关键词: 景观指数, 连通区域算法, MPI, 并行计算

Abstract:

Landscape metrics have been widely used to quantitatively evaluate the spatial patterns of landscapes and to analyze the temporal dynamics of landscapes and their effects. However, when dealing with massive amounts of data, the calculation of landscape metrics requires large amount of computing time and extremely large memory size, which greatly decreases the feasibility in real-world applications. This paper presents a parallel algorithm to improve the performance of landscape metric calculation. First, the classical two-pass connected component labeling (CCL) algorithm was modified: (1) the calculations of some basic geometrical metrics of patches, such as areas and perimeters, are integrated into the second pass of the data for the calculation of landscape metrics; and (2) continuous patch IDs are generated along the second pass, to reduce the overheads for re-labeling. Then, a parallel algorithm consisting of a master process and a set of worker processes is designed and implemented using the C++ programming language and Message Passing Interface (MPI). In our parallel algorithm, the whole spatial domain is decomposed into multiple sub-domains and assigned to a set of concurrent processes. Each process uses the modified CCL algorithm to identify the patches within its assigned sub-domain and calculates the basic geometrical metrics of the patches. After gathering and merging the basic metrics from other processes, the master process calculates the final landscape metrics. The experiments using the land-use dataset of California showed that the computing time of landscape metrics was largely reduced using multiple processes. In conclusion, our parallel algorithm provides a high-performance solution for landscape metric calculation using massive and high-resolution datasets.

Key words: landscape metrics, connected component labeling algorithm, MPI, parallel computing