Mapping erosion susceptibility by a multivariate statistical method: A case study from the Ayvalık region, NW Turkey

Erosion is one of themost important natural hazard phenomena in theworld, and it poses a significant threat to Turkey in terms of land degredation and desertification. To cope with this problem, we must determine which areas are erosion-prone. Many studies have been carried out and different models and methods have been used to this end. In this study, we used a logistic regression to prepare an erosion susceptibility map for the Ayvalık region in Balıkesir (NW Turkey). The following were our assessment parameters: weathering gradesof rocks, slopegradient, structural lineamentdensity, drainagedensity, land cover, streampower index (SPI) and profile curvature. These were processed by Idrisi Kilimanjaro GIS software. We used logistic regression analysis to relate predictor variables to the occurrence or non-occurrence of gully erosion sites within geographic cells, and then we used this relationship to produce a probability map for future erosion sites. The results indicate that lineament density, weathering grades of rocks and drainage density are the most important variables governing erosion susceptibility. Other variables, such as land cover and slope gradient,were revealedas secondary importantvariables.Highlyweatheredbasalt, andesite, basalticandesite and lacustrine sediments were the units most susceptible to erosion. In order to calculate the prediction accuracy of the erosion susceptibilitymapgenerated,we compared itwith themap showing the gully erosion areas. On the basis of this comparison, the area under curvature (AUC) valuewas found to be 0.81. This result suggests that the erosion susceptibility map we generated is accurate. & 2010 Elsevier Ltd. All rights reserved.