Avalanche Hazard Assessment Based On Multi-Source Data

Avalanches are one of the natural disasters caused by global warming in the cryosphere.Relevant research on it can enhance the understanding of the operation and interaction mechanism of the cryosphere and other spheres,and has important theoretical significance for research on global change.Avalanches will threaten the safety of human life and property,destroy transportation,electricity and other basic public facilities that humans rely on for survival,cause secondary disasters such as mudslides,and have a profound impact on the surrounding environment and ecosystem.Therefore,carrying out avalanche risk assessment research has important practical significance for disaster prevention and mitigation.Currently,there are few researches on avalanche risk evaluation,and there are the following problems:(1)Most of them only carry out evaluation method system studies,with few specific case studies or small test areas;(2)Incomplete evaluation factors or important factors such as snow cover status and Few meteorological elements are used,leading to insufficient initial characterization capabilities,which affects the accuracy of subsequent evaluations;(3)Evaluation models are mostly based on expert experience methods,which are not objective enough,resulting in poor model generalization ability;(4)Lack of long-term series risk distribution maps and corresponding method systems from the perspective of "prevention".Therefore,how to extract key evaluation factors and establish an objective factor representation model to construct a regional scale and long-term avalanche risk evaluation method is a key scientific issue to improve the avalanche risk evaluation accuracy and avalanche prevention capabilities.In response to this problem,this article takes the core area of the “Belt and Road”and the northern part of Xinjiang where avalanches is high as a study,and has carried out three aspects of research and The following conclusions were obtained:(1)Cloud removal algorithm for MODIS snow productsTwo cloud removal algorithms for snow products are proposed:(1)Improved Snow L cloud removal algorithm based on elevation partition and ground temperature threshold:Snow L algorithm uses the average elevation of land and snow area to divide clouds into land or snow to achieve cloud removal.However,in areas with large topographic changes,the average elevation of the snow-covered area may be "too high" and the average land elevation may be "too low",so that clouds that are actually snow are classified as land,and clouds that are actually land are classified as Snow accumulates,resulting in the loss of accuracy of snow products even though it goes to the cloud.Therefore,based on the actual situation of the study area,it is proposed to use the elevation to partition,and extract the average land and snow area elevation in the partition;use the average ground temperature in the elevation partition as the threshold to comprehensively partition the cloud.(2)Cloud removal algorithm based on single-month ground temperature extreme value filtering: using the upper limit and lower limit of the single-month ground temperature interval to further correct the misdivision of clouds.The experimental results show that the proposed cloud removal algorithm improves the accuracy by up to 20%.The obtained daily cloudless snow products provide an accurate range for subsequent snow depth inversion and avalanche risk evaluation.(2)The snow depth inversion model that takes into account different elevation zones and land cover types uses ground weather station snow depth data,microwave brightness temperature data in different frequency bands,different polarization characteristics,and multiple linear regression,random forest(RF)and The CHANG algorithm constructs a snow depth inversion model that takes into account different elevation zones and land cover types.The experimental results show that the inversion model is the best when based on the RF algorithm and taking into account the elevation zone,the root mean square error(RMSE)is 6.9cm;the model based on the RF algorithm is the second(RMSE is 8.0cm);And it is better than the algorithm based on multiple linear regression and CHANG.Combining the optimal inversion model and daily cloudless snow products,a highprecision daily snow depth distribution map is obtained.As a key factor for subsequent avalanche risk evaluation,Xueshen products can evaluate the initial characterization ability of the factor.(3)Long-term serial avalanche risk assessment method taking into account snow depth Based on historical avalanche point data,factors such as snow depth,topography,vegetation,and weather,as well as Analytic Hierarchy Process(AHP),Analytic Network Process(ANP)and binary logistic regression algorithms,we built A deep long time series avalanche risk assessment method.The experimental results show that based on AHP,ANP and binary logistic regression algorithms,78.13%,84.38%,and 87.50% of avalanche points are located in high-risk zones,respectively;their daily average prediction accuracy rates are 76.09%,82.61% and respectively.86.96%.The innovation of this paper is:(1)Two cloud removal algorithms for snow products are proposed.One is the improved Snow L cloud removal algorithm based on elevation partition and ground temperature threshold: based on the actual situation of the study area,the elevation is partitioned,and the average land and snow area elevations in the partition are extracted;the average ground temperature in each elevation partition is used as the threshold,Comprehensively divide the cloud.The other is a cloud removal algorithm based on a single-month ground temperature extreme value filter: the upper and lower limits of the single-month ground temperature interval are used to further correct the misdivision of clouds.(2)a regional scale and long-term series avalanche risk evaluation method based on daily cloud-free snow products and snow depth data is proposed,which improves the accuracy of avalanche risk evaluation and the level of avalanche prevention.