A GPGPU-accelerated implementation of groundwater flow model in unconfined aquifers for heterogeneous and anisotropic media

The application of computer simulation models plays a significant role in the understanding of water dynamics in basins. The recent and explosive growth of the processing capabilities of General-Purpose Graphics Processing Units (GPGPUs) has resulted in widespread interest in parallel computing from...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Environmental modelling & software : with environment data news 2018-03, Vol.101, p.64-72
Hauptverfasser: Carlotto, T., da Silva, R.V., Grzybowski, J.M.V.
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The application of computer simulation models plays a significant role in the understanding of water dynamics in basins. The recent and explosive growth of the processing capabilities of General-Purpose Graphics Processing Units (GPGPUs) has resulted in widespread interest in parallel computing from the modelling community. In this paper, we present a GPGPU implementation of finite-differences solution of the equations of the 2D groundwater flow in unconfined aquifers for heterogeneous and anisotropic media. We show that the GPGPU-accelerated solution implemented using CUDA1 C/C++ largely outperforms the corresponding serial solution in C/C++. The results show that the GPGPU-accelerated implementation is capable of providing up to a 56-fold speedup in the solution using an ordinary office computer equipped with an inexpensive GPU2 card. The code developed for this research is available for download and use at http://modelagemambientaluffs.blogspot.com.br/. •The model solves groundwater flow in unconfined aquifers with moving.boundary.•The model is easily tailored to be coupled to an overland flow model.•The GPGPU-accelerated implementation provides over 50-fold speedup.•Transient solutions in grids with 5 million points obtained in about 5 min.•Our high-performance implementation is open-source.
ISSN:1364-8152
1873-6726
DOI:10.1016/j.envsoft.2017.12.004