Open AccessSUPPORT VECTOR MACHINE FOR LANDSLIDE ACTIVITY IDENTIFICATION BASED ON VEGETATION ANOMALIES INDICATOR
Author(s) -
Mohd Radhie Mohd Salleh,
Muhammad Zulkarnain Abdul Rahman,
Zamri Ismail,
Mohd Faisal Abdul Khanan,
Huey Tam Tze,
Ismaila Usman Kaoje,
Mohamad Jahidi Osman,
Mohd Asmadi
Publication year - 2022
Publication title -
journal of information system and technology management
Language(s) - English
Resource type - Journals
ISSN - 0128-1666
DOI - 10.35631/jistm.725012
Subject(s) - landslide , support vector machine , vegetation (pathology) , terrain , remote sensing , geology , identification (biology) , normalized difference vegetation index , artificial intelligence , computer science , cartography , geomorphology , geography , climate change , medicine , oceanography , botany , pathology , biology
Landslide activity identification is critical for landslide inventory mapping. A detailed landslide inventory map is highly required for various purposes such as landslide susceptibility, hazard, and risk assessments. This paper proposes a novel approach based on vegetation anomalies indicator (VAI) and applying machine learning method namely support vector machine (SVM) to identify status of natural-terrain landslides. First, high resolution airborne LiDAR data and satellite imagery were used to derive landslide-related VAIs, including tree height irregularities, canopy gap, density of different layer of vegetation, vegetation type, vegetation indices, root strength index (RSI), and distribution of water-loving trees. Then, SVM is utilized with different setting of parameter using grid search optimization. SVM Radial Basis Function (RBF) recorded the best optimal pair value with 0.062 and 0.092 misclassification rate for deep seated and shallow translational landslide, respectively. For landslide activity classification, SVM RBF recorded the best accuracy value for both deep seated and shallow translational landslides with 86.0 and 71.3, respectively. Overall, VAIs have great potential in tackling the landslide activity identification problem especially in tropical vegetated area.