Parallel Processing for Modeling Reactive Transport in Groundwater

Abstract

Natural attenuation and biotransformation are processes that can potentially control the transport and enhance the remediation of contaminants in groundwater. It is necessary to develop computer simulations that not only model the physical transport (advection and dispersion) of contaminants, but that can also accurately depict chemical reactions and some of these more complex processes, in order to determine the type and extent of contaminant plumes and to analyze potential remediation strategies. Modeling these systems effectively is becoming possible with a growing understanding of the chemical and biological processes that occur in groundwater. However, more accurate and more involved models come with much higher memory and computational requirements. Parallel processing provides the computational resources needed to employ reactive transport simulations effectively and more efficiently.

N2D-H2 is a FORTRAN code that simulates two-dimensional reactive solute transport in groundwater. More specifically, it simulates the biotransformation of nitrate into the end products of denitrification. A parallel version of the N2D-H2 code is developed using the Message-Passing Interface (MPI), a library of sequences and routines that can be called from FORTRAN programs. Using MPI to develop the parallel version of the code involves decomposing the computational domain among processors, defining the computational roles of each processor, and implementing the required communication between processors by using the message-passing procedures that allow the processors to exchange data.

Several test problems are developed to analyze the performance of the parallel code. The test problems are used in the benchmarking procedure to demonstrate that the parallel code returns results identical to the sequential code. The CPU time required and the speedup achieved by running the simulation on parallel processors is presented for multiple test problems with varying physical processes and computational grid sizes. For a two-dimensional plume simulation of five solutes, with a finite difference grid of 490 nodes x 99 nodes, the total CPU time is decreased from 410 seconds on one processor to 220 seconds on two processors, and 75 seconds on ten processors. The speedup achieved gets closer to the ideal speedup as the problem size increases. Although the speedup observed with the parallel version of N2D-H2 is not 100% of the ideal speedup because of communication requirements, the parallel simulation demonstrates the benefits of parallel processing and the possibility of expansion that it provides for modeling reactive transport in groundwater.