Trends between zooplankton communities and fish communities?

33 views
Skip to first unread message

Hartman, Rosemary@DWR (she/her)

unread,
Jun 23, 2026, 2:14:30 PMJun 23
to Bay-Delta Data Science Fun

Hi Data Science PWT,

 

I’m hoping someone can help me out here. I’m looking for some sort of statistical method to look at associations between two different communities – specifically zooplankton and fish. I know lots of multivariate statistics for comparing fish OR zooplankton communities over space and time, but I want to see if certain species of fish are more or less associated with certain species of zooplankton. Has anyone tried to do a similar analysis before? I’ve seen a few potentially useful functions for looking at co-occurrence, but I’d ideally like to incorporate abundance as well as occurrence.

 

Thanks in advance for any ideas!

 

---
Rosemary Hartman,
PhD, she/her
Environmental Program Manager

Rosemary...@water.ca.gov
California Department of Water Resources
M:
916-882-2926

saveourwater.com

 

Kyle Hardage

unread,
Jun 24, 2026, 11:56:17 AMJun 24
to Hartman, Rosemary@DWR (she/her), Bay-Delta Data Science Fun
I shamelessly plugged your query into Gemini to see what the AI would recommend (my logic of choosing Gemini is that perhaps Google has trained it on Google Scholar data). It had four recommendations, and I am curious if this community thinks these are valid approaches: 


1. Co-Clustering or Symmetric Co-Correspondence Analysis (CoCA)

2. Redundancy Analysis (RDA) or Canonical Correspondence Analysis (CCA)

3. Joint Species Distribution Models (JSDMs)

4. Procrates Analysis / Mantel Test (Matrix-Level)


I've really only used PERMANOVA but Gemini indicates this is for a single community composition which was my application. 🤷‍♂️ Perhaps these methods at least provide a starting point for further research. Sounds like an interesting IEP topic. 

Kyle

--
https://interagencyecologicalprogram.github.io/DataScience/
---
You received this message because you are subscribed to the Google Groups "Bay-Delta Data Science Project Work Team" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bay-delta-datasc...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/bay-delta-datascience/CO6PR09MB8151D7541B2187C9EE67BE45BCEE2%40CO6PR09MB8151.namprd09.prod.outlook.com.

Hartman, Rosemary@DWR (she/her)

unread,
Jun 24, 2026, 12:37:30 PMJun 24
to Kyle Hardage, Bay-Delta Data Science Fun

Thanks!

 

I’ve done JSDMs with single communities, but haven’t tried them for looking at correlations between different communities. It’s worth a try! And someone else suggested mantel tests too.

 

A number of people have replied just to me, so I thought I’d share responses I’ve got so far:

 

    1. Co-Inertia Analysis (CoIA) with the ade4 package. I’ve already tried that and got it to give me output… now I just need to understand what it says!
    2. CCA – This is more understandable, but it requires that you assume zooplankton abundance is driving fish abundance OR vice versa, whereas co-intertia apparently doesn’t make that assumption.
    3. Abundance-Weighted Network Analysis (Pairwise Correlation). You calculate pairwise correlations (using Spearman's rank to account for non-normal, overdispersed ecological counts) between every fish species and every zooplankton species across all sites. You then apply a false discovery rate (FDR) correction to filter out spurious correlations. Standard cor() functions combined with the igraph and ggraph packages in R. I haven’t tried this yet.
    4. Principal tensor analysis, thought this would need to have all the fish and zooplankton in one community and assessing taxa that occur together, but wouldn’t separate out the zoops and fish.
    5. Partial Least Squares Correlation (PLSc). Also known as Singular Value Decomposition (SVD) of the cross-covariance matrix. LSc identifies latent variables that explain the maximum covariance between the fish block and the zooplankton block. It is highly robust to datasets where there are more species than sampling sites (a common issue in ecology). The plsdepot or mixOmics packages in R. Haven’t tried this yet either
    6. Dynamic Structural Equation Models • dsem
    7. Something similar to this paper: Joint-species analysis reveals potential displacement of native fishes by non-native fishes within …

Kyle Hardage

unread,
Jun 24, 2026, 12:42:20 PMJun 24
to Hartman, Rosemary@DWR (she/her), Bay-Delta Data Science Fun
Oh cool, thanks for the summary Rosie! Always nice to keep such things in the back pocket.

