Search

The traditional Stability INdex MAPping (SINMAP) model does not perform detailed divisions of study areas and neglects differences caused by the asymmetrical spatial distribution of geotechnical parameters; thus, the accuracy of the evaluation results is insufficient. In this study, the evaluation results of the SINMAP model were improved based on a combination with the certainty factor (CF) model, and the proposed method is referred to as the CF-SINMAP model. The Wuling Mountain area in Cili County of Hunan Province (China) was selected to verify the CF-SINMAP model. First, eight geological environmental factors in the region were analyzed by the CF method, including the slope, distance from fault, slope direction, distance from water, rock and soil type, elevation, distance from road and vegetation coverage. The rock and soil type, vegetation coverage and human engineering activities were determined as the key factors underlying landslide hazards. Then, the study area was divided into six regions based on the key factors, and the physical and mechanical parameters of each region were refined by the natural environment, formation lithology and human activities. Finally, the CF-SINMAP model was used to calculate and analyze the landslide hazard assessment results under different rainfall conditions. The results show that the CF-SINMAP model is more sensitive to rainfall compared with the traditional method and the unstable areas are mainly distributed along river valleys, reservoir banks and areas with continual human engineering activities. The area under the receiver operating characteristic (ROC) curve values was 0.75 and 0.61 for the CF-SINMAP and SINMAP models, respectively. Compared with the traditional SINMAP model, the CF-SINMAP model produces more reliable results. The rainfall threshold that induced the landslide disaster in Cili County, Hunan Province, was 90 mm/d. In summary, the CF-SINMAP model provides new ideas for the prediction of regional rainfall-induced landslides.

Machine learning models are regarded as efficient and popular models for natural disaster susceptibility prediction. However, few studies have focussed on the applications of the latest popular machine learning models in collapse susceptibility assessment (CSA). This paper proposes a 3S (RS, GPS and GIS) technology-based support vector machine (SVM) to map collapse susceptibility in the Nantian area of China. The 3S technology-based back-propagation neural network (BPNN) and radial basis function neural network (RBFNN) models are also proposed for comparison. First, 44 recorded collapses are identified through field investigation, and fourteen collapse-related causal factors are acquired using 3S technology. Second, among these recorded collapses and randomly selected ‘non-collapses’, 70% of the collapse and non-collapse grid cells are used to train the three models, while the remaining 30% are used to test the models. Third, the collapse susceptibility maps of the Nantian area are produced using the three models. Finally, the prediction accuracies of these models are evaluated. The results indicate that the SVM model has the highest prediction accuracy, while the RBFNN model has the lowest prediction accuracy for CSA. In addition, the distribution characteristics of collapse susceptibility in the Nantian area are produced well by all three models.

Landslide displacement time series can directly reflects landslide deformation and stability characteristics. Hence, forecasting of the non-linear and non-stationary displacement time series is necessary and significant for early warning of landslide failure. Traditionally, conventional machine learning methods are adopted as forecasting models, these forecasting models mainly determine the input and output variables experientially and does not address the non-stationary characteristics of displacement time series. However, it is difficult for these conventional machine learning methods to obtain appropriate input-output variables, to determine appropriate model parameters and to acquire satisfied prediction performance. To deal with these drawbacks, this study proposes the wavelet analysis (WA) to decompose the displacement time series into low- and high-frequency components to address the non-stationary characteristics; then proposes thee chaos theory to obtain appropriate input-output variables of forecasting models, and finally proposes Volterra filter model to construct the forecasting model. The GPS monitoring cumulative displacement time series, recorded on the Shuping and Baijiabao landslides, distance measuring equipment monitoring displacements on the Xintan landslide in Three Gorges Reservoir area of China, are used as test data of the proposed chaotic WA-Volterra model. The chaotic WA-support vector machine (SVM) model and single chaotic Volterra model without WA method, are used as comparisons. The results show that there are chaos characteristics in the GPS monitoring displacement time series, the non-stationary characteristics of landslide displacements are captured well by the WA method, and the model input-output variables are selected suitably using chaos theory. Furthermore, the chaotic WA-Volterra model has higher prediction accuracy than the chaotic WA-SVM and single chaotic Volterra models.

Landslide is a common natural hazard and responsible for extensive damage and losses in mountainous areas. In this study, Longju in the Three Gorges Reservoir area in China was taken as a case study for landslide susceptibility assessment in order to develop effective risk prevention and mitigation strategies. To begin, 202 landslides were identified, including 95 colluvial landslides and 107 rockfalls. Twelve landslide causal factor maps were prepared initially, and the relationship between these factors and each landslide type was analyzed using the information value model. Later, the unimportant factors were selected and eliminated using the information gain ratio technique. The landslide locations were randomly divided into two groups: 70% for training and 30% for verifying. Two machine learning models: the support vector machine (SVM) and artificial neural network (ANN), and a multivariate statistical model: the logistic regression (LR), were applied for landslide susceptibility modeling (LSM) for each type. The LSM index maps, obtained from combining the assessment results of the two landslide types, were classified into five levels. The performance of the LSMs was evaluated using the receiver operating characteristics curve and Friedman test. Results show that the elimination of noise-generating factors and the separated modeling of each landslide type have significantly increased the prediction accuracy. The machine learning models outperformed the multivariate statistical model and SVM model was found ideal for the case study area.

It is important to monitor the displacement time series and to explore the failure mechanism of reservoir landslide for early warning. Traditionally, it is a challenge to monitor the landslide displacements real-timely and automatically. Globe Position System (GPS) is considered as the best real-time monitoring technology, however, the accuracies of the landslide displacements monitored by GPS are not assessed effectively. A web-based GPS system is developed to monitor the landslide displacements real-timely and automatically in this study. And the discrete wavelet transform (DWT) is proposed to assess the accuracy of the GPS monitoring displacements. Wangmiao landslide in Three Gorges Reservoir area in China is used as case study. The results show that the web-based GPS system has advantages of high precision, real-time, remote control and automation for landslide monitoring; the Root Mean Square Errors of the monitoring landslide displacements are less than 5 mm. Meanwhile, the results also show that a rapidly falling reservoir water level can trigger the reactivation of Wangmiao landslide. Heavy rainfall is also an important factor, but not a crucial component.

Displacement predictions are essential to landslide early warning systems establishment. Most existing prediction methods are focused on finding an individual model that provides a better result. However, the limitation of generalization that is inherent in all models makes it difficult for an individual model to predict different cases accurately. In this study, a novel coupled method was proposed, combining the long short-term memory (LSTM) neural networks and support vector regression (SVR) algorithm with optimal weight. The Shengjibao landslide in the Three Gorges Reservoir area was taken as a case study. At first, the moving average method was used to decompose the cumulative displacement into two components: trend and periodic terms. Single-factor models based on LSTM neural networks and SVR algorithms were used to predict the trend terms of displacement, respectively. Multi-factors LSTM and SVR models were used to predict the periodic terms of displacement. Precipitation, reservoir water level, and previous displacement are considered as the candidate factors for inputs in the models. Additionally, ensemble models based on the SVR algorithm are used to predict the optimal weight to combine the results of the LSTM and SVR models. The results show that the LSTM models display better performance than SVR models, the ensemble model with optimal weight outperforms other models. The prediction accuracy can be further improved by also considering results from multiple models.