Application of GIS-based fuzzy logic and rock engineering system (RES) approaches for landslide susceptibility mapping in Selelkula area of the Lower Jema River Gorge, Central Ethiopia

Abstract

Landslide susceptibility mapping is an important tool for disaster management and development activities such as planning of transportation infrastructure, settlement and agriculture. Selelkula area of the Jema River Gorge in Central Ethiopian Highland was chosen as a study area. Fuzzy logic (FL) and rock engineering system (RES) in a GIS environment were employed to prepare landslide susceptibility maps for this area. Data sets including slope, aspect, profile curvature, plan curvature, lithology, land use, distance from river and distance from lineament were generated using the field data, remote sensing and GIS. In FL, the fuzzy membership values were calculated by normalizing the frequency ratio values of each factor’s class in a range between 0 and 1. Then, the fuzzy membership values were combined by fuzzy AND, fuzzy OR, fuzzy gamma, fuzzy sum and fuzzy product operators. A total of twelve landslide susceptibility maps were produced using FL and the best map was chosen based on area under the curve (AUC) of the receiver operating characteristic (ROC) curve and the highest difference between minimum and maximum susceptibility index values. Based on these criteria, the landslide susceptibility map produced using fuzzy gamma (γ = 0.8) operator was selected. RES is a semi-quantitative method which assigns each factor’s classes a value between 0 and 4 based on the principles of Hudson (Rock engineering systems, theory and practice. Ellis Horwood, Chichester, 1992) and uses an interaction matrix through a coding system of five values in previous researches but a coding system of nine values has been used in this study to provide a wider range of values for interactions among each pair of landslide factors and each landslide factor with a landslide. Unlike other methods, RES has a freedom to evaluate and assign each interaction based on expert’s knowledge and experience. But it also bears the problem of subjectivity in assigning the interaction values. On the other hand, FL is completely data driven and does not show any subjectivity in assigning values for each factor class. The prediction accuracies of FL and RES, which can be determined from ROC curves, were found to be 87.2 and 88.6 %, respectively. For validation purpose, the existing landslides were overlaid over the two landslide susceptibility maps and the percentage of landslides in each susceptibility class was calculated. The percentages of landslides that fall in the high and very high susceptibility classes are slightly higher in RES than FL method and hence the former is more accurate in predicting the future landslide occurrence than the later.