Precise and efficient landslide displacement prediction is crucial for improving the effectiveness of landslide warning systems. Numerous time series decomposition and machine learning (ML) methods have been proposed and applied in landslide displacement prediction. Nevertheless, most ML methods display individual biases when applied to landslide displacement datasets, and the effect of different methods for time series decomposition on prediction results has not been systematically studied. Therefore, this paper adopts four methods commonly used for time series decomposition to decompose the accumulated displacement into a trend term and a periodic term. The double exponential smoothing is utilized to predict the trend displacement. After the grey relation analysis between the periodic displacement and the external cyclical influencing factors, the ensemble algorithm is used to integrate six commonly used ML algorithms for the prediction of periodic displacement, so as to eliminate the bias of individual artificial intelligence method and enhance the accuracy and stability of prediction results. Furthermore, Bayesian optimization is employed to optimize the base-learners, ensuring the integration fairness. The typical step-like landslides (i.e., Bazimen landslide, Caojiatuo landslide) in the Three Gorges area are selected to compare the performance of different methods for time series decomposition and illustrate the effectiveness of the framework of the ensemble algorithm with the evaluation indices of mean absolute error, mean absolute percentage error and root mean square error. The prediction results indicate that the ICEEMDAN method has the best performance in displacement decomposition. In addition, the prediction results of Bayesian optimized ensemble method are more robust than those of individual ML method, facilitating more accurate and stable landslide displacement prediction and more effective reference for landslide early warning.

Since the impoundment of Three Gorges Reservoir (TGR) in 2003, numerous slopes have experienced noticeable movement or destabilization owing to reservoir level changes and seasonal rainfall. One case is the Outang landslide, a large-scale and active landslide, on the south bank of the Yangtze River. The latest monitoring data and site investigations available are analyzed to establish spatial and temporal landslide deformation characteristics. Data mining technology, including the two-step clustering and Apriori algorithm, is then used to identify the dominant triggers of landslide movement. In the data mining process, the two-step clustering method clusters the candidate triggers and displacement rate into several groups, and the Apriori algorithm generates correlation criteria for the cause-and-effect. The analysis considers multiple locations of the landslide and incorporates two types of time scales: long-term deformation on a monthly basis and short-term deformation on a daily basis. This analysis shows that the deformations of the Outang landslide are driven by both rainfall and reservoir water while its deformation varies spatiotemporally mainly due to the difference in local responses to hydrological factors. The data mining results reveal different dominant triggering factors depending on the monitoring frequency: the monthly and bi-monthly cumulative rainfall control the monthly deformation, and the 10-d cumulative rainfall and the 5-d cumulative drop of water level in the reservoir dominate the daily deformation of the landslide. It is concluded that the spatiotemporal deformation pattern and data mining rules associated with precipitation and reservoir water level have the potential to be broadly implemented for improving landslide prevention and control in the dam reservoirs and other landslide-prone areas.

External causes like changes in reservoir level and intense rainfall can cause reservoir landslides. Exploring the factors that govern landslide deformation and analyzing its stability evolution is essential in mitigating the associated risks. The Sanzhouxi landslide, which has experienced ongoing movements and has been implemented a professional monitoring system, is chosen for analysis in this paper. A combination of geological survey and analysis of monitoring data is utilized to explore landslide deformation characteristics. A data mining method, grey relation analysis (GRA), is subsequently performed to determine the causes of landslide deformation. Furthermore, the stability of the Sanzhouxi landslide in response to reservoir level fluctuation and rainfall for each day over an entire year is assessed using the Morgenstern-Price (MP) approach in 2D GeoStudio software. Such a process illustrates clearly how the landslide stability alters with external triggers changing. The findings reveal that the landslide deforms variably in spatial and temporal. The reservoir level rising contributes to landslide deformation primarily, while rainfall has a secondary impact. The factor of safety (FS) of the Sanzhouxi landslide drops from 1.17 to 1.07 during high reservoir water level periods and remain the same or increase in other periods except for some transitory moments while decreasing only by about 2% under the effect of rainfall. The daily FS results validate the dominant influence of reservoir level fluctuation on the stability of the landslide. The comprehensive understanding of landslide movement based on deformation characteristics, triggering factor identification, and daily stability validation, contributes to realizing nearly real-time prediction and evaluating the risk due to slope movements in similar geological settings worldwide.

