A Remote Sensing-Based Approach for Debris-Flow Susceptibility Assessment Using Artificial Neural Network in Uttarkashi, India.

Abstract

This study aims at identifying, mapping the distribution of vulnerable debris flow sites, and generating a corresponding susceptible map of the study area using Remote Sensing, Geographic Information system (GIS), and Artificial Neural Network (ANN). The above sites are identified using visualization technique in Google Earth to be later utilized in ANN-based model setup. The model is constructed, optimized, and validated using identified sites. The controlling factors of debris flow: Topographic Position Index (TPI), Normalized Vegetation Index (NDVI), Slope, Aspect, Topographic Wetness Index (TWI), Stream Power Index (SPI), Distance to Drainage and Digital Elevation Model (DEM) derived from Remote Sensing and observed datasets are utilized for generation of a best-fit model (Area Under Curve (AUC) - 0.77). The observations analyzed from generated susceptibility map include 1) area along the river with less vegetation, high slopes, barren land are more susceptible 2) consistent field validation results made this technique potentially reliable.