Discretization-based reconstruction of watertight building geometry from raw GIS data for urban CFD simulation

Urban wind simulations require CFD-ready building geometries, yet real-world GIS datasets often contain geometric inconsistencies that hinder automated preprocessing and mesh generation. This study presents an automated geometry generation pipeline that converts heterogeneous GIS building data into watertight LOD1 models suitable for urban CFD applications. The workflow combines topology-neutral surface sampling, footprint reconstruction, geometry simplification, and automated regularization within a single preprocessing framework. A district-scale case study of approximately 15,000 buildings in Seoul was used to evaluate geometry quality, fidelity, and computational efficiency. The generated models achieved 100% watertightness while preserving building-scale geometric characteristics with median volume fidelity and footprint IoU values of 0.994 and 0.992, respectively. Compared with manual preprocessing workflows, the proposed approach reduced processing time from several days to approximately 190 seconds. The framework provides a practical foundation for scalable generation of CFD-ready urban geometries, supporting future large-scale airflow simulations and data-driven modeling.