to evaluate the assessment results obtained from above models. The Kappa coefficients were 0.72,0.69 and 0.55 by artificial neural network model,logistic regression method and information value model,respectively. The empirical probity(namely accuracy of prediction results) proposed by Sridevie Jadi of above 3 models was 87.48%,74.26% and 69.54%,respectively. The AUC values were 0.805,0.724 and 0.684,respectively. These evaluations proved that the assessment result obtained by artificial neural network model was more realistic to the fact. As a result,artificial neural network model performed the highest level of accuracy in the 3 models. Secondly,there could be one zone and 3 areas according to the landslide assessment results in Ningqiang County. They were: the zone of two sides of Mianxian County-Yang pingguan-Jin shanshi fault,volcanic area of Da n-Miaoba-Gongjiahe-Daijiaba,shale,siltstone and slate area of Tiesuoguan-Hujiaba,phyllite,slate and sandstone area of Anlehe-Guangping,respectively. The area of high-susceptibility area was 1 300.85 km2which accounted for 39.96% of the county area. Landslide in this area was well developed which was affected by Jinshansi-Yangpingguan-Mianxian fault obviously. The area of medium and low susceptibility was 1 227.34 km2and 727.02 km2which accounted for 37.7% and 22.33% respectively.

[49]
Ramani S E, Pitchaimani K, Gnanamanickam V R, 2011. GIS based landslide susceptibility mapping of Tevankarai Ar sub-watershed, Kodaikkanal, India using binary logistic regression analysis.Journal of Mountain Science, 8(4): 505-517.

DOI

[50]
Regmi N R, Giardino J R, Vitek J D, 2010. Modeling susceptibility to landslides using the weight of evidence approach: Western Colorado, USA.Geomorphology, 115(1/2): 172-187.Seventeen factors that cause landslides were measured and weighted using the WOE method to create a map of areas susceptible to landslides. The maps of weighted factors were summed on a pixel-by-pixel basis after performing chi-square tests to determine factors that are conditionally independent of each other. By combining factors that represent topography, hydrology, geology, land cover, and human influences, six models were developed. The performance of each model was evaluated by the distribution of the observed landslides. The validity of the best map was checked against landslides, which were not entered in the analysis. The resulting map of areas susceptible to landslides has a prediction accuracy of 78%.

DOI

[51]
Sabokbar H F, Roodposhti M S, Tazik E, 2014. Landslide susceptibility mapping using geographically-weighted principal component analysis.Geomorphology, 226: 15-24.Landslide susceptibility mapping (LSM) documents the extent of probable landslide events in a region to investigate the distribution, pattern, recurrence and statistics of slope failure and consequent mass movement. Similar to other analyses of quantitative sources of spatial data, LSM sometimes uses principal component analysis (PCA), a form of multivariate statistical analysis. This approach helps identify susceptibility by grouping locations or by measuring the variation between groups. The present study outlines the principles and examines the capability of the proposed methodology for landslide mapping, considers optimized shapes for spatial units, estimates an efficient kernel size using alternating least squares (ALS) analysis confirmed by cross-validation, and uses geographically-weighted principal component analysis (GWPCA) to calculate landslide susceptibility using a fuzzy gamma operator. RMSE and PBIAS statistical estimators were then used to assess operational efficiency of all LSMs using fuzzy gamma operators (0.1 to 0.9). ROC curves were drawn for the best result for LSM using a landslide inventory containing 82 landslide points, with an area under curve of 0.889. The new tools can improve the quality of landslide-related analyses, including erosion studies and landscape modeling, susceptibility and hazard assessments, and risk evaluation.

DOI

[52]
Samarasundera E, Hansell A, Leibovici D et al., 2014. Geological hazards: From early warning systems to public health toolkits.Health Place, 30: 116-119.Extreme geological events, such as earthquakes, are a significant global concern and sometimes their consequences can be devastating. Geographic information plays a critical role in health protection regarding hazards, and there are a range of initiatives using geographic information to communicate risk as well as to support early warning systems operated by geologists. Nevertheless we consider there to remain shortfalls in translating information on extreme geological events into health protection tools, and suggest that social scientists have an important role to play in aiding the development of a new generation of toolkits aimed at public health practitioners. This viewpoint piece reviews the state of the art in this domain and proposes potential contributions different stakeholder groups, including social scientists, could bring to the development of new toolkits.

