Closed brianstock-NOAA closed 2 years ago
@brianstock @kimberly-bastille The tech doc describing the cold pool index methodology probably needs to have more detailed explanation to address some of Brian's comments, especially since the tech doc cites the same literature that Brian has
Thanks @brianstock-NOAA and @andybeet for pointing this out. I'd like to have a more standardized approach to which years are included for anomaly calculation for as many indices within ecodata as possible. Since the 2020 SOEs are out, I think we should have this discussion to see how much we can standardize for 2021.
The CPI data we receive for ecodata
is preprocessed so it may be a good idea to loop in Chris Melrose to answer the specific questions about how he derived this data.
I forwarded the issue to Chris.
From Chris...
On item 1: I can push it back to 1973 if desired. The 1977 cutoff was just to match the start of Marmap when there was higher quality data available.
On Item 2: Yes, I believe Brian is correct about the main reason for the differences from the old Miller version. I was able to reproduce the Miller paper figures with Jon's code if I use the same input data/period so the math itself is definitely the same. I made no changes to the calculations in Jon's code, I essentially only changed the input data so basically any difference is related to the differences in input data and time period. Probably could alter how it does the anomalies to keep a consistent baseline so that the prior years stay the same as new ones are added.
On item 3: I think it would be pretty easy to add the SE to the output.
All that said, I think it would be nice to do something more sophisticated than this index in the future as we have discussed before. This was just the easy way since we had the code from Jon.
Thanks everyone for the quick feedback. ecodata
is looking good, I'm excited to use it!
I'm not concerned about replicating the CPI from Miller et al. (2016), so no problem if you wanted to change the time period to use for the mean in the anomaly calculation to be standardized across ecodata. Also ok if you change the method to something more sophisticated, as long as it's documented :)
(Posting here bc it's similar enough I think)
Trying to use environmental covariates from ecodata
in stock assessments via wham
, we have similar issues for the Gulf Stream Index that we have for the CPI:
These are more general than just the CPI or GSI:
For discussion. I know the primary purpose of ecodata
is not to interface with stock assessments or wham
, but seems like it could with some relatively minor changes.
Nye et al. 2011. https://www.nature.com/articles/ncomms1420 O'Leary et al. 2019. https://www.nrcresearchpress.com/doi/abs/10.1139/cjfas-2018-0092 Xu et al. 2018. https://onlinelibrary.wiley.com/doi/full/10.1111/fog.12236
This might be a good discussion to have with the larger LTL/habitat group at the SOE debrief on Wednesday (5/20/20). I sent you an invite @brianstock. One of the requests from the council is to expand the CPI, I think a discussion of GSI could also be useful.
@brianstock -- thank you for challenging the purpose of ecodata
. Your feedback is excellent and will definitely help us improve the product. I think it will be good to define if the purpose is it just for the SOE reports or as a broader ecological indicator hub.
@slucey -- do you recall the CRAN package that DFO produced and highlighted during WGNARS? It might help guide our thinking, here.
Also, as the issue of reference periods for anomalies has come up before -- do the pros outweigh the cons of making it a user-defined argument (e.g., ecodata::as_anomaly(data = "CPI", reference_period = c(1980, 1989))
). The calculation obviously isn't the hard part, but it would require us to fundamentally change the structure of the package.
I don't recall off the top of my head but will check with Jamie.
@slarge I think the idea of having a broader ecological indicator hub is exactly why we need a database and ERRDAP. This way the SOE and others can pull from the exact same database.
I also think we need to have a discussion on how we define our anomaly calculations because I don't believe there is a one size fits all solution depending on what you want the anomaly data to portray. Do you want to compare the anomaly to a baseline or look at inter-annual variation?
I'm happy to stoke discussion, that's the easy part! The idea of an EDAB-curated package with ecological indicators that interfaces with a PopDy-curated package that fits stock assessments using these indicators is attractive. And we're 95% of the way there already. It seems silly for me to distribute CPI.csv
and GSI.csv
files in WHAM when ecodata
also has them, in addition to dozens of other variables that could be used as covariates in assessment models.
I don't recall off the top of my head but will check with Jamie.
The package is called marindicators: https://cran.r-project.org/web/packages/marindicators/index.html
This is awesome that you've updated the CPI and made it available via ecodata. However, the CPI in ecodata doesn't match the data file I have from Miller et al. (2016), which is distributed here in the WHAM package. It'd be nice for WHAM to just get the CPI (and other environmental data) from ecodata. But first, there are a couple issues to work out:
CP_year_r_se
, saved as theSE_Cold_Pool_Index
column in the output csv file. Note that we can estimate the SE of an environmental covariate within WHAM (example using the GSI), but it's an additional parameter(s) and better to use the calculated SE.Chris's note on lines 155-165 seems relevant to both (1) and (2).