Processing math: 100%
+ - 0:00:00
Notes for current slide
Notes for next slide

When first hatched, and before the disappearance of the yolk sac, the larvae (European) feed on larval snails and crustaceans, on diatoms, and on peridinians, but they soon begin taking copepods, and depend exclusively on these for a time after they get to be 12 mm. long, especially on the little Pseudocalanus elongatus. As they grow older they feed more and more on the larger copepods and amphipods, pelagic shrimps, and decapod crustacean larvae.

Developing zooplankton indices in VAST

Herring Research Track
10 March 2025

Sarah Gaichas, Adelle Molina

1 / 26

Does food drive recruitment of Atlantic herring?

Atlantic herring, Clupea harengus

Atlantic herring illustration, credit NOAA Fisheries

"The herring is a plankton feeder.... Examination of 1,500 stomachs showed that adult herring near Eastport were living solely on copepods and on pelagic euphausiid shrimps (Meganyctiphanes norwegica), fish less than 4 inches long depending on the former alone, while the larger herring were eating both." (Collette et al., 2002)

2 / 26

When first hatched, and before the disappearance of the yolk sac, the larvae (European) feed on larval snails and crustaceans, on diatoms, and on peridinians, but they soon begin taking copepods, and depend exclusively on these for a time after they get to be 12 mm. long, especially on the little Pseudocalanus elongatus. As they grow older they feed more and more on the larger copepods and amphipods, pelagic shrimps, and decapod crustacean larvae.

Objective

Create zooplankton indices to evaluate changes in food for Atlantic herring larvae, juveniles, and adults over time and in space in the Northeast US continental shelf ecosystem.

Two applications:

  1. Addressing uncertainty in the stock assessment for Atlantic herring (Clupea harengus)

  2. Describing zooplankton species and group trends for integrated ecosystem assessment

Atlantic herring illustration, credit NOAA Fisheries

copepods

https://www.naturepl.com/stock-photo-copepods-calanus-finmarchicus-aggregated-at-the-surface-reverse-diel-nature-image01407135.html

3 / 26

Which indicators are relevant for recruitment?

4 / 26

Which indicators are relevant for recruitment?

Boosted regression tree (Molina 2024) investigated relationships between environmental indicators and Atlantic herring recruitment estimated in the assessment.

Larval and juvenile food (zooplankton), egg predation, and temperature always highest influence

5 / 26

Northeast US existing zooplankton indicators

Abundance anomalies in ecodata

Ecosystem reporting areas: Ecosystem Production Units (EPUs) used for zooplankton

EPU by ten minute squares

6 / 26

Northeast US existing zooplankton indicators

Calanus stages, Gulf of Maine

Calanus finmarchicus time series by EPU

7 / 26

Outline

  1. Vector Autoregressive Spatio-Temporal (VAST) modeling

  2. Identify key zooplankton species: what are herring prey?

  3. Which RE model is best? Selection

  4. Spatial bounds for indices

  5. Spatial and temporal bounds specific to herring larvae

  6. Index trends

xkcd maps

https://xkcd.com/1138

8 / 26

Vector Autoregressive Spatio-Temporal (VAST) modeling (Thorson et al., 2017; Thorson, 2019)

VAST is a Geostatistical generalized linear mixed effects model (GLMM) that models two linear predictors for an index: 1. encounter rate, and 2. positive catch (amount in stomach)

A full model for the first linear predictor ρ1 for each observation i can include:

  • fixed intercepts β1 for each category c and time t,
  • spatial random effects ω1 for each location s and category,
  • spatio-temporal random effects ε1 for each location, category, and time,
  • fixed vessel effects η1 by vessel v and category, and
  • fixed catchability impacts λ1 of covariates Q for each observation and variable k:

ρ1(i)=β1(ci,ti)+ω∗1(si,ci)+ε∗1(si,ci,ti)+η1(vi,ci)+nk∑k=1λ1(k)Q(i,k)

The full model for the second linear predictor ρ2 has the same structure, estimating β2, ω2, ε2, η2, and λ2 using the observations, categories, locations, times, and covariates.

We modeled aggregate small pelagic prey as a single category, and apply a Poisson-link delta model to estimate expected prey mass per predator stomach as in (Ng et al., 2021).