DOI PMID

[53]
Shekhar S, Schrater P R, Vatsavai R R et al., 2002. Spatial contextual classification and prediction models for mining geospatial data.IEEE Transactions on Multimedia, 4(2): 174-188.Modeling spatial context (e.g., autocorrelation) is a key challenge in classification problems that arise in geospatial domains. Markov random fields (MRF) is a popular model for incorporating spatial context into image segmentation and land-use classification problems. The spatial autoregression (SAR) model, which is an extension of the classical regression model for incorporating spatial dependence, is popular for prediction and classification of spatial data in regional economics, natural resources, and ecological studies. There is little literature comparing these alternative approaches to facilitate the exchange of ideas. We argue that the SAR model makes more restrictive assumptions about the distribution of feature values and class boundaries than MRF. The relationship between SAR and MRF is analogous to the relationship between regression and Bayesian classifiers. This paper provides comparisons between the two models using a probabilistic and an experimental framework.

DOI

[54]
Su Shiliang, Xiao Rui, Zhang Yuan, 2012. Multi-scale analysis of spatially varying relationships between agricultural landscape patterns and urbanization using geographically weighted regression.Applied Geography, 32(2): 360-375.Scientific interpretation of the relationships between agricultural landscape patterns and urbanization is important for ecological planning and management. Ordinary least squares (OLS) regression is the primary statistical method in previous studies. However, this global regression lacks the ability to uncover some local-specific relationships and spatial autocorrelation in model residuals. This study employed geographically weighted regression (GWR) to examine the spatially varying relationships between several urbanization indicators (urbanization intensity index, distance to urban centers and distance to road) and changes in metrics describing agricultural landscape patterns (total area, patch density, perimeter area ratio distribution and aggregation index) at two block scales (5km and 10km). Results denoted that GWR was more powerful than OLS in interpreting relationships between agricultural landscape patterns and urbanization, since GWR was characterized by higher adjust R 2 , lower Akaike Information Criterion values and reduced spatial autocorrelations in model residuals. Character and strength of the relationships identified by GWR varied spatially. In addition, GWR results were scale-dependent and scale effects were particularly significant in three aspects: kernel bandwidth of weight determination, block scale of pattern analysis, and window size of local variance analysis. Homogeneity and heterogeneity in the relationships between agricultural landscape patterns and urbanization were subject to the coupled influences of the three scale effects. We argue that the spatially varying relationships between agricultural landscape patterns and urbanization are not accidental but nearly universal. This study demonstrated that GWR has the potential to provide references for ecological planners and managers to address agricultural landscapes issues at all scales.

DOI

[55]
Tan Yumin, Guo Dong, Bai Bingxin et al., 2015. Geological hazard risk assessment based on information quantity model in Fuling District, Chongqing City, China.Journal of Geo-Information Science, 17(12): 1554-1562. (in Chinese)Using geospatial technologies to assess geological hazard risk has been proved feasible, effective and important in the southwest of China, which is featured by mountainous landscape and the population density is very large. The main objective of this study is to make the risk assessment of the geological hazards in Fuling district using information quantity model, and eight triggering factors are used, including slope, aspect, cumulative catchment area, formation lithology, distances to water, precipitation, vegetation, and land use/land cover type respectively. GaoFen-1 image of December 24, 2013 is used to extract two dynamic triggering factors, vegetation and land use, and precipitation is also taken as a dynamic triggering factor. All triggering factors were then used to construct an information model to assess and predict the geological hazards in the study area in December 2013, producing a geological hazard risk distribution map. Finally, ROC curve was used to validate the information model. The statistical results indicate that the areas with high risk zone is about 9.73% of the entire area and that the percentage of the geological hazards sites is about 52.7% of the entire geological hazards sites. And it shows a satisfactory consistency between the susceptibility map and the geological hazard locations. The AUC of success-rate ROC of 0.796 and the AUC of prediction-rate ROC of 0.748 demonstrate the robustness and relatively good reliability of the information quantity model. Above all, the model can be applied to interpret and predict the geological hazard occurrences in the study area.

