Landslide susceptibility mapping using geological data, a DEM from ASTER images and an Artificial Neural Network (ANN)
Introduction
The occurrence of landslides is the result of the interaction of complex and diverse environmental factors. These factors are divided into the trigger and the primary cause. Landslide occurrence triggers include weathering, earthquakes, rainfall and snow melting. Human activity like construction of roads and buildings on steep slopes and dispersal of water from supply systems and sewers could also trigger the occurrence of the phenomena (Cubito et al., 2005). Important primary causes include geomorphic and geologic features, rock outcropping, rock types and vegetative cover (Zêzere et al., 1999a, Fernandes et al., 2004, Cubito et al., 2005, Moreiras, 2005). Primary causes of landslides could include a wide range of factors like flow accumulation and distance to roads (Dahal et al., 2008), topographic-wetness and stream-power indices (Gokceoglu et al., 2005) and land use, presence of old landslides and human activity (Zêzere et al., 1999b). Studies on the dynamics and interactions of the different factors affecting landslide occurrence are very important for successful landslide risk assessment. Several studies have been conducted to determine the relationship between various environmental factors and landslide occurrences (Anbalagan, 1992, Lee and Min, 2001, Iwahashi et al., 2003, Ayalew and Yamagishi, 2005).
Landslides are one of the most destructive geological hazards affecting Japan every year. Major landslides are normally triggered by strong earthquakes, like the ones that devastated Niigata Prefecture in Honshu Island in 2004. Several studies have been conducted on landslides after the 2004 earthquake. Some of them concentrated on the contribution of the geologic and geomorphic factors to landslides (Chigira and Yagi, 2006, Yagi et al., 2007). Yagi et al. (2007) found a strong relationship between landslide occurrence and geologic and geomorphic factors (slope and aspect) in landslide concentrated areas. However, analyzing the results of these studies for predicting future occurrence of landslides using conventional statistical analytical tools is very important. Indeed, landslide occurrence prediction requires a quantitative methodology to model these complex phenomena from past events using data gathered in the field or in the laboratory (Melchiorre et al., 2006). However, the complicated non-linear relationships between landslide occurrence and its contributing factors require the use of a complex modeling method for more accurate prediction.
Artificial Neural Network (ANN) has recently been an analytical tool for a wide range of applications in the fields of natural sciences. These applications include speech recognition (Bengio, 1993), human face recognition (Soulie et al., 1993), satellite image classification (Civco, 1993, Atkinson and Tatnall, 1997, Bandibas and Kohyama, 2001) and shape and texture recognition (Khotanzad and Lu, 1991). One of the advantages of using an ANN for qualitative modeling of natural phenomena is that it can handle data at any measurement scale ranging from nominal, ordinal to linear and ratio, and any form of data distribution (Wang et al., 1995). In addition, it can easily handle qualitative variables making it widely used in integrated analysis of spatial data from multiple sources for prediction and classification. A number of authors have described the basic principles and applications of ANNs for pattern recognition (Rumelhart et al., 1986, Alexander and Morton, 1990, Sethi and Jain, 1991, Guyon and Wang, 1993, Nigrin, 1993, Haykin, 1994). ANNs are data-driven models and universal non-linear function approximators. Consequently, ANNs have been important modeling tools for landslide susceptibility zonation (Lee et al., 2003, Lu and Rosenbaum, 2003, Ermini et al., 2005, Gomez and Kavzoglu, 2005). ANNs' ability to learn non-linear functions from the data is an important feature in the problem of classifying landslide-prone areas (Melchiorre et al., 2006).
This study focuses on the use of an ANN to quantitatively model the relationship between landslide occurrence and geologic/geomorphic factors and to accurately and efficiently generate landslide susceptibility maps. Data obtained from the landslide areas in Niigata Prefecture, Japan, were used in this study. This research also aims to determine the feasibility of developing an efficient method of generating landslide susceptibility mapping system by using the data gathering system of Japan's Advanced Spaceborne Thermal Emission Radiometer (ASTER) satellite (Fig. 1). This research uses ASTER images to generate important inputs such as geomorphic information for landslide modeling.
Section snippets
Error Back-Propagation ANN computing
Neural computing is the study of networks of adaptable nodes which, through a process of learning from task examples, store experiential knowledge and make it available for use (Alexander and Morton, 1990). The aim of ANN computing is to build a new model of the data generating process so that it can generalize and predict outputs from inputs (Atkinson and Tatnall, 1997). One of the most important neural computing methodologies is the Error Back-Propagation Neural Network (EBPNN) computing. In
Study area
The study area is located in the mountainous region of Honshu Island, Japan (Fig. 3). The site covers a 680 km2 area in Chuetsu, Niigata Prefecture. The area has a mean elevation of 220 m above sea level and a maximum elevation of 765 m on the mountain range that traverses the central part of the site from NNE to SSW. The Uono River runs across the mountain range. The mean and average slopes are 11° and 64°, respectively. The area normally receives around 2000 mm precipitation/year. Human
Results and discussion
The initial phase of the study involves the analysis of the relationship between geomorphological and geological parameters and landslide occurrence using some measured frequencies of landslide occurrence. The analysis is important to have a general idea about the relationship between the individual landslide conditioning factors used and landslide occurrence. Fig. 8 shows the relationships of altitude, slope angle and slope aspect with landslide occurrence in terms of frequency distribution,
Conclusion
This study has successfully generated a trained ANN to produce a landslide susceptibility index map. The six geomorphic and geologic factors are sufficient to quantitatively model the relationship between landslide occurrence and the factors. The results of the study also indicate that an efficient landslide mapping system could be developed using ASTER images as the main source of geomorphic data. However, the accuracy of the trained ANN to predict the non-occurrence of landslides is
Acknowledgements
The authors wish to express their sincere gratitude to Dr. Minoru Urai of the Institute of Geology and Geoinformation, Geological Survey of Japan (GSJ), AIST for providing the ASTER images used in this study. They are also most grateful to Dr. Shinsuke Kodama of the Grid Computing Center, AIST, for his valuable contribution in the revision of the manuscript. The authors also wish to express their gratitude to Dr. Junko Iwahashi of the Geographical Survey Institute of Japan (GSI), for providing
References (38)
Landslide hazard evaluation and zonation mapping in mountainous terrain
Engineering Geology
(1992)- et al.
The application of GIS-based logistic regression for landslide susceptibility mapping in the Kakuda-Yahiko Mountains, Central Japan
Geomorphology
(2005) - et al.
Geological and geomorphological characteristics of landslides triggered by the 2004 Mid Niigata prefecture earthquake in Japan
Engineering Geology
(2006) - et al.
Landslide hazard in the Nebrodi Mountains (Northeastern Sicily)
Geomorphology
(2005) - et al.
Predictive modeling of rainfall-induced landslide hazard in the Lesser Himalaya of Nepal based on weights-of-evidence
Geomorphology
(2008) - et al.
Artificial neural network to landslide assessment
Geomorphology
(2005) - et al.
Topographic controls of landslides in Rio de Janeiro: filed evidence and modeling
Catena
(2004) - et al.
The 17 March 2005 Kuzulu landslide (Sivas, Turkey) and landslide susceptibility map of its near vicinity
Engineering Geology
(2005) - et al.
Assessment of shallow landslide susceptibility using artificial neural networks in Jabonosa River Basin, Venezuela
Engineering Geology
(2005) - et al.
Mapping from ASTER stereo image data: DEM validation and accuracy assessment
ISPRS Journal of Photogrammetry & Remote Sensing
(2003)