Distributed Parallel D8 Up-Slope Area Calculation in Digital Elevation Models

TL;DR

This paper introduces a parallel algorithm for D8 up-slope area calculation in digital elevation models that minimizes inter-node communication, enabling processing of large DEMs efficiently on various hardware.

Contribution

The proposed algorithm achieves strict bounds on inter-node communication, allowing scalable processing of large DEMs using OpenMP and MPI models with minimal data exchange.

Findings

Supports large DEM processing on desktops and supercomputers

Uses only two communication steps per node

Compatible with OpenMP and MPI parallel models

Abstract

This paper presents a parallel algorithm for calculating the eight-directional (D8) up-slope contributing area in digital elevation models (DEMs). In contrast with previous algorithms, which have potentially unbounded inter-node communications, the algorithm presented here realizes strict bounds on the number of inter-node communications. Those bounds in turn allow D8 attributes to be processed for arbitrarily large DEMs on hardware ranging from average desktops to supercomputers. The algorithm can use the OpenMP and MPI parallel computing models, either in combination or separately. It partitions the DEM between slave nodes, calculates an internal up-slope area by replacing information from other slaves with variables representing unknown quantities, passes the results on to a master node which combines all the slaves' data, and passes information back to each slave, which then computes its final result. In this way each slave's DEM partition is treated as a simple unit in the DEM as a whole and only two communications take place per node.