[56]
Tang Bangxing, Wu Jishan, 1990. Mountain natural hazards dominated (mainly debris flow) and their prevention.Journal of Geographical Sciences, (2): 202-209. (in Chinese)China is a country with serious mountain natural hazards, especially debris flow and landslide. Since 1980's, the frequent occurrences of mountain hazards such as debris and landslide have made an economic loss of more than 3 billion yuan. This paper analyzes the cause, characteristics of distribution as well as forecast and prevention of the mountain natural hazards.

[57]
Tang Y, Atkinson P M, Wardrop N A et al., 2013. Multiple-point geostatistical simulation for post-processing a remotely sensed land cover classification.Spatial Statistics, 5: 69-84.A post-processing method for increasing the accuracy of a remote sensing classification was developed and tested based on the theory of multiple-point geostatistics. Training images are used to characterise the joint variability and joint continuity of a target spatial pattern, overcoming the limitations of two-point statistical models. Conditional multiple-point simulation (MPS) was applied to a land cover classification derived from a remotely sensed image. Training data were provided in the form of “hard” (land cover labels), and “soft” constraints (class probability surfaces estimated using soft classification). The MPS post-processing method was compared to two alternatives: traditional spatial filtering (also a post-processing method) and the contextual Markov random field (MRF) classifier. The MPS approach increased the accuracy of classification relative to these alternatives, primarily as a result of increasing the accuracy of classification for curvilinear classes. Key advantages of the MPS approach are that, unlike spatial filtering and the MRF classifier, (i) it incorporates a rich model of spatial correlation in the process of smoothing the spectral classification and (ii) it has the advantage of capturing and utilising class-specific spatial training patterns, for example, classes with curvilinear distributions.

DOI

[58]
Trigila A, Iadanza C, Esposito C et al., 2015. Comparison of logistic regression and random forests techniques for shallow landslide susceptibility assessment in Giampilieri (NE Sicily, Italy).Geomorphology, 249: 119-136.The aim of this work is to define reliable susceptibility models for shallow landslides using Logistic Regression and Random Forests multivariate statistical techniques. The study area, located in North-East Sicily, was hit on October 1st 2009 by a severe rainstorm (225 mm of cumulative rainfall in 7 h) which caused flash floods and more than 1000 landslides. Several small villages, such as Giampilieri, were hit with 31 fatalities, 6 missing persons and damage to buildings and transportation infrastructures. Landslides, mainly types such as earth and debris translational slides evolving into debris flows, were triggered on steep slopes and involved colluvium and regolith materials which cover the underlying metamorphic bedrock. The work has been carried out with the following steps: i) realization of a detailed event landslide inventory map through field surveys coupled with observation of high resolution aerial colour orthophoto; ii) identification of landslide source areas; iii) data preparation of landslide controlling factors and descriptive statistics based on a bivariate method (Frequency Ratio) to get an initial overview on existing relationships between causative factors and shallow landslide source areas; iv) choice of criteria for the selection and sizing of the mapping unit; v) implementation of 5 multivariate statistical susceptibility models based on Logistic Regression and Random Forests techniques and focused on landslide source areas; vi) evaluation of the influence of sample size and type of sampling on results and performance of the models; vii) evaluation of the predictive capabilities of the models using ROC curve, AUC and contingency tables; viii) comparison of model results and obtained susceptibility maps; and ix) analysis of temporal variation of landslide susceptibility related to input parameter changes. Models based on Logistic Regression and Random Forests have demonstrated excellent predictive capabilities. Land use and wildfire variables were found to have a strong control on the occurrence of very rapid shallow landslides.

