[2H1-OS-28] OS-28

10:00 AM - 10:15 AM JST(1:00 AM - 1:15 AM UTC)

[2H1-OS-28-05] Geospatial Data Integration Using 3D Voxel Based on Spaital ID

〇Hironobu Kasuga¹, Shun Nakayama², Wanglin Yan¹ (1. Keio University , 2. Senshu University)

Keywords:

GeoAI,Machine Learning,3D Urban Information

In GeoAI, machine learning models are predominantly designed for 2D raster inputs, and standard 3D grid representations that directly handle the vertical dimension remain underdeveloped. This study proposes a framework that uses Spatial ID based 3D voxels as a common spatial index to integrate 2D and 3D geospatial data. For 2D datasets, we first (1) extract and structure height information, (2) harmonize heterogeneous vertical datums, and (3) assign a 3D spatial reference system. Diverse 3D datasets are then intersected with voxels discretized by Spatial ID, and a link table records the many to many correspondences between features and voxels. For each dataset, arbitrary attribute values and aggregation functions are specified to compute voxel level summaries, yielding a 3D raster representation. Application to three datasets including POI and human mobility data confirmed that data with different vertical datums can be integrated into a single voxel space and that variable resolution aggregation keyed on Spatial ID is achievable. By treating categorical attributes and continuous variables as channels on the same 3D grid, this framework can serve as a GeoAI preprocessing infrastructure that provides a common format for generating input tensors for machine learning models.