Download ModelSummary - North Pacific Research Board

One model to rule them all?
Talk is about current framework, context.
Many slides of details, equations, set
aside for discussion.
Focus on groundfish, groundfish issues and predator/prey models on “whole
ecosystem” scale and climate links to “whole ecosystem” models and ones
that have been LINKED TO MANAGEMENT PROCESS with Statistical
considerations. “Lessons learned”.
(May discuss some IBMs following Al’s presentation).
Groundfish-centric history of AFSC
“ecosystem” models and issues
• Laevastu 1970s – 1980s
– Stock assessment focus:
– Surplus production Limits on total fisheries production
– (Indirectly) instrumental in developing Optimal Yield
cap (imposed outside of model calculation)
– Model next moves to Polovina to become ECOPATH.
– Meanwhile…
Groundfish-centric history of AFSC
“ecosystem” models and issues
• Stock assessments
– Predation mortality (cannibalism or multispecies) key
to unpredictable interactions (i.e. pollock). Several
examples, beginning Honkalehto 1988.
– Climate becomes an obvious player, incorporated
(correlations, mechanisms)
Groundfish-centric history of AFSC
“ecosystem” models and issues
• Then, all at once: EBFM, Steller Sea Lions, PSEIS
• MSVPA (mainly for predation mortality of key species) JuradoMolina and Livingston
• Pollock prediction models (FOCI) : most valuable species was
least predictable, most eaten by everything.
• Bormicon/Gadget spatial model (never advanced) Livingston
• Ecosim started as S.S.L focused-investigation (EBS found limited
effects on scale of model) Trites and Livingston Now Ecosim for
EBS, GOA, ALEUTIANS used for PSEIS style analyses.
•
Multispecies Bycatch Model
• Steller models and IBMs
• MRM models: Cod/Crab. Atka/Cod/Pollock/SSL.Quinn et al. GOA
Current issues
• Programatic Supplemental Environment Impact Statement
– Development of Ecosystem Considerations, designed to include
trends but also model analyses.
– Ecosystem Considerations are part of stock assessment
process, regular reports to Council.
• Modeling?
– Include CUMULATIVE EFFECTS, OY analyses, etc.
– Model policy tradeoffs: TAC EA and EIS
– BiOp for Stellers
•(Was a draft of a proposal for EcoFore, wasn’t submitted).
Integration with management is key
• Developing and
integrating multiple
models with different
scales and different
simplifying assumptions
Insights gained
• NEED CONTEXT: Integrating ecosystem and habitat
information into current single species management
process
• Integrating ecosystem information with more spatial
habitat information
“Non believer”
“Realist”
Belief in model
“Believer”
Research survey
Abundance data
Biological data:
Catch at age, size
Life history
Commercial fishery
Catch data
Stock
assessment
Ecosystem
Stuff???
Plan Team Review
Initial ABC OFL
Scientific & Statistical
Committee
Final ABC OFL
Advisory Panel
Initial TAC
Public
input
North Pacific Fishery
Management Council
Public
input
Final TAC specifications
Ecosystem Considerations
Organization
I.
–
–
–
Ecosystem Assessment
Summarize historical climate and fishing effects on BSAI and
GOA
Summarize possible future effects of climate and fishing on
ecosystem structure and function (using multi-species and
ecosystem models)–in progress
MODELS for: what do the current indicators mean and
what UNCOLLECTED indicators might be important and
create testable hypotheses.
II. Ecosystem Status Indicators
–
Historical and current status and trends
III. Ecosystem-Based Mngt Indices
–
Early signals of direct human effects
How will multispecies models affect
a catch level?
• Crises
– BiOp (endangered, threatened)
• Strategic – never done explicitly
– Cumulative effects of harvest rates
– Optimum Yield (2,000,000 mt?)
– Tradeoffs between species
• Single species – several avenues, none explict or mandated
–
–
–
–
–
–
Currently part of “ecosystem considerations”, may affect Tiers.
Reference points (F-40)
Informational (qualitative MSE)
Management Strategy Evaluation
Addition of mortality to single species
Adjustment of uncertainty
Needed for ecosystem assessment
• Issue driven or framework driven? (preparation for rapid scenario
investigation in response to issues)
•
Many components or too many compoments? (140+ species, and
not enough)
• Precision or range of hypotheses?
