Study Of Geological Disasters Hazard Assessment In Sinan County Of Guizhou Province

Sinan county in guizhou province is a key geological disasters monitoring area.There are many unfavourable factors which are conducive to the occurrence of geological disasters,for example,great regional terrain relief,fault and other adverse geological characteristics and thick quaternary accumulation layer.The main disaster type in Sinan is landslide,followed by collapse.Hence,on the basis of in-depth analysis of the occurrence mechanism of geological disasters(especially landslides)and the collapse and landslide susceptibility prediction in Sinan county,it is significant to carry out the rainfall-induced disasters hazards warning analysis.Regional geological disasters and their hazard early warning have received a lot of attention from researchers,it is a hot topic in the field of geological disasters.However,the existing literature shows that,the instability mechanism of landslides in guizhou province is not fully understood.The acquations of disasters related environmental factors are not studied in detail.The analysis of advanced machine learning models for landslide susceptibility assessment(LSA)is not carried out in depth.At the same time,The mapping of the geological disasters susceptibility to collapse within the county area of guizhou province has not been carried out effectively.Furthermore,no in-depth analysis has been carried out on the rainfall-induced disasters hazard assessment within the county areas of Guizhou province.Hence,this article sticks to this theme of “geological disasters hazard assessment in Sinan County of Guizhou Province”,comprehensively applies the theories of engineering geology,geospatial informatics and computer to systematically study the geological disasters spatial distribution rules,the disasters related environmental factors,the landslide instability mechanism,geological disasters susceptibility and hazard assessment in Sinan county.Finally,the following conclusions are obtained:(1)In this study,the topography,weather,geology,hydrological characteristics and human engineering activities in Sina County of Guizhou province are researched.Then taking the remote sensing,groundsurface surveying,engineering prospecting as technologies,the disasters including landslides and collapses in Sina County are analyzed in detail,to realize the deep understanding of the evolution features,temporal-spatial distribution and formation rules of geological disasters.(2)The geographic spatial analysis of geological disasters-related environmental factors in Sinan County is carried out using the “3S” technology.A total of 14 environmental factors are acquired,including topographical factors(elevation,slope,aspect,plan curvature,profile curvature,relief amplitude and slope length),hydrological factors(distance to river and modified normalized difference water index),land cover and land use factors(bare land index,normalized difference vegetable index and normalized difference building index)and geology factors(rock types and distance to fault).These environmental factors are used as the input variables of disasters susceptibility and hazard assessment.Meanwhile,the basic theory and model building processes of geological disasters hazard assessment are discussed in depth.(3)Some drawbacks existed in the conventional machine learning for landslide susceptibility assessment(LSA),for example,the inherent features in the input variables have not been extracted fully by the machine leanring,and self-learning abilities are limited.These drawbacks decrease the predictive ability of machine learning.To overcome these drawbacks,a novel deep learning model namely sparse connected autoencoder is proposed.The radial basis function neural networklogictical(RBFNN)and regression(LR)model are used as comparisons.Results show that sparse connected autoencoder has higher LSA accuracy than RBFNN and LR models.The drawbaccks in conventional machine learning are overcomed by deep learning models.In addition,the high and very high susceptible areas mainly distribute in areas where DEM are higher than 600 m,the slope are relatively great and the rock type is soft rock class.On the contrary,low and very low susceptible areas mainly locate in the areas with high DEM,low slopes and hard rock class.(4)In this study,Guanzai landslide in Sinan county is used as case study to study the geological disaster unstability mechanism.The results show that: in the four rainfall types of continuous five-day rainfall,the occurrence probability of rainfall that first rising and then falling is largest while the occurrence probability of rainfall that first falling and then rising is smallest;the probability of rainfall decreases along with the increase of cumulative rainfall.When cumulative rainfall is constant,the change raw of landslide seepage line is consistent with the rainfall raw of corresponding rainfall type.Landslide stability decreases gradually when the rain continues;When rainfall is constant,the landslide stability coefficient decreases the most seriously in the former four days under the effect of falling-type rainfall while the landslide stability decreases the least under the effect of rising-type rainfall.In addition,the change rate of landslide stability gradually rises under rising-type rainfall while the change rate of landslide stability decreases under falling-type rainfall,with the other two rainfall types fall somewhere in between.(5)Focuing on the problems that only a few grid cells of collapses are investigated and the collapse susceptibilities of too many grid cells are needed to be predicted in Sinan County,a particle swarm optimizedsupport vector machine(PSO-SVM)model is proposed to overcome these small sample problems to improve the accuracy of collapse susceptibility assessment(CSA).The PSO-SVM model is compared with two typical machine learning namly Logistic regression(LR)model,radial basis function neural network(RBFNN).Results show that the PSO-SVM model has higher CSA accuracy than LR and RBFNN models,and the distribution rules of collapse disasters in Sinan County are predicted very accurately using both models.The high and very high collapse susceptibility mainly distributed areas with slope greater than 31.1° and profile curvature greater than 14.937.Meanwhile,the collapse disasters are also affected by the soft rock types,unreasonable engineering activitaties and high surface wetness.On the contract,the low and very low colloapse susceptibility mainly distributed in areas with small slope and high vegetable cover rates.(6)The existed researches mainly focus on the single type of geological disaster susceptibility assessment,such as LSA or CSA.Almost no researches focus on the susceptibility assessment of two types of geological disasters at the same time.To overcome this problem,a probability statistical method is proposed to do geological disasters susceptibility assessment(GSA).This method is built based on the both LSA and CSA,then a new geological disasters susceptibility index is calculated based on the probability statistics.(7)The GSA can only be used to determine the spatial probability of disasters occurrence,not the dynamic disasters hazard assessment.To overcome this problem,The time probability of geological disasters occurrence under rainfall conditions and geological disasters susceptibility map are superimposed through spatial overlay in ARCGIS software.As a result,the rainfall-induced geological disasters can be predicted both in perspectives of space and time.First,the LSA produced by the sparse connected autoencoder and the CSA produced by the PSO-SVM model are used to calculate the geological disasters susceptibility map based on the probability statistics method.Second,the I-D threshold can be used to determine the time probability of geological disasters occurrence.Thirdly,the rainfall-induced geological disasters warning hazard levels can be obtained through overlaying the I-D threshold curves and disasters susceptibility map.Finally,the accuracy of disasters hazard warning is assessed.Results show that,geological disasters hazard warning is consistent well with the actual rainfall-induced disasters.In addition,the temporal-spatial probability of rainfall-induced disasters are reflected well by the disasters hazard warning.