Ecosim automated fitting

[Thomas Seyer]

Hello again, every one,

Here is my new problem: to force my ecosim, I'm using 46 catch series (type -6) and 5 biomasses for my plankton groups (type -1)(as reference, I have 26 series of biomasses, type either 1 or 0). After I got familiar with the two fitting interfaces, I realised I have a trouble with the automatic (stepwise) one: when testing for baseline hypothesis (so no trophic interactions and no fishing), it basically works with no forcing so on the steady state.

Is there a way to specify to the automated routine to try the vulnerability/PP anomaly combinaisons with a defined subset of the time series?

Another question, about data formating:

for the plankton forcing biomasses (type -1), I couldn't find if they must be absolute or relative. What I using now is the variation around the average value over the period of my ecopath that I multiplied with the initale (ecopath module) value. So it is an absolute value centered on that initale value. Is it correct? Or should it be only the variations (relative), centered on 1?

As you migh understand with all these questions, no one is doing EwE in my lab so thanks for your help!

Best,

Thomas

[Thomas Seyer]

Additional question about the AIC data points:

depending on if I include my fishing forcing series or not, the number of AIC data points vary greatly. How can I compare the different fitting configurations?

Should I reduce the number of AIC points to the lowest for all the configurations. I don't understand why the number of points depends also of the number of forcing series and not only of the reference series.

[Marc Taylor]

Hi Thomas,

I feel your pain regarding the automatic fitting routine. I have several statistical issues with the method and do not advise its use in my collaborations.

My critiques are the following: 1. The number of observations (n) considered by the AIC calculation are based on the number of time series, independent of their length (e.g. number of years), 2. The model complexity (i.e. the number of parameters estimated, K) is based on the number (and complexity) of forcing functions and the number of vulnerability, V, parameters that are tuned. In my opinion, it is incorrect to consider the number of  V tuned given that the model has the same number of Vs, whether they are tuned or not; untuned Vs are still in the model, but simply use the default value of 2.0.

I would rather recommend focussing on the sum of squares, SS, as the main metric for measuring model fit. If you could develop an AIC test for evaluating best model fits, then I would think that increasing complexity, K, should mainly consider the addition of forcing functions, not the tuning of Vs.

Hope that helps,

Marc

--
You received this message because you are subscribed to the Google Groups "EwE users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to eweusers+u...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/eweusers/936a49b2-1594-42ee-a280-3bb994869702n%40googlegroups.com.

[EwE user support]

Hi Marc and Thomas,

Great to see some discussion!

The benefit of the stepwise fitting routine is its integration within
EwE, but that does not mean that the methodologies underneath are set in
stone. As with every feature in EwE, if you have better ideas as to how
things can be done, please join the fray and help us (the dev team) fix
it. We're always open to collaborations and ideas.

Best,
Jeroen

> <https://groups.google.com/d/msgid/eweusers/936a49b2-1594-42ee-a280-3bb994869702n%40googlegroups.com?utm_medium=email&utm_source=footer>.

>
> --
> You received this message because you are subscribed to the Google
> Groups "EwE users" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to eweusers+u...@googlegroups.com.
> To view this discussion on the web, visit

> https://groups.google.com/d/msgid/eweusers/CACg2Sf1dGHcj6u8jvwpZp2TJohPcsA7_GP_BDQU00M-NyUQaOA%40mail.gmail.com
> <https://groups.google.com/d/msgid/eweusers/CACg2Sf1dGHcj6u8jvwpZp2TJohPcsA7_GP_BDQU00M-NyUQaOA%40mail.gmail.com?utm_medium=email&utm_source=footer>.


--
This email has been checked for viruses by Avast antivirus software.
https://www.avast.com/antivirus

[Thomas Seyer]

Hi Marc & Jeroen,

Well, my point here is not about the statistic as I'm not quite there yet and I don't think I have the mathematical background to be relevant. It is more practical (see my questions above). I've watched several time the different videos of the Ecopath channel on Youtube (with Mr Christensen) on how to deal with the fitting, but I still have many interrogations I couldn't find answer even the user manual or original paper. If I could make a suggestion, it would be to make some "tutorial" videos with the last version of EwE, so we know exactly what to do with every parameter, button, ect...

Marc, would you have any answer, about the series formating for example? My planktonic biomasses (for the forcing) are in months so I'm using the timestep paramater in the serie upload form. But my catches are in years, so I copy-pasted my yearly values for every month. Is that correct?

If anyone reading this that is well familiar with ecosim would accept a quick visio call (lets say max 30min), that would be great to answer all those small yet blocking questions.

Best

Thomas 

[Marc Taylor]

Hi Thomas - Sorry, but I am a bit out of practice regarding the set-up of the forcing functions using monthly data. I'm sure that this is possible given that EwE usually iterates at monthly time steps. My main comments were regarding your question on comparing AIC values between fittings.

Hi Jeroen - I would be happy to help conceptualize a change to the automatic fitting. I'm sure it would be relatively straightforward to do, but a consensus is needed regarding what constitutes an observation / estimated parameter. From my communications with folks involved in it's original set-up, this was not a straightforward decision. I'm also not a statistician, and only know how to program in R, so this would require some additional brain-power :-) 

Cheers,

Marc

To view this discussion on the web, visit https://groups.google.com/d/msgid/eweusers/7c7e540a-f0cb-49f5-995e-65dfb31baa78n%40googlegroups.com.

[Thomas Seyer]

Hello everyone,

Here are some answers provided by Jeroen (thank you!) that I'd like to share. You might find them usefull:

The time series are absolute unless specified otherwise, so in case of doubt, ABSOLUTE!

Here is a complete list of data types:

     BiomassRel = 0
     BiomassAbs = 1
     BiomassForcing = -1
     TimeForcing = 2
     FishingEffort = 3
     FishingMortality = 4
     TotalMortality = 5
     ConstantTotalMortality = -5
     Catches = 6
     CatchesForcing = -6
     CatchesRel = 61
     AverageWeight = 7
     DiscardProportion = 10
     DiscardMortality = 11
     Landings = 12
     Discards = 13
     FishingMortalityRef = 104

Concerning the use of timestep mode (when importing a time serie csv file), if you have annual catches data as forcing (t/km²/y) you may copy 12 times each year value (as the forcings needs entry for every timestep) but you MUST NOT scale the values in t/km²/month (by dividing by 12).

I also would like to remember to everyone but especially the more experiencied users of EwE that for a forum like that to work, we need that people take a bit of there time to see if there are questions they can answer in other feeds and not only theirs. Let take a bit of pressure of Jeroen's shoulders!

Cheers

Thomas