Landslide displacement prediction has garnered significant recognition as a pivotal component in realizing successful early warnings and implementing effective control measures. This task remains challenging as landslide deformation involves not only temporal dependency within time series data but also spatial dependence across various regions within landslides. The present study proposes a landslide spatiotemporal displacement forecasting model by introducing attention-based deep learning algorithms based on spatiotemporal analysis. The Maximal Information Coefficient (MIC) approach is employed to quantify the spatial and temporal correlations within the daily data of Global Navigation Satellite System (GNSS) observations. Based on the quantitative spatiotemporal analysis, the proposed prediction model combines a convolutional neural network (CNN) and long short-term memory (LSTM) network to capture spatial and temporal dependencies individually. Spatial–temporal attention mechanisms are implemented to optimize the model. Additionally, we develop a single-point prediction model using LSTM and a multiple-point prediction model using the CNN-LSTM without an attention mechanism to compare the forecasting capabilities of the attention-based CNN-LSTM model. The Outang landslide in the Three Gorges Reservoir Area (TGRA), characterized by a large and active landslide equipped with an advanced monitoring system, is taken as a studied case. The temporal MIC results shed light on the response times of monitored daily displacement to external factors, showing a lagging duration of between 10 and 50 days. The spatial MIC results indicate mutual influence among different locations within the landslide, particularly in the case of nearby sites experiencing significant deformation. The attention-based CNN-LSTM model demonstrates an impressive predictive performance across six monitoring stations within the Outang landslide area. Notably, it achieves a remarkable maximum coefficient of determination (R2) value of 0.9989, accompanied by minimum values for root mean squared error (RMSE), absolute mean error (MAE), and mean absolute percentage error (MAPE), specifically, 1.18 mm, 0.99 mm, and 0.33%, respectively. The proposed model excels in predicting displacements at all six monitoring points, whereas other models demonstrate strong performance at specific individual stations but lack consistent performance across all stations. This study, involving quantitative deformation characteristics analysis and spatiotemporal displacement prediction, holds promising potential for a more profound understanding of landslide evolution and a significant contribution to reducing landslide risk.

Reliable and accurate prediction of landslide displacement is essential for early warning systems, as well as for disaster prevention and mitigation. Machine learning and deep learning algorithms are capable of modeling the relationship between step-like deformation and causal factors for displacement prediction. The ratio of the training set to the testing and the randomness of model parameters affect the assessment of model prediction capacity, which has been neglected in the literature. To take a more integrated view of the displacement prediction model and results, this study develops a hybrid approach that combines random training set length, Monte Carlo simulation, and performance function to evaluate the prediction model. The proposed approach is applied to a typical step-like case of the Jiuxianping landslide in the Three Gorges reservoir area. The uncertainty and variability of two extensively explored neural network models in the prediction of Jiuxianping landslide are systematically explored. The results showed that the probability of failure (and coefficient of variation (COV)) values for the LSTM and RNN models were 16.20% (COV = 0.19%) and 17.12% (COV = 0.42%), respectively. The LSTM model outperformed the RNN model in terms of failure probability, COV, and error distribution. The proposed scheme is worthy of reference and allows for a comprehensive evaluation of the prediction model and results. (c) 2023 International Association for Gondwana Research. Published by Elsevier B.V. All rights reserved.

