1 of 56

Developing climate-integrated stock assessment models: An American plaice example

Amanda Hart

11/8/2022

Models for Marine Ecosystem-Based Management

2 of 56

Recap

  • Chapter 3: Extended stock assessment models
  • Chapter 7: Multivariate analysis of Ecosystem Indicators
  • Chapter 9: State-space models revisited – Woods Hole Assessment Model
    • State-space population dynamics models
    • WHAM
    • TODAY: Climate-linked stock assessment models

​

American plaice case study

3 of 56

3

Management process

4 of 56

5 of 56

Terms of Reference (ToR)

ToR 1: Identify relevant ecosystem and climate influences on stock dynamics

​

Identify relevant ecosystem and climate influences on the stock. Characterize the uncertainty in the relevant sources of data and their link to stock dynamics. Consider findings, as appropriate, in addressing other TORs. Report how the findings were considered under impacted TORs. 

5

6 of 56

ToR 1: Identify relevant ecosystem and climate influences on stock dynamics.�

ToR 1 is a larger-scale issue that influences how we approach other ToRs

ToR 1. Identify relevant ecosystem and climate influences on stock dynamics...

ToR 3. Identify the appropriate survey data to be used in the assessment.

ToR 4. Use appropriate assessment approach to estimate annual F, R, SSB.

ToR 5. Update or redefine status determination criteria.

ToR 6. Define appropriate methods for producing projections…

ToR 2. Estimate catch from all sources including landings and discards.

ToR 7. Review, evaluate, and report on the status of research recommendations…

​

ToR 8. Develop a backup assessment approach…

7 of 56

ToR 1: Identify relevant ecosystem and climate influences on stock dynamics.�

7

Science Review (Ecosystem Profile) &

Fishermen’s Ecological Knowledge

Exploratory Modeling

Climate-informed models

Characterize State of Knowledge

Ecosystem context

Improved model assumptions and parameterization

Acknowledgement of non-stationarity

Science

Assessment Modeling

Conduct

Integrated Modeling

Develop Relationships among Climatic-Ocean-Stock Variables

Application

General Approach

8 of 56

8

American plaice case study

9 of 56

American plaice working group & contributors

  • Northeast Fishery Science Center
  • New England Fishery Management Council
  • University staff & students

​

​

  • Non-profits
  • Stakeholder groups

Working Group Overview

10 of 56

American plaice working group & contributors

  • Northeast Fishery Science Center
  • New England Fishery Management Council
  • University staff & students

​

​

  • Non-profits
  • Stakeholder groups

Working Group Overview

Term of Reference 1 Subgroup

11 of 56

American plaice biology

  • Demersal flatfish
  • 1 US stock separate from neighboring Canadian stocks
  • Prefer sandy mud habitats 10-700m deep
    • Shallower distribution in spring (50-100m) when migrating inshore to spawn vs. fall (100-180m)
  • Oldest plaice in US waters 24 years old
    • Research track aggregated ages 11+ into plus group

Working Group Overview

12 of 56

American Plaice fishery & assessment

(NOAA 2019, Assessment Update Report, Working Group Overview)

FMSY proxy

SSBtarget = SSBMSY proxy

SSBthreshold = 1/2 SSBMSY proxy

VPA

irtual

opulation

nalysis

Retrospective adjustment

  • Beam trawl fishery in 1880s
  • Foreign fleets 1960s-1970s
  • Directed fishery beginning in 1970s
  • Catch from 1970s on mostly from mixed-species groundfish trips

13 of 56

Terms of Reference 1 for American Plaice

  1. Conduct literature review
  2. Incorporate fishermen’s ecological knowledge
  3. Exploratory modelling
  4. ID strong drivers to consider in stock assessment

14 of 56

1) Lit review: What climate and ecosystem effects are impacting the stock?

Potential Driver 

Decadal-scale  climate  

variability

Climate  

warming trend

Unaccounted  for predation

Fish  

distribution  

shift

Changes in  

currents 

Fishing  

pressure 

Pulled from literature review

  • U.S. plaice stock & neighboring Canadian
  • 1918-2021

​

Behan et al. 2022 & ToR1 overview

15 of 56

1) Lit review: How do these drivers impact stock dynamics?

