Import Data from ParSWMS
Three-dimensional applications often require a large number of finite elements to discretize realistically large transport domains. Even with the fast personal computers currently available, it is virtually impossible to solve within a reasonable computational time problems having more than about half a million nodes or more. To decrease the required computational time, Hardelauf et al. (2007) parallelized SWMS_3D to develop PARSWMS that distributes problems with a large number of elements over multiple processors working in parallel. The PARSWMS code was developed for the LINUX or UNIX workstations using the installed free-wares MPI, PETSc and PARMETIS. Hardelauf et al. (2007) demonstrated that doubling the number of processors may decrease the computational time by up to nearly 50%.
The SWMS_3D model (Šimůnek et al., 1995) was a direct extension of the SWMS_2D code (version 1.2; Šimůnek et al., 1995; direct predecessor of HYDRUS-2D) to three-dimensional flow and transport problems. The model uses the finite element method with tetrahedral linear finite elements to solve the Richards equation for water flow and the advection-dispersion equation with linear sorption for solute transport in three-dimensional transport domains. As such ParSWMS does not consider all processes (e.g., nonlinear and nonequilibrium solute transport, hysteresis of soil hydraulic properties, heat transport) that are considered by HYDRUS (2D/3D) and represents its simplified version.
Since ParSWMS uses input and output files that have different formatting than files for HYDRUS, two new commands were implemented into the HYDRUS (2D/3D) GUI. The first command, i.e., “Export Data to ParSWMS” (called using the menu command “File->Export-> Export Data to ParSWMS”), creates input files (Selector.in and Grid.in) required by ParSWMS from a HYDRUS project, while checking which processes are supported by ParSWMS. The second command, i.e., “Import Data from ParSWMS” (called using the menu command “File->Import-> Import Data from ParSWMS”), reads output files created by ParSWMS and converts them into the HYDRUS (2D/3D) output files.
Newly created input files for ParSWMS (during the “Export Data to ParSWMS” process) are written into the ParSWMS subfolder located in the HYDRUS project working directory (that should be made permanent), from which they can be copied to a required location on a parallelized computer. ParSWMS output files (should be copied into this subfolder (ParSWMS) after calculations are carried out so that the “Import Data from ParSWMS” command can be carried out.
Since the ParSWMS program runs on multiple processors (j – number of processors), each processor creates its own output files for each variable (i.e., pressure head h, water content th, velocities v_x, and concentrations conc) and for nodes that are solved on a particular processor. These output files are called, for example, h.0000j.out (where j=0, …, number of processors). The format is as follows (nodal number n, z, y, a z coordinates, and a particular variable):
Time *** 0 ***
n x(n) y(n) z(n) h(n)
4006 8.00000e+01 2.00000e+01 2.50000e+01 2.38500e+00
4040 9.00000e+01 2.50000e+01 2.50000e+01 2.30800e+00
4074 1.00000e+02 3.00000e+01 2.50000e+01 2.23100e+00
4174 1.10000e+02 4.50000e+01 2.50000e+01 2.15400e+00
. . . . . . . . .
12128 1.20000e+02 5.00000e+01 0.00000e+00 2.70800e+01
12129 1.30000e+02 5.00000e+01 0.00000e+00 2.70000e+01
Time *** 2.5000e-01 ***
n x(n) y(n) z(n) h(n)
4006 8.00000e+01 2.00000e+01 2.50000e+01 2.37142e+00
4040 9.00000e+01 2.50000e+01 2.50000e+01 2.28621e+00
Similarly, files for other variables are called th.0000j.out, conc.0000j.out, vx.0000j.out, vy.0000j.out, a vz.0000j.out.
The process of conversion is relatively complex to insure that it is applicable also for large problems with millions of nodes (nNodes) and hundreds (thousands) of print time layers (nPrint) without breaching memory limits. The output files for a particular variable and a particular processor (e.g., h.0000j.out) are first divided into temporary smaller files, each containing information for one particular print time layer (iPrint). These smaller files for all processor and a single print time layer are then copied (merged) into the main file (e.g., h.out) one time layer at a time. These temporary files are then deleted. Doing this, the program does not need to read multiple (nNodes) potentially large output files simultaneously.
ParSWMS also creates multiple output files with information for observation nodes (ObsNod.0000j.out), depending on whether a particular observation node is located in the transport domain solved by a particular processor. Information from these files is also merged into a single output file ObsNode.out.
Other output files with additional output information (i.e., CumQ.out – cumulative water fluxes across boundaries, v_mean.out - actual water fluxes across boundaries, hMean.out – average pressure heads on boundaries, Solute.out – solute fluxes, and Run_inf.out – run time information) are also converted to have a compatible format with HYDRUS and can be thus read and displayed by its GUI.
After the file conversion (using the “Import Data from ParSWMS” command), ParSWMS output can be viewed using the HYDRUS (2D/3D) GUI, since its output files were fully converted into the HYDRUS output file format.
Return to File Menu.