局部型地形因子并行计算方法研究
收稿日期: 2012-11-01
修回日期: 2012-12-01
网络出版日期: 2012-12-25
基金资助
国家"863"项目(2011AA120303);国家自然科学基金项目(40930531);"资源与环境信息系统国家重点实验室"开放基金项目(2010KF0002SA);江苏省普通高校研究生科研创新计划(CXZZ12_0391);南京师范大学研究生优秀学位论文培育计划(2011BS0007)。
Study on Parallel Calculation Method of Local Terrain Parameters
Received date: 2012-11-01
Revised date: 2012-12-01
Online published: 2012-12-25
随着分析区域的扩展及需求精度的提高,数据-计算密集型地形分析亟需通过并行化来满足用户的时间响应需求。局部型地形因子是以一定半径的分析窗口(通常为3×3)计算且具有单元计算结果独立性的地形信息,是数字地形分析的基本参数。本文在分析局部型地形因子串行算法特征的基础上,以坡度算法为样本,对局部型地形因子的并行计算方法进行了深入研究。从数据并行的角度,对并行计算环境下的数据划分粒度、方式及结果融合策略进行了分析,构建了局部型地形因子的并行计算方法。利用SRTM陆地表面地形DEM数据,设计了坡度并行计算的实验以验证其方法的正确性和实用性。实验结果表明,本文提出的并行计算方法顾及了任务、数据及计算环境,可快速对局部型地形因子串行算法进行并行化改造,提高算法的执行效率,具有较好的并行性能。
江岭, 汤国安, 刘凯, 宋效东, 阳建逸, 张刚 . 局部型地形因子并行计算方法研究[J]. 地球信息科学学报, 2012 , 14(6) : 761 -767 . DOI: 10.3724/SP.J.1047.2012.00761
As the analysis region becomes wider and accuracy requirement becomes higher, the parallel method is necessary for digital terrain analysis (DTA) which is data-intensive to meet the time response requirement of customs. Local terrain factor, the fundamental parameter of digital terrain analysis, is usually calculated based on the analysis window with a certain radius (the usual value is 3×3). Its calculation result of each pixel is independent and could reflect terrain information. After analyzing of serial algorithm features of local terrain parameter, extensive study on parallel method of local terrain factor is performed in this paper taking slope for example. From the aspect of data parallelism, the strategies of the way of data division, partition granularity model and data fusion of parallel calculation of local terrain factor are analyzed, and the parallel method has been constructed. To verity the correctness and practicality of the parallel method for local terrain factor in this paper, the parallel experiment of slope algorithm is designed by using SRTM DEM with 16 300×17 400 and it has been implemented and tested on a PC cluster system. The experiment results show that: (1) with the increase of process number, the execution time of parallel computing decreases significantly for different partition granularities. When the task number equals to processing node number, the execution time is similar for the whole DEM could be read for computation task by parallel computing system at a time. (2) The parallel speedup of slope algorithm rises gradually with the increase of partition granularity. When the granularity gets growth to a certain value, the changes of speedup and efficiency are basically identical. (3) With the increase of processing node, the execution time of slope algorithm without I/O consumption decreases gradually, meanwhile the change for different granularity is consistent. (4) The main influence factor of execution time is caused by reading and writing data. The efficiency of I/O determines the parallel efficiency to a great extent. So, the research indicates that the parallel method is efficient in completing the parallelization of sequential algorithms of local terrain factor, and the execution efficiency of algorithms could be increased greatly by using the parallel method which processes a good performance. The presentment and implementation of the parallel method can also provide a reference for the parallelization of the algorithm with the similar matrix type data.
Key words： digital terrain analysis; local type; DEM; parallel computation; terrain factor
[1] Hengl T, Reuter H. Geomorphometry concepts, software, applications[M]. Amsterdam, Boston: Elsevier, 2009.
[2] Wilson J P. Digital terrain modeling[J]. Geomorphology, 2012, 137(1): 107-121.
[3] Floriani L D, Montani C, Scopigno R. Parallelizing visibility computations on triangulated terrains[J]. International Journal of Geographical Information Systems, 1994, 8(6): 515-531.
[4] Mills K, Fox G, Heimbach R. Implementing an intervisibility analysis model on a parallel computing system[J]. Computers & Geosciences, 1992, 18(8): 1047-1054.
[5] Kidner D B, Railings P J, Ware J A. Parallel processing for terrain analysis in GIS: Visibility as a case study[J]. Geoinformatica, 1997, 1(2), 183-207.
[6] Lanthier M, Nussbaum D, Sack J R. Parallel implementation of geometric shortest path algorithms[J]. Parallel Computing, 2003, 29 (10): 1445-1479.
[7] Gong J, Xie J. Extraction of drainage networks from large terrain datasets using high throughout computing[J]. Computers & Geosciences, 2009, 35(2): 337- 346.
[8] Qin C, Zhan L. Parallelizing flow-accumulation calculations on graphics processing units: From iterative DEM preprocessing algorithm to recursive multiple-flow-direction algorithm[J]. Computers & Geosciences, 2012, 43: 7-16.
[9] Shang Y, Wu B, Li T, et al. Fault-tolerant technique in the cluster computation of the digital watershed model[J]. Tsinghua Science and Technology, 2007, 12 (s1): 162-168.
[10] Vrolijk B, Post F H. Interactive out-of-core isosurface visualization in time-varying data sets[J]. Computers & Graphics, 2006, 30(2): 265-276.
[11] 孙敏,薛勇,马蔼乃. 基于格网划分的大数据集DEM三维可视化[J]. 计算机辅助设计与图形学学报, 2002,14(6): 566-570.
[12] Barros S R, Kauranne T. On the parallelization of global spectral weather models[J]. Parallel Computation, 1994, 20(9): 1335-1356.
[13] Coumou D, Matthai S, Geiger S, et al. A parallel FE-FV scheme to solve fluid flow in complex geologic media[J]. Computers & Geosciences, 2008, 34(12): 1697-1707.
[14] Xie J, Yang C, Zhou B, et al. High-performance computing for the simulation of dust storms[J]. Computers, Environment and Urban Systems, 2010, 34(4): 278-290.
[15] Shelestov A, Kussul N, Skakun S. Grid technologies in monitoring systems based on satellite data[J]. Journal of Automation and Information Science, 2006, 38(3): 69-80.
[16] 张立立,朱江,曾志明,等. 适于大规模地形分析计算格网应用[J]. 地球信息科学,2006,8(2):14-17.
[17] 刘学军,龚健雅,周启鸣,等. 基于DEM坡度坡向算法精度的分析研究[J]. 测绘学报,2004,33(3):258-263.
[18] 周海芳,刘光明,郑明玲,等. 遥感图像自动配准的串行与并行策略研究[J]. 国防科技大学学报,2004,26(2): 56-61.
[19] Clarke K C. Geocomputation's future at the extremes: High performance computing and nanoclients[J]. Parallel Computing, 2003, 29(10):1281-1295.
/
| 〈 |
|
〉 |