VAST model code and documentation: https://github.com/James-Thorson-NOAA/VAST

9 / 26

Spatial and spatio-temporal correlation decay with increasing distance estimated as κ in a Matern function with fixed smoothness and geometric anisotropy (directional correlation, optionally estimated by the model).

Initial model selection consistently supported the inclusion of spatial and spatio-temporal random effects and anisotropy across all datasets: fall, spring, and annual.

Spatial estimation assumptions

VAST has built-in "extrapolation grids" for many surveyed areas, including the Northeast US.

Observations in space are used to define fixed locations ("knots") covering the full extent of the model. We assume constant variation within a timestep at a knot.

Modelers specify the number of knots to group observation locations, to balance computation time and resolution. We used 500 knots:

spring lgcopeALL knots

10 / 26

Extrapolation grid: "area over which densities will be extrapolated to calculate derived quantities" encompassing survey area allows comparison with design-based estimates

Knots are defined that "minimize the average distance between samples and knots" using k-means clustering, which results in knots in proportion to sampling intensity across the area.

THIS MEANS DIFFERENT DATASETS HAVE DIFFERENT GRIDS

Modeling spatial and spatio-temporal variation

Observations correlated in space and in space over time due to unmeasured processes are modeled as multivariate normal Gaussian Random Fields (GRF):

ω1 ~ MVN(0, R1); ω2 ~ MVN(0, R2)
ε1(,t) ~ MVN(0, R1); ε2(,t) ~ MVN(0, R2)

Spatial and spatio-temporal correlation decay with increasing distance d estimated as κ in a Matérn function with fixed smoothness ν and geometric anisotropy H (directional correlation).

Correlation function between locations s and s′:

R1(s,s′)=12ν−1Γ(ν)×(κ1|d(s,s′)H|)ν×Kν(κ1|d(s,s′)H|)

Estimation uses stochastic partial differential equation (SPDE) approximation.

example spring lgcopeALL aniso

11 / 26

Estimating density in space and the index

Observation model "Index2", Gamma distribution for positive catches and Alternative "Poisson-link delta-model" using log-link for numbers-density and log-link for biomass per number. It is intended for continuous data, which includes biomass data and “numbers standardized to a fixed area.”

Probability of encounter Poisson: p(i)=1−exp[−n(i)] set to 1 for groups found at all stations, e.g. small copepods or zooplankton volume

Number of zooplankton cells per volume given encounter: r(i)=n(i)p(i)w(i)

Probability for numbers per volume B where g is a Gamma function. :

Pr[b(i)=B]={1−p(i),B=0p(i)×g[B|r(i),σ2b],B>0,

Density b at a location (knot) s for year t is then the predicted number of cells per volume (linear predictor for encounter * linear predictor for cells/vol given encounter):

ˆbs,t=ˆns,tˆws,t

Index based on area a weighting for each of 500 knots (or subsets):

It=500∑s=1asˆbs,t

Bias correction as in (Thorson et al., 2016)

12 / 26

I am interpreting zooplankton abundance per 100 cubic meters as numbers standardized to a fixed area (volume) in applying the Gamma observation model.

Herring prey from adult stomachs: a mix of zooplankton

copepods euphausiids

Northeast Fisheries Science Center Diet Data Online: https://fwdp.shinyapps.io/tm2020/

13 / 26

Zooplankton groups

Focus is on recruitment, food for larvae through juveniles.

Models developed for the Herring RT are for the following copepod categories:

  • Large copepods ALL: Calanus finmarchicus, Metridia lucens, Calanus minor, Eucalanus spp., Calanus spp.

  • Small copepods ALL: Centropages typicus, Pseudocalanus spp., Temora longicornis, Centropages hamatus, Paracalanus parvus, Acartia spp., Clausocalanus arcuicornis, Acartia longiremis, Clausocalanus furcatus, Temora stylifera, Temora spp., Tortanus discaudatus, Paracalanus spp.

14 / 26
  • Calanus finmarchicus (calfin_100m3) = "calfin" in table below,
  • Large copepods (calfin_100m3, mlucens_100m3, calminor_100m3, euc_100m3, calspp_100m3) = "lgcopeALL",
  • Small copepods (all) (ctyp_100m3, pseudo_100m3, tlong_100m3, cham_100m3, para_100m3, acarspp_100m3, clauso, acarlong_100m3, fur_100m3, ost_100m3, temspp_100m3, tort_100m3, paraspp_100m3) = "smallcopeALL" and
  • Small copepods (SOE) (ctyp_100m3, pseudo_100m3, tlong_100m3, cham_100m3) = "smallcopeSOE".

