A GIS-based back-propagation neural network model and its cross-application and validation for landslide susceptibility analyses Biswajeet Pradhan a, * , Saro Lee b,1 , Manfred F. Buchroithner a a Institute of Cartography, Faculty of Forestry, Geo and Hydro-Science, Dresden University of Technology, 01062 Dresden, Germany b Geoscience Information Center, Korean Institute of Geoscience and Mineral Resources (KIGAM), 30 Kajung-Dong, Yusung-Gu, Taejon, South Korea article info Keywords: Landslide Susceptibility Neural network model Cross-application Malaysia abstract Landslide-susceptibility mapping is one of the most critical issues in Malaysia. These landslides can be systematically assessed and mapped through a traditional mapping framework that uses geoinformation technologies (GIT). The main purpose of this paper is to investigate the possible application of an artificial neural network model and its cross-application of weights at three study areas in Malaysia, Penang Island, Cameron Highland and Selangor. Landslide locations were identified in the study areas from the interpretation of aerial photographs, field surveys and inventory reports. A landslide-related spatial database was constructed from topographic, soil, geology, and land-cover maps. For the calculation of the relative weight and importance of each factor to a particular landslide occurrence, an artificial neural network (ANN) method was applied. Landslide susceptibility was analyzed using the landslide occur- rence factors provided by the artificial neural network model. Then, the landslide-susceptibility analysis results were validated and cross-validated using the landslide locations as study areas. Different training sites were randomly selected to train the neural network, and nine sets of landslide-susceptibility maps were prepared. The paper then illustrates the verification of those maps using an ‘‘area under the curve” (AUC) method. The verification results show that the case of the weight using the same test area showed slightly higher accuracy than the weight used for the cross-applied area. Among the three studied areas, the verification results showed similar accuracy trends while using the weight for the study area itself. Cameron showed the best accuracy and Penang showed the worst accuracy. Generally, the verification results showed satisfactory agreement between the susceptibility map and the existing data on the land- slide location. Ó 2009 Elsevier Ltd. All rights reserved. 1. Introduction In Malaysia, frequent landslides often result in significant dam- age to people and property; the most recent such events occurred in 1991, 1996, 1998, 1999, 2002, 2005, 2006, 2007, 2008 and 2009. In the past, the Penang Island, Cameron Highland and Selangor areas have been faced with numerous landslide events, which re- sulted in substantial damage. Most of these landslides have been triggered by heavy rainfall. The landslides that occurred along the new Klang Valley express highways (NKVE) in the year 2003 alerted the highway authorities and other governmental organiza- tions of the seriousness of landslide management and prevention. The October 2002 landslide in Kuala Lumpur, which completely destroyed few houses and killed six members of a family, is still in the public’s memory. Landslides in Malaysia are mainly trig- gered by tropical rainfall and flash floods causing failure of the rock surface along fracture, joint and cleavage planes. The geology of the country is quite stable but continuous development and urbaniza- tion lead to deforestation and erosion of the covering soil layers thus causing serious threats to the slopes. However, there was little effort subsequently made to assess or predict these events. Through a scientific analysis of landslides, we can assess and predict landslide-susceptible areas and thus de- crease landslide damage through proper preparation. Therefore, understanding landslides and preventing them is one of the serious challenges facing Malaysia and its people. In this paper, we have applied and cross-validated an artificial neural network model (ANNs) for a landslide-susceptibility analysis in three test areas in Malaysia. This paper is organized in four major parts. The first part de- scribes the previous literature related to this research; the second part describes the methods, development and training of ANNs algorithms; the third part presents application and cross-valida- tion of ANNs for landslide-susceptibility analysis in three test areas 0198-9715/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.compenvurbsys.2009.12.004 * Corresponding author. Tel.: +49 351 463 33099; fax: +49 351 463 37028. E-mail addresses: Biswajeet.Pradhan@mailbox.tu-dresden.de, biswajeet@mail- city.com (B. Pradhan). 1 Tel.: +82 42 868 3057; fax: +82 42 861 9714. Computers, Environment and Urban Systems 34 (2010) 216–235 Contents lists available at ScienceDirect Computers, Environment and Urban Systems journal homepage: www.elsevier.com/locate/compenvurbsys