Regional Rainfall-induced Landslide Hazard Assessment Method Based On Data-driven And Forming Mechanism

Due to the complex topographic and geological conditions,frequent extreme weather and high-intensity earthquakes,dramatic influence of engineering activities on the geological environment,China is one of the countries with the most serious geological disasters and the largest population at risk in the world.There are a large number of landslides in the southeast region of China with the features of high frequency,wide distribution and group-occurring,which pose a major threat to the protection of residents’ lives and property.Regional landslide susceptibility,stability and hazard assessment can accurately predict the spatial location,occurrence time and damages of potential landslides,which has important practical significance for the prevention and control of landslide disasters and can effectively reduce the casualties and economic losses.Nowadays,there are still many difficult issues to be solved in the research of regional landslide susceptibility,stability and hazard assessment.For example,it is difficult to guarantee the efficiency and precision of slope unit division.The spatial variability of conditioning factors within the mapping unit,the uncertainty of the prediction model and sample selection,the timevariance property of landslide susceptibility have not been taken into account in the landslide susceptibility modeling.The stability of regional slopes under rainfall conditions is analyzed based on one-dimensional grid infinite slope models,geotechnical models and hydrological models,which simplify the forming mechanism and morphological features of landslides,and fail to fully consider the response law of landslides under rainfall conditions.In the landslide hazard assessment system,the forming mechanism and mechanical characteristics of rainfallinduced landslides cannot be fully considered,and it is difficult to extend the failure mechanical mechanism from slope scale to regional scale.As a result,all these issues seriously restrict the effective application of intelligent prediction results.To solve the above issues,the rainfall-induced landslide in the southeast region of China is selected.Some studies have been carried out based on a series of “intelligent” and “digital”technologies,such as the multi-scale segmentation algorithm,data-driven models,threedimensional geological model and numerical simulation method.A method has been proposed to divide slope units.The influence of the heterogeneity of conditioning factors within slope units,machine learning model selection uncertainty and non-landslide sample selection uncertainty on landslide susceptibility modeling has been discussed.A two-dimensional slope stability assessment system at the regional scale has been established based on the landslide forming mechanism.The landslide hazard assessment system based on the data-driven and forming mechanism under typhoon rainfall conditions has been effectively constructed by integrating regional landslide susceptibility prediction and individual landslide stability analysis.The results show that:(1)The slope unit can be efficiently and automatically divided at the regional scale by the multi-scale segmentation method proposed in this study,which has the advantages of a more uniform slope unit shape,higher efficiency and accuracy,and stronger applicability.The relevant parameter combination can be quickly and accurately determined using the modified trial-and-error method.(2)A landslide susceptibility prediction framework considering the heterogeneity of conditioning factors within slope units has been proposed based on the data-driven models.The heterogeneity of conditioning factors within slope units can be scientifically represented by the mean,range and standard deviation variables of conditioning factors,which fully reflect the variation features of conditioning factors.The prediction performance of landslide susceptibility modeling can be greatly improved when considering the heterogeneity of conditioning factors within slope units.(3)The effects of uncertainty from the data-driven models and non-landslide sample selection on landslide susceptibility prediction results have been discussed.The effects of machine learning model uncertainty can be effectively reduced by voting and average methods.In addition,the effects of the non-landslide sample selection uncertainty can be reduced by randomly selecting non-landslide samples from non-landslide areas multiple times to conduct susceptibility modeling.The landslide susceptibility index can be expressed as a normal probability distribution rather than a value.The influence of uncertainty from non-landslide sample selection on landslide susceptibility prediction results can be effectively reduced by statistical analysis and the maximum probability method to obtain more accurate results.(4)A landslide susceptibility updating system considering the time-varying features of landslide susceptibility has been proposed.The normalized spatial distance index can be used to scientifically represent the spatial correlation between landslide and slope units.The landslide susceptibility index is confirmed to be a time-varying variable that can be updated in real time using the multi-phase landslide data.(5)The slope stability assessment system considering the landslide forming mechanism at the regional scale has been proposed based on the slope unit,a three-dimensional geological model and the finite element method.The thickness of soil at the regional scale has been predicted using field survey data and machine learning models.Meanwhile,the threedimensional geological model has been constructed.Based on the forming mechanism of rainfall-induced landslides at the slope scale,a two-dimensional stability analysis of slope units at the regional scale has been carried out under the conditions of typhoon heavy rainfall.The seepage characteristics and the stability variation of regional slopes have been revealed.The landslide stability evaluation collaborating regional and individual scale,and the mechanical mechanism analysis of slopes at the regional scale are realized.(6)The regional landslide hazard assessment system under typhoon rainfall conditions has been established based on the data-driven and forming mechanism of landslides,which integrates the regional landslide susceptibility results and individual slope stability assessment results.In this system,the spatial occurrence probability at the regional scale and the forming mechanism at the individual scale can be simultaneously taken into account.The issue that the forming mechanism of landslides cannot be considered at the regional scale can be solved to a certain extent.Moreover,taking the “Lichima” typhoon as an example,the regional landslide hazard maps of Huangtan town under the natural condition(0 mm),heavy rainfall condition in2019.8.5(91.5 mm /24 h),an extremely heavy rainfall condition in 2019.8.9(274 mm /24 h)and extremely heavy rainfall condition in 2019.8.10(306 mm /24 h)are obtained.