Research On The Evaluation Of Debris Flow Hazard In Lushui City Yunnan Province Based On GIS

Taking Lushui City in Yunnan Province as the study area,this thesis systematically analyses the characteristics of debris flow development and formation conditions on the basis of comprehensive and systematic collection of regional geological data and geological hazards,combined with remote sensing interpretation and field survey results.The factors and watershed units applicable to the evaluation of mudslide hazards in the study area were selected,and the evaluation models and optimal combinations of evaluation factors were selected from qualitative-quantitative hierarchical evaluation model→quantitative hierarchical evaluation model,from single statistical model→coupled statistical model,from linear→non-linear quantitative evaluation model,with the help of Arc GIS spatial analysis technology.To provide reference for the prevention and control of mudslide hazards in Lushui City,the understanding and results achieved are as follows:(1)The basic geological conditions for the occurrence of mudslides in Lushui City were identified,and the distribution pattern of mudslides in the study area was summarised and analysed:72 mudslides were developed in Lushui City,with a surface density of 2.36 mudslides/km~2,mainly of small to medium size,all of which were gully-type mudslides.The mudflows in the region are characterized by a patchy distribution along townships and a zonal distribution along rivers;the solid sources of mudflows in the region mainly come from landslides and piles,followed by piles of alluvial deposits in gully beds and discards.(2)Based on the characteristics of debris flow and the accuracy and practicality of the risk evaluation map,watershed units were selected as the basic units for evaluation.After trial calculations,the watershed unit with a flow threshold of 20,000 was found to be more suitable,and a total of 487 watershed units were divided,with an average watershed unit area of 6.15km~2.11 factors related to the formation of mudslides in the study area were initially screened out,and after grading,statistical and correlation tests between the factors,the watershed area,watershed geomorphological information entropy,watershed cut density,engineering geological rock group,distance from fracture,normalized vegetation index,annual average rainfall index,and the risk of mudslides were finally determined.A total of nine evaluation factors,including watershed area,watershed geomorphic information entropy,watershed incision density,engineering geological rock formation,distance from fracture,landslide density,normalized vegetation index,average annual rainfall and maximum daily rainfall,were identified for the mudflow evaluation modelling in the study area.(3)Qualitative-quantitative(hierarchical analysis),single statistical model(weighted informativeness),coupled statistical model(informativeness-logistic regression)and non-linear statistical model(BP neural network)methods were used to construct the mudflow hazard evaluation models for the study area.The results show that the hierarchical analysis method relies mainly on the researcher’s empirical judgement based on the conditions of the disaster-forming bodies in the study area,which is highly subjective and has unsatisfactory evaluation results.The evaluation effect is not ideal;the coupled evaluation model is overall more accurate than the single evaluation model;the non-linear evaluation model,BP neural network,has somewhat higher evaluation accuracy than the linear coupled model and is the method considered more appropriate by the study.(4)The effects of the 9-factor,8-factor,7-factor and 6-factor combinations on the modelling of the BP neural network model were discussed according to the weight size of each factor in the BP neural network.The results showed that the area under the ROC curve AUC values for the accuracy of the neural network model built by the 7-factor were higher than those of the models of other methods,and the prediction model was more reliable.This indicates that when using BP neural network modelling for the study area and surrounding areas,the optimal 7-factor combination should be preferred for modelling to obtain a more desirable evaluation accuracy,with the 7-factors ranked in order of importance as rock group,flow city cut density,geomorphological information entropy,normalised vegetation index,distance from fault,area,and mean annual rainfall.(5)Lushui City mudslides are classified into 5 levels:very low risk,low risk,medium risk,high risk and very high risk,including 108 watershed units in the very low risk area,with an area of 794.22 km~2,accounting for 26.50%;116 watershed units in the low risk area,with an area of 728.05 km~2,accounting for 40.84%;94 watershed units in the medium risk area,with an area of 554.77 km~2,or 31.12%;77 high-risk watershed units,covering 420.29 km~2,or 23.58%;and 92 low-risk watershed units,covering 499.85 km~2,or 28.04%.