GIS Application in Landslide Hazard Analysis – An Example from the Shihmen Reservoir Catchment Area in Northern Taiwan

This study used GIS as a tool to map storm-induced landslides from SPOT5 images. Digital elevation model (DEM) of 10m x 10m resolution was used to extract geomorphic landslide causative factors: slope gradient, slope roughness, tangential curvature, relative slope height, total slope height and wetness index etc. Digital geological map was used to extract geologic causative factors: lithology and distance to fault trace. SPOT image taken prior to a typhoon was also used to calculate an environmental factor NDVI (normalized differential vegetation index). Triggering factor was tested to be the maximum rainfall intensity. These causative factors and triggering factors were used to build a landslide susceptibility model via logistic regression. Validation result shows that this model could be used for the prediction of future landslides. GIS is a useful tool for the construction of landslide prediction model and for application in regional planning, hazard mitigation, and sediments yield estimation. Introduction Advances in Geographical Information Systems (GIS) technology and the mathematical/statistical tools for modeling and simulation, have led to the growing application of quantitative techniques in many areas of the earth sciences (Carrara & Pike, 2008). The study of landslide hazard also applied these basic tools frequently with intensively use of digital elevation models (DEMs) and SOPT images. This study firstly used GIS to map storm-induced landslide distributions from SPOT5 images taken prior and after a typhoon storm, and an event-based landslide inventory was built. For the purpose of constructing a landslide hazard model, high resolution DEM was used to extract geomorphic landslide causative factors, such as: slope gradient, slope roughness, tangential curvature, relative slope height, total slope height and wetness index etc. These factors were processed in a raster GIS – Erdas Imagine. Digital geological map was used to extract geologic causative factors: lithology and distance to fault trace. They were processed in vector GIS – MapInfo, and were transferred to raster cells in the raster GIS. SPOT image taken prior to the typhoon was also used to calculate an environmental factor NDVI (normalized differential vegetation index) in Erdas Imagine system. Hourly rainfall data from 26 rain gauge stations in and around the catchment area were used to process rainfall factors at each station, and then these point data were spatially interpolated to each raster cell in the study area. Landslide causative factors and triggering factors were used to build a landslide susceptibility model via logistic regression. A validation work was also done to check the performance of using this model for prediction. Result shows that this model could be used for the prediction of future landslide occurrences during a scenario event. It may be useful for the decision making of slope remedial measure, regional planning and hazard mitigation policy. Regional Setting The Shihmen Reservoir is an important water resources reservoir in northern Taiwan (Fig. 1). It has a catchment area of 763.4 km. Elevations in the watershed range from 135 m in the northwest to 3,524 m in the southeast, with generally rugged topography. Slopes with gradient greater than 55% covers 60.5% of the catchment area, slopes gradient of 30~55% occupies 29.3% area, only 10.2% is gentle slopes with gradient less than 30%. Rocks are composed of folded and faulted Miocene and Paleogene indurate sandstone and mudrocks (Fig. 1). Terraces on the river sides are composed of sandy gravels and they may be covered by lateritic soils on high terraces. Slopes are commonly mantled by shallow slope washes or colluvium. Nearly 90% of the study area is forested. The climate is influenced by typhoons in summer and the northeast monsoon in winter. The mean annual temperature is 20°C, with a mean monthly temperature of 27.5 °C in July and 14.2 °C in January. The annual precipitation averages 2,370 mm. Because of the visiting of typhoons, large rainfall events usually happen from May to September. From 23 to 25 August 2004, Typhoon Aere crossed the northern part of Taiwan. The passage of Typhoon Aere brought a maximum recorded rainfall of 1,578 mm and a maximum rainfall intensity of 88 mm/hr in the study area. During the Typhoon Aere, there occurred numerous landslides in the catchment area and caused the reservoir water become turbid, and the sediments even blocked the water intake and suspend water supply for 20 days. Millions of people and thousands of factories suffered this disaster. Fig. 1 Geological map of the Shihmen Reservoir catchment area (modified after CGS, 2007). Methodology and Working Procedure The methods and working procedure utilized in the present study generally follow those of Lee et al. (2008a, 2008b). The first step includes image and data collection. Then an event-based landslide inventory is established. In parallel