Potential Driver 

Hypothetical Impact

Decadal-scale  climate  

variability

Time-varying  

growth or natural  mortality,  

recruitment

Climate  

warming trend

Time-varying  

growth or natural  mortality,  

recruitment, ssb,  center of biomass  location, maturity

Unaccounted  for predation

Time-varying  

natural mortality

Fish  

distribution  

shift

Time-varying  

catchability in  

survey or fishery

Changes in  

currents 

Larval transport  (recruitment),

Fishing  

pressure 

Growth, maturity,  abundance (which  also affects  

growth)

Focused literature review on drivers for:

  • Natural mortality
  • Growth & maturity
  • Recruitment

16 of 56

1) Lit review: What data is available to describe the climate driver?

16

Potential Driver 

Hypothetical Impact

Time series 

Decadal-scale  climate  

variability

Time-varying  

growth or natural  mortality,  

recruitment

Climate indices such  as North Atlantic  

Oscillation, Atlantic  Multidecadal  

Oscillation, and Gulf  Stream Index

Climate  

warming trend

Time-varying  

growth or natural  mortality,  

recruitment, ssb,  center of biomass  location, maturity

Sea surface or  

bottom temperature  anomaly for region

Unaccounted  for predation

Time-varying  

natural mortality

Predator abundance  time series (e.g.,  

spiny dogfish) or diet  information

Fish  

distribution  

shift

Time-varying  

catchability in  

survey or fishery

Sea surface or  

bottom temperature  anomaly for region

Changes in  

currents 

Larval transport  (recruitment),

Current direction,  

speed, or anomalies?

Fishing  

pressure 

Growth, maturity,  abundance (which  also affects  

growth)

Fishing effort time  series

17 of 56

2) Fishermen’s Ecological Knowledge

Summary of Fishermen’s Feedback:

Impacts from Management

  • Declining catch of plaice is not reflective of declining biomass
    • due to increasing regulatory measures that have prevented targeting the stock in specific areas at specific times of the year.
  • Decreased Otter trawling from implementation of annual catch limit.
    • resulted in increased fixed gear (lobster traps) in areas traditionally trawled
    • Led to less available trawlable areas
  • Increased minimum mesh size regulations impacted ability to catch plaice.

Distribution Changes & Catch Rates

  • Plaice traditionally came inshore in spring, but are now not caught inshore.
  • Plaice abundance, size, and age have been impacted by changes in water temperature.
  • Suggested examining CPUE indices for plaice may be useful to consider in the research track assessment.

Survey Catches

  • Fishery-independent surveys have low sampling intensity.
    • NEFSC and MA DMF trawl surveys cannot adequately sample inshore areas due to fixed gear
    • The NEFSC survey gear is not effective for catching flatfish species and there is low survey catch efficiency for plaice and other flounder species.

ToR1 overview & Working paper on fishermen’s ecosystem knowledge

18 of 56

3) Exploratory modelling: Catchability GAMs

(NOAA Fisheries, 2021)

Depth & distribution shifting in response to temperature

(Behan and Kerr 2022)

19 of 56

3) Exploratory modelling: Recruitment GAMs

Recruitment linked to temperature

  • North Atlantic Oscillation (NAO)
  • Atlantic Multidecadal Oscillation (AMO)
  • Surface & bottom temperature

(Behan and Kerr 2022)

20 of 56

3) Exploratory modelling: Growth GAMs

  • Growth: AMO was the most prevalent significant variable between both the condition index and WAA analyses.
    • Significant variables in the condition index model also included bottom temperature anomaly and SSB.
    • Significant variables in the WAA anomaly models also included GSI, SSB, and bottom temperature anomaly

(Behan and Kerr 2022)

Significant WAA anomalies by age class

21 of 56

3) Exploratory modelling: VAST

  • VAST modelling
    • Accounts for changing distribution & catchability
      • Considered depth & temperature
    • Can incorporate multiple survey indices
      • NEFSC bottom trawl
      • ME-NH inshore
      • MADMF survey
    • Depth is main driver

ToR 3. Identify the appropriate survey data to be used in the assessment.