DOI

[59]
Tsangaratos P, Ilia I, 2016. Comparison of a logistic regression and Naïve Bayes classifier in landslide susceptibility assessments: The influence of models complexity and training dataset size.Catena, 145: 164-179.The comparison and validation of the outcomes of each model was achieved using statistical evaluation measures, the receiving operating characteristic and the area under the success and predictive rate curves. The results indicated that model's complexity and the size of the training dataset influence the accuracy and the predictive power of the models concerning landslide susceptibility. In particular, the most accurate model with high predictive power was the eighth model (five variables and 92 training data), with the Na ve Bayes classifier having a slightly higher overall performance and accuracy than the Logistic Regression classifier, 87.50% and 82.61% on the validation datasets, respectively. The highest area under the curve was achieved by the Na ve Bayes classifier for both the training and validating datasets (0.875 and 0.806 respectively) while the Logistic Regression classifier achieved a lower AUC values for the training and validating datasets (0.844 and 0.711, respectively). When limited data are available it seems that more accurate and reliable results could be obtained by generative classifiers, like Na ve Bayes classifiers. Overall, landslide susceptibility assessments could serve as a useful tool for the local and national authorities, in order to evaluate strategies to prevent and mitigate the adverse impacts of landslide events.

DOI

[60]
Uitto J I, Shaw R, 2016. Sustainable development and disaster risk reduction: Introduction. In: Sustainable Development and Disaster Risk Reduction. Springer Japan, 1-12.

[61]
Varnes D J, 1984. Landslide hazard zonation: A review of principles and practice.Natural Hazard, 3: 63.Of all natural hazards, slope failures are the most amenable to measures directed towards avoidance, prevention or correction. The causes inherent in the terrain are relatively well understood, hence the possibility of approaching landslide hazard mitigation from an areal zonal point of view. This publication gives the definitions and principles of landslides, and identifies causative conditions and processes (inherent or basic conditions, geology, geomorphology, hydrologic conditions and climate, vegetation, factors that change stress conditions and strength of materials). Investigation planning is described, including preliminary studies, techniques for identifying unstable and potentially unstable areas, e.g., remote sensing and ground studies, and individual factor maps. For the analysis of data, primarily cartographical methods, numerical rating of contributing factors and numerical-cartographical methods are outlined. Mention is made of some governmental and private responses to slope stability hazards. A summary of operational precepts closes the report. (TRRL)

[62]
Wang H B, Sassa K, 2005. Comparative evaluation of landslide susceptibility in Minamata area, Japan.Environmental Geology, 47(7): 956-966.Landslides are unpredictable; however, the susceptibility of landslide occurrence can be assessed using qualitative and quantitative methods based on the technology of the Geographic Information Systems (GIS). A map of landslide inventory was obtained from the previous work in the Minamata area, the interpretation from aerial photographs taken in 1999 and 2002. A total of 160 landslides was identified in four periods. Following the construction of geospatial databases, including lithology, topography, soil deposits, land use, etc., the study documents the relationship between landslide hazard and the factors that affect the occurrence of landslides. Different methods, namely the logistic regression analysis and the information value model, were then adopted to produce susceptibility maps of landslide occurrence. After the application of each method, two resultant maps categorize the four classes of susceptibility as high, medium, low and very low. Both of them generated acceptable results as both classify the majority of the cells with landslide occurrence in high or medium susceptibility classes, which could be believed to be a success. By combining the hazard maps generated from both methods, the susceptibility was classified as high–medium and low–very low levels, in which the classification of high susceptibility level covers 6.5% of the area, while the areas predicted to be unstable, which are 50.5% of the total area, are classified as the low susceptibility level. However, comparing the results from both the approaches, 43% of the areas were misclassified, either from high–medium to low–very low or low–very low to high–medium classes. Due to the misclassification, 8% and 3.28% of all the areas, which should be stable or free of landsliding, were evaluated as high–medium susceptibility using the logistic regression analysis and the information value model, respectively. Moreover, in the case of the class rank change from high–medium susceptibility to low–very low, 35% and 39.72% of all mapping areas were predicted as stable using both the approaches, respectively, but in these areas landslides were likely to occur or were actually recognized.

