基于DEM的Priority-flood并行填洼算法优化

Optimization of Parallel Priority-flood Depression-filling Algorithm Based on Digital Elevation Model

针对大尺度数字高程模型无法适应单机内存,导致单机串行填洼算法无法计算的情况,对Barnes提出的并行PF填洼算法加以优化。基于spark实现了由Barnes提出的并行PF填洼算法,同时针对单张DEM未切分的情况,对该算法加以改进,设计了带光环的切分策略等一系列方法,将原算法一二阶段的同步处理变为异步,节约了原算法耗时。在进行2600亿单元(10 m数据集)的填洼实验中,该方法与原方法填洼结果一致,且比原算法缩短了37%的处理时间,提高了并行填洼的计算效率。

In view of the fact that the large-scale digital elevation model cannot adapt to the memory of a single machine,which leads to the inability of the single machine serial algorithm to calculate,this paper optimizes the parallel priority-flood(PF)algorithm proposed by Barnes.This paper implemented Barnes’s parallel PF depression filling algorithm based on spark.Furthermore,to address the situation that a single DEM is not segmented,a series of methods such as the halo segmentation strategy are designed to turn the synchronous processing of the first and second stages of the original algorithm into asynchronous processing,saving computation time.In the 260-billion-unit depression-filling experiment,the proposed method achieves results consistent with those of the original method while shortening the processing time by 21%,which improves the computing efficiency of parallel depression-filling.