It is crucial to predict landslide displacement accurately for establishing a reliable early warning system. Such a requirement is more urgent for landslides in the reservoir area. The main reason is that an inaccurate prediction can lead to riverine disasters and secondary surge disasters. Machine learning (ML) methods have been developed and commonly applied in landslide displacement prediction because of their powerful nonlinear processing ability. Recently, deep ML methods have become popular, as they can deal with more complicated problems than conventional ML methods. However, it is usually not easy to obtain a well-trained deep ML model, as many hyperparameters need to be trained. In this paper, a deep ML method-the gated recurrent unit (GRU)-with the advantages of a powerful prediction ability and fewer hyperparameters, was applied to forecast landslide displacement in the dam reservoir. The accumulated displacement was firstly decomposed into a trend term, a periodic term, and a stochastic term by complementary ensemble empirical mode decomposition (CEEMD). A univariate GRU model and a multivariable GRU model were employed to forecast trend and stochastic displacements, respectively. A multivariable GRU model was applied to predict periodic displacement, and another two popular ML methods-long short-term memory neural networks (LSTM) and random forest (RF)-were used for comparison. Precipitation, reservoir level, and previous displacement were considered to be candidate-triggering factors for inputs of the models. The Baijiabao landslide, located in the Three Gorges Reservoir Area (TGRA), was taken as a case study to test the prediction ability of the model. The results demonstrated that the GRU algorithm provided the most encouraging results. Such a satisfactory prediction accuracy of the GRU algorithm depends on its ability to fully use the historical information while having fewer hyperparameters to train. It is concluded that the proposed model can be a valuable tool for predicting the displacements of landslides in the TGRA and other dam reservoirs.

Quantifying the uncertainties in the prediction of landslide displacement is important for making reliable predictions and for managing landslide risk. This study develops a novel approach for the interval prediction (i.e. uncertainty) of landslide with step-like displacement pattern in the Three Gorges Reservoir (TGR) area using Density-Based Spatial Clustering of Applications with Noise (DBSCAN), Synthetic Minority Oversampling Technique and Edited Nearest Neighbor (SMOTEENN) based Random Forest (RF) and bootstrap-Multilayer Perceptron (MLPs). DBSCAN was employed to carry out clustering analysis for different deformation states of the landslide with step-like displacement pattern. The SMOTEENN based RF classifier was trained to deal with imbalanced classification problems. A dynamic switching prediction scheme to construct high-quality Prediction Intervals (PIs) using bootstrap-MLPs was established. The concepts of Pareto front and Knee point were adopted to select the PIs that could provide the best compromise between reliability and accuracy. The proposed DBSCAN-RF-bootstrap-MLP method is illustrated and verified with one typical landslide with step-like displacement pattern, the Bazimen landslide from the TGR area in China. The method showed to perform well and provides the uncertainties associated with landslide displacement prediction for decision making.

在库水位波动和降雨作用的共同影响下,库岸滑坡的变形规律往往更为复杂.以三峡库区麻柳林滑坡为例,基于野外调查、钻探编录、深部位移监测以及数值模拟等手段,分析了库水位波动和降雨作用下滑坡变形特征及演化规律.结果表明:麻柳林滑坡在粉质黏土层和块石层交界处发育一个次级滑带,目前该滑坡主要沿次级滑带运动,导致次级滑动的原因与坡体物质的差异性有关;Si(Sf)指标分析法揭示滑坡的滑带还未完全破坏,滑坡仍处于蠕变状态;根据三峡水库水位调度规律,将一个完整水文年划分为6个阶段,数值模拟结果表明滑坡在库水位缓慢下降阶段变形速率较小、在快速下降阶段和低水位阶段变形速率持续增大、在快速上升阶段和缓慢上升阶段以及高水位阶段变形速率则保持平稳.其中,降雨的直接影响和降雨导致库水位波动进而对滑坡变形造成的间接影响,使得麻柳林滑坡在低水位阶段的变形显著增加、稳定性最差,应加强该时段内滑坡的监测和预警.