DOI

[63]
Wang L, Guo M, Sawada K et al., 2015. Landslide susceptibility mapping in Mizunami City, Japan: A comparison between logistic regression, bivariate statistical analysis and multivariate adaptive regression spline models.Catena, 135: 271-282.Landslides are dangerous natural hazards. Because of their threat, a comprehensive landslide susceptibility map should be produced to reduce the possible damages to people and infrastructure. The quality of landslide susceptibility maps is influenced by many factors, such as the quality of input data and the selection of mathematical models. This study aimed to identify the optimal quantitative method for landslide susceptibility mapping in Mizunami City, Japan. Three mathematical methods, logistic regression (LR), bivariate statistical analysis (BS), and multivariate adaptive regression spline models (MARSplines), were used to create landslide-susceptibility maps by comparing the past landslide distribution and the conditioning factor thematic maps. A landslide inventory map with a total of 222 landslide locations was extracted from aerial photographs provided by NIED (National Research Institute for Earth Science and Disaster Prevention, Japan). Then, the landslide inventory was randomly divided into two datasets: 50% was used for training the models and the remaining 50% for validation purposes. The landslide inventory map provided by NIED and an area under the ROC curve were used to evaluate model performance. We found that the MARSpline method resulted in a better prediction rate (79%) when compared to LR (75%) and BS (77%). In addition, a higher percentage of landslide polygons were found in the high to very high classes using the MARSpline method. Therefore, we concluded that the MARSpline method was the most efficient method for landslide susceptibility mapping in this study area.

DOI

[64]
Wen Xueze, Ma Shengli, Xu Xiwei et al., 2008. Historical pattern and behavior of earthquake ruptures along the eastern boundary of the Sichuan-Yunnan faulted-block, southwestern China.Physics of the Earth and Planetary Interiors, 168(1/2): 16-36.The eastern boundary of the Sichuan-Yunnan faulted-block in southwestern China, which contains four major fault zones named Xianshuihe, Anninghe, Zemuhe, and Xiaojiang, behaves as a huge left-lateral strike-slip active fault system, where the most frequent seismicity in continental China occurs. To uncover the history of earthquake ruptures of this fault system, we developed an empirical relation between rupture extent and seismic intensity distribution based on data of those earthquakes whose rupture extents and intensity distributions are well known. Through analyzing various types of data, including distributions of damage or intensity of historical earthquakes as well as surface ruptures and aftershock zones of modern earthquakes, we determined systematically the locations and spatial extents of ruptures for 36 main earthquakes, and built up a spatial-temporal pattern of the rupture history of the fault system for the last several hundred years. The spatial-temporal rupture pattern contains information of multi-cycle and broad-scale ruptures, from which we find that for most fault segments, scales of segment-ruptures are time- or cycle-variable and only in two or three cases do scales of segment-ruptures in successive cycles seem to be characteristic. In infrequent cases, several adjacent and connecting fault units may rupture simultaneously, resulting in cascading ruptures. Triggering of ruptures is common along the fault system but differs in various periods even on a single fault zone. Triggered ruptures may not occur on fault-segments adjacent to a preceding rupture of major earthquake but on other segments at some distance away from the preceding rupture, implying that the rupture history determines whether a segment can be triggered or not. Recurrence intervals of major segment-ruptures are longer on the Anninghe and Zemuhe fault zones and on the southernmost segment of the Xiaojiang fault zone than that on the Xianshuihe fault zone and the northern and middle segments of the Xiaojiang fault zone, probably due to the partitioning of slip along major and secondary faults and their complicated fault geometry. Four seismic gaps along the fault system are recognized, where major earthquakes have been absent for a long time.

DOI

[65]
Wu Wei, Zhang Lianjun, 2013. Comparison of spatial and non-spatial logistic regression models for modeling the occurrence of cloud cover in north-eastern Puerto Rico.Applied Geography, 37: 52-62.We compared a non-spatial logistic regression model (LRM) with two spatial models, i.e., logistic mixed model (LMM) and geographically weighted logistic model (GWLM), to model/predict the occurrence of orographic cloud cover at the Luquillo Experimental Forest in north-eastern Puerto Rico. The cloud cover data were derived from two Landsat images on a relatively clear day and a relatively cloudy day, respectively. In these models, the response variable was binary with 0 representing non-cloudy areas and 1 representing cloudy areas. The covariates included slope, aspect, and the difference between elevation and lifting condensation level. The results indicated that the spatial LMM did not improve the prediction of probability of cloud cover over the non-spatial LRM. One possible explanation is that the LMM was not able to account for the anisotropy in the non-spatial LRM residuals. In contrast, the GWLM with the bandwidth close to the effective range of the semivariogram of the LRM residuals showed the best model fitting among the three types of models, resulting in the lowest Akaike Information Criterion (AIC) and sum of squared errors (SSE), and the smallest spatial autocorrelation and heterogeneity in the model residuals. The GWLM model coefficients were spatially nonstationary, ranging from negative to positive depending on locations. The significance of the covariates also varied spatially. Our study demonstrated that GWLM is a useful and effective tool to account for spatial heterogeneity in modeling/predicting the occurrence or probability of cloud cover, using the spatial data derived from satellite imagery and GIS.

