Landslide Detection Using Remote Sensing

Abstract

Landslides are a major natural hazard that continues to endanger lives and infrastructure globally. Although advancements in artificial intelligence (AI) have improved the automated mapping of landslides using satellite imagery, many models struggle to generalize across diverse geographic regions due to the limited availability of high-resolution and diverse datasets. To address this issue, the High-Resolution Global Landslide Detector Database (HR-GLDD) is introduced—a publicly available dataset constructed from 3-meter resolution PlanetScope satellite imagery. HR-GLDD comprises data from ten significant landslide events across Asia, South America, and Oceania, triggered by both rainfall and earthquakes. Each image is paired with carefully annotated landslide labels, making the dataset highly suitable for training and evaluating AI-based detection models. Strong performance across a range of terrains and unseen occurrences was revealed by the evaluation of five sophisticated deep-learning models on HR-GLDD, underscoring the dataset's potential to allow precise, generalizable, and scalable landslide mapping on a global scale.