• Statistical preditions/fitting (must be “right”)
• MSE operational models (must be inclusive of hypotheses)
• Uncertainty, and also risk assessment
• Data sensitivity
Conclusions I:
Forecasting and Contingency Planning
• Ecosystem models provide
–
–
–
–
Context of species relationships for current management
Fuller evaluation of uncertainty than single species models alone
Ability to simulate policy options under our control and
Ecosystem change outside our control
• Developing a suite of integrated models
– Can address specific questions at appropriate scales
– Can evaluate uncertainty across and between models
• Current models are in place for contingency planning, and quicker
response during “crisis management”
– Natural disaster
– Lawsuit, EIS analysis
NOAA 5-year plan
• Research milestone under “Scenario Development to Support
Specific Management Actions and Decisions”
“Develop the next generation of multi-species fisheries and food
web production models (3-5 years).”
• WHAT ABOUT NOW?
– This talk focuses on operational or near-operational models (Alaska
examples).
– Operational = quantitative, for direct use in management context
(either tactical or strategic).
– Primarily predator prey or predator/prey/climate, stock or regional
scale, limited spatial resolution.
– MANY exploratory, in-development, not predictive in stock-assement
context models exist or are in development.
State of the art
Similar predictive issues to stock assessment
• Functional response / model error?
• Similar to issues Ricker v. Beverton-Holt in singlespecies management, fitting and robust statistical
techniques are part of the solution.
• Recruitment, climate drivers, gear/bycatch changes,
management/economic effects not predictable (but these models
can be driven by such inputs if the scenarios are described).
• Current solutions: test hypotheses with “external forcing”, test
management responses for robustness.
• Next improvements should explore these forcings with COUPLED
models of proposed mechanisms. But needs careful consideration
of WHAT is predictable and WHEN it’s predictable! (Flatfish may be,
pollock sometimes, herring ever? Some key characteristics and
tools shown later may define which will work right now).
Model Definitions
•
Biomass dynamics
– Ecopath with Ecosim
– Elseas (AFSC version of Ecosim) – CIE Review
– GEEM
•
Age-structured interactions
– MSVPA –CIE Review
– Bycatch Interaction Model
•
Individual interactions
– IBMs (Hinckley et al.)
– Bioenergetics
– Minimum realistic models: Atka/Cod, Cod/crab, GOA (Terry Quinn)
•
Spatial interaction models (currently don’t exist)
– Atlantis
– GADGET
•
•
Highlighed are most advanced “statistically” (e.g. in stock assessment
sense)
SLIDES CAN BE SHOWN AS NEEDED FOR SEVERAL Others OF THESE
<-statistical rigor but model error
SS -> MRM ->Bycatch -> MSVPA/MSM -> Ecosim -> Gadget -> Atlantis
detail and hypothesis inclusiveness ->
• All of the above could develop more statistically.
• Do we need to choose one, or do we need a framework for
uncertainty, to make blended model predictions?
• Why general models? You can’t start building specific models in a
crises, but can spawn them off a big one.
“This Generation”
Multispecies & multimulti-fisheries management
Fisheries
• Multispecies Bycatch Model
– Gear interactions, age structured single
species dynamics, simple bycatch, complex
management scenarios
– (no predator/prey links)
• MSVPA/MSFOR/MSSAM
– Multispecies age structured predator/prey
for 7 target species, adds explicit predation
effects to recruitment hindcasts
– (no bycatch or full system effects)
• Ecopath/Ecosim
– Complex system effects on non-target and
protected species biomass dynamics, gear
interactions, simple management scenarios
– ELSEAS: full age structure for key species
(forward fitting).
Multiple species/stocks
•
Current multispecies models
– “Stock assessment” scale.
– Calibrated from annual or quarterly data, may be subregional parameter
partitioning (but no “true” spatial movement).
– Main units are “stocks” on a “management/survey region” scale.
• IBM or NEMURO/Fish: Explicit spatial movement, no population
closure.