Kyle

Swift, Ted@DWR (he/his)

unread,
Jun 24, 2026, 1:15:48 PMJun 24
to Kyle Hardage, Hartman, Rosemary@DWR (she/her), Bay-Delta Data Science Fun
Stepping back from the analytical approaches, what are the assumptions about the species association(s)? For instance, are you assuming that the fish are eating the zoops, in which case you'd expect a negative relationship? Are the fish eating one species (or "guild") of zoops, releasing the other species of zoops from competition and producing a positive correlation? Are the fish eating copepods that would otherwise be praying on cladocerans? 
-Ted

__________________________________________

Ted Swift, Ph.D. (He, Him, His), Senior Environmental Scientist (Specialist),

Quality Assurance Program, Division of Integrated Science and Engineering

Department of Water Resources; New phone number: 279/789-4554

“Reality must take precedence over public relations, for Nature cannot be fooled.”

    -Richard Feynman

https://water.ca.gov/images/dwr-logo-new.png

 


From: bay-delta-...@googlegroups.com <bay-delta-...@googlegroups.com> on behalf of Kyle Hardage <kyleh...@gmail.com>
Sent: Wednesday, June 24, 2026 9:42 AM

Hartman, Rosemary@DWR (she/her)

unread,
Jun 24, 2026, 1:55:28 PMJun 24
to Swift, Ted@DWR (he/his), Kyle Hardage, Bay-Delta Data Science Fun

Hi Ted,

 

This project is currently in the data exploration phase, but the goal is to look for patterns between fish and their prey, and whether there may be food limitation for post-larval fish. WE are assuming fish are eating zooplankton, but that might result in a negative correlation (if they are eating so many zooplankton that they are controlling the population), or a positive correlation (if fish are seeking out their preferred food source and surviving better where food is plentiful). Inter-guild changes in zooplankton abundance based of fish predation pressure are possible, but we do not have a-priori hypotheses about what those would be.

 

Rosie

 

From: Swift, Ted@DWR (he/his) <Ted....@water.ca.gov>
Sent: Wednesday, June 24, 2026 10:15 AM
To: Kyle Hardage <kyleh...@gmail.com>; Hartman, Rosemary@DWR (she/her) <Rosemary...@water.ca.gov>
Cc: Bay-Delta Data Science Fun <bay-delta-...@googlegroups.com>
Subject: Re: [bay-delta-datascience] Trends between zooplankton communities and fish communities?

 

Stepping back from the analytical approaches, what are the assumptions about the species association(s)? For instance, are you assuming that the fish are eating the zoops, in which case you'd expect a negative relationship? Are the fish eating one species (or "guild") of zoops, releasing the other species of zoops from competition and producing a positive correlation? Are the fish eating copepods that would otherwise be praying on cladocerans? 

-Ted

 

__________________________________________

Ted Swift, Ph.D. (He, Him, His), Senior Environmental Scientist (Specialist),

Quality Assurance Program, Division of Integrated Science and Engineering

Department of Water Resources; New phone number: 279/789-4554

“Reality must take precedence over public relations, for Nature cannot be fooled.”

    -Richard Feynman

 

Tobias, Vanessa@Wildlife

unread,
Jun 24, 2026, 2:13:06 PMJun 24
to Hartman, Rosemary@DWR (she/her), Swift, Ted@DWR (he/his), Kyle Hardage, Bay-Delta Data Science Fun

Hi Rosie,

 

I’m sure what you want is MORE suggestions at this point, but your response to Ted’s questions made me think of the Lotka-Volterra predator-prey model. I think the familiar two-species example (e.g., lynx and hares) equations can be extended for multiple prey species, but I don’t have a specific example to share. That might just be a dynamic linear model. Anyway, if you want to think about fish and zoops as timeseries, those are a couple of options that might fit.

