Abstract:
Accurate identification of potential snow avalanche release areas or the snow avalanche high-proning areas are crucial in disaster mitigation and prevention of engineering construction in alpine mountains. This paper uses the Yining-Aksu railway in the Tianshan region of Xinjiang Uygur Autonomous Region as case study. It collectes 154 snow avalanche starting zones as assessment samples,and conductes snow avalanche susceptibility assessment based on machine learning algorithm,to construct the assessment system along the railway. We identify the potential snow avalanche release areas(PRA)based on data superposition,to map the distribution illustration of its PRA. Then we test and compare both results using Kappa coefficient and
AUC values. The results show that the Kappa coefficients by SVM,MLP and PRA are 0.806,0.774 and 0.600,respectively,while
AUC values are 0.993,0.961 and 0.802,respectively. Machine learning algorithms in snow avalanche susceptibility assessment outperform traditional PRA identification algorithm based on data superposition. Both machine learning algorithm models have high accuracy,with SVM performing the best,outperforming MLP. The assessment results are more consistent with the actual situation of snow avalanche development in the study area,and it can provide a basic scientific basis for snow avalanche prevention and mitigation in major engineering construction in alpine mountains. Albeit the automatic algorithm of PRA identification is relatively defective,the assessment results are basically consistent with the actual situation of snow avalanche development in the study area,making it meaningful for assessing in alpine mountain where available data is limited.