**3. Susceptibility analysis of earthquake induced Landslides**

Landslide susceptibility analysis (LSA) is necessary and important for land use planning and disaster mitigation. Many researchers have made great effort to identify the relationships of landslide characteristics such as distribution pattern, type, area coverage and volume with the triggering factors such as the magnitude, intensity and peak ground acceleration (PGA) of the earthquake, coseismic fault rupture (e.g. Lee et al., 2008; Rodriguez et al., 1999; Miles and Keefer, 2009; Keefer, 1984, 2000, 2002; Papadopoulos and Plessa, 2000). Some researchers have studied the relationships of landslide distribution with geoenvironmental factors such as lithology, morphology, presence of secondary active or inactive faults (e.g. Chigira and Yagi, 2006; Jibson et al., 2000; Khazai and Sitar, 2003; Keefer, 2000; Yagi et al., 2009).

The Wenchuan earthquake induced landslides has been carried out by several researchers. For example, Huang et al. (2011a) studied the characteristics and failure mechanism of Daguangbao Landslide, the largest scale landslide, and suggested a classification system. Tang et al. (2011b) studied the effect of the quake on the landslides induced by the subsequent strong rainfall after earthquake by a case study in the Beichuan area. Qi et al. (2010) built a spatial database of landslides by using the remote sensing (RS) results which cover 11 counties seriously damaged by the earthquake. Yin et al (2009a, b) analyzed the landslide distribution, the mechanisms of some typical landslides, and evaluated the potential hazards of the landslide dams. Gorum et al. (2011) presented the preliminary results of an extensive study of the mapping the distribution of landslides by using a large set of optical high resolution satellite images. Yin et al. (2010) presented a quantitative result of the number and area of the landslides from Anxian to Beichuan. Dai et al. (2011) mapped over 56,000 landslides using aerial photographs and satellite images and characterized the spatial distribution of landslides by correlating landslide-point density and landslide-area density with the physical parameters that control the seismic stability of slopes.

In this chapter, we show some results from landslide susceptibility analysis carried out in Qingchuan County. Our analysis was based on slope units rather than the traditional grid units. At first, the relationship of landslide distribution with an individual causative factor is analyzed. And then, landslide susceptibility is analyzed by using artificial neural network (ANN) method. Finally, a landslide susceptibility map is made based on the ANN results.

#### **3.1 Study area and data source**

Qingchuan County is located at the north-western part of the earthquake zone as shown in Fig. 7. The landslides in the area of 3,271km2 are studied.

#### **3.2 Slope unit**

Up to now, most of such studies were carried out based on the grid units. There is a problem in grid-based study that a grid may contain different slopes and a large slope may contain several grids with different slope grades. Despite the problem, the grid units were still used just because the slope units are difficult to be indentified for a wide range in the past.

Nowadays, it becomes possible and easy to indentify slope units by using GIS-based hydrologic analysis tool (David, 2002), which is based on the watershed divide and drainage lines. The slope unit size should be determined when the tool is used. We suggest that the appropriate slope unit size should match the average size of the landslide bodies in the study area.

Earthquake Induced a Chain Disasters 395

1. More than 90% of the slopes have the slope gradient larger than 20o. The landslides

5. The number of the landslides in the slopes in N direction is as twice as the slopes in the

6. The number of the landslides in the slopes with the distances to the fault less than

7. The number of the landslides in the slopes with the distances to a stream less than 5km

The landslide susceptibility analysis is carried out by using artificial neural network (ANN)

ANN program is a "computational mechanism able to acquire, represent, and compute a mapping from one multivariate space of information to another, given a set of data representing that mapping, which is independent of statistical distribution of the data, can resolve the nonlinear problem and get high prediction accuracy for classification problem especially for large amount samples (Garrett, 1994). The applications of ANN to landslide susceptibility evaluation have been made by many researchers (e.g. Ermini,L., et al., 2005; S. Lee et al.,2006; Pradhan.B et al.,2010). Nefesilioglu et al. (2008) showed that ANN could give a more optimistic evaluation of landslide susceptibility than logistic regression analysis. Ermini et al. (2005) compared two neural architectures: probabilistic neural network and

In this study, the neural network tool SPSS clementine is used since very few parameters are required. One group of the total slopes are randomly selected for training. 611 collapsed slopes of 885 landslides (70%) and 3300 of 55014 un-collapsed slopes (6%) are randomly

2. The landslides occurred majorly in the area with the elevations less than 1,200m. 3. The landslides occurred majorly in the slopes with slope ranges from 200 to 400m. 4. There is no clear relationship between landslides and specific catchment area.

Fig. 8. Slope unit division of Qingchuan area

other directions.

selected for this group.

From the statistical analysis, the following results can be found.

0.5km is as twice as the slopes in other categories.

8. There is no clear relationship between landslides and lithology.

multi-layered perceptor, and obtained a better prediction result.

**3.4 Landslide susceptibility analysis using artificial neural network** 

is as 3 times as the slopes in other categories.

based on the above statistical analysis results.

occurred majorly in the slopes with gradients between 20o to 35o.

A total of 55,899 slope units were indentified in Qingchuan County (Fig. 8). They will be used for landslide susceptible analysis in this study.

Fig. 7. Location of the study area

The basic data include a 1:100,000 geological map and a 10m grid digital elevation model (DEM) made from the available topographic map with 5m contour line interval. 885 landslides were identified from field investigations and RS results.
