Application of Neural Networks for Landslide Susceptibility Mapping in Turkey

Abstract

Landslides are a major natural hazard in many areas of the world, and globally they cause hundreds of billions of dollars of damage, and hundreds of thousands of deaths and injuries each year. Landslides are the second most common natural hazard in Turkey, and the Black Sea region of that country is particularly affected. Therefore, landslide susceptibility mapping is one of the important issues for urban and rural planning in Turkey. The reliability of these maps depends mostly on the amount and quality of available data used, as well as the selection of a robust methodology. Although statistical methods generally have been implemented and used for evaluating landslide susceptibility and risk in medium scale studies, they are distribution-based and cannot handle multi-source data that are commonly collected from nature. These drawbacks are responsible for the on-going investigations into slope instability. To overcome these weaknesses, the desired technique must be able to handle multi-type data and its superiority should increase as the dimensionality and/or non-linearity of the problem increases - which is when traditional regression often fails to produce accurate approximations. Although neural networks have some problems with the creation of architectures, processing time, and the negative “black box” syndrome, they still have an advantage over traditional methods in that they can deal with the problem comprehensively and are insensitive to uncertain data and measurement errors. Therefore, it is expected that the application of neural networks will bring new perspectives to the assessment of landslide susceptibility in Turkey. In this paper, the application of neural networks for landslide susceptibility mapping will be examined and their performance as a component of spatial decision support systems will be discussed.