Council areas
First: Information requirements
Standard stock assessment data
– Biomass or abundance index
– Productivity information
Fishery observation
– Commercial catch
– Incidental catch and discards
Food habits collections
– Multiple species and trophic levels
– Multiple seasons
Spatial resolution of data
Why Multispecies? SHIFTING BASELINES
Biomass at F0 (change from reference B2002)
70%
Single Species
Ecosim
MSFOR
60%
% change
50%
40%
30%
20%
10%
0%
-10%
pollock
cod
g turbot
yf sole
r sole
herring
at flounder
-20%
Species
Key to uncertainty: this is MEDIAN % change over multiple climate
scenarios and error range for inputs (error bars not shown).
MSM Estimates of M2 (age-1 pollock)
Predation mortality (t-1)
0.6
0.5
0.4
0.3
0.2
0.1
0
1975
1980
1985
1990
Year
1995
2000
2005
Population-scale: dB/dt = ???
Top-down
control
Bottom-up
control vs.
prey
switching
Life-history:
Age structure and
energetics
Life-history:
Recruitment (younger
ages than SS)
MSFOR
Set by age
structure
100% prey
switching
Explicit age-structure,
energetics shift with agestructure
External, S-R relationship
possible.
Ecosim
Set by foraging
parameter
100% bottomup
Juvenile/adult to year class
for some species,
energetic bias of “current”
age structure for others
“Emergent” from
parameter apportioning
food to growth vs.
fecundity
Blended
dynamics/
bioenergetics
Age structure
proxy plus prey
selectivity
parameters.
Set by satiation
parameter.
Explicit age-structure,
energetics shift with agestructure
External, S-R relationship
possible.
GEEM
Economic choice model produces Type
II and Type III as emergent properties
of energy maximization.
Biomass dynamics only, no
age structure.
No explicit recruitment.
ALL RESPONSES ARE NONLINEAR, HARD TO SIMPLFY.
Only way to evaluate these is to fit historical data, or test results for
robustness over a range of parameter values.
SENSITIVITY RESULTS Ecopath food webs
(EBS, GOA, AI)
EBS food web
Quantifying uncertainty
Data
quality
Results for use in
stock assessments
Web-based format for results
• Example: The
ecosystem role of
Pacific cod varies
between systems,
especially in the AI
(not in proportion
to biomass).
– Quantifying impacts
and uncertainties from
removal: Pacific cod
example.
– Quantifying factors
influencing target
species: Pacific cod
example.
Forage fish
v.
squid
Common data issues
•“Tier 4-6” resolution of individual stocks
OBSERVATION
•Need continued food habits input (update ~5 years).
•Need off-survey food habits data, especially late summer/early fall.
•Need bycatch reconstruction, updating. Bycatch in non-NMFS fisheries?
PROCESS
•Recruitment and climate drivers.
•Euphausiid variability/uncertainty is a key bottom-up effect in all systems.
•Deepwater food web (sablefish, grenadier, squid) poorly understood.
•Squids are key, poorly understood component.
Not all species are equally predicable! (Why did we start with pollock??)
Scales differ: (switch driven species by local studies, longer life-histories by
correlative studies, etc.)
LEADS TO PREDICTABILITY STUDIES…
Gulf of Alaska groundfish
4,000,000
Pollock
P. cod
Arrowtooth
Halibut
1990-1993
snapshot
2,000,000
1,000,000
year
2000
1990
1980
1970
0
1960
biomass (t)
3,000,000
Fitting using fishing, pollock recruitment—all series
Predictability 1: hypotheses testing?
• GOA – all have support – Final AIC under review, each fit in
“different ways”, each give wildly different “best” functional
response types and fits (ecosystem states).
– Climate link to primary production (Ecosim)
– Climate link to pollock recruitment (Elseas)
– Fishing with arrowtooth recovery from POP fishery (assumes
present-day bycatch q in that fishery)
• EBS has food data available to do functional fitting and
distingusih hypotheses, GOA data is limited for this purpose.