 

Vanessa

 

---

Vanessa D. Tobias, PhD (she/her)

Smelt Supplementation

Environmental Program Manager

Water Branch

California Department of Fish and Wildlife

Cell phone: (916) 516-2176

 

From: 'Hartman, Rosemary@DWR (she/her)' via Bay-Delta Data Science Project Work Team <bay-delta-...@googlegroups.com>

Sent: Wednesday, June 24, 2026 10:55 AM
To: Swift, Ted@DWR (he/his) <Ted....@water.ca.gov>; Kyle Hardage <kyleh...@gmail.com>
Cc: Bay-Delta Data Science Fun <bay-delta-...@googlegroups.com>

Subject: RE: [bay-delta-datascience] Trends between zooplankton communities and fish communities?

 

WARNING: This message is from an external source. Verify the sender and exercise caution when clicking links or opening attachments.

 

Peter Nelson

unread,
Jun 25, 2026, 2:30:23 PMJun 25
to Tobias, Vanessa@Wildlife, Hartman, Rosemary@DWR (she/her), Swift, Ted@DWR (he/his), Kyle Hardage, Bay-Delta Data Science Fun
Interesting discussion! It'd be a great IEP topic, for sure as Kyle suggested. 

Because more options are always better...RLQ + fourth-corner is worth thinking about too. If you're willing to treat one community as "environmental context" for the other, RLQ links the two species tables through site scores, and the fourth-corner tests specific species-level associations. I've some experience trying to correlate spp composition and spp traits to different years (bad idea) and to different parts of the Delta (better idea) using this approach. I've attached a relevant paper plus the tutorial. 

Pete 


Peter Nelson
Institute of Marine Sciences
University of California, Santa Cruz


Dray13 - suppl1.pdf
Dray14Ecology - Combining the fourth‐corner and the RLQ methods for assessing trait responses to environmental.pdf

Hartman, Rosemary@DWR (she/her)

unread,
Jun 25, 2026, 2:35:53 PMJun 25
to Nelson, Peter@USCS, Tobias, Vanessa@Wildlife, Swift, Ted@DWR (he/his), Kyle Hardage, Bay-Delta Data Science Fun

Thanks! RLQ looks worth looking into.

 

Predator-prey dynamics via a lotka-volterra equation might be a bit tricky, because I’m looking at abundance of larval/post-larval fishes and zooplankton in the spring, but zoops have 2-week to 2-month life spans, and fish are annual to multi=annual, so without incorporating population-level estimates (as opposed to trawl-level CPUE), I might be in trouble. Plus a lot of the adult fish eat totally different things from the larval fish. My “whole food web” model is another project 😝

Alex Vaisvil

unread,
Jun 25, 2026, 2:52:01 PMJun 25
to Bay-Delta Data Science Project Work Team
Hi Rosie et al. I wanted to provide a little context to my DSEM suggestion. While I have not had an opportunity yet to apply them, DSEMs might be a good fit for your project because they can jointly model abundance (not just occurrence) for fish and zooplankton, estimate both simultaneous and lagged effects (so you can detect immediate predation or delayed bottom‑up/top‑down responses), and partition direct, indirect, and total effects to test food‑limitation hypotheses. They also handle measurement error and missing data, allow inclusion of environmental covariates or latent habitat factors, and produce interpretable path coefficients you can use to test whether associations are positive (fish track prey) or negative (fish suppress zooplankton).  You obviously need to think about the causal structure of the system and identify confounders and all that fun stuff if wanting to tease out effects (I like the way Richard McElreath presents it in Statistical Rethinking).

Alex

Reply all
Reply to author
Forward
0 new messages