Soil thickness has a great importance in many processes such as slope stability, seismic local effects, landscape evolution, soil moisture distribution. It is a fundamental parameter in many environmental models. In local scale applications, direct or indirect measurements can be easily used to accurately measure soil thickness. Nevertheless, in large scale applications, it is often difficult to obtain a reliable distributed soil thickness map and existing methods have been applied only to test sites with shallow soil depth. In this research, we cope with this limitation showing a first attempt to test the applicability of some state-of-the-art soil thickness models in a test site characterized by a complex geological setting and soil thickness values extending from zero to forty meters. Two different approaches were used to derive distributed soil thickness maps: a modified version of the Geomorphologically Indexed Soil Thickness (GIST) model, purposely customized to better take into account the peculiar setting of the test site, and a regression performed with a machine learning algorithm, the Random Forest (RF), combined with the geomorphological parameters of GIST. The proposed models are implemented in a geographic information system environment on a pixel-by-pixel basis. Finally, validation quantifies errors of the two models and a comparison with geophysical data is carried out. The results showed that the GIST model is not able to fully grasp the high spatial variability of soil thickness of the study area: mean absolute error was is 10.68 m with 7.94 m standard deviation, and the frequency distribution of residuals showed a proneness to underestimation. In contrast, RF returned a better performance (mean absolute error is 3.52 m with 2.92 m standard deviation), and the derived map could be considered to be used in further analyses to feed models that require a distributed soil thickness map as a spatially distributed input parameter.

Landslides represent major threats to life and property in many areas of the world, such as the landslides in the Three Gorges Dam area in mainland China. To better prepare for landslides in this area, we explored how several machine learning algorithms (long short term memory (LSTM), random forest (RF), and gated recurrent unit (GRU)) might predict ground displacements under three types of landslides, each with distinct step-wise displacement characteristics. Landslide displacements are described with trend and periodic analyses and the predictions with each algorithm, validated with observations from the Three Gorges Dam reservoir over a one-year period. Results demonstrated that deep machine learning algorithms can be valuable tools for predicting landslide displacements, with the LSTM and GRU algorithms providing the most encouraging results. We recommend using these algorithms to predict landslide displacement of step-wise type landslides in the Three Gorges Dam area. Predictive models with similar reliability should gradually become a component when implementing early warning systems to reduce landslide risk.

A good prediction of landslide displacement is an essential component for implementing an early warning system. In the Three Gorges Reservoir Area (TGRA), many landslides deform distinctly and in steps from April to September each year under the influence of seasonal rainfall and periodic fluctuation in reservoir water level. The sliding becomes more uniform again from October to April. This landslide deformation pattern leads to accumulated displacement versus time showing a step-wise curve. Most of the existing predictive models express static relationships only. However, the evolution of a landslide is a complex nonlinear dynamic process. This paper proposes a dynamic model to predict landslide displacement, based on time series analysis and long short-term memory (LSTM) neural network. The accumulated displacement was decomposed into a trend term and a periodic term in the time series analysis. A cubic polynomial function was selected to predict the trend displacement. By analyzing the relationships between landslide deformation, rainfall, and reservoir water level, a LSTM model was used to predict the periodic displacement. The LSTM approach was found to properly model the dynamic characteristics of landslides than static models, and make full use of the historical information. The performance of the model was validated with the observations of two step-wise landslides in the TGRA, the Baishuihe landslide and Bazimen landslide. The application of the model to those two landslides demonstrates that the LSTM model provides a good representation of the measured displacements and gives a more reliable prediction of landslide displacement than the static support vector machine (SVM) model. It is concluded that the proposed model can be used to effectively predict the displacement of step-wise landslides in the TGRA.

Since the impoundment of the Three Gorges Reservoir in mid-2003, slope displacements have occurred and existing landslides have been reactivated. The movement are a threat to the community and the environment. High precipitation in the area increases landslide susceptibility. The relationships among landslide displacement, rainfall and reservoir drawdown were analyzed for the Baijiabao landslide with the Grey relational analysis approach and using eight years of monitored displacement. The results suggest that rainfall triggered the landslide deformation in the earlier years. Then reservoir drawdown and the combined effect of rainfall and drawdown became preponderant. The identification of the triggers is important for selecting mitigation measures and evaluating the risk due to slope movement.