(Hansell et al. 2022)

22 of 56

3) Exploratory modelling: VAST

  • Vector Autocorrelated Spatio-Temporal (VAST) modelling
    • Accounts for changing distribution & catchability
      • Considered depth & temperature
    • Can incorporate multiple survey indices
      • NEFSC bottom trawl
      • ME-NH inshore
      • MADMF survey
    • Depth is main driver

Spring

Fall

(Hansell et al. 2022)

23 of 56

4) ID strong drivers to consider in stock assessment

  • Does the relationship with stock dynamics change over time?
  • Does recent literature support the driver?
  • Is there available data to support the driver’s use in a stock assessment context?

Bottom temperature anomalies drive catchability

Temperature indices drive recruitment

24 of 56

Climate-integrated stock assessment modelling

ToR 1. Identify relevant ecosystem and climate influences on stock dynamics...

ToR 3. Identify the appropriate survey data to be used in the assessment.

ToR 4. Use appropriate assessment approach to estimate annual F, R, SSB.

ToR 5. Update or redefine status determination criteria.

ToR 6. Define appropriate methods for producing projections…

ToR 2. Estimate catch from all sources including landings and discards.

ToR 7. Review, evaluate, and report on the status of research recommendations…

​

ToR 8. Develop a backup assessment approach…

25 of 56

Climate-integrated stock assessment modelling

WHAM

The oods

ole

ssessment

odel

VPA

irtual

opulation

nalysis

ASAP

ge

tructured

ssessment

rogram

SS

tock

ynthesis

Path to incorporate environmental data depends on:

  • Explicit (link environmental driver to stock dynamic)
  • Implicit (inform modeling assumption/set-up)

​

Approach depends on:

  • Stock assess model
  • Type of environmental driver

​

​

26 of 56

Environmental link to recruitment

Implicit

Explicit

Autocorrelated stock recruit relationship

Model environmental process

  • Random walk
  • AR1
  • May or may not include environmental lags

Explore deviations from mean R vs. proportional to stock size

Possible indicators:

  • North Atlantic Oscillation
  • Atlantic  Multidecadal Oscillation
  • Gulf  Stream Index
  • Sea surface temperature anomaly 
  • Bottom temperature anomaly
  • Current direction,  speed, or anomalies?

​

​

(Behan et al. 2022)

27 of 56

Environmental link to natural mortality

Implicit

Explicit

Time-varying M

Model environmental process

  • Effect on mean M
  • Specify ages impacted by environmental covariates (e.g. juvenile mortality)

Age-varying M

Time- and age-varying M

Possible indicators:

  • North Atlantic Oscillation
  • Atlantic  Multidecadal Oscillation
  • Gulf  Stream Index

​

​

​

​

  • Sea surface temperature anomaly
  • Bottom temperature anomaly
  • Predator abundance  time series (e.g., spiny dogfish) or diet  information

(Behan et al. 2022)

28 of 56

Environmental link to catchability

Implicit

Explicit

Time-varying catchability

Model environmental process

  • Impact catchability
  • Can impact a subset of survey indices

Possible indicators:

  • Sea surface anomaly
  • Bottom temperature  anomaly

(Behan et al. 2022)

29 of 56

Environmental link to selectivity

Possible indicators: None

​

BUT evidence of time-variation

30 of 56

Environmental link to selectivity

Implicit

Explicit

Select survey indices that capture trends

No option in WHAM

Time-varying selectivity

Alternative approaches

  • Spatial modeling (e.g. VAST)

(Behan et al. 2022)

Possible indicators: None

​

BUT evidence of time-variation

Selectivity

time blocks

Time-varying selectivity

31 of 56

Environmental link to growth & maturity

Implicit

Explicit

Time-varying growth

Option in WHAM still in development

Time-varying maturity

Alternative approaches

  • Spatial modeling (e.g. VAST)

Possible indicators:

  • North Atlantic  Oscillation
  • Atlantic Multidecadal Oscillation
  • Gulf Stream Index
  • Sea surface anomaly

​

​

​

  • Bottom temperature  anomaly
  • Fishing effort time series

(Behan et al. 2022)

32 of 56

Other paths for including environmental impacts:

  • Model-based indices:
    • VAST
    • CPUE/LPUE
  • Data inputs:
    • Temperature-linked discard mortality (e.g. Weltersback & Strehlow 2013 study on Baltic Sea Atlantic cod)
    • Maturity & weight-at-age

​

ToR 3. Identify the appropriate survey data to be used in the assessment.

