Construction Of Earthquake Landslide Susceptibility Assessment Model Based On Machine Learning

Landslides are the most common secondary resultant effects of earthquakes in mountainous areas,and the loss caused by a landslide is sometimes significantly greater than that caused by earthquakes.Beijing is not only China's political,economic,and cultural center,but also a national super city with high intensity VIII.It has a complicated but diversified geological development history and geomorphic type,with complex seismogeological environment,widely developed active faults,and high sensitivity and vulnerability to earthquake disasters.Once an earthquake occurs,the earthquaketriggered landslides in Beijing's mountainous areas will inevitably result in huge economic losses and casualties.Therefore,the analysis of the susceptibility of seismic landslide in Beijing's mountainous areas and the compilation of corresponding thematic maps will undoubtedly provide guidance for disaster prevention and mitigation in Beijing.Among all the maps,the susceptibility zoning map of seismic landslides has been considered a salient tool in effectively evaluating the probability of earthquake landslides occurring in a region.Therefore,analyzing the susceptibility of earthquake and landslide in mountainous areas of Beijing and compiling the corresponding thematic map can provide guidance for disaster prevention and reduction in Beijing.Among numerous methods for analyzing the susceptibility of earthquake and landslide,machine learning is the mainstream method at present.However,traditional machine learning requires big data as the basis,and it is usually difficult to obtain the desired results in areas where there is a lack of data.This is the case with the establishment of the seismic landslide susceptibility evaluation model in mountainous areas of Beijing.The lack of historical earthquake landslide data and the difficulty of collecting complete earthquake landslide data have become the major problems to be solved in the modelling process.Due to the data,transfer learning has become the key technology to resolve such problems.In this paper,transfer learning was set as the theoretical basis,and the pretraining model trained on the basis of the big data set collected by Wenchuan earthquake was migrated to the mountainous areas of Beijing.Finally,an evaluation model of seismic landslide sensitivity in the mountainous areas of Beijing was formed.The following work has been done in this paper:(1)Using the spatial distribution of landslides induced by Wenchuan earthquake as a reference and taking the strike of surface fracture zones as the long axis direction of ellipses,the boundary of the study area for the pre-training model was established.With the geomorphic evolution process of Beijing as a reference,the northwest of Beijing was taken as the target area of the model migration.(2)A total of 14 influencing factors,including intensity,elevation,slope,aspect,curvature,distance from fault,distance from drainage,distance from road,land use,annual precipitation,normalized difference vegetation index(NDVI),stream power index(SPI),and topographic wetness index(TWI)were selected and the basic data of seismic landslide susceptibility database in the study area was established.The data sets of earthquake landslide susceptibility in Wenchuan earthquake area and Beijing's mountainous area were established respectively.(3)Using the data set of earthquake landslide susceptibility in Wenchuan earthquake area.Firstly,the correlation between the influencing factors was clarified by combining the correlation coefficient,and then the data set was analyzed in detail through the combination of the principal component analysis and KL divergence.(4)Using BP neural network of artificial neural network to build the pre training model of earthquake landslide susceptibility assessment in Wenchuan earthquake area.(5)Combined with the example of the earthquake landslide in Jiuzhaigou's seismic area,the migration of the pre training model was analyzed in detail.The analysis showed that the evaluation model has a strong migration.(6)Combined with the pre training model,a sample expansion method based on transfer learning was proposed,and the sample number of Beijing's mountainous area was expanded.Subsequently,the pre training model was "transferred" to the mountainous area of Beijing,the evaluation model of earthquake landslide susceptibility in the mountainous area of Beijing was constructed,and a preliminary analysis of earthquake landslide susceptibility in the mountainous areas of Beijing was established...