The Contribution of Coupled PhysicalBiological Models CPBM to Understanding Fish Recruitment: Critic

1
The Contribution of Coupled Physical-Biological
Models (CPBM) to Understanding Fish Recruitment
 Critical Review and Prospects

• Thomas J. Miller
• Chesapeake Biological Laboratory
• University of Maryland Center for Environmental
  Science
• Solomons, MD 20688
• USA

2
Why CPBM?

• Hypotheses regarding recruitment
• Food-mediated
• Hjorts critical period hypothesis
• Cushings match-mismatch
• Laskers stable ocean hypothesis
• Rothschild and Osborns plankton contact
  hypothesis
• Transport-mediated
• Hjorts second hypothesis
• Harden-Jones migration triangle hypothesis
• coastal conveyor belt hypothesis
• Sinclairs member vagrant hypothesis
• Predation-mediated
• Bigger is better hypothesis
• Stage duration hypothesis

3
Approach

• Literature review based on keyword search of ISI
  database
• Unbiased but not comprehensive
• Focus on larval stages
• 64 articles (1993 2005)
• Categorized articles with respect to
• Modeling format
• System
• Model application
• Explanatory
• Inferential
• Hypothesis-generating

4
Trends in publications
5
Geographic distribution of studies
6
Species representation
7
Model format
8
Model resolution
9
Biological processes

• Most models track the age and position of eggs
  and larvae
• Feeding
• Often implicit
• When explicit, prey field often not dynamic
• Growth
• Temperature-mediated
• Bioenergetic (feeding)
• Mortality
• Age-, size- or growth-dependent
• No functional response
• Behaviour
• Vertical migration
• Light-dependent or turbulent-dependent feeding

10
Modelling approach
H0-generating 11

• Explanatory models
• Comparison between observed and predicted often
  qualitative
• Assessment of sensitivity rare
• Inferential models
• Alternative mechanisms / explanations considered
• Hypothesis-generating models
• Formal testable H0
• Tested within model
• Tested external to model

Inferential 31
Explanatory 58
11
ExplanatoryWerner et al. 1993

• 3D FEM with adaptive mesh grid and sigma depth
  coordinates
• Virtual cod eggs released on NE crest of
  Georges bank at range of depths
• Passive and swimming particle
• No formal statistical comparisons with field
  observations, but release at 50 m lead to
  retention on bank

12
ExplanatoryVoss et al. 1999

• 3D FEM model with 5 km grid and 28 depth levels
  (6m)
• Virtual cod released at times observed in field
  surveys and tracked for 21 d.
• Contours of observed and predicted larval
  distributions compared visually.

13
InferentialBrickman and Frank 2000

• 3D finite difference (QUODDY) model
• Tracked distribution, development and abundance
  of virtual haddock
• Two alternative mortality models
• Discrepancy between constant and stage-specific
  mortality model for Browns Bank implicates role
  for stage-dependent mortality

14
InferentialFiksen and MacKenzie 2002

• IBM foraging model with readily interpretable
  biological and physical parameters
• Implemented within a coupled biological physical
  model of Georges Bank
• 3D FEM model with adaptive mesh grid and sigma
  depth coordinates
• Model predictions contrasted to output from
  Werner et al. 1996
• Inferences regarding likelihood of prey
  limitation

15
Hypothesis-generatingMullon et al. 2002

• 3D ROMS of Benguela ecosystem
• Spawning period and site has evolved to maximize
  transport to nursery ground
• Hypotheses
• Transport
• Competency
• Thermal preference

16
Hypothesis-generatingQuinlan et al. 1999

• 3D FEM model with adaptive mesh
• Menhaden should recruit to mid-Atlantic estuaries
  from the same site at predictable ages
• Test Otolith microchemistry, age and birthdate
  frequency

17
Rationale for individual-based approaches

• Recruitment is determined by how many grow and
  survive to be in the right place.
• IBMs developed in ecological literature to model
  recruitment, recognizing the importance of
  individual heterogeneity in vital rates and
  states, e.g. size-selective mortality in
  determining fates
• However, in CPBM
• Little attention to intracohort variability
• Most variability is spatially-derived
• Growth trajectories not fully developed
• Little attention to mortality

18
Spatial differences Is the grid scale sufficient

• What is the appropriate spatial and temporal
  scale?
• Pepin and Helbig (2002) resampled HF radar data
  to drive a circulation model at different grid
  scales
• Conclude precision depends on spatial more than
  temporal scale the finest scale features must
  be resolved
• Assuming spatial scale is appropriate, spatial
  variability in larvae is generated by sub-grid
  scale processes
• Accuracy of sub-grid predictions

19
Spatial differences Larval behaviour repertoire

• IBM results indicate individual behaviours matter
• Yet, larval behaviour in the field is largely
  unknown
• Most data come from depth-stratified samples
  which track the average, not the individual
• Some laboratory data is available, but on limited
  ontogenetic and taxonomic scales

20
Parameterizing growth

• Critical importance of growth given the
  prevalence of size-dependent processes
• Yet only 38 of studies included growth
• Two approaches
• Heathian use temperature to drive growth
  because of sub-grid scale concerns
• Wernerian specifically model encounter and
  feeding process

From Rice et al. (1993)
21
Heathian growth modeling

• Advantage
• No need to model prey
• Temperature likely well predicted at sub-grid
  scales
• Disadvantages
• Implicitly assumes no prey limitation
• But see Bartsch (2002)
• Accuracy of underlying temperature-growth models
  Folkvord (2005)

22
Wernerian growth modeling

• Advantages
• Explicitly links prey and environment with
  growth
• Generated inferential approaches
• Disadvantages
• Sub-grid scale concerns
• Encounter and prey selectivity processes not well
  understood
• Coupling larval and prey models

23
The mortality problem

• The average larval fish is dead!
• Field studies indicate strongly size-selective
• Measuring mortality rates sufficiently accurately
  to predict recruitment is impossible
• Hindcasting approach characteristics of
  survivors
• Mortality in models
• Only 30 of reviewed papers included mortality.
  Why?
• Increases the number of particles needed to be
  tracked to get a valid sample (U-I problem -
  Brickman and Smith, 2002)
• Resampling algorithms (e.g., super-individuals
  (Sheffer et al. 1995). Used in only 2 reviewed
  papers

24
Characteristics of survivors
91-92

• Are survivors lucky or adapted?
• Birthdate reconstruction
• Growth rate reconstruction (e.g., Meekan and
  Fortier, 1996)
• Longitudinal analysis of survivors (Miller 1997)
• Few CPBM have learned this lesson
• Overcomes Brickman and Smiths (2002) U-II
  problem
• Removes the mortality problem, if forecast and
  hindcast models are compared

92-93
Growth rate (mm/d)
Age (d)
25
Prognosis

• Coupled biological-physical modeling approaches
  are maturing
• Early focus on exploratory approaches
• Inherent risk in exploratory approaches in
  getting the right answer for the wrong reason
• In the future we should
• Focus beyond distribution and transport to
  include non-passive stages
• Improved understanding of reliability of
  algorithms to model sub-grid scale processes
• Use hierarchies of models to determine the
  contribution of different processes to outcomes
• Quantify effects of uncertainty in parameter
  estimates
• Increased used of experimental design approaches
  to understanding sensitivity
• Shift toward inferential and H0 generating
  approaches that yield testable hypotheses
  relating to recruitment processes is desireable

26
(No Transcript)
27
Mullon et al. 2002 - ctd