Sampling for zooplankton taxa

Between 1982-2022 there were:

  • 27774 NEFSC zooplankton survey stations.
    • 52% between January-June (Spring).
    • 48% between July-December (Fall).
  • 21524 with Calanus finmarchicus.
    • 52% between January-June (Spring).
    • 48% between July-December (Fall).
  • 23813 stations with large copepods.
    • 49% between January-June (Spring).
    • 51% between July-December (Fall).
  • 25255 stations with small copepods.
    • 49% between January-June (Spring).
    • 51% between July-December (Fall).
  • 11710 stations with euphausiids
    • 50% between January-June (Spring).
    • 50% between July-December (Fall).
  • 3815 stations with herring larvae
    • 28% between January-June (Spring).
    • 72% between July-December (Fall).

Zooplankton stations, January-June Northeast Fisheries Science Center surveys

15 / 26

Model selection

Result: Always best to estimate spatial and spatio-temporal random effects

However, the small copepod fall model had the spatial random effects parameter for encounter rate, ω1, approach 0. We allowed it to be estimated at 0 rather than fixing it at 0. Resulting indices were identical.

Only spring large zooplankton did better with a day of year catchability covariate

Did not have time to explore other catchability covariates or habitat covariates

Next steps might be to look at depth and or temperature

16 / 26
Maps of key areas for Herring assessment indices. The full VAST model grid is shown in brown.Maps of key areas for Herring assessment indices. The full VAST model grid is shown in brown.Maps of key areas for Herring assessment indices. The full VAST model grid is shown in brown.

Maps of key areas for Herring assessment indices. The full VAST model grid is shown in brown.

Indices for zooplankton can be calculated for any subset of the full model domain. Bias correction of the resulting indices is then applied (Thorson et al., 2016).

17 / 26

NEFSC survey strata definitions are built into the VAST northwest-atlantic extrapolation grid already. We defined additional new strata to address the recreational inshore-offshore 3 mile boundary. The area within and outside 3 miles of shore was defined using the sf R package as a 3 nautical mile (approximated as 5.556 km) buffer from a high resolution coastline from thernaturalearth R package. This buffer was then intersected with the current FishStatsUtils::northwest_atlantic_grid built into VAST and saved using code here. Then, the new State and Federal waters strata were used to split NEFSC survey strata where applicable, and the new full set of strata were used along with a modified function from FishStatsUtils::Prepare_NWA_Extrapolation_Data_Fn to build a custom extrapolation grid for VAST as described in detail here.

Results: Compare ecodata and VAST annual Calanus finmarchicus

GB = Georges Bank, GOM = Gulf of Maine, MAB = Mid Atlantic Bight. Different calculation method, spatial definitions

18 / 26

Results: Jan-Jun (Spring) large copepods

Model is bias corrected, includes a day of year "catchability" covariate which improved the fit.

19 / 26

Results: Jul-Dec (Fall) small copepods

Model is bias corrected, does not include covariates.

20 / 26

When and where are herring larvae present?

Herring larvae VAST model, timing for presence of larvae Sept. - Feb. (Richardson et al., 2010)

VAST estimated Fall forage biomass density →

21 / 26

Based on herring larvae collected on the same surveys as the zooplankton Year for January-February in this dataset was shifted to match September - December; this is the same cohort.

Defining the larval herring area

Herring larvae Sep-Feb VAST output:

  • Sum density by cell over all years
  • Evaluate quantiles of density in space
  • Select quantile with largest continuous area
  • Outline "concave hull" of area
  • Integrate as new strata set into VAST grid
  • Run small copepods model with larval strata
  • 27774 NEFSC zooplankton survey stations.
    • 51% between September-February.
    • 49% outside larval time.
  • 3815 stations with herring larvae
    • 93% between September-February.
    • 7% outside larval time.
  • 25255 stations with small copepods.
    • 52% between September-February.
    • 48% outside larval time.

22 / 26

Small copepods available to larvae

Model is bias corrected with no covariares.

