High-performance watershed delineation algorithm for GPU using CUDA and OpenMP

Abstract

Watershed delineation is one of the fundamental tasks in hydrological studies. Tools for extracting watersheds from digital elevation models and flow direction rasters are commonly implemented in GIS software packages. However, the performance of available techniques and algorithms often turns out to be far from sufficient, especially when working with large datasets. While modern hardware offers high computing performance through massive parallelism, there is still a need for algorithms that can effectively use these capabilities. This paper proposes an algorithm for rapid watershed delineation directly from flow direction rasters, using the possibilities offered by modern GPU devices. Performance measurements show a significant reduction in execution time compared to other parallel solutions proposed for this task in the literature. Moreover, this implementation makes it possible to delineate multiple watersheds from the same dataset simultaneously, each having one or more outlet cells, with virtually no additional computational cost.