ToR 2. Estimate catch from all sources including landings and discards.

33 of 56

Climate-informed modelling using WHAM

0) Model diagnostics

1) Develop a baseline model

2) Explore alternative catch and indices of abundance

3) Incorporate environmental covariates based on ToR1

findings

​

ToR 4. Use appropriate assessment approach to estimate annual F, R, SSB.

34 of 56

0) Model diagnostics for climate-informed models

  • AIC
    • WHAM has features to fit to environmental covariate without specifying a link to stock dynamics as a baseline model for comparison with climate-integrated models
  • One step ahead (OSA) residuals
  • Mean Absolute Scaled Error (MASE)
  • Self-tests & simulations

​

​

35 of 56

0) Model diagnostics for climate-informed models: OSA

One step ahead (OSA) residuals

  • provides independent residuals for correlated observations
  • available for aggregate catch and indices
  • available for age composition

​

ToR4 miller_wham_intro.pptx

36 of 56

0) Model diagnostics for climate-informed models: MASE

 

Naïve forecast residual

Prediction residual

Eq3 from Carvalho et al. 2021, MASE described in Carvalho et al. 2021 and Kell et al. 2021, part of slide borrowed from Brian Stock

37 of 56

0) Model diagnostics for climate-informed models

  • AIC
    • WHAM has features to fit to environmental covariate without specifying a link to stock dynamics as a baseline model for comparison with climate-integrated models
  • One step ahead (OSA) residuals
  • Mean Absolute Scaled Error (MASE)
  • Self-tests & simulations

​

​

38 of 56

1) Develop a baseline model

  • Selectivity
    • Fleet random effects
    • NEFSC dome-shaped selectivity
  • Abundance-at-age random effects (includes Recruitment)
  • Split NEFSC bottom trawl survey in two:
    • Albatross (1980-2008)
    • Bigelow (2009-2019)
  • Logistic-normal age composition likelihood

Fleet selectivity-at-age

Albatross fall age composition

Bigelow fall age composition

(Hart et al. 2022)

39 of 56

2) Explore alternative catch and indices of abundance: VAST

(Hansell et al. 2022)

Incorporates:

  • NEFSC bottom trawl
  • ME-NH inshore
  • MADMF survey
  • Depth

40 of 56

2) Explore alternative catch and indices of abundance: VAST

Fit to spring & fall VAST indices with logistic selectivity assumed

Spring

Fall

Selectivity-at-age when fit to design-based Albatross/Bigelow indices

Spring

Fall

41 of 56

2) Explore alternative catch and indices of abundance: VAST

Design-based Albatross & Bigelow indices:

Fleet OSA residuals

Model-based VAST indices:

Fleet OSA residuals

42 of 56

2) Explore alternative catch and indices of abundance: VAST

Design-based Albatross & Bigelow indices:

Fleet age composition OSA residuals

Model-based VAST indices:

Fleet age composition OSA residuals

43 of 56

2) Explore alternative catch and indices of abundance: LPUE

Optimal model used to generate Landings per unit effort (LPUE) index included:

  • Year
  • Statistical area
  • Quarter
  • Vessel tonnage class
  • Depth
  • Price

(Hankowsky et al. 2022)

44 of 56

2) Explore alternative catch and indices of abundance: LPUE

Fit to NEFSC only (with R random effects)

Fit to NEFSC + LPUE:

Fleet OSA residuals

  • Fleet OSA residuals less normally distributed
  • Catch observed-predicted residuals larger in magnitude
  • Model only converged when recruitment random effect (not full state-space NAA) implemented

45 of 56

Incorporate environmental covariates: catchability linked to BT anomalies�

Shift offshore & into deeper water over time

(Behan and Kerr 2022)

