validation in SUFI-2

[Lingfeng Zhou]

Dear Dr. Abbaspour,

I have questions about validation in SUFI-2. In the user manual, it says that we should make one complete iteration with the calibrated parameter ranges to validate the model. I wander if SWAT-CUP could output the simulated data (every simulation in calibration) of validation period, then the 95PPU of validation period could be determined. So we just need  the measured data of validation period to calculate the p value and t value. 

If we make one complete iteration with the calibrated parameter ranges, then we may get a new best parameter set, predictably. Besides the p value and t value, we need to report statistics like R2 and NS, so  I am confused  about how to calculate the R2 and NS of validation period? is it still based on the best parameter set got from calibration?  

Thnaks

Lingfeng Zhou

[Abbaspour, Karim]

All of the validation statistics are calculated for you already. When you do validation run, execute post processing and you get everything. Goal_fn.txt contains parameters and their objective function values.

Did I misunderstand something?

 

Karim

--
You received this message because you are subscribed to the Google Groups "SWAT-CUP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to swat-cup+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Lingfeng Zhou]

Dear Dr. Abbaspour,

Sorry for my unclear statements, to sum up, what I am confused is "is it necessary to do a new iteration for validation?". 

When in the calibration process in SUFI-2, if SWAT-CUP is allowed to import the measured data of validation period, then, I think, p value, t value and other statistics of validation period can be determined. If we do a new iteration for validation (user manual suggests), then the 95PPU of calibration and validation period are based on different parameter sets (parameters ranges are the same, but Latin hypercube samplings are random)...

Thanks 

Lingfeng Zhou

在 2017年4月17日星期一 UTC+8下午2:44:17，Abbaspour写道：

[Abbaspour, Karim]

Yes, this is clearly explained in the manual.

[Lingfeng Zhou]

Dear Dr. Abbaspour,

Thank you for your time and patience, I read the operation method of validation in SUFI-2 before, but I still can not understand the principles behind.

For example, if I do validation as manual suggests, I will get a new best parameters set (In Best_Par.txt, based on validation process), which is different from the original best parameters set (based on calibration process).  So it means that, when calculating  the statistics like R2, NS, they are based on the different best parameters sets obtained from calibration and validation, respectively. I feel it (two best parameters sets) hard to accept  if considering the model as a whole.

So is it also right and effective to do validation without a new iteration,  if SWAT-CUP is allowed to import the measured data of validation period, then, p value, t value and other statistics of validation period can be determined immediately when calibration is done. What's more, p value,t value, 95PPU and other statistics are based on same parameters sets.

Thanks again

Lingfeng Zhou

在 2017年4月17日星期一 UTC+8下午5:48:40，Abbaspour写道：

[Abbaspour, Karim]

Don't worry about the parameters. Parameters are those you obtained by calibration. Only look at the 95ppu and the best validation result.

Karim