Authors: Yasin Wahid Rabby*, University of Tennessee, Yingkui Li, University of Tennessee, Knoxville
Topics: Hazards, Risks, and Disasters, Hazards and Vulnerability, Earth Science
Keywords: Landslides, Rangamati, Susceptibility, Mahalanobis distance, logistic regression
Session Type: Paper
Start / End Time: 8:00 AM / 9:40 AM
Room: Washington 1, Marriott, Exhibition Level
Presentation File: No File Uploaded
This research presents the landslide susceptibility map for Rangmati Municipality using logistic regression with three sampling strategies: random sampling, systematic sampling, and Mahalanobis distance method. Eleven factors were considered in the landslide susceptibility analysis: elevation, slope, aspect, plan curvature, profile curvature, rainfall, land use land cover, distance from road network, distance from drainage network, normalized difference vegetation index and lithology. An inventory of 152 landslide points occurred during summer 2017 were used for model training (80% of the landslides) and validation (20% of the landslides). Landslides cover less than 1% of the pixels of the area because they were recorded as points. For the logistic regression model, we use the same ratio of landslide absence and presence pixels for training based on random sampling and systematic random sampling. We also apply the Mahalanobis distance method to determine non-landslides pixels to reduce the bias caused by the fact that recorded landslides are close to roads and settlements. The Mahalanobis distance method is usually used to determine outliers, but in this study, we use it to detect a threshold for landslide and non-landslide pixels. These landslide and non-landslide pixels will be used in the logistic regression model to produce landslide susceptibility map of the study area. Susceptibility maps produced from these three sampling methods will be validated by using success, prediction and ROC curves and the results would suggest which sampling method in logistic regression model helps to produce accurate landslide susceptibility map for the study area.