Vegetation-based Approach for Rotational Landslide Activity Classification: Case Study of Kundasang, Sabah

Abstract

Abstract Background Mapping and monitoring the state of activity of landslides is crucial for effective landslide management and risk assessment. This study presents a novel approach using vegetation anomalies indicator (VAI) derived from high-resolution remotely sensed data for landslide state of activity mapping. The study focuses on the Kundasang area in Sabah, Malaysia, known for its tectonic activity. High-resolution remotely sensed data were utilized to assist in the manual inventory process of landslide activities and to generate VAIs as input for modeling. Results The landslide inventory process identified active, dormant, and relict landslides. The resulting inventory map was divided into training (70%) and validation (30%) datasets for modeling purposes. Seven main VAIs, including canopy gap, mature woody vegetation, primary forest, Root Strength Index (RSI), and water-loving tree, were produced and used as the input for the classification process using Support Vector Machine (SVM) and Artificial Neural Network (ANN) methods. The result showed that SVM outperforms ANN for both deep-seated and shallow rotational landslides, with an overall accuracy of 68.6% and 80.7%, respectively. Furthermore, an evaluation of SVM revealed that the radial basis function (RBF) kernel yielded the highest accuracies, whereas ANN performed best with a hyperbolic tangent (tanh) activation function. Conclusion The accurate classification of landslide state of activity using VAI provides several benefits, including the ability to map and classify landslide activity in forested areas, characterize vegetation characteristics specific to each activity state, and enable continuous monitoring in areas where field monitoring is impractical. This research opens new possibilities for more accurate landslide activity mapping and monitoring, thereby improving disaster risk reduction and management in tectonically active regions.