Abstract: Pointer octree, linear octree and other octree models have low efficiency either on storage or processing of massive dataset, which is a key problem in true three dimensional geo science modeling. The authors present a large virtual octree model which is based on multi linear structure and has no pointer or location code in its octants. The new model is both efficient on storage by a multilevel compression method which is based on regularly partitioned blocks, and is efficient on octant accessing by a multilevel index method which is based on fast morton segmentation and array operation. The new model is consistent both in core and out of core, and implementation details about block swap is presented finally. The large virtual octree model is effective to process high resolution and massive property dataset, and will promote the true three dimensional geoscience modeling.

Key words: true three dimensional, geo science modeling, massive data, octree, volumetric data

摘要： 高分辨率和海量数据是真三维地学属性建模所面临的关键问题，传统的指针八叉树、线形八叉树及其变种模型在这方面存在不足，存储空间大、处理效率低。作者提出一种多级线性结构、无指针/无位置码的海量虚拟八叉树模型，采用基于规则分块的多级压缩机制实现了海量属性数据的高效压缩和组织，采用基于快速Morton码分解和数组访问的多级索引方式实现了属性数据的高效查询。新模型在内存和外存中具有统一的压缩原理、索引结构和查询机制，以数据块为基本单位进行数据交换，给出了具体的实现方法。应用实例表明，新模型能够有效处理海量属性数据，对于大范围、高精度的真三维地学属性建模及其分析具有重要的应用价值。

关键词: 真三维, 地学建模, 海量数据, 八叉树, 体数据