DOI

[66]
Wu Zizi, 2015. Analysis of cause and study on liability assessment of landslide and debris flow hazards in Yunnan province. Beijing: China University of Geosciences. (in Chinese)

[67]
Xu Chong, Dai Fuchu, Xu Xiwei et al., 2012. GIS-based support vector machine modeling of earthquake-triggered landslide susceptibility in the Jianjiang River watershed, China.Geomorphology, 145/146: 70-80.Support vector machine (SVM) modeling is based on statistical learning theory. It involves a training phase with associated input and target output values. In recent years, the method has become increasingly popular. The main purpose of this study is to evaluate the mapping power of SVM modeling in earthquake triggered landslide-susceptibility mapping for a section of the Jianjiang River watershed using a Geographic Information System (GIS) software. The river was affected by the Wenchuan earthquake of May 12, 2008. Visual interpretation of colored aerial photographs of 1-m resolution and extensive field surveys provided a detailed landslide inventory map containing 3147 landslides related to the 2008 Wenchuan earthquake. Elevation, slope angle, slope aspect, distance from seismogenic faults, distance from drainages, and lithology were used as the controlling parameters. For modeling, three groups of positive and negative training samples were used in concert with four different kernel functions. Positive training samples include the centroids of 500 large landslides, those of all 3147 landslides, and 5000 randomly selected points in landslide polygons. Negative training samples include 500, 3147, and 5000 randomly selected points on slopes that remained stable during the Wenchuan earthquake. The four kernel functions are linear, polynomial, radial basis, and sigmoid. In total, 12 cases of landslide susceptibility were mapped. Comparative analyses of landslide-susceptibility probability and area relation curves show that both the polynomial and radial basis functions suitably classified the input data as either landslide positive or negative though the radial basis function was more successful. The 12 generated landslide-susceptibility maps were compared with known landslide centroid locations and landslide polygons to verify the success rate and predictive accuracy of each model. The 12 results were further validated using area-under-curve analysis. Group 3 with 5000 randomly selected points on the landslide polygons, and 5000 randomly selected points along stable slopes gave the best results with a success rate of 79.20% and predictive accuracy of 79.13% under the radial basis function. Of all the results, the sigmoid kernel function was the least skillful when used in concert with the centroid data of all 3147 landslides as positive training samples, and the negative training samples of 3147 randomly selected points in regions of stable slope (success rate = 54.95%; predictive accuracy = 61.85%). This paper also provides suggestions and reference data for selecting appropriate training samples and kernel function types for earthquake triggered landslide-susceptibility mapping using SVM modeling. Predictive landslide-susceptibility maps could be useful in hazard mitigation by helping planners understand the probability of landslides in different regions.

DOI

[68]
Xu Zengwang, 2001. GIS and ANN model for landslide susceptibility mapping.Journal of Geographical Sciences, 11(3): 374-381.Landslide hazard is as the probability of occurrence of apotentially 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 radient, slope aspect, slope curvature, elevation, the characteristic value, the rainstorms corresponding to the landslide, and distance to drainage line are considered to be related to landslide usceptibility 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 nural network model is a flexible and powerful approach to identify the spatial probability of hazards.

DOI

[69]
Yang Wen, Liu Jie, Cheng Jia et al., 2015. The impact of the 24 March 2011 Myanmar earthquake (MS 7.2) on seismic structure of the Yunnan region.Tectonophysics, 649: 165-175.The velocity structure and the velocity variation in the sixty days before and after the Myanmar earthquake(MS 7.2) on March 24,2011 are inverted by the ambient noise and cross-correlation method,using the continuous seismic waveform records from January to June,2011 in Yunnan region.Meanwhile,the spatial response ratio distribution of strain energy release is obtained using moderate and small earthquakes,and the coulomb stress impacts in different faults caused by the Myanmar earthquake are calculated using the source parameters of the Myanmar earthquake.The results show that in Luquan-Huaping,Yongding-Lushui,and Tonghai-Jianshui regions,the velocities increase,and the seismic activity and coulomb stress in corresponding fault zones of these regions are enhanced,reflecting the positive impact on tectonic activities;in Malong-Xuanwei and Lincang-Jinghong regions,the seismic activity and coulomb stress in fault zone of these regions are decreased,reflecting the negative impact on tectonic activities.

