Large-scale 3D geospatial data visualization has become increasingly critical for the development of the digital society infrastructure in Japan. This study conducted a comprehensive performance evaluation of two major WebGL-based web mapping libraries, CesiumJS and MapLibre GL JS, using large-scale 3D point-cloud data from the VIRTUAL SHIZUOKA and PLATEAU building models. The research employs standardized 3D Tiles 1.1, and Mapbox Vector Tiles (MVT) formats, comparing performance across different data scales (2nd and 3rd grid levels) using Core Web Vitals metrics, including First Contentful Paint (FCP), Largest Contentful Paint (LCP), Speed Index, Total Blocking Time (TBT), and Cumulative Layout Shift (CLS). The results demonstrate that MVT-based building visualization with MapLibre GL JS achieves optimal performance (FCP 0.8s, TBT 0ms), whereas MapLibre GL JS combined with deck.gl shows superior performance for large-scale point cloud processing (TBT: 3ms, CesiumJS: 21,357ms). This study provides data-driven selection guidelines for appropriate technology choices according to use cases, establishing reproducible performance evaluation frameworks for 3D web mapping technologies during the WebGPU and OGC 3D Tiles 1.1 standardization era.

翻译：大规模三维地理空间数据可视化对日本数字社会基础设施建设日益重要。本研究利用VIRTUAL SHIZUOKA和PLATEAU建筑模型的大规模三维点云数据，对两种主流的基于WebGL的Web地图库CesiumJS和MapLibre GL JS进行了全面的性能评估。研究采用标准化的3D Tiles 1.1和Mapbox矢量切片（MVT）格式，通过Core Web Vitals指标（包括首次内容绘制时间、最大内容绘制时间、速度指数、总阻塞时间和累积布局偏移）比较了不同数据规模（二级和三级网格层级）下的性能表现。结果表明：基于MVT的建筑可视化在MapLibre GL JS中实现了最佳性能（首次内容绘制时间0.8秒，总阻塞时间0毫秒）；而MapLibre GL JS结合deck.gl在大规模点云处理方面表现出更优性能（总阻塞时间：3毫秒，CesiumJS：21,357毫秒）。本研究为根据使用场景选择适宜技术提供了数据驱动的选择指南，在WebGPU和OGC 3D Tiles 1.1标准化时代建立了可复现的三维Web地图技术性能评估框架。