(NOAA Fisheries, 2021)

46 of 56

Incorporate environmental covariates: catchability linked to BT anomalies�

Generally positive relationship between bottom temperature anomaly on catchability

  • As fish move deeper into colder water, catchability decreases – aligns with expectations

47 of 56

Incorporate environmental covariates: catchability linked to BT anomalies�

Generally positive relationship between bottom temperature anomaly on catchability

  • As fish move deeper into colder water, catchability decreases – aligns with expectations

Bigelow spring estimated negative relationship

48 of 56

Incorporate environmental covariates: catchability linked to BT anomalies�

Generally positive relationship between bottom temperature anomaly on catchability

  • As fish move deeper into colder water, catchability decreases – aligns with expectations

Uncertainty based on process assumption

49 of 56

Incorporate environmental covariates: recruitment linked to temperature (e.g. SST, NAO, AMO)

(Behan et al. 2022)

Red = estimated

Blue = observed

SST anomaly linked to recruitment via AR1 process

SST Anomaly

SST Anomaly

Greatest recruits per spawner occurred at extreme cold temperature (Brodziak and O’Brien 2005)

50 of 56

Incorporate environmental covariates: Diagnostic challenges

�

Very different scale compared to runs with link to environmental covariate

Challenging to select between models with environmental drivers

  • AIC not comparable for models fit to different data
  • Fitting to environmental data without link so AIC comparable lead to overdispersed residuals & poor predictions
    • Unable to resolve given time constraints
    • No baseline to compare models with link against

(Hart et al. 2022)

51 of 56

Projections & Reference points

  • Reference points
    • F40% and SSBF40%
    • Full recruitment time series informed

​

​

Not overfished

No overfishing

ToR5 Overview

ToR 5. Update or redefine status determination criteria.

52 of 56

Projections & Reference points

  • Projections with environmental effects
    • Process assumptions
      • Variance of projections asymptotes for AR1 process (recommended)
      • Variance of projections goes to infinity for random walk process
      • Process mattered for catchability models but not for recruitment models
    • Projection assumptions
      • Full time series of used to inform mean recruitment

​

​

(Alade 2022)

ToR 6. Define appropriate methods for producing projections…

53 of 56

Research track take-aways:

53

Science Review (Ecosystem Profile) &

Fishermen’s Ecological Knowledge

Exploratory Modeling

Climate-informed models

Characterize State of Knowledge

Ecosystem context

Improved model assumptions and parameterization

Acknowledgement of non-stationarity

Science

Assessment Modeling

Conduct

Integrated Modeling

Develop Relationships among Climatic-Ocean-Stock Variables

Application

General Approach

X

54 of 56

Research track take-aways:

  • Identified promising mechanistic relationships between climate drivers and stock dynamics
    • Catchability linked to BT anomaly = promising once diagnostic problems resolved
    • Recruitment linked to temperature = may reassess if future recruitment trends change
  • Demonstrated WHAM’s flexibility to account for environmental effects in a number of ways
  • No environmental drivers included in candidate models
  • Passed peer review in July 2022
  • Moved into management track (operational use) Fall 2022

​

55 of 56

Research track take-aways:

​

​

  • New research recommendations
    • Exploration of spatiotemporal integration of federal and state surveys should continue.
    • Continue to monitor shifts in distributions of plaice, particularly the noted increase depth associated with increasing temperature, to evaluate whether there are improvements to model performance by including an environmental covariate on catchability.
    • The relationship between recruitment and ocean temperature should continue to be monitored in future analyses and considered in the context of model fitting (i.e., including environmental covariates) and recruitment assumptions for reference points and projections.
    • To ensure that future model explorations with environmental covariates are comparable via AIC, problems with calculation and comparability of AIC with and without environmental covariate linkage should be resolved.

​

ToR 7. Review, evaluate, and report on the status of research recommendations…

WG Report

56 of 56

Ongoing work

  • Tim Miller et al.: Resolve issue with OSA residuals for environmental covariates

​

  • State-space research track working group:
    • Further testing of WHAM climate-integrated models

​

  • My work: test climate-integrated models for plaice via MSE to ID possible climate thresholds

56