Building a community-curated, open-access platform of global landslide datasets to support reliable and scalable AI models.

Recent advances in machine learning and deep learning have significantly advanced landslide-related applications, including detection, early warning, and susceptibility mapping. Generative AI further offers new opportunities to accelerate landslide research through rapid prototyping and iteration of ML/DL workflows.

However, most existing models are trained on datasets specific to certain regions or landslide types, resulting in poor or untested generalization across different geographic and environmental settings. Open-access landslide datasets remain fragmented across individual publications, institutional repositories, and project-specific websites — researchers spend substantial time locating, retrieving, and preparing data, and progress remains constrained by the lack of high-quality, high-volume, standardized, and accessible datasets.

To address this gap, TerraMosaic aggregates global landslide inventories and related geospatial data, with detailed metadata for every dataset — inventory type, record count, spatial resolution, geographic coverage, input features, ML/DL models used, evaluation settings, and whether cross-regional generalization was tested. Users can search, filter, and download datasets through an interactive map-based interface, and contribute new data via an easy-to-use upload flow. It serves as a central hub supporting the development and benchmarking of reliable, scalable, and generalizable AI models for both fundamental research and real-world applications.