The Study On The Susceptibility Statistical Model Of Earthquake-induced Landslides Based On Geospatial Data

The Wenchuan M8.0Earthquake occurred in Longmenshan Fault Zone of westernSichuan province on May12th，2008， which ranks the first with regard to itsdestructiveness，affected area，disaster relief difficulty and reconstruction in relativelydense population mountain area in mainland of China. Three provinces (i.e. Sichuan，Gansu， Shanxi) suffered the extreme malevolent earthquake. The area of seismicintensity IV or above reached440，000km2， including heavy disaster zones andextreme heavy zones (51counties with132，596km2). The earthquake inducedcountless geological hazards， collapse， landslide， debris flow， barrier lake et al.，thus aggravated the disaster losses. Based on GIS this dissertation takes Wenchuancounty as the research object， employing logic regression model to analyze landslidesusceptibility zone concerning the impact of the earthquake， and the rationality of theanalysis results is tested. The main research work and results are as follows：①Intelligent analysis methods (i.e. LR, ANN, AHP et al.) of earthquake-inducedlandslide are discussed in terms of fundamental principle, especially the LR method isstudied on its advantage and feasibility when applied to earthquake-induced landslideanalysis. In addition， the application of its key techniques and implementation steps isproposed through R software.②Impact factors spatial database of Wenchuan earthquake-induced landslide andlandslide is built on the basis of GIS. Height, slope gradient, slope direction， slopecurvature, distance from water system, distance from faults, distance from highways,formation lithology, NDVI, micro-topography, rainfall and PGA (13factors in total) arechosen as the impact factors， afterwards these factors are rasterized in special graphlayers, and further on platform of GIS they are presented as geospatial data byextraction of fishnet attribute data in the research area.③The logic regression model is set up through R software，20specimens arecalculated， regression coefficients of13factors in each grade and earthquake inducedlandslide logic regression model are obtained. By means of earthquake inducedlandslide LR model the correlation between the impact factors and earthquake inducedlandslide is analyzed， then all the geospatial data is handled in next process ofsimulation calculation. The probability of landslide is gained by using the coefficients，finally susceptibility zoning analysis of earthquake-induced landslide is achieved. ④The LR model analysis results are tested through ROC curve and mathematicalstatistical method respectively. With the AUC value0.962， the test results have shownthat susceptibility zoning analysis basically tallies the actual landslides development，different risk level also has high accuracy， which is feasible and ideal.Considering both the scientific in theory and the high accuracy in analysis ofearthquake-induced landslide， the study of this dissertation proves that LR methodcombining with GIS has great advantage in its research of earthquake-induced landslideon spatial prediction， meanwhile， the LR model has great application prospect inother aspects. The research of this dissertation about earthquake-induced landslide mayprovide reference for similar studies about earthquake-induced landslide warning.