三峡水库建成后,库水位周期性涨落和暴雨产生的渗流作用导致大量古滑坡的复活或新滑坡的发生.以库区近水平层状结构的四方碑滑坡为例,依据库水位实际调动,将水位从175 m至145 m不同降速与50年一遇暴雨进行工况组合,计算4种工况下滑坡的稳定性及破坏概率.然后采用Geo-studio软件的Sigma模块对滑坡进行变形模拟,运用R/S分析方法判断滑坡的变形持续性,并结合野外调查情况,综合评价分析四方碑滑坡的稳定性.结果表明:滑坡在各工况下整体均处于基本稳定状态,具有低危险性;变形模拟结果显示滑坡前缘位移最大,与野外调查情况一致;各监测点Hurst指数均介于0.5～1,表明时间序列具有正持续性,在研究的时间限度内滑坡的局部破坏增强,应在汛期加强对滑坡前缘的巡查和预警.

针对滑坡演化的动态特性和传统静态预测模型的不足,提出一种基于时间序列与长短时记忆网络(long and short term memory neural network,LSTM)的滑坡位移动态预测模型.该模型首先采用移动平均法将滑坡累积位移分解为趋势项位移和周期项位移.然后采用多项式函数预测趋势项位移;基于滑坡变形特征与诱发因素的响应分析,建立LSTM模型进行周期项位移预测.最后将各分项位移叠加,即实现滑坡累积位移的预测.以三峡库区典型阶跃型滑坡——白水河滑坡为例,并与支持向量机模型(support vector machine,SVM)进行对比分析.结果表明,与静态模型SVM相比,动态模型LSTM的预测精度较高,在阶跃式变形期的预测优势尤为突出,且不依赖于训练数据时效性的分析.该模型为三峡库区阶跃型滑坡位移预测提供了新的思路和探索.

Considering the strong nonlinear dynamic characteristics of dam deformation, the prediction model of dam deformation is investigated. Support vector machine (SVM) is combined with other methods, such as phase space reconstruction, wavelet analysis and particle swarm optimization (PSO), to build the prediction model of dam deformation. Firstly, the chaotic characteristics and the predictable time scale of dam deformation are identified by implementing the phase space reconstruction of observation data series on dam deformation. Secondly, a SVM-based prediction model of dam deformation is proposed. The reconstructed phase space of observed deformation and the Morlet wavelet basis function are selected as the input vector and the kernel function of SVM. Thirdly, the PSO algorithm is improved to implement the parameter optimization of SVM-based prediction model of dam deformation. Finally, the displacement of one actual dam is taken as an example. The results demonstrate the modeling efficiency and forecasting accuracy can be improved. (C) 2018 Elsevier Ltd. All rights reserved.

The GPS monitoring cumulative displacement on reservoir landslides in the Three Gorges Reservoir area shows step-like characteristics and is a probable chaotic time series under the influences of the seasonal rainfall and reservoir water level fluctuation.Traditionally,the uni-variable chaotic model is commonly used to predict the landslide displacement;and all exist-ing multivariable models select the input variables empirically without theoretical exploration of the nonlinear dynamic evolution process of landslide displacement and its inducing factors.A new combined model based on double exponential smoothing (DES),multivariable chaotic model,extreme learning machine(ELM)is proposed in this study.First,the chaos characteristic of landslide displacement is identified by the combined DES and multivariable chaotic ELM.Second,the DES method is used to predict the cumulative displacement.The predictive results are the trend displacement,and the periodic displacement is ob-tained by reducing the trend displacement from the cumulative displacement.Third,the multivariate phase space reconstruction method of chaotic theory is used to explore the dynamic relationship between the periodic displacement and its inducing factors, and the ELM model is established to predict the periodic displacement.Finally,the total forecast cumulative displacement is obtained by adding the predictive trend and periodic displacement.The GPS monitoring cumulative displacement on the Baishuihe landslide is used as case study.In addition,the proposed model is compared with the combined DES and multivariable chaotic particle swarm optimized support vector machine model,the combined DES and uni-variable chaotic ELM model.The results show that the prediction accuracy of the proposed model is higher than that of other models.The proposed model ex-plores the nonlinear characteristic of landslide displacement and its dynamic relationship with inducing factors.The model also reflects the physical meaning of the nonlinear evolution of the landslide displacement.

