Research On Data Partition Method Of Parallel Digital Terrain Analysis

The update of spatial information acquisition technologies makes DEM data present multi-scale, multi-resolution and massive characteristics, which leads to the problem that the low efficiency of digital terrain analysis algorithm based on stand-alone environment is increasingly prominent. Data parallel technology brings new opportunities for massive spatial data processing, but so far the parallel model of digital terrain analysis hasn’t been formed a unified and effective understanding. As we all know, data partitioning is the basis of data parallel. Taking into account the DEM structure and parallel characteristics of digital terrain analysis, how to build adaptive data partitioning, task scheduling and result fusion strategies is an urgent issue to be solved. This article adopts master-slave parallel structure at multi-core cluster for the research of data partitioning method of digital terrain analysis based on massive DEM. The main research results are as follows:1. The classification of digital terrain analysis based on data parallelData partitioning strategy of parallel algorithm for digital terrain analysis based on DEM is significantly affected by the type of algorithm. From the aspect of data dependent characteristic for digital terrain analysis and data communications during data parallel, this paper classifies the parallel algorithms for digital terrain analysis as local and global terrain algorithm. In the meantime, this paper establishes the target dependencies of the same classification system in digital terrain analysis algorithms.2. Data partitioning methods for parallel digital terrain analysisFirstly, for the computing process of local terrain algorithm has the relatively independent characteristic, this paper proposes the row data partitioning strategy, which uses row data redundancy strategy to eliminate data communications caused by window analysis and achieves the seamless integration of result datasets based on token control. Secondly, in order to remove the absolute dependence on visual line data for terrain visibility analysis, which uses triple data structure to achieve a low data redundancy and high efficiency storage, and uses the data buffer strategy based on boundary shift to ensure the correctness of calculation results. Thirdly, because the watershed analysis algorithm has fuzzily dependent relationship between computational grid and global grid, this paper elaborates the parallel design pattern based on double stack for basin analysis algorithm. This article, according to the examples of the parallelization of insolation-duration algorithm, viewshed algorithm and flow accumulation algorithm, analyzes the parallel efficiency of different data partitioning strategy, and acquires a good speedup to effectively improve the execution efficiency of parallel algorithm.3. Improved data partitioning method considering memory limitThis paper expounds the necessity of memory limit, puts forward the improved data partitioning method, and analyzes its impact on parallel efficiency of digital terrain analysis based on different data partitioning. Experimental result shows that the optimal data partition granularity is about60～120MB for the using multi-core cluster to get the optional parallel execution efficiency of digital terrain analysis.In conclusion, the data partitioning method proposed by this paper effectively improves the ability of digital terrain analysis about real-time processing of massive DEM data and makes it comes to a practical level, which provides technical support for the research of physiographic macro regions and virtual geographical environments.