DEVELOPMENT OF A NEURO-FUZZY SET BASED LANDSLIDE HAZARD ZONATION AND RISK ASSESSMENT APPROACH

Abstract

Landslides are one of the indicators of the geomorphological modifications taking place in this active and fragile Himalayan terrain. These processes of mass movement and landslides have been constantly modifying the landscape. Therefore, landslide hazard (LHZ) becomes important. The key task in LHZ studies is the determination of weights and ratings giving relative importance of causative factors and their categories respectively for landslide occurrence. Therefore, an objective approach for determination of weights and ratings may be appropriate for LHZ mapping. Further, most of the landslide related studies culminate at providing LHZ maps only. These maps should further be appropriately used for carrying out landslide risk assessment (LRA). The main aim of this dissertation is to explore the potential of distribution-free quantitative approaches and to develop a GUI based software to implement these approaches for landslide hazard zonation and risk assessment within remote sensing and GIS framework. The study area belongs to Chamoli district and is bounded by 30° 19' N and 30° 30' N latitudes and 79° 15' E and 79° 21' E longitudes covering an area of about 600 2 km . The datasets used to generate various thematic data layers are remote sensing data-IRS-1C LISS-III multispectral and PAN data, IRS-P6 LISS-IV multispectral data, Survey of India topographic maps at 1:50,000 scale, geological map at 1:326,000 scale and field data on existing landslides. For LHZ mapping, five different approaches ANN black box, fuzzy set based, ratings derived from ANN, combined neuro-fuzzy with subjective ratings and fuzzy ratings, and logistic regression. In the ANN black box approach, the weights and ratings remain hidden and are not known which, is its major limitation. Therefore, in this study, fuzzy and combined neuro-fuzzy approaches have also been attempted to analyze the importance of weights and ratings