Deformation Characteristics And Displacement Prediction Of Colluvial Landslides In Wanzhou County,Three Gorges Reservoir

Since the impoundment of the Three Gorges Reservoir in 2003,the hydrogeological conditions of the entire region was dramatically changed and triggered many landslides.The majority of the landslides were colluvial landslides.It is critical to study on deformation characteristics and displacement prediction of colluvial landslides along the reservoir banks,which can provide scientific guidelines for landslide risk assessment,avoidance and mitigation.In this thesis,the colluvial landslides along the banks of the Three Gorges Reservoir in the Wanzhou County are studied.Based on the latest geological investigation materials and monitoring data,the statistical analysis of development laws for colluvial landslide is carried out.Then the internal and external factors influencing landslide stability are summarized.The Tangjiao NO.1 landslide and Sanzhouxi landslide are chosen as cases to study on the deformation characteristics and mechanism of colluvial landslides under the effect of water level fluctuation.The deformation of these two landslides are induced by drawdown and raise of reservoir water level,respectively.Main controlling factors are selected according to the deformation mechanism of the two cases,and are used to predict landslide displacement.The main conclusions are summarized as follows.(1)The development regularities of colluvial landslides including distribution,development and deformation characteristics are studied by statistics.The distribution laws include location,scale and thickness.The development laws involve material composition,morphological feature and hydrogeological conditions.The deformation characteristics include ground surface displacement,deep displacement and deformation mode.The results show that most of the colluvial landslides along Wanzhou County are large-sized or medium-sized with the thickness of 10~25 m.For most of the landslides,the front parts are submerged in the reservoir water and thus can be effected by the water level fluctuation.The elevation of upper edge is mainly between205m and 265 m.The slope height ranges from 40 m to 120 m.Both of the length and width of landslides mainly range from 200 m to 600 m,while the areas are within 30×10~4m~2.Most of the landslides are dustpan shape and rectangle in the plane form,and are stepwise and convex in the sections.The slope is mainly between 10°and 30°.The landslides with medium permeability account for 90%.The sliding surface develops mainly as a curve.The ground surface displacements of the landslides show step-like behavior.The statistic results of deep displacements reveal that the front parts of landslides move along sliding surfaces while the upper parts still keep creep deformation.The deformation modes of landslides are mainly retrogressive.(2)The Tangjiao NO.1 landslide is taken as an example to analyze the deformation characteristics induced by reservoir water level drawdown.The intense deformation occurs during the period of reservoir water level rapid drawdown.But the landslide deformation lags behind water level fluctuation and the annual accumulated displacement decreased gradually.The loading-unloading response ratio theory(LURR)is used to analyze the response of landslide deformation to reservoir water fluctuation.This method is applied to two time-scales including the whole period of step-like deformation and weekly.The results of LURR show that the Tangjiao NO.1landslide tends to fail during the period of step-like deformation.An abrupt jump occurs in the RULL curve during the third week.This means that the front part of the landslide can lose stability and more warning work needs to be done during this period.The front part of the Tangjiao NO.1 landslide deforms more distinctly during the rapid drawdown period than in the slow drawdown period.The potential cause is the increasing of hydrodynamic pressure in the slope due to the raise of hydraulic gradients in rapid drawdown stage.The front part of the Tangjiao NO.1 landslide is unstable during the periods of reservoir level slow drawdown,rapid drawdown and 145 m water level.As the front part of the landslides is mainly submerged in the reservoir water,the effect of rainfall on landslide deformation is limited.The deformation mechanism of the Tangjiao NO.1 landslide is summarized from the effect of internal and external factors.The landslide is located in the southeast of Wanzhou Synclinore,which may cause collapse due to apparent gravity.The front part,left and right boundaries can provide enough space for landslide sliding.The sliding materials contain montmorillonite and kaolin,which results in shear strength decreasing when there is water.However,the step section and gentle topography contribute to improving landslide stability.The deformation of the front part is mainly induced by reservoir water level rapid drawdown.The middle and upper edge are in creep deformation stage and the deformation is mainly induced by rainfall and human activity.The Tangjiao NO.1landslide deforms as a retrogressive landslide,but the sliding body on different platforms are independent.(3)The Sanzhouxi landslide is taken as an example to analyze the deformation characteristics induced by reservoir water level increasing.The grey relational analysis approach is used to identify the key effect factor of landslide deformation in different evolution stages.The results suggest that rainfall triggered the landslide deformation in the earlier stage and reservoir water level raise became the preponderant factor in the later stage.Using the S_i(S_f)index analysis method to evaluate landslide stability,the results show that the Sanzhouxi landslide is stable and keeps creep deformation at present.The landslide displacement increases obviously when reservoir water level keeps 175 m,but it has negative horizontal displacement in others periods.The landslide is stable or basic stable.The factor of safety keeps decreasing during the high water level period(175 m).With the effect of rainfall,the factor of safety decreases 0.018.The deformation mechanism of the Sanzhouxi landslide is summarized from the effect of internal and external factors.As the landslide is linear shape in section,it keeps creep deformation under the effect of gravity.The sliding material contains montmorillonite and develops a large number of micro-cracks and micro-pores,resulting in the descend of shear strength when there is water.The front part deforms under the effect of reservoir water level raise and rainfall,in which the former one is the dominating triggering factor.The deformation of the upper part is mainly induced by rainfall.The secondary sliding of the front part is caused by landslide toe excavation.The deformation mode of the Sanzhouxi landslide changes from“advancing”to“retrogressive”under the effect of reservoir water level fluctuation.(4)A dynamic model based on wavelet transform(WT)and long short-term memory(LSTM)neural network is used to predict the displacements of the Tangjiao NO.1landslide and Sanzhouxi landslide.Most of the existing prediction models are static models,which consider landslide displacement prediction as a static regression problem.However,landslide evolution is a complex nonlinear dynamic process.A dynamic model based on wavelet transform(WT)and long short-term memory(LSTM)neural network is proposed.The accumulated displacement is divided into trend term and periodic term using WT.The trend displacement is predicted by univariate LSTM.The periodic displacement is predicted by multivariate LSTM,while two static models,support vector machine(SVM)and back propagation neural network(BPNN)are used for comparison.The results show that the multivariate LSTM has better prediction accuracy than the univariate LSTM.As a complex nonlinear system,the landslide displacement is induced by many factors.It is reasonable to take the influencing factors into consideration in the construction of the prediction models.The selection of influencing factors should be based on the analysis of landslide deformation mechanism.The dynamic LSTM approach is found to provide a good representation of the measured displacement and give a more reliable prediction of landslide displacement than the SVM and BPNN,especially during the critical early warning period.The state feedback is realized in the LSTM approach,which can properly model the dynamic characteristics of landslides.The LSTM model can establish connections between landslide conditions at different time and learn rules from previous deformation time steps.Thus,a LSTM model can make full use of the historical information.This information can be filtered and the accuracy of the prediction will not be disturbed by past information from a long time ago.These characteristics of the LSTM model contribute to improving the predictions.In addition,the application of the proposed model in the displacement prediction of the Tangjiao NO.1 landslide shows better performance than the Sanzhouxi landslide.The cause is the difference in the deformation mechanism.For the Sanzhouxi landslide,the raise of reservoir water level does not synchronize with rainfall,which makes the deformation characteristics more complex.It is thus harder to select reasonable deformation influencing factors and the prediction accuracy of the proposed model can be disturbed.