The Global Very-High-Resolution Remote Sensing Landslide Mapping (GVLM) dataset

GVLM is a large-scale benchmark dataset designed for the intelligent interpretation of landslide disasters in remote sensing. It covers 24 representative landslide events across 17 countries on five continents, encompassing diverse triggering mechanisms, landslide morphologies, and land-cover types. The dataset includes multi-temporal high-resolution optical imagery, multispectral data, and SAR imagery, along with fine-grained pixel-level annotations, supporting a wide range of tasks such as change detection, semantic segmentation, and landslide extraction. The total image coverage is approximately 860.8 square kilometers. Addressing challenges such as the scarcity of landslide remote sensing data, difficulties in spatiotemporal localization of landslide events, and the high cost of precise annotation, GVLM systematically tackles the long-standing issue of the limited availability of high-quality landslide samples for disaster monitoring and assessment. It provides a high-precision dataset spanning multiple regions and diverse landslide types, and is currently the largest, highest-resolution, and most event-rich benchmark dataset for landslide remote sensing monitoring.