How to set variables for Predictive analysis in FePEST

Dear All,

I have calibrated my model and now want to perform predictive analysis using FePEST but I don't see too much help in manual. I have to set PD0, PD1, PD2 on the basis of my calibrated model results but confused how to decide about these values. Moreover, what should be NOPTMAX value for this type of analysis when I have found best calibrated parameters already.

Best Regards,
Ali :(

Dear Ali,

A run under the PEST predictive mode, it means that PEST would start modifying the parameters in other to minimize/maximize certain observation, but keeping the model "calibrated" at the same time.

PD0 is the target objective function. This is the value what you would expect for a calibration model. Normally, you can take this value from a previous PEST run (i.e. standard estimation mode).
PD1 is the acceptable objective function. Since you want to keep the model "calibrated", then you would need to say what "calibrated" status means. This is simply few percentages above PD0 (e.g. 1-2%).
PD2 tells PEST when to change to the next iteration after having try several Marquardt Lambdas.

You can find all the details in the PEST Manual (section 6-14, page 192).

Regards,
Carlos

Dear Carlos,

Thanks for sharing information.

Yes, I had got information from PEST manual. After setting some suitable values of PDo, PD1 and PD2, I run my model in predictive mode for different piezometric heads at some different observation times, but I am wondering that heads and parameter values are similar for all optimized, maximum and minimum cases. Can you please let me know what mistake I am making or setting predictive model bad. I am really stuck!!

Kind regards,
Ali

Dear Ali,

There are two possible explanations:
1) The value of PD1 is not so large compared to PD0. Therefore, PEST cannot maximize/minimize the observation. You need to extend your definition of a "calibrated" model in the parameter space.
2) The hydraulic heads are not so sensitive to the calibrated parameters (i.e. the parameters are indeed changing, but the observations do not).

You may try to go further in your analysis and carry out a Monte Carlo analysis with FePEST. You need a couple of manipulations in the command line to get started in FePEST (see the post [color=blue][i][b]"Null space Monte Carlo with FePest?"[/b][/i][/color])

Regards,
Carlos