Landslide susceptibility assessment of SE Bartin (West Black Sea region, Turkey) by artificial neural networks


Creative Commons License

Ercanoglu M.

NATURAL HAZARDS AND EARTH SYSTEM SCIENCES, cilt.5, sa.6, ss.979-992, 2005 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 5 Sayı: 6
  • Basım Tarihi: 2005
  • Doi Numarası: 10.5194/nhess-5-979-2005
  • Dergi Adı: NATURAL HAZARDS AND EARTH SYSTEM SCIENCES
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Sayfa Sayıları: ss.979-992
  • Hacettepe Üniversitesi Adresli: Evet

Özet

Landslides are significant natural hazards in Turkey, second only to earthquakes with respect to economic losses and casualties. The West Black Sea region of Turkey is known as one of the most landslide-prone regions in the country. The work presented in this paper is aimed at evaluating landslide susceptibility in a selected area in the West Black Sea region using Artificial Neural Network (ANN) method. A total of 317 landslides were identified and mapped in the area by extensive field work and by use of air photo interpretations to build a landslide inventory map. A landslide database was then derived automatically from the landslide inventory map. To evaluate landslide susceptibility, six input parameters (slope angle, slope aspect, topographical elevation, topographical shape, wetness index, and vegetation index) were used. To obtain maps of these parameters, Digital Elevation Model (DEM) and ASTER satellite imagery of the study area were used. At the first stage, all data were normalized in [0, 1] interval, and parameter effects on landslide occurrence were expressed using Statistical Index values (Wi). Then, landslide susceptibility analyses were performed using an ANN. Finally, performance of the resulting map and the applied methodology is discussed relative to performance indicators, such as predicted areal extent of landslides and the strength of relation (r(ij)) value. Much of the areal extents of the landslides (87.2%) were classified as susceptible to landsliding, and rij value of 0.85 showed a high degree of similarity. In addition to these, at the final stage, an independent validation strategy was followed by dividing the landslide data set into two parts and 82.5% of the validation data set was found to be correctly classified as landslide susceptible areas. According to these results, it is concluded that the map produced by the ANN is reliable and methodology applied in the study produced high performance, and satisfactory results.