Ensemble approach to develop landslide susceptibility map in landslide dominated Sikkim Himalayan region, India
- PDF / 7,850,256 Bytes
- 28 Pages / 595.276 x 790.866 pts Page_size
- 31 Downloads / 258 Views
ORIGINAL ARTICLE
Ensemble approach to develop landslide susceptibility map in landslide dominated Sikkim Himalayan region, India Indrajit Chowdhuri1 · Subodh Chandra Pal1 · Alireza Arabameri2 · Phuong Thao Thi Ngo3 · Rabin Chakrabortty1 · Sadhan Malik1 · Biswajit Das1 · Paramita Roy1 Received: 24 March 2020 / Accepted: 23 September 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract The landslide is a downward movement of soil and rock, and one of the most destructive geo-hazards that causes losses in lives, environment, and economy all over the world. Landslide susceptibility mapping is a scientific method to evaluate the landslide probability zones and causative factors. The main objective of the present study was to introduce ensemble landslide susceptibility models which are developed on the basis of two statistical models (evidential belief function and geographically weighted regression) and one machine learning model (random forest) for spatial prediction of landslide of the Upper Rangit River Basin, Sikkim, India. Totally, 102 landslide locations have been identified and randomly classified into 70% and 30% as training and validating database, respectively. Total 16 landslide causative factors are considered and grouped into four categories: geomorphological, hydrological, geological, and environmental factors. The evidential belief function (EBF), geographically weighted regression (GWR), and random forest (RF) method and their ensemble methods, RF-EBF, and RF-GWR models have been applied with the help of training landslide and non-landslide dataset and spatial database of landslide causative factors. Five landslide susceptibility maps have been generated by the said model, and the maps have been validated by validating dataset with the help of sensitivity, specificity, accuracy, Kappa index, and area under curve (AUC) of receiver operating characteristic (ROC) tools. The ensemble methods have the best degree-of-fit and prediction performance than single methods, i.e., RF-EBF and RF-GWR model have 91.8% and 89.9% prediction capabilities. The result of the relative importance of factor showed that land use land cover (LULC), distance to river, soil, drainage density, and road density factors have played the key role in the occurrence of the landslide. The result of the study can be used by local planning, dicession makers, and the methods of landslide susceptibility can be applied in other areas. Keywords Landslide susceptibility mapping · Evidential belief function · Random forest · Geographically weighted regression · Ensemble methods
Introduction Landslide is a destructive geo-hazard which takes serious detrimental disaster impact on the manmade and natural environment (Bui et al. 2012). According to Mandal and Maiti (2015), Landslide is defined as falling, sliding, and * Subodh Chandra Pal [email protected] 1
Department of Geography, The University of Burdwan, Burdwan, West Bengal 713104, India
2
Department of Geomorphology, Tarbiat Modares University, 14117‑1311
Data Loading...
