Uncertainty in landslide occurrence prediction using Dempster–Shafer theory

Abstract

Landslide as one of the natural hazards has always caused huge financial losses and fatalities. Hence the goal of the present article is improvement in the prediction results of landslide occurrence in Tutkabon region in Gilan Province (Iran). For this purpose the Dempster–Shafer theory of evidence together with analyses and techniques of geospatial information systems (GIS) have been implemented for modeling and considering the uncertainties inherent in data selection. Also the parameters of slope, height, morphological conditions, earth curvature and distance to river and proximity to faults are taken as effective factors in landslide occurrence. Using the Dempster–Shafer theory, the belief, unbelief and uncertainty values for sub classes of each parameter are calculated separately and in continuation, utilizing the spatial information system, the landslide occurrence risk maps for each of these values are prepared at the study area. Finally for assessment of the results, the locations of landslide occurrence at the study area and the risk belief map are compared to each other. The results indicate that 65% of the landslides occur at the very high hazard class. Also assessing the results a value of AUC = 0.74 was obtained for the area under the prediction rate curve of the belief map.