with this, the causative factors of the landslides are processed and the triggering factors determined. These factors are then statistically tested, and the effective factors selected for susceptibility analysis. The working procedure is shown in Fig. 2. Logistic regression allows us to determine a linear function of factors for interpreting the landslide distribution from a set of training data. The methodology has been well established in many previous studies (e.g., Carrara, 1983; Atkinson & Massari, 1998; Dai et al., 2001; Ayalew & Yamagishi, 2005; Eeckhaut et al., 2006; Greco et al., 2007). The linear function is used to calculate the landslide susceptibility index (LSI) for each cell. The LSIs are then used to establish a probability of failure to LSI curve and determine the spatial probability of landslide occurrence at each cell. The probability of failure used here is the ratio of landslide cells to total cells in an LSI bin (Lee et al., 2008a, 2008b). It is called the proportion of landslide cells in Jibson et al. (2000). The spatial probability of landslides is then used for landslide hazard mapping. Fig. 2 Working procedure for event-based landslide susceptibility analysis in this study. Fig. 3 Spatial distribution of landslides induced by Typhoon Aere in the study region. Data Acquisition and Processing The basic data utilized in this study included a 5m x 5m grid DEM, SPOT5 images, 1/5,000 photo-based contour maps, 1/50,000 geologic maps, and hourly rainfall data. The DEMs were collected from Central Geological Survey (CGS), Taiwan. They were transferred to a color-shaded image and were visually checked. Defects were replaced by re-digitizing from a 1/5,000 scale photo-based contour map. Other abnormal points were corrected using a median filter. Finally the DEMs were reduced to 10m x 10m grid-cells for use. SPOT5 images taken before and after the typhoon event were received, processed and rectified by the Center for Space and Remote Sensing Research, National Central University, Taiwan. Both multi-spectral (XS) and panchromatic (PAN) images were used. A fusing technique was utilized to produce a higher resolution (2.5 m) false-color composite image to facilitate landslide recognition. 1/50,000 geological maps were collected from the CGS. Each map was overlaid with a shaded DEM and visually inspected in a GIS. Some abnormal boundaries, mostly associated with alluvial and terrace deposits, were corrected. The ERDAS IMAGINE system was used to transform the geologic vector map to a raster image of 10m x 10m grid-cells. The hourly rainfall data in and around the study region were collected from the Central Weather Bureau and the Water Resources Agency, Taiwan. These data were first plotted and visually inspected to compare individual records for consistency with neighboring gauge stations; abnormal data was deleted. Rainfall data were finally interpolated into 10m x 10m grid-cells data. All later processing and analysis for each susceptibility factor and for logistic analysis are based on the 10m x 10m grid-cells unit. Event-based Landslide Inventory Landslides triggered by Typhoon Aere were interpreted and delineated by comparing SPOT5 images taken before and after the typhoon. Landslides found in both inventories were examined very carefully for changes in tone and/or enlargement of extent. Typhoon Aere triggered 1,624 landslides, of which 663 were enlarged or reactivated old landslides. Most observed slope failures were shallow landslides on soil mantled slopes with depths less than 2 m. To develop our susceptibility model, we only considered new landslides triggered by typhoons. Fig. 3 shows the spatial distribution of landslides triggered by Typhoon Aere. A landslide area is composed of source area and deposit area. Landslide deposits were identified by comparing the GIS landslide layer with the 1/5,000 scale photo-based contour map. The slope angle or concentration of contour lines was used to differentiate deposits from sources. Only landslide source area was used in building the susceptibility model. Selection of Factors for Modeling There are more than fifty different landslide-related factors commonly used (both in Taiwan and worldwide) for LSA (Lin, 2003). In the present storm event-induced landslide study, we first selected sixteen of the most frequently used, based on data abundance and availability. These factors were further tested, including the normality of each factor, standardized differences (Davis, 2002) between the landslide group and non-landslide group for each factor, probability of failure curve, success rate curve (Fig. 4) and correlation coefficient between any two factors. A final selection of effective factors was decided based upon the evaluation and test results. In this study, 8 causative factors: lithology, slope gradient, NDVI, slope roughness, profile curvature, relative slope height, total slope height, topographic wetness index, distance to a fault, and a triggering factor maximum rainfall intensity, were selected for building the susceptibility model.