Machine Learning Based Identification Of Potential Low-frequency Debris Flow Catchments In The Bailong River Basin

Debris flow is one of the major geohazards in mountainous areas.For debris flows with different frequencies,medium-high frequency debris-flow catchments are highly valued and prevention measures for them are relatively complete.However,potential low-frequency debris-flow catchments are easily be ignored.Especially with the increasing of the population in mountainous areas,flat places at the outlets of some debris flow catchments become the ideal living places.Once a rare heavy rain triggers the debris-flow,it will cause major disasters.This type of debris-flow catchment is often overlooked because of its concealment,and related research is weak.Especially under the strenuous conditions of mountain exploration,how to identify them quickly and effectively is extremely urgent.In this paper,we identified and predicted the potential low-frequency debris-flow catchments in the Bailong River basin based on Machine Learning technology through methods such as data collection,field surveys,data database construction,and model building.The main work and conclusions are as follows:(1)Based on the analysis of the debris-flow formation conditions,this paper considers that the geomorphological conditions are the "relatively stable main controlling factors",the material conditions are "relatively dynamic controlling factors",and the triggering conditions are "relatively random excitation factors".Based on the geomorphic parameters,the development stages of debris flow catchment are quantitatively divided.(2)A remote sensing automatic identification model for debris-flow fans was constructed based on spatial deep-learning model.The recognition results were that the model had a recall rate of 93.8% in the training area,in the detection area,the recall rate was 90.9%,and 20 new debris flows were found.It shows that the debris-flow fan identification model has a good overall recognition effect,especially the identification of slope debris-flow,which fills the deficiency of the existing methods.(3)A debris flow frequency prediction model was built based on the regression model,and a good simulation prediction effect was obtained.It was shown that the factor that has the greatest influence on the frequency of debris flow is the 10-minute average precipitation,followed by the vegetation cover index.In general,the most important influence on the frequency of debris flow in the study area is the excitation condition,followed by the material condition.The frequency distribution map of debris-flow in the study area predicted by the model can provide scientific guidance for debris-flow disaster reduction.(4)A low-frequency debris flow identification model was constructed based on the classifier model,which can quickly identify low-frequency debris flows in the study area.It was found that the comprehensive material conditions are the main influencing factors of the low-frequency debris flows,which are mainly developed in places with weak lithology,few landslides and high vegetation coverage.Low-frequency debris flow distribution in the study area predicted by the model can provide technical support for the investigation and prevention of hidden danger points of debris flow.In summary,the identification method and model of potential low-frequency debris-flow catchments established in this paper are of great significance for understanding the characteristics of the development conditions of low-frequency debris flows,and are effective countermeasures for the prevention of potential lowfrequency debris flows.It can quickly and effectively identify the potential lowfrequency debris flow catchments hidden in the study area,make up for the shortcomings of traditional methods,and provide new technical and theoretical support for the investigation and prevention of hidden danger points of debris flow.