GIS-supported evaluation of landslide susceptibility in the karst mountainous area ： A case study in Wudang, Guiyang

. Landslide is the main disaster in the mountainous area. Based on landslide information content models of remote sensing, the work used the aerial and space remote sensing of UltraCamXp WA, Beijing-1 and Landsat images in Wudang, Guiyang to obtain the relative relief, slope, curvature of bedding slope, LUCC, geology and TWI. Finally, we analyzed the spatial susceptibility in the research area. Results showed that there were 42, 56 and 46 potential landslide groups in the high, higher and medium risk regions. The controlling factors of landslides in Wudang, Guiyang refer to the precipitation and precipitation intensity. The densely-populated regions also have the high risk of landslide, and the risk of landslide generally decreases from cities to rural areas. Through the space prediction research of landslide disasters, it is expected to provide valuable protection for regional security and harmonious development, and then sustainable development of Guizhou Province.


Introduction
As the main type of disasters in mountain areas, landslide can be predicted in the space and time dimensions. There are qualitative and quantitative methods for landslide space prediction. Depending on expert experiences, the qualitative and semi-qualitative methods have strong subjective severance [1][2] . In contrast, the quantitative method shows the advantages of objectivity and efficiency by exploring the quantitative factors in landslides. Probability and statistical methods are commonly used in the landslide space prediction [3][4][5] . According to comparative analysis of three methods, ANN has better application effect than logistic regression and frequency ratio methods [5][6][7][8][9] . Landslide prediction is greatly affected by the selection of landslide influencing factors, which varies depending on the region, type and researcher [9][10][11] . General landslide influencing factors consist of precipitation, slope, aspect, land type, lithology and structure. Besides, the precipitation characteristics affect the landslide spatial prediction [12][13][14] . Tongzhen Yan [15] divided the landslide prediction theory into deterministic, statistical analysis and information models, where the first two essentially belong to the white and black box models; the rest is somewhere between white and black boxes. The information content model was used to analyze the landslide space prediction in Wudang( fig.1), Guiyang with urban and rural characteristics in Karst mountainous areas. Through spatial prediction research of landslide disasters, it is expected to provide valuable protection for regional security and harmonious development.

Landslide factors
This study area is located at Northeast of Guiyang, and coordinates of latitude and longitud are 106.718°E-107.049°E, 26.570°N-26.911°N, respectively.The area of this study area is 686km 2 .The factor analysis method is used to determine the landslide factors and find the main factors. The m variables are expressed as the linear combinations of factors as follows.  (2020) a ij is the feature vector p (j=1,2,... ,m) corresponding to feature value λ (j=1,2,...m), expressed as a unit vector; j λ is the square root of the j -th feature value, indicating the importance of factor.

Landslide information model
The information content model is used to analyze the landslide space prediction in research area (Equation (3)). … … 1 2 1 2 2 P(y, , , , ) I(y, , , , ) log P(y) As the terms in statistical probability, P(y,x 1 ,x 2 ,…,x n ) and P(y) represent the information content by the landslide factors x 1 ,x 2 ,…, x n . When P(y,x 1 ,x 2 ,…, x n )>P(y), then the information content I(y, x 1 ,x 2 ,…, x n )>0, indicating the high probability of landslide occurrence. When I(y,x 1 ,x 2 ,…,x n )<0, there is low probability of landslide occurrence. Geographical cells are classified according to geology, topography, hydrology and LUCC. In general, landslide is the comprehensive effect of regional multi-factors. Considering the multi-factors, Equation (4) can be transformed into Equation (5).
After screening, 6 key landslide factors are determined, including relative height difference, slope, forward slope

Landslide factors
After vectorizing the 1:200000 geological map, the 30m×30m lattice was used for meshing according to the 1:50000 topographic map and digital elevation model (DEM, 30m). Before the information extraction and subsequent processing, other basic data were conducted with relative geometric correction based on the topographic map. After that, the DEM was used to extract relative height difference, slope, forward slope curvature and TWI. Landsat 8 OLI was used for the extraction of LUCC.

