Distributed memory parallel computing of three-dimensional variable-density groundwater flow and salt transport

J Verkaik,J Van Engelen,S Huizer,M.F.P Bierkens,H.X Lin,G.H.P Oude Essink

doi:10.1016/j.advwatres.2021.103976

J Verkaik, J Van Engelen + Show 4 more

Open Access

https://doi.org/10.1016/j.advwatres.2021.103976

Copy DOI

Abstract

Fresh groundwater reserves, being of vital importance for more than a billion of people living in the coastal zone, are being threatened by saltwater intrusion due to anthropogenic activities and climate change. High resolution three-dimensional (3D), variable-density (VD), groundwater flow and salt transport (FT) numerical models are increasingly being used to support water managers and decision makers in their strategic planning and measures for dealing with the problem of fresh water shortages. However, these computer models typically require long runtimes and large memory usage, making them impractical to use without parallelization. Here, we parallelize SEAWAT, and show that with our parallelization 3D-VD-FT modeling is now feasible for a wide range of hydrogeologists, since a) speedups of more than two orders of magnitude can be obtained as illustrated in this paper, and b) large 3D-VD-FT models are feasible with memory requirements far exceeding single machine memory.

Highlights

We parallelize SEAWAT, and show that with our parallelization 3D-VD-FT modeling is feasible for a wide range of hydrogeologists, since a) speedups of more than two orders of magnitude can be obtained as illustrated in this paper, and b) large 3D-VD-FT models are feasible with memory requirements far exceeding single machine memory
Saltwater intrusion caused by anthropogenic activities and climate change threatens fresh groundwater reserves that are of vital importance for more than a billion of people living in the coastal zone (Neumann et al, 2015)
Our parallelization differs from other MODFLOW and MT3DMS parallelization efforts done by other researchers in a way that a) salt transport is parallelized using a physical overlap to account for e.g. Total Variation Diminishing (TVD) advection; b) our approach fully distributes memory including parallel input/output; c) we apply the robust orthogonal recursive bisection partitioning method to address irregular model boundaries

Summary

Introduction

Saltwater intrusion caused by anthropogenic activities and climate change threatens fresh groundwater reserves that are of vital importance for more than a billion of people living in the coastal zone (Neumann et al, 2015). High-resolution, three-dimensional (3D), groundwater flow and salt transport models become more and more important instruments to support water coastal management and policy development (Harbo et al, 2011). Such models are generally very computational demanding since they often consist of many cells and many timesteps and are generally not capable of using the available computer resources efficiently. The large inertia of variable-density groundwater flow systems makes that often large simulation times are needed to accurately estimate future groundwater salinities, e.g. for paleo-hydrogeological reconstruction of the fresh-saline groundwater distribution (Delsman et al, 2014) This makes transient variable-density groundwater flow and salt transport simulations very computationally challenging

Objectives

Methods

Results

Discussion

Conclusion