Study On Dynamic Landslide Susceptibility Mapping Based On Multi-source Remote Sensing Imagery

The geological environment contains a wide range of contents,including climate,hydrology,geology,tectonic activities and biological activities,all of them are part of the geological environment.These contents influence and interact with each other,and promote the continuous change of geological environment.With the large-scale development of the economy,the scope of human activities has gradually expanded,and the impact on the geological environment is expanding.Human activities play an important role in changing the geological environment.Human activities will not only make the geological environment more conducive to the human's life,but also bring great disasters to the human being at some time.Geological environment may cause geological disasters,when natural processes,biological and human engineering activities drastically changes.Geological disasters have caused great damage to human's life and properties.Geological disasters can be divided into artificial geological disasters,natural geological disasters and the combination of them,according to the man-made factors or natural factors.China is vast in territory with complex geological conditions and active tectonic activities.Moreover,China's economy is developing at a high speed,and human engineering activities are very strong,which has a strong impact on the geological environment.Under the comprehensive action,China suffers serious geological hazards,among which the landslide disaster is especially prominent.Because of the complex terrain and enormous water systems,the landslide disasters occur extensively in the Three Gorges reservoir area.It is a huge threat to local people's life and properties and influences the regional economic development and social stability.The risk assessment of landslide disaster under the condition of the surface coverage changes and the human engineering activities changes,is of great importance to scientific and realistic significance.This paper explores the relationship and distribution of the impact factor and the landslide disasters,based on multi-source remote sensing images,field investigation data,professional monitoring data,geological data and existing research results.Extracting land use and land use changes,major human engineering activities and their variations and exploring the rule of landslide disasters and the impact factor changes directly or indirectly,based on multi-temporal remote sensing imageries.Establishing a single landslide prediction dynamic model and a regional landslide prediction dynamic model based on data-driven machine learning methods.The dynamic and spatial risk prediction of landslide hazard in Zigui to Badong is realized finally.The main conclusions and results are as follows:The land use and their changes,the major human engineering activities and the major human engineering changes are considered as very important input factors in the prediction models.The process is extracting the land use through the medium resolution remote sensing image,and extracting the change information by comparing the results.At the same time,the high spatial resolution remote sensing image is used to interpret the major human engineering activities and their changes.And these changes are included in the evaluation system of the landslide disaster prediction models.The fusion and application of remote sensing image and other data are realized by combining with other factors such as topography,geological conditions,earthquake and rainfall.The hysteresis effect problem exists in the influence factors and the landslide displacements is solved by using the long-short term memory neural network method.By using the long-short term memory neural network,the information flow can be propagated further,and the network can obtain the influences of the rainfall and the reservoir level fluctuations in the past.The results show that this method can improve the performances when data is sufficient.Dynamic landslide hazard prediction models are established based on shallow machine learning methods and deep learning methods.It is impossible to use an algebraic expression to describe the complex relationship between the landslide influence factors and the spatial distributions of landslide disasters.Data-driven intelligent algorithms can automatically extract the complex relationship between factors and landslide disasters.In this paper,processing unit selection,hyper parameters optimization and class imbalanced problem are considered in the prediction process.The application of convolution neural network and long-term memory neural network in regional landslide risk assessment is explored.At last,the landslide risk assessments in eight phases are carried out by using the shallow dynamic models and the deep dynamic models,and the accuracies of the predictions are compared and analyzed.