sensitivity analysis

[vivian]

hi there!

every article talks about a ranking parameter for sensitivity analysis, but where can I find the P-values and ranking values in the output if I performed a sensitivity analysis?

[Soham Adla]

On the project explorer side-pane. In "Sensitivity Analysis -> One-at-a-time" if you carried out parameterization one at a time, or in "Sensitivity Analysis -> Global Sensitivity" if you carried out calibration using multiple parameters. You can look at them both in a graph view or grid view (if you need the numbers), and they are additionally present in the "SUFI.OUT" folder of your swatcup project.

Goodluck!

[chantal]

If I want to perform a sensitivity analysis- one-at-a-time I run the model for each parameter 3 times(n+2) but I can only see the graph and I cannot find the numbers or should I based the ranking on calibration outputs, summary_stat?

I am looking forward to your reply!

[Abbaspour, Karim]

What numbers? The ones given in the text view?

--
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.

[chantal]

the P and t values to make a ranking for the one-at-a-time sensitivity analysis. I can only see the graph..

[Abbaspour, Karim]

There is also a text view with numbers

Karim

[chantal]

Where can I find them? in sufi.out?

[Abbaspour, Karim]

I thought you are talking about the global sensitivity when all parameters are changing. For one-at-a-time you only have one parameter so you cannot use the one-at-a-time option in swatcup.

What you can do is to let ONE parameter change 100 times, run post processing, then run the global sensitivity. Then you need to do this for the other parameters, one at a time.

Hope this is clear.

Karim  

 

 

 

 

 

From: swat...@googlegroups.com [mailto:swat...@googlegroups.com] On Behalf Of vivien Nooij
Sent: Wednesday, June 15, 2016 3:56 PM
To: swat...@googlegroups.com
Subject: Re: sensitivity analysis

 

where can i find it?

 

--
You received this message because you are subscribed to a topic in the Google Groups "SWAT-CUP" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/swat-cup/SaHpdnXXZF0/unsubscribe.
To unsubscribe from this group and all its topics, send an email to swat-cup+u...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[chantal]

Dear Karim, 

Thank you so much for all your help! 

No I would like to investigate which parameters I should use for my calibration. So if I understand correctly, you use a one-at-a-time sensitivity analysis for that. 

Do you mean you perform one iteration with 100 simulations, see the P and t value at global sensitivity. And based on the t-value you determine if the parameter is sensitive enough to use for calibration? And then you select the one with the highest t-value. 

Is this correct?

Thank you! 

Regards, Chantal

 

Op woensdag 15 juni 2016 16:49:41 UTC+2 schreef Abbaspour:

To unsubscribe from this group and all its topics, send an email to swat-cup+...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Abbaspour, Karim]

Sensitivity of one parameter often depends on the values of other parameters. For this reason it is not an absolute thing and both one-at-a-time and global procedures have conceptual issues. You can perform one-at-a-time analysis at the beginning to see sensitivity of different parameters and infer a reasonable range for them, but I would not leave a parameter completely out just based on that. Luckily in most cases we already know what the sensitive parameters are, so this really eliminates having to do such an analysis at the beginning. For global sensitivity, I am not surprised that each time you get a different ranking!

 

Karim

p.s. please write to swatcup google group for more help

 

 

 

 

 

From: vivien Nooij [mailto:vno...@gmail.com]
Sent: Friday, June 17, 2016 8:03 AM
To: Abbaspour, Karim
Subject: Re: sensitivity analysis

 

Dear Sir Karim, 

 

Thank you already for your reply on my post sensitivity analysis in swat-cup user group. Could you please confirm if this is the right way to perform a sensitivity analysis before calibration? I read somewhere that you have to use n+2 (n=amount of parameters) for sensitivity, but this will be after your calibration right? 

I already run the model 6 times for each parameters seperatly and then use 100x simulations. However, I get a different ranking compared to most of the articles where CN2 is very sensitive for predicting flow.. So that is why I'm not sure if I do it right. 

 

Looking forward to your reply!

 

Regards, Chantal

[Abbaspour, Karim]

No, I don’t mean that. Please refer to the email I just sent to google group.

[chantal]

Thank you very much for your repons!