Stochastic Mode Input

The keywords shown here control the behavior of the stochastic mode including the EnKF. They are specialized input. Nevertheless, they should be written in the standard Input File.

Table of Contents

• Stochastic Mode Input
  • Table of Contents
• Ensemble Size
• Subsample (not clear)
• Parameters
  • Logarithm conversion
  • Split units
  • Explanation: Parameters
• Boundary conditions
• Postcomputation
• Global EnKF Iterations
• Output Directories

Ensemble Size

There are two ways to specify a number of ensemble members (used for example in MC or EnKF).

# samples
6

# simulate
6

The internal number of ensemble members is set as follows:

The number of samples is forced to be the number of simulation-samples plus one. Exception: For the pure forward run (runmode = 1), the numbers are equal.

TIP: Just use # simulate, then the number of samples will automatically be adjusted to be the simulate value plus one. The first n samples will be filled with realizations and the last sample with the mean value of the realizations.

TIP: For a forward run, always set # simulate to 1.

Subsample (not clear)

Sample Average Approximation

# subsample
2

Reads in

Does not appear to be used in the code.

Parameters

Preparation for reading the parameters:

# parameter group, records=2
4 parfile=sgsim1.par
2 1 lin
3 1 lin
4 1 lin
5 1 lin
1 parfile=sgsim2.par
1 4 log

The first three lines of the keyword input are now explained in a little more detail:

# parameter group, records=2

The # parameter group is the standard catch phrase of this input. records=2 states that parameters from two files will be read.

4 parfile=sgsim1.par

4 sets the number of different parameters (either different unit or different physical parameter) read in from the parfile. Under the keyword parfile SHEMAT-Suite reads the name of the SGSIM Input File.

2 1 lin

2 refers to the second unit defined in the input file. 1 refers to parameter number one (porosity, see table in Rock Matrix Chapter of the Input File documentation for a list of all 17 parameters). The addition lin specifies to use the distribution from the SGSIM Input File as probability distribution of the parameter.

1 4 log

Here, the addition log specifies to use the probability distribution from the SGSIM Input File as distribution of the logarithm of the parameter (for example if the permeability distribution is log-normal, the distribution in the SGSIM Input File would be normal).

Logarithm conversion

# standard deviation
0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 ! unit 1
0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 ! unit 2
0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 0.5d0 ! unit 3
...

17 deviations are given in different lines (one for each unit in the unit parameter field of the SHEMAT Input File). Shown in the example above: The default of the program invoked if the keyword # standard deviation is not found. The naming of this input is confusing, because the values are not used as standard deviations inside the code. The input is important if there is a logarithmic variable (such as permeability), which is transformed from a natural logarithm to a logarithm of base 10. It is recommended to use log(10) ≈ 2.30258509299d0 instead of 0.5d0 for the fields which will be randomized by SGSIM and 0.0d0 for every other field.

The following example consisting of four units randomizes the porosity field in unit one, the permeability field in unit two, leaves unit three untouched and randomizes both the porosity and permeability fields in unit four:

# standard deviation
2.30258509299d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 
0.0d0 0.0d0 0.0d0 2.30258509299d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 
0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 
2.30258509299d0  0.0d0 0.0d0 2.30258509299d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 0.0d0 

Standard deviations/variances are read in through the EnKF Input File.

Split units

There is another input-command controlling units:

# split units, records=2
1
2

This command delivers the information, which units to split up, so that each cell formerly belonging to this unit is set to be a unit itself (so a unit consisting of 100 cells will then be turned into 100 units, each consisting of one cell). This command is most important, when using SGSim to simulate different parameter values at each cell inside a certain unit.

Explanation: Parameters

In the stochastic mode, parameters are now specified by their mean and a variation. The mean is saved in the array

propunit ( maxunits , npropload, nsample )

and the variation is calculated according to

lin:  < Standard deviation > * < random number (normed Gaussian) > \
log: exp[ < Standard deviation > * < random number (normed Gaussian) > ].

The standard deviations are saved in the array

d_propunit ( maxunits , npropload ).

NOTE: In the end of simul_wrapper, the numbers in d_propunit are used as factors for the simulated values taken from SGSIM simulations, NOT AS STANDARD DEVIATIONS. They are instead used to handle conversion between natural and base-10 logarithms or to change a whole field by a common factor.

Boundary conditions

Boundary condition data and structure:

# bc group, records=1
1 parfile=sgsim1.par
1 lin temp split

In more detail:

# bc group, records=1

# bc group is the keyphrase. records=1 means that there is only one parameter file read in.

1 parfile=sgsim1.par

1 means that there is one line of information following parfile=sgsim1.par defines the name of the parameter file.

1 lin temp split

1 is the unit-index of the boundary condition lin is the specification how the value is given (other possibility: log) temp defines the physical property (other possibilites: head, conc, etc.) split is the keyword to look into ibc_data and propunit (see array.f90, where a lot of the variables concerning boundary conditions are declared)

Standard deviation for the boundary conditions

# bc standard deviation, records=1
1 0.01d0 temp

In the second line: 1 is the unit-index. 0.01d0 is the value for the standard deviation. temp is the physical property. Default: 1.0d0 for all boundary conditions, no matter, which unit of property.

Postcomputation

A postcomputation switch can be activated.

# enkf postcompute
mean

The variable enkf_post is set to a value depending on input:

Global EnKF Iterations

Number of iterations in the EnKF. This should be set to one, otherwise more than one full EnKF simulation is calculated one after the other, which is usually not what you would want.

# enkf iter
1

Variable maxiter_enkf is set to the number of global assimilation iterations given as input. Experience showed that a number less or equal three should be chosen.

Output Directories

In the end the directories

enkf_output/
05_output/

whose names are specified in mod_simul.f90, are created (if runmode (see Runmode Documentation on Input File) is greater or equal to two).