Landslide Susceptibility Assessment Method And Application In Wenchuan Earthquake Disaster Area

Influenced by Wenchuan’s complex geographical environment, varied geological structures, storms, earthquakes and other geological factors, the geological disaster of the Wenchuan earthquake disaster area has damaged the life, property and other survival conditions of local people. Therefore, carrying out landslide susceptibility studies to reduce landslide disaster caused casualties, to protect people’s lives and property, to achieve the goal of disaster prevention and mitigation, has great scientific and practical significance.By taking the construction of evaluation index system, uncertainty processing and the traditional models’ limited using conditions, which are rarely considered by current studies, into account, we studied two landslide susceptibility assessment methods and applied these methods in the Wenchuan earthquake disaster area to verify the effectiveness of them. The main works and contributions of this dissertation include the following sections:(1)We improved the method integrating C4.5 decision tree algorithm and m-branch probability smoothing technique to evaluate the landslide susceptibility. The cloud model of the method was used to achieve discretization of the uncertainty of the continuous property data; the rough set theory was used to analyze and calculate the significance of all landslid-related factors to obtain the core controlling factors; C4.5 decision tree algorithm was used to build classifier; m-branch probability smoothing technique was used to estimate probability which is the measurement of landslide susceptibility. At the same time, the statistical methods, Receiver Operating Characteristic(ROC) curves and Area Under the Curve(AUC) values were used to evaluate the performance of the model.(2)The Dempster-Shafer theory of evidence was adopted to study the uncertainty of landslide susceptibility. In this method, the mass function was defined by the likelihood ratio function. The method measured the uncertainty of landslide susceptibility evaluation through the belief function, plausibility function, uncertain function and disbelief function.(3)Geographic information system and remote sensing technology were adopted to extract the landslide-related factors. We analyzed the response relation and variation pattern between each landslide-related factors and landslide.(4)The method integrating C4.5 decision tree algorithm and m-branch probability smoothing technique was used to assess the landslide susceptibility of the Wenchuan earthquake disaster area. The rough set theory was adopted to quantitatively analyze and calculate the significance of elevation, slope, aspect and some other 10 landslide-related factors. The result shows the significance of the landslide-related factors we prepared are greater than 0 and all landslide-related factors would be regarded as core controlling factors which cannot be rejected. The performance of the model was evaluated and the AUC value reached to 81.80 %. By running statistics on how many validation sample cases fell into each susceptibility area, we found that 48.59% of the validation samples fell into high susceptibility area(10% of the total area), 27.58% fell in the medium susceptibility area(20% of the total area), 23.83% of the validation samples fell into the low susceptibility areas(70% of the total area). After using the statistical analysis method to evaluate the uncertainty of the model, the result showed that the model is robust, stable and the predict results were valid and available. Meanwhile, the model also has some uncertainty, and the uncertainty improves with increase of the predictive value.(5)The Dempster-Shafer theory of evidence was adopted to analyze the landslide susceptibility of the Wenchuan earthquake disaster area. The belief function map was used as the susceptibility map and its AUC value reached to 75.96%. By running statistics on how many validation sample cases fell into each susceptibility area, we found that 40.53% of the validation samples fell into high susceptibility areas(10% of the total area), 26.23% fell in the medium susceptibility area(20% of the total area), 33.24% of the validation samples fell into the low susceptibility area(70% of the total area).