一种景观指数的GPU并行算法设计

A GPU-Based Parallel Algorithm for Landscape Metrics

空间数据量的迅猛增长给传统单机模式的空间分析软件带来了巨大挑战,如景观格局分析软件FRAGSTATS已无法处理省级尺度的高分辨率土地覆盖数据。在两次遍历连通域标记算法的基础上,充分利用单机图形处理器的并行运算特性,提出一种改进的景观指数并行算法。该算法针对斑块尺度的斑块周长、斑块面积景观指数指标,实现了大规模区域景观指数的高效运算。应用该算法及串行算法,对不同分辨率下的土地利用分类栅格图像进行斑块尺度景观指数计算,结果表明,在大数据量的情况下,该算法能够大幅度提高景观指数的计算性能,相较串行算法效率提升了5倍,为海量数据的景观分析提供了更好的选择。

Massive spatial data poses increasing challenges to traditional analysis software.For example,landscape pattern analysis software FRAGSTATS has been unable to process provincial-level high-resolution land cover data.Based on Two-Pass connected component labeling algorithm,this paper provides an improved parallel algorithm with GPU programming to solve the landscape metrics computation problem about massive land use data.This parallel algorithm for massive landscape metrics calculation takes full advantage of a general computer,and focuses on patch perimeter and area calculation.It can also accelerate computation speed by multithreading and iteration times reduction to decrease computation time than traditional serial algorithms.We apply the proposed algorithm and serial algorithm to calculate landscape metrics of the land use classification raster images at different resolutions under patch scale.The experiment result shows great improvement of calculation performance of landscape metrics,and the efficiency has been improved by 5 times comparing with the serial algorithm,which proves that our proposed algorithm is a better choice for landscape analysis of massive data.