Results and analysis
Based on the spatial analysis function of ArcGIS, the single factor information content of each landslide factor was calculated by Equation (4). Only the information content calculation and classification parts were listed for the LUCC and geological factors. Then, the single factor information content calculated by Equation (4) was synthesized to obtain the comprehensive landslide information content in research area. After calculation, the landslide information content was 0-10.76 in research area. Combined with the Nature Break method and the group-occurring landslides, the landslide information content was divided into higher risk region For LUCC, the medium-coverage, low-coverage grassland and sparse forest land had the higher landslide risk and large landslide density, with the information contents of 0.932, 1.621 and 1.240, respectively. Oppositely, the forest land was the most stable region. Therefore, the vegetation was important to the stability of landslides. The region with higher TWI value had greater development potential of saturation region. It easily reached saturation and formed a weak region, leading to the high risk of landslide. In Wudang, the region of TWI>7 had the highest risk of landslide, with the landslide information content of 0.200. The region with the information content between 6.0 and 7.0 had the higher risk of landslide, with the landslide information content of 0.167. The concave and composite slopes had the high risk of landslide, with the information contents of 0.474 and 0.167, respectively. The relative height  168, respectively. The information content was only -0.806 in the region with relative height difference of less than 100 m. The landslide risk was directly related with the relative height difference. The larger height difference meant the greater landslide information and higher landslide risk. After the field verification of landslide prediction by GPS and SLR cameras, we analyzed the impact of slope landslide risk on regional human activities based on slope stability. The project took the schools with the highest population density and quantity as the analytic targets. Superimposed landslide prediction, school location and 3D terrain (2.5 times magnification) showed that Wudang Shunhai Primary School (106.626N, 26.625E The main reason is that these schools are either built on steep slopes or too close to steep slopes. Shunhai Primary School is built on an earthy steep slope in Shunhai. The roads pass through the school gate and behind the school, causing the high risk of landslides in the region. Xintian Driving School and Quanlin Driving School are located at the front edge of soil slope, with high risk of landslide. Shuijinhuadu Kindergarten is also located at the front edge of the soil slope. In contrast, the trailing edge of the Wudang Party School is a stone slope, which is easily affected by the collapse. Landslide significantly affects the Xintian First Primary School and Guizhou Normal College in Wudang. For Xintian First Primary School, the risk of landslide comes from the mountain behind the teaching building. For Guizhou Normal College, the risk is from the mountain behind Jianhu Lake. Two small landslides have occurred in this region. Although with the low overall risk of landslide in Guizhou Normal College, the region near the mountain has high risk and probability of landslides. Urban night lighting area map of Wudang is obtained by extracting the positive region from DMSP/OLS night lighting data of NASA in 2013 (Fig. 3). The lighting area is distributed in the part of city expansion in Guiyang, covering the area of nearly 200 km 2 . Nearly the half landslide risk regions in Wudang are distributed in the urban regions with high population density. With the highest landslide risk, Xintian and Gaoxin Community Service Centers have the population densities of 1,509 and 2,705 per km 2 , respectively. High vulnerability of landslides is caused by high population and economic density in urban regions.
Based on the detailed vector boundary file of administrative division, we analyzed the landslide space prediction to collect the landslide prediction data of each township in Wudang. Wherein, Xintianzhai has the highest risk of landslide, with the higher risk and higher risk regions accounting for 5.3 and 21.3% of total area, respectively. Xiaba Town has the second highest risk of landslide, with the higher risk and high risk regions accounting for 1.3 and 17.9%, respectively. Dongfeng Town has the high risk of landslide, with the high risk and high risk regions accounting for 0.3 and 6.5%, respectively. Shuitian Town has the significant risk of landslide, with the high risk and high risk regions accounting for 0.13 and 7.1%, respectively. In Xinchang and Yangchang Town, the high risk regions are less than 0.1%; the high risk areas are 2.6 and 2.3%, respectively. In Baiyi, Xinpu and Pianpo Town, the high risk regions

Discussion
Taking Wudang, Guiyang of Guizhou Province as an example, the work conducted the space prediction of landslide disasters in Karst region by remote sensing, GIS and information-based landslide prediction method. The principal component quantitative method was used to select the relative height difference, slope direction, gradient, type, TWI and LUCC. After extracting the LUCC by SVM-based computer classification and visual interpretation method, we obtained other landslide factors from the DEM based on ArcGIS. The information content of single-factor landslide was calculated by the spatial analysis function of ArcGIS. Then, the information contents of different single-factor layers were summed to obtain the comprehensive information content. Based on the group-occurring landslides and Nature Break method, the landslide information content was divided into high, higher, medium, lower and low risk regions. Wherein, the higher, high and medium risk regions had 42, 56 and 46 potential landslide groups, respectively.
Finally, the landslide space prediction was restricted by the data accuracy and landslide sample data. The former was obtained by the high-resolution remote sensing images and DEM, and the latter should strengthen the landslide monitoring to provide the detailed landslide samples. In addition, the landslide interpretation was conducted with multi-temporal remote sensing images. Due to the contradiction between people and land in the research area, the typicality of most landslides is destroyed after disasters. Therefore, few landslides can be interpreted by the remote sensing image of a phase, which increases the cost of remote sensing interpretation.