DOI

[70]
Yao X, Tham L G, Dai F C, 2008. Landslide susceptibility mapping based on support vector machine: A case study on natural slopes of Hong Kong, China.Geomorphology, 101(4): 572-582.The Support Vector Machine (SVM) is an increasingly popular learning procedure based on statistical learning theory, and involves a training phase in which the model is trained by a training dataset of associated input and target output values. The trained model is then used to evaluate a separate set of testing data. There are two main ideas underlying the SVM for discriminant-type problems. The first is an optimum linear separating hyperplane that separates the data patterns. The second is the use of kernel functions to convert the original non-linear data patterns into the format that is linearly separable in a high-dimensional feature space. In this paper, an overview of the SVM, both one-class and two-class SVM methods, is first presented followed by its use in landslide susceptibility mapping. A study area was selected from the natural terrain of Hong Kong, and slope angle, slope aspect, elevation, profile curvature of slope, lithology, vegetation cover and topographic wetness index (TWI) were used as environmental parameters which influence the occurrence of landslides. One-class and two-class SVM models were trained and then used to map landslide susceptibility respectively. The resulting susceptibility maps obtained by the methods were compared to that obtained by the logistic regression (LR) method. It is concluded that two-class SVM possesses better prediction efficiency than logistic regression and one-class SVM. However, one-class SVM, which only requires failed cases, has an advantage over the other two methods as only ailed case information is usually available in landslide susceptibility mapping.

DOI

[71]
Yi Lixin, Ge Lingling, Zhao Dong et al., 2012. An analysis on disasters management system in China.Natural Hazards, 60(2): 295-309.This paper introduced the principle, institutional framework, and legal construction of Chinese disaster management system, and operating mechanisms of disaster management departments in pre-disaster, response and post-disaster phases were also demonstrated. Although China has basically built the disaster management system, formed national emergency plan system, and gained achievements in some aspects, the disasters management system is still a segmental model and is not an integrated management system. This article analyzes problems of Chinese disaster management system, and puts forward some suggestions for improving and optimizing this system. This can make Chinese disaster management system better respond and handle to disasters risk, and reduce the social and economic losses of disasters caused.

DOI

[72]
Yu Wenjin, Shao Mingyang, Ren Manliang et al., 2013. Analysis on spatial and temporal characteristics drought of Yunnan Province.Acta Ecologica Sinica, 33(6): 317-324.Since the beginning of the 21st century, the global warming trend is an increasing cause for concern. Frequent extreme weather events occur under the conditions of global change. Exploring the relationship between climate change and drought disasters using climate change performance characteristics has become a prime research problem. It is increasingly important to study the temporal and spatial distribution of aridity and drought causes in the southwest region. We collected meteorological data from 16 meteorological stations from 1956 to 2010 in Yunnan Province of Southwest China, and used a composite index (CI) of meteorological drought to analyze temporal and spatial characteristics of droughts in the province. Based on daily CI values of each station during the 50 years, drought processes there were confirmed individually. Occurrence frequencies, scopes and severities of meteorological droughts were computed and analyzed to reveal their temporal and spatial occurrence patterns in different parts of the province. The results are as follows: (1) Temperature in the province, showing the well-being of the turn, has an obvious 4-year principal cycle. Temperature gradually rose after 1980. Fluctuation of precipitation was relatively stable, with a cycle of about 2 years. (2) On the spatial scale, drought occurred over wide areas in the province. Its overall distribution showed a gradual increase from northwest to southeast. Drought days were the most numerous in the southeast, and fewer in the southwest. (3) Examining seasonal variation, the occurrence of spring droughts was high. The multi-year average of drought frequency exceeded 70%. The frequency of summer and autumn drought was less, and that of winter drought was highest. (4) The causes of drought are complex, comprising the combined effects of atmospheric circulation, geography, and human factors. There was a strong negative correlation between the interannual variation of drought days and precipitation anomaly values. With warming temperatures, the chances for regional drought significantly increased, but the interaction mechanism remains unexplained and should be explored in the future. The climate in Yunnan Province has typical characteristics. The temperature and extreme weather of recent years has certainly changed in response to global climate change, and has caused regional disasters. The aforementioned mechanism represents the next research direction.

