Not all models can be calibrated with PEST because PEST might take too long. The time required to calibrate a model with PEST depends on several factors.
•The number of layers in the model.
•The time required for the model to run one time.
•Time required by PEST to prepare new input files and read the output.
•The number of parameters being used.
•The number of parameters using pilot points.
•The number of data sets using parameters.
•The number of candidate pilot points.
•Where and how the parameters are applied.
If you don't understand parameters or pilot points read the next two sections Using Parameters with Data Sets and Using PEST Parameters with Model Features and then come back here.
PEST will run the model many times for each parameter estimation iteration. PEST will treat each pilot point for each base parameter that uses pilot point on each layer as a separate parameter. It will run the model at least once for each parameter in each parameter estimation iteration. As an example, lets estimate the time for one parameter estimation iteration for the following case.
Note: The RunModel.bat used by PEST to run the model will display its elapsed time each time it is run. The time it displays includes both the model run time as well as the time required to generate new input files, run the model, and to read the output.
•Number of layers = 6
•Time required for the model to run once = 3 minutes
•Time required by PEST to prepare new input files and read the output: negligible.
•Number of parameters = 3
•Number of parameters using pilot points = 2
•Parameters are applied to one data set (Kx).
•Number of candidate pilot points = 60
•Where and how the parameters are applied.
oOne parameter (K1_2) with pilot points is applied in layers 1 and 2.
oThe other parameter (K3_5) with pilot points is applied in layers 3, 4 and 5.
oNo parameters are applied to layer 6.
oThe parameter without pilot points (RchRate) is used with recharge.
The number of pilot points for parameter K1_2 is 60 for layer 1 plus an additional 60 for layer 2 for a total of 120. Each of those pilot points will be treated as a separate parameter by PEST.
The number of pilot points for parameter K3_5 is the number of candidate pilot points times the number of layers to which K3_5 is applied for a total of 180.
RchRate isn't used with pilot points, so it is treated as a single parameter.
The total number of parameters that PEST will modify is thus 120 + 180 + 1 = 301.
PEST will run the model at least once for each of the 301 parameters plus a small number of additional runs. Towards the end of the parameter estimation process, it may run it more times. The number of times it runs the model for each parameter depends on FORCEN which is a property of the parameter groups.
With 301 parameters and a 3-minute run time per model, one parameter estimation iteration will take approximately 900 minutes or 15 hours. However, the run time of your original model may differ from the run time of the model with modified parameters. The run time might be either shorter or longer in the modified model. If there are 10 parameter estimation iterations, PEST might run for 6 days or more. You might or might not consider that a reasonable amount of time.
There are some things you can do to reduce the time PEST requires.
1.Use SVD Assist (see Using SVD Assist).
2.Use a parallel version of PEST such as Parallel PEST or BEOPEST. (See chapter 11 of the PEST documentation.) There is an extra input file, the run management file, required to use Parallel PEST. ModelMuse doesn't create it for you, but it isn't a complicated file. BEOPEST does not require a run management file.
