Landslide Displacement Prediction And Susceptibility Assessment Based On 3S And Artificial Intelligence

China is a country with frequent occurrence of geological disasters.Among these geological disasters,landslide is one of the most serious disasters because of its high occurrence frequency,its expanding sphere of influence and a long duration.The landslide disasters not only seriously affect the local human life and property,but also have significant impact on the social stability and natural environment.For example,it can be seen from the National Geological Disaster Bulletin in 2015,there are total 8224 geological disasters occurring in the nationwide.Among the 8224 geological disasters,there are about 5616 landslides with the proportion of 68.3%,which shows the severity of the landslide disasters.The Three Gorges Reservoir Area(TGRA)in the Yangtze river basin is one of the key areas of the national geological disaster prevention and treatment.The geological conditions in TGRA are very complicate,and TGRA is affected by various factors such as human engineering activities,heavy rainfall and reservoir water level fluctuation.As a result,landslide disasters occurred frequently in TGRA,which is also verified by environmental geological disaster monitoring in the three decades.Especially after the impoundment of the Three Gorges Reservoir in 2003,the landslide disasters increase significantly in TGRA.Therefore,there are important theoretical and practical significances to monitor and predict the landslides in TGRA.In recent years,the ?3S? technology consisted by the geographical information system(GIS),remote sensing(RS)and globe positioning system(GPS),has been widespread concerned in the areas of landslide basic data acquisition,landslide monitoring and prediction.Meanwhile,based on the ?3S? technology,the machine learning algorithm-based artificial intelligences(AI)such as simple nearest neighbor classification algorithm,artificial neural network(ANN)and support vector machine(SVM),has been widely used in the landslide time series prediction and regional landslide susceptibility prediction.The AI methods are one of the main landslide prediction models.In addition,the AI combined with the nonlinear methods such as chaos theory,fractal theory are sometimes used to predict the landslide displacement time series,and the good prediction results are obtained.However,there are some drawbacks when some studies use the combined ?3S? technology and AI to monitor and predict the landslide.For example,there is limit effective case study about using the combined ?3S? technology and AI to monitor reservoir landslide;the landslide displacement prediction models based on AI are failed to consider the characteristics of nonlinear in the landslide displacement time series;the AI models used for the regional landslide susceptibility prediction have drawbacks of low prediction efficiency and selecting non-landslide grid cells randomly.Therefore,based on the high-resolution remote sensing images of the Tangjiao landslide in the TGRA in recent 12 years,GPS monitoring landslide displacement,rainfall and the landslide geological information in the Wanzhou district,some novel combined ?3S? and AI models have been proposed to overcome the drawbacks present above.The present models are mainly used to monitor signal and regional landslides,to predict landslide displacement and regional landslide susceptibility indexes.To date,the studies obtain the following important results and conclusions:(1)The object-oriented change detection(OOCD)used for the high-resolution images,the spatial analysis function of the GIS and the static real-time GPS monitoring network are combined to monitor the deformation and damage of the individual reservoir landslide.The OOCD approach has recently become more popular than traditional pixel-oriented methods for high-resolution image analysis.However,few studies have applied the OOCD approach to the land cover change detection(LCCD)and the damage assessments of individual landslide.The OOCD approach,simple nearest neighbor classification method and GIS are applied to the multi-temporal high-resolution images taken in 2002,2005,2010 and 2013 throughout the Tangjiao Landslide in the Three Gorges Reservoir Area.The object-oriented classification results show that the overall classification accuracies of the 2005,2010 and 2013 images were greater than 92%,with Kappa Index of Agreement values of at least 89%.The IKONOS image taken in 2002 is an exception to both of these values.Furthermore,the damage mechanisms are also discussed based on the integrated monitoring of the LCCD,and GPS.Then deformation and damage characteristics in the study area are explored based on the land cover change maps and the GPS monitoring displacement.And the trigger factors of the reactivation of the Tangjiao Landslide are also discussed.(2)The detail geological conditions in the Wanzhou district are described.Then the 3S technology is used to acquire the basic geographic data in the Wanzhou district,and the spatial analysis and field investigation are also used to analysis the geographic data.A total of 639 landslides are verified through the landslide inventories.Nine environmental factors of elevation,slope,profile curvature,plan curvature,relief amplitude,lithology,geological structure,Normalized Difference Vegetation Index(NDVI)and distance to river are determined.Then the AI models are used to assess the regional landslide susceptibility.(3)The cumulative displacement monitored by GPS on reservoir landslides in TGRA show step-like characteristics and is a probable chaotic time series under the influences of seasonal rainfall and reservoir water level fluctuation.Traditionally,univariate chaotic models that do not consider rainfall and reservoir water level changes are used to predict the displacement of landslides.And some conventional multivariate models are also commonly used to predict the displacement,where the input and output variables are selected empirically,and the evidence of chaos in the displacement is not identified.This paper proposes a multivariate chaotic Extreme Learning Machine(ELM)model to identify the evidence of chaos and predict the displacement.The displacement time series of the Baishuihe and Bazimen landslides in the Three Gorges Reservoir Area in China are used as data sets.The results show that there are evidences of chaos in the displacement time series.The univariate chaotic ELM model and the multivariate chaotic model based on Particle Swarm Optimization and Support Vector Machine(PSO-SVM)model are also applied for the purpose of comparison.The comparisons show that the multivariate chaotic ELM model achieves higher prediction accuracy than the univariate chaotic ELM model and the multivariate chaotic PSO-SVM model.(4)In recent years,the machine learning models are widely used to determine regional landslide susceptibility indexes.Among these machine learning models,the support vector machine model is commonly used,which is time-consuming and has inadequate prediction accuracy.In addition,the non-landslide grid cells are selected randomly and subjectively,which may result in unreasonable training and testing data for the machine learning models.This study proposes to use the self-organizing-map network to select the non-landslide grid cells reasonably,and use the extreme learning machine which has high prediction efficiency and accuracy to calculate the landslide susceptibility indexes.The nine environmental factors are also used as input variables.The SOM network is used to produce landslide susceptibility map in the Wanzhou district,and the non-landslide grid cells are selected from the area with very low susceptibility.Then the ELM model is used to produce final landslide susceptibility map based on the nine environmental factors,the landslide and non-landslide grid cells.The single ELM model which selects the non-landslide grid cells randomly,and the SOM network based SVM model are used as comparisons.It is concluded that the landslide susceptibility are classified well using the SOM network,and the SOM network has a higher classification accuracy than the k-means clustering method.It is also concluded that the SOM-ELM model possesses higher success and prediction rates than the single ELM and SOM-SVM models,and ELM has a much higher prediction efficiency than SVM.