DOI

[73]
Zhang Yongshuang, Zhao Xitao, Lan Hengxing et al., 2011. A Pleistocene landslide-dammed lake, Jinsha River, Yunnan, China.Quaternary International, 233(1): 72-80.On both sides of the Jinsha River valley near Benzilan, Dêqên, northwest of Yunnan, China, a series of varied lacustrine deposits have been discovered under the fourth terrace (Q 3 2 ). Landslide debris has been identified underlying the lacustrine deposits, indicating that the lacustrine deposits were caused by a landslide damming the Jinsha River. U-series dating shows that the age of the upper and middle portions of the lacustrine deposits on the right bank were 55.4±3.5 and 82.1±6.6ka respectively. Lacustrine deposits on the left bank have been dated to 122.0±12.4ka at the bottom, which indicates that the large-scale landslide was formed before 122.0±12.4ka and the paleo-lake was developed within the early stage to the mid-stage of the Late Pleistocene. The characteristic of spore-pollen and high content of CaCO 3 in the lacustrine clays reveal that the paleoclimate was semi-arid with less heavy rainfall. Considering the regional neotectonic background, the authors prefer to interpret that the Benzilan ancient landslide-dammed lake was caused by an ancient earthquake. The discovery of the ancient landslide-dammed lake provides significance to the understanding of the formation and evolvement of calamitous geological hazards along large rivers.

DOI

[74]
Zhou Jinxing, Wang Lixian, Xie Baoyuan et al., 2002. A study on the early-warning technique concerning debris flow disasters.Journal of Geographical Sciences, 12(3): 363-370.According to the principle of the eruption of debris flows, the new torrent classification techniques are brought forward. The torrent there can be divided into 4 types such as the debris flow torrent with high destructive strength, the debris flow torrent, high sand-carrying capacity flush flood torrent and common flush flood by the techniques. In this paper, the classification indices system and the quantitative rating methods are presented. Based on torrent classification, debris flow torrent hazard zone mapping techniques by which the debris flow disaster early-warning object can be ascertained accurately are identified. The key techniques of building the debris flow disaster neural network (NN) real time forecasting model are given detailed explanations in this paper, including the determination of neural node at the input layer, the output layer and the implicit layer, the construction of knowledge source and the initial weight value and so on. With this technique, the debris flow disaster real-time forecasting neural network model is built according to the rainfall features of the historical debris flow disasters, which includes multiple rain factors such as rainfall of the disaster day, the rainfall of 15 days before the disaster day, the maximal rate of rainfall in one hour and ten minutes. It can forecast the probability, critical rainfall of eruption of the debris flows, through the real-time rainfall monitoring or weather forecasting. Based on the torrent classification and hazard zone mapping, combined with rainfall monitoring in the rainy season and real-time forecasting models, the debris flow disaster early-warning system is built. In this system, the GIS technique, the advanced international software and hardware are applied, which makes the system performance steady with good expansibility. The system is a visual information system that serves management and decision-making, which can facilitate timely inspect of the variation of the torrent type and hazardous zone, the torrent management, the early-warning of disasters and the disaster reduction and prevention.

DOI

[75]
Zhuang Jianqi, Cui Peng, Wang Gonghui et al., 2015. Rainfall thresholds for the occurrence of debris flows in the Jiangjia Gully, Yunnan Province, China.Engineering Geology, 195: 335-346.61The different rainfall thresholds for the occurrence of debris flow in JJG are built.61The rainfall thresholds can't be exported to neighboring similar areas.61The antecedent precipitation is not a significant factor in JJG.61Debris flows in JJG are predominantly triggered by intraday precipitation.

DOI

Options
Outlines

/

〈 〉