Landslide Susceptibility Analysis In The Middle Reach Of Bailong River Basin Based On Statistical Models And In-situ Experiment

The Bailong River basin located in southern Gansu,is one of the four areas in China most severely affected by landslide and debris flow hazards,and the seriously developed geohazards has become one of the biggest bottleneck restricting the local social-economic development,whereas the practice of landslide suscptibility assessment(LSA)in the area based on comparing and combining different statistical models bears two defects,i.e.,the lack of field observation data under rainfall conditions to verify the result of LSA and the insufficient interpreting ability on the process-based physical mechanics of landslide of LSA.In order to provide more reliable quantitative basis for land use planning and risk management,and to improve the methodological development of LSA,and further to provide theoretical basis and practical guidance for constructing the affordable,efficient and reliable regional monitoring and early warning system,this paper firstly made quantitative analysis of the landslides prior and post the "5.12" Wenchuan earthquake and their relationships with the influence factors,including topographical,geological,meteorological and hydrological,and human activity in Zhouqu-Wudu segment along the middle reach of the Bailong River;secondly,different statistical models were compared and finally combined into one regional LSA model;lastly,the susceptibility of typical landslides were studied by combining in-situ artificial rainfall test with monitoring methods.The following three main research results were obtained:(1)Weak lithology group and well-developed faults are the most important internal factors for landslide in the middle reach of Bailong River Basin,whereas earthquakes and rainfall are the significant external triggering factors.The distribution characteristics of landslides have changed insignificantly before and after the Wenchuan Earthquake,which can be seen from the similar numbers of landslides under various classes of each external factor prior and post the earthquake,except that more landslides were distributed in the higher elevations and steeper slopes due to the influence of the amplification effect of seismic wave in the mountain area.(2)To further express the quantitative relationship between landslide and its influence factors and to deepen the understanding on the mechanism of landslide formation,study based on comparing and combining different statistical models were conducted.Results show that the semi-quantitative AHP(analytic hierarchy process)and multi logistic regression have enhanced interpreting ability,whereas the artificial intelligence ANN(artificial neural network)and SVM(support vector machine)are models with better zonation and prediction capability.In order to combine the advantage of different models effectively,new strategy that integrated the expert knowledge and statistical methods on the basis of comparing different models were proposed.And further to overcome the underestimations of economic cost caused by misclassification of LSA,cost curve were introduced for establishing the optimal LSA.It is indicated that the zonation ability and prediction accuracy were apparently improved and the uncertainty was reduced to a certain degree,and thus the optimal LSA that encompasses the cost curve could provide more reliable basis for land use planning in the study area.(3)To make up the deficiency of statistical models in expressing the physical evolutionary processes of landslides under various rainfall conditions,and to fill the gap between the results of LSA and the field observational data,typical colluvial landslide were selected to implement the artifical rainfall simulating test.Results have showed that the pore water pressure,earth pressure,groundwater level and displacement within the colluvial slope exhibited rapid response to the rainfall,and this quick response was ascribed to the preferential infiltration pathway such as cracks,fissures and macropores.The forming process of colluvial landslide could be generalized as:rainfall infiltration induced deformation-accelerated movement causing shear dilatant strengthening-reconsolidation and strength decreasing.The spatial migration process of moisture could be effectively monitored by the high-density electrical method.Among the numerous monitored indexes,the earth pressure showed best correlation with the displacement,therefore it could be selected as the critical monitoring and early warning indicators for the widely distributed colluvial landslides in the Bailong River basin.In combination with the earth pressure and the accelerated velocity,threshold index system for colluvial landslide was established and could provide scientific guidance for the landslide early warning in the study area.