Landslides in the Three Gorges Reservoir (TGR) are widely distributed and are a serious threat to the environment and the local people. Since the impoundment of the reservoir in 2003, many of the landslides have been reactivated which were triggered by water level fluctuation and rainfall. Taking the Maliulin landslide in the TGR as a case study, field investigation and displacement monitoring are conduced to study the characteristics of the landslide. According to the annual variation, the fluctuation of reservoir water is divided into four periods. The cumulative rainfall corresponding to different rainfall return periods is computed by Gumbel model. The variation of landslide stability and failure probability under the effect of water level fluctuation and rainfall in a complete annual cycle is calculated in terms of the Morgenstern–Price and Monte Carlo model. Based on the monitoring by inclinometer, a secondary shallow sliding surface is detected which controls its current activities. The annual variation of landslide stability tends to coincide with the change of reservoir water level. The minimum factor of safety occurs during the period of water level drawdown. Combining with the effect of extreme rainfall 50-year return period and water level dropdown, the calculated minimum factor of safety is below unit and the landslide is unstable. The scenario of annual failure probability of landslide is completed in the paper that is the basis for further risk evaluations.

Landslide displacement system is generally characterized by non-stationary and nonlinear characteristics. Traditionally, many artificial neural network (ANN) models have been proposed to forecast landslide displacement. However, the underlying non-stationary characteristics in the landslide displacement are not captured, and the input–output variables of the ANN models are not selected nonlinearly. To overcome these drawbacks, this paper proposes the chaos theory-based discrete wavelet transform (DWT)–extreme learning machine (ELM) model to predict landslide displacement. The DWT method is adopted to decompose the landslide displacement into several low- and high-frequency components to address the non-stationary characteristics. And chaos theory is used to determine the input–output variables of the ELM model. The cumulative displacement time series of the Baishuihe and Baijiabao landslides in the Three Gorges Reservoir Area, China, are used as data sets. The results show that the chaotic DWT-ELM model accurately predicts landslide displacement. The chaotic DWT–support vector machine (SVM), chaotic DWT–back-propagation neural network (BPNN) and single chaotic ELM models are used for comparisons. The comparison results show that the chaotic DWT-ELM model achieves higher prediction accuracy than do the chaotic DWT-SVM, chaotic DWT-BPNN and the single chaotic ELM models.

库水位升降和降雨通过改变三峡库区库岸滑坡岩土体的抗剪强度和应力状态,影响库岸滑坡的稳定性.为探讨白家包滑坡在库水位升降和降雨联合作用下的稳定性变化特征,本文首先根据GPS监测数据定性分析白家包滑坡变形规律,再采用Geo-studio软件计算4种工况下滑坡的稳定性系数,最后采用R/S分析法计算各GPS监测点累积位移的Hurst指数,并将Hurst指数值与Geo-studio数值模拟结果进行对比分析.结果表明:白家包滑坡累积位移曲线呈“阶跃状”特征;滑坡稳定性受库水位升降和降雨的综合影响,库水位下降时稳定性系数减小,上升时稳定性系数增大,降雨也能在一定程度上降低滑坡稳定性;175～145 m加降雨工况下滑坡最小稳定性系数为1.034,处于欠稳定状态;各监测点Hurst指数均介于0.5～1之间,表明未来滑坡变形将持续加剧,与滑坡变形定性分析及稳定性数值模拟结果一致.