The Debris Flow Susceptibility Mapping Based On Machine Learning Arthimetic

As a typical geological disaster,debris flow is widely distributed all over the world.The development of these debris flows not only causes huge economic losses,but also endangers the lives of people.Therefore,people must take reasonable measures to deal with the occurrence of debris flow.Debris flow susceptibility mapping can give the occurrence probability of debris flow.In engineering construction and daily life,according to the debris flow susceptibility map,the government and people can avoid debris flows.At present,machine learning algorithms have made remarkable achievements in many industries.Machine learning algorithms have also been widely used in debris flow susceptibility assessment.The application of machine learning algorithms for debris flow susceptibility mapping includes the following steps: data preprocessing,model construction,partitioning the susceptibility index map and evaluation of model performance.However,most scholars only focus on building new evaluation models to optimize the results,and they lack specialized research on other aspects.As a result,there are many non-standard operations in data preprocessing,model performance evaluation and other aspects,which will affect the final results.When using machine learning algorithm for debris flow susceptibility mapping,this paper optimized the process reasonably in order to improve the performance of the final results.We optimized the data processing part,the model building part,the partition part,and the model performance evaluation part.As an example,we evaluated the debris flow susceptibility in Yongji County,Jilin Province.In particular,this paper has carried out the following aspects.(1)In data processing part,the classification of debris flow influencing factors is very important.For debris flow influencing factors,different classification methods lead to different classification results,which makes models’ input data different.Machine learning algorithms are data-driven algorithms,and differences in input data will lead to differences in final results.Therefore,combining the inflection point method and the natural discontinuity method,this paper proposed a new method to determine the optimal classification number and the optimal classification interval of continuous influencing factors.Compared with the previous empirical method,this method is a quantitative method,which reduces results’ uncertaintys.This method can well represent the clustering characteristics of debris flow influencing factors.(2)When constructing debris flow susceptibility evaluation models,it is necessary to select reasonable evaluation units and good models.Because debris flow occurs in a certain basin,the basin unit was taken as the evaluation unit.In order to optimize the evaluation results,this paper selected the convolutional neural network model as the evaluation model.Therefore,combined with watershed units and convolutional neural network,a new debris flow susceptibility assessment model was constructed in this paper.Compared with other models,this model can consider the interaction between adjacent regions,and can make more effective use of its spatial information.In order to verify the performance of the model,this paper compared its performance with the results of the multi-layer feedforward neural network.The results showed that the model constructed in this paper has good performance.(3)The machine learning model can output the debris flow susceptibility index map.By partitioning index map,the debris flow susceptibility partition map can be obtained.According to the receiver operating characteristic curve,a new method to partition the debris flow susceptibility index map was proposed.This method took into account the economic and social costs of misclassification.To verify the performance of this model,this paper made statistical analysis on the results of the susceptibility partition,the distribution of debris flows and the distribution of influencing factors,and proved the rationality of debris flow susceptibility partition map.(4)In order to simplify the flow of debris flow susceptibility mapping,a debris flow assessment software was constructed in this paper.The software was written in python language and exists as a toolbox of Arc GIS Pro.The software can realize data preprocessing,model construction,model performance evaluation and so on.