Research on the Display of Ultra-Large Point Cloud Data Using a 3DWebGIS Distributed Rendering System

Abstract. This paper presents a method for ultra-high-resolution visualization of large-scale 3DWebGIS data using ChOWDER, a web-based scalable display system, combined with the open-source 3DWebGIS application iTowns. Web browsers impose a heap memory limit of about 4 GB, which restricts conventional WebGIS from handling very large 3D datasets. ChOWDER distributes 3DWebGIS rendering across multiple browser-based display clients, expanding both memory space and screen space. As a case study, we convert cloud data from the Himawari meteorological satellite into approximately 500 million 3D points, generate 3DTiles (about 6.1 GB, eight-level octree), and render them on a tiled display wall composed of fifteen 4K displays (20K horizontal resolution). Each display browser renders only its portion of the scene, enabling detailed inspection of cloud structures while preserving an overview.

During visualization, we observe triangular non-rendered regions in the point cloud. Analysis shows that these artifacts arise from the use of an Earth-centered Cartesian coordinate system (EPSG:4978), which causes 3DTiles bounding voxels to intersect the Earth’s surface, combined with WebGIS behavior that occasionally fails to display tiles at the requested zoom level. We argue that such issues are inherent when mapping global-scale 3D data onto WebGIS platforms. As a mitigation strategy, we propose presegmenting global data into multiple regions and generating 3DTiles separately so that bounding volumes do not span the Earth’s surface. Future work includes implementing and validating this region-segmented workflow, and comparing 3DTiles with alternative point-cloud formats such as Potree and COPC for performance and artifact behavior.