Spatial datasets for benchmarking machine learning-based landslide susceptibility modelsMendeley Data

Abstract

This article presents a comprehensive dataset developed for benchmarking machine learning-based landslide susceptibility models. The dataset includes landslide polygons delineated through manual interpretation of high-resolution satellite imagery and controlling factors data extracted from topographic maps and Indonesia's national digital elevation model (DEMNAS). Landslide events were mapped by comparing pre- and post-event satellite imagery from Tropical Cyclone (TC) Cempaka, which occurred from 27 to 29 November 2017, and verified through field surveys. Pre-event landslides were mapped using Google Earth imagery, while post-event landslides were mapped using Pleiades Pan-sharpened Multispectral Natural Color Band imagery sourced from the European Space Agency (ESA) via Indonesia's National Institute of Aeronautics and Space (LAPAN). The landslide polygons identify areas with confirmed landslide activity, while the controlling factors dataset includes topographic attributes such as slope, aspect, elevation, profile curvature, plan curvature, terrain wetness index, stream power index, land use, distance to road, and distance to river. The dataset is publicly available and aims to promote transparency, reproducibility, and collaboration in landslide research. It offers significant reuse potential for researchers across diverse domains and regions, enabling comparative studies, model benchmarking, and validation efforts. This dataset provides a valuable resource for advancing machine learning applications in landslide susceptibility modeling and supporting a wide range of geospatial analyses.

Keywords

• Landslide area data
• Landslide point data
• Conditioning factors data
• Controlling factors data
• Pixel
• Raster