Spatial Biomass and Density Dependence

[Joseph Caracappa - NOAA Federal]

Hi All,

We are looking to implement density-dependent processes (ddpend 2 or 3) for a few key species. A question that arose is how to adequately measure that we're doing a good job replicating survey biomass.

Has anyone come across a method or diagnostic to test how close an Atlantis output distribution is to some reference spatial distribution?  Ideally we'd have a few calibration runs and we'd want to rank candidate runs on how well they reproduce observations.

I figured I'd ask around before coming up with something new. 

Thanks,

Joe

[Owen Liu - NOAA Federal]

Hey Joe,

I don't have a diagnostic for you, but I'm sharing a piece of R code I used to parse and plot biomass by box from the BoxBiomass.txt output file. It organizes output into a data frame that's easy to work with or visualize things like biomass or proportion of biomass by time and space. This might be useful in calculating the diagnostic you're developing?

Cheers,

Owen

--
You received this message because you are subscribed to the Google Groups "Atlantis ecosystem model" group.
To unsubscribe from this group and stop receiving emails from it, send an email to atlantis-ecosystem...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/atlantis-ecosystem-model/be36046e-6aaf-429f-9c9e-aed1b1316842n%40googlegroups.com.

[Fulton, Beth (Environment, Hobart)]

G’day

 

None of my bespoke scripts will be overly helpful, though Javier and co’s AtlantisTool is likely a decent place to build off.

 

I typically use netcdf file output and compare that to relative distributions (or absolute if good survey info) from surveys – using the same scripts used to map into polygons in the first place

 

Beth

To view this discussion on the web, visit https://groups.google.com/d/msgid/atlantis-ecosystem-model/CABHsB0nW%3DwLAoDnzHZjsaS-32EwZM%2B0XENrjb52oL7Rr9TnYuw%40mail.gmail.com.

[Holly Perryman]

Hei!

I just wanted to add that I have some R scripts on my github for getting biomass via the NC out file:

https://github.com/hollyannperryman/atlantis_tools_data_processing.git

Also, I would be careful with data reported in the BoxBiomass.txt file. A little while back, I saw some discrepancies between BoxBiomass.txt data and data reported from the NC out file. Since then, I have stuck with using data from the NC out file. So, I would suggest double checking to make sure BoxBiomass.txt is giving you accurate info. before proceeding forward with using it. 

Cheers!

Holly

To view this discussion on the web, visit https://groups.google.com/d/msgid/atlantis-ecosystem-model/C886F867-F63B-4858-BFF4-3422DA541C7B%40csiro.au.

--

<><   <><   <><   <><   <><

Holly A. Perryman, Ph.D.

Postdoctoral Researcher

University of South Florida

Tampa, FL, USA

[Joseph Caracappa - NOAA Federal]

Thanks for the supporting scripts! Currently I'm doing something similar to what was mentioned by visually comparing box biomass values. I'll make sure to share the diagnostic when we settle on one.

[Gavin Fay]

Hi Joe,

If you are not looking for something complicated, you could treat the proportions by space as one might for any other type of composition data.

That would open the door for computing the likelihood assuming (say) a multinomial for the proportions by space (with the data from the surveys being the observed proportions). 

e.g.

where p_{t,s} is the observed proportion for box s in time t, phat_{t,s} is the predicted (Atlantis) proportion for the same cell/time slice, and N_{t} is the assumed effective sample size for the given observation vector.

With most of these spatial observed distributions in the Northeast US being derived from fishery independent surveys, you will have a decent first approximation to the sample sizes (the number of tows).

This would also allow you to inspect the correspondence between distributions across time slices via say Pearson residuals to identify any patterning (e.g. particular boxes with proportions that are consistently over- or under-estimated)

Gavin

-- 
Gavin Fay, Ph.D. [he/him/his]

Associate Professor
Department of Fisheries Oceanography
School for Marine Science and Technology
University of Massachusetts Dartmouth

phone: +1-508-910-6363
web: www.thefaylab.com
mailing address:

836 South Rodney French Boulevard
New Bedford, MA 02744-1221 USA

---

From: atlantis-eco...@googlegroups.com <atlantis-eco...@googlegroups.com> on behalf of Holly Perryman <holly.p...@gmail.com>
Sent: Friday, July 19, 2024 5:02 AM
To: Fulton, Beth (Environment, Hobart) <Beth....@csiro.au>
Cc: Owen Liu - NOAA Federal <owen...@noaa.gov>; Joseph Caracappa - NOAA Federal <joseph.c...@noaa.gov>; Atlantis ecosystem model <atlantis-eco...@googlegroups.com>

Subject: Re: Spatial Biomass and Density Dependence

 

[EXTERNAL SENDER]

To view this discussion on the web, visit https://groups.google.com/d/msgid/atlantis-ecosystem-model/CAJUC1OZa8NPoeriubzX_QSkfBbcfteMbV1yEi1rvPycaj91Obg%40mail.gmail.com.

[Isaac Kaplan - NOAA Federal]

These are neat topics -- both the idea of turning on the foraging movement, and the comparison of model-projected spatial distributions to observed spatial distributions. 

As you dig into this, I'll be interested to see if you want to give some weight to cases where the model predicts foragers to be near the correct grid cell, if not in precisely the right place. I.e. partial weighting based on distance. There was an old literature on this using the fuzzy kappa statistic, and something called the Map Comparison Kit, for instance here: https://www.tandfonline.com/doi/full/10.1080/13658810802570317  . Maybe there is a more modern way of handling this concept. 

To view this discussion on the web, visit https://groups.google.com/d/msgid/atlantis-ecosystem-model/LV8PR20MB69827DA55080D6DF31CBCE4AA4AD2%40LV8PR20MB6982.namprd20.prod.outlook.com.

--

Isaac Kaplan

Northwest Fisheries Science Center
National Marine Fisheries Service, NOAA
Conservation Biology Division, Room 205 South
2725 Montlake Blvd. E., Seattle WA 98112

cell phone 1(206)384-2043