23 / 26

Extensions: Other zooplankton indices could link to adult growth or survival (...next RTA)

24 / 26

Potential risk criteria:

Risk Level Definition
Low Prey availability high (not limiting) and/or good fish condition past 5 years
Low-Moderate Aggregate prey available for this species has stable or increasing trend, moderate condition
Moderate-High Aggregate prey available for this species has significant decreasing trend, poor condition
High Managed species highly dependent on prey with limited and declining availability, poor condition

Next steps

These were implemented as recruitment covariates in WHAM

Presentation tomorrow

Discussion

Further tailoring for herring?

These and other zooplankton indices were used for

  • 2025 ecosystem reporting

  • Time series for fitting or forcing food web models

  • Council risk indicators

Ideas for improvement welcome

25 / 26

Thank you! References

Collette, B. B. et al. (2002). Bigelow and Schroeder's Fishes of the Gulf of Maine, Third Edition. 3rd ed. edition. Washington, DC: Smithsonian Books. ISBN: 978-1-56098-951-6.

Ng, E. L. et al. (2021). "Predator stomach contents can provide accurate indices of prey biomass". In: ICES Journal of Marine Science 78.3, pp. 1146-1159. ISSN: 1054-3139. DOI: 10.1093/icesjms/fsab026. URL: https://doi.org/10.1093/icesjms/fsab026 (visited on Sep. 01, 2021).

Richardson, D. E. et al. (2010). "Development of long-term larval indices for Atlantic herring (Clupea harengus) on the northeast US continental shelf". In: ICES Journal of Marine Science 67.4, pp. 617-627. ISSN: 1054-3139. DOI: 10.1093/icesjms/fsp276. URL: https://doi.org/10.1093/icesjms/fsp276 (visited on Aug. 30, 2024).

Thorson, J. T. (2019). "Guidance for decisions using the Vector Autoregressive Spatio-Temporal (VAST) package in stock, ecosystem, habitat and climate assessments". En. In: Fisheries Research 210, pp. 143-161. ISSN: 0165-7836. DOI: 10.1016/j.fishres.2018.10.013. URL: http://www.sciencedirect.com/science/article/pii/S0165783618302820 (visited on Feb. 24, 2020).

Thorson, J. T. et al. (2017). "Comparing estimates of abundance trends and distribution shifts using single- and multispecies models of fishes and biogenic habitat". In: ICES Journal of Marine Science 74.5, pp. 1311-1321. ISSN: 1054-3139. DOI: 10.1093/icesjms/fsw193. URL: https://doi.org/10.1093/icesjms/fsw193 (visited on Nov. 04, 2021).

Thorson, J. T. et al. (2016). "Implementing a generic method for bias correction in statistical models using random effects, with spatial and population dynamics examples". En. In: Fisheries Research 175, pp. 66-74. ISSN: 0165-7836. DOI: 10.1016/j.fishres.2015.11.016. URL: https://www.sciencedirect.com/science/article/pii/S0165783615301399 (visited on Jul. 29, 2022).

Slides available at https://noaa-edab.github.io/presentations
Contact: Sarah.Gaichas@noaa.gov

26 / 26

Does food drive recruitment of Atlantic herring?

Atlantic herring, Clupea harengus

Atlantic herring illustration, credit NOAA Fisheries

"The herring is a plankton feeder.... Examination of 1,500 stomachs showed that adult herring near Eastport were living solely on copepods and on pelagic euphausiid shrimps (Meganyctiphanes norwegica), fish less than 4 inches long depending on the former alone, while the larger herring were eating both." (Collette et al., 2002)

2 / 26

When first hatched, and before the disappearance of the yolk sac, the larvae (European) feed on larval snails and crustaceans, on diatoms, and on peridinians, but they soon begin taking copepods, and depend exclusively on these for a time after they get to be 12 mm. long, especially on the little Pseudocalanus elongatus. As they grow older they feed more and more on the larger copepods and amphipods, pelagic shrimps, and decapod crustacean larvae.

Paused

Help

Keyboard shortcuts

↑, ←, Pg Up, k Go to previous slide
↓, →, Pg Dn, Space, j Go to next slide
Home Go to first slide
End Go to last slide
Number + Return Go to specific slide
b / m / f Toggle blackout / mirrored / fullscreen mode
c Clone slideshow
p Toggle presenter mode
t Restart the presentation timer
?, h Toggle this help
Esc Back to slideshow