09-11, 13:30–14:00 (America/Chicago), Grand A
For any hydrologic studies, it is a crucial step to delineate watersheds to identify study areas. This talk presents the new Memory-Efficient Watershed Delineation (MESHED) OpenMP parallel GIS algorithm for delineating a large number of watersheds.
The Memory-Efficient Watershed Delineation (MESHED) is a new OpenMP GIS algorithm for delineating a large number of watersheds in parallel. It uses different computing techniques to save memory shared among multiple threads and can solve 50% larger problems than its benchmark algorithm can. It first uses a union data structure to store both input and output data in the same matrix and utilizes statistics about incoming flow cells to reduce computational resources for cell discovery. It then parallelizes delineation of a large number of watersheds in an embarrassingly parallel manner. In this talk, I will present typical challenges in this type of large-scale watershed delineation problems and some details about the new techniques in the algorithm. Finally, I will share benchmark results to highlight the memory efficiency and performance of the new algorithm.