Food web networks
GOA 2001:
307 nodes
1011 links
EBS 2000:
277 nodes
1027 links
Predictability 2: Structural type  Biological Regime:
“Climate” forcing experiment
-0.4
0.00
0.05
Power (cv)
0.10 0.15
forcing anomaly
-0.2
0.0
0.2
0.20
0.4
0.25
• Band-pass filtered (white noise at annual and higher
timescales) for all species
0
20
40
60
years
80
100
0.2
0.5
2.0
5.0
20.0
Period (years)
50.0
200.0
0
20
40
60
Time (years)
80
100
0
20
40
60
Time (years)
80
100
Structural type  Biological Regime: The Punchline
7
GOA
0
0
1
10
Density (t/km^2)
2
3
4
5
Density (t/km^2)
20
30
6
40
EBS
40
60
Time (years)
80
100
0
20
40
60
Time (years)
45
16
40
14
35
80
100
12
30
10
25
8
20
6
15
4
10
2
5
1999
1994
1989
1984
1979
1974
1969
2000
1996
1992
1988
1984
1980
1976
1972
1964
-
0
1968
20
1964
0
EBS
EBS
GOA
GOA
0.00 0.05 Power
0.10 (cv)0.15 0.20
0.2 0.4 0.6 0.8 1.0 1.2
Bottom-up
Combined
Power (cv)
Juv. pollock
1
1
2
2
5
5
10
20
50
10
20
50
Period
Period(years)
(years)
100 200
100 200
500
500
10
20
50
Period (years)
100
500
Adu. pollock
0.0
Power (cv)
0.5
1.0
GOA food web is less predictable
than EBS for adult pollock: this is
emergent property.
SCALE for data-needs differs and is
species and ecosystem specific!
Combined
1.5
EBS
GOA
1
2
5
200
Linking models: Ecosim and
scale??
SCALE doesn’t work for our purposes (linking
ecosystems, investigating climate interactions in
species with strong seasonal, migratory, or
bottleneck dynamics).
Ecosim on its own works well for stock-scale top-down
perturbations (fishing) on relatively closed
populations, NOT for mechanisms: in particular,
VERY poor for Pacific salmon.
Latitude (°N)
NEMURO vs. ECOSIM
Alaska Gyre
Alaska Stream
Alaska Current
'OSP'
Subarctic Current
Subarctic Frontal Zone
California
Current
Longitude (°W)
Biomass/Average Biomass
2.5
Copepods
2
Microzoop.
Large phyto.
1.5
Small phyto.
NEMURO output drives
seasonality in primary and
secondary production in Ecosim
1
0.5
0
J
F
M
A
M
J
J
A
S
O
N
D
J
F
M
A
M
J
J
A
S
O
N
D
J
Month
Biomass/Average Biomass
2.5
NEMURO Pred. zoop
Ecosim Euphausiids
2
1.5
RESULT:
NEMURO predatory zooplanktion
vs. Ecosim Euphausiids
1
0.5
0
J
F
M
A
M
J
J
A
S
O
N
D
J
Month
F
M
A
M
J
J
A
S
O
N
D
J
Latitude (°N)
NEMURO vs. ECOSIM
Alaska Gyre
Alaska Stream
Alaska Current
'OSP'
Subarctic Current
Subarctic Frontal Zone
California
Current
Longitude (°W)
2.5
Biomass/Average Biomass
Micronek. squid
Mesopel. fish
Lg. Jellyfish
2
Ctenophores
Salps
1.5
1
Ecosim gelatinous zooplankton
and forage fish
0.5
0
J
F
M
A
M
J
J
A
S
O
N
D
J
F
M
A
M
J
J
A
S
O
N
D
J
Month
Biomass/Average Biomass
2.5
Sharks
Flying squid
2
Pink
Coho
1.5
1
Large predators, and salmon
(???)
0.5
0
J
F
M
A
M
J
J
A
S
O
N
D
J
Month
F
M
A
M
J
J
A
S
O
N
D
J
Didn’t capture salmon well, so
made model linkages
Pink salmon bioenergetics model,
predicts daily pink salmon growth and
numerical mortality based on input
ration.
Consumption and mortality
rates for Pink salmon based
on predator and prey biomass.
Pink salmon body weight and
numbers used to set Ecosim biomass
for predator and prey equations in
next timestep.
Ecosim (ecosystem
biomass dynamics model),
run on a daily timestep.
No direct feedback to NEMURO:
Ecosim parameters for predatory
zooplankton (euphausiids) tuned to
match NEMURO predictions for same
species.
Daily biomass density of phytoplankton,
microzooplankton, large zooplankton
(copepods).
NEMURO (nutrient-phytoploankton-zooplanktondetritus): 1-dimensional water column model
integrated on an hourly timestep.
Pink salmon growth
1800
1600
Body
N
Body weight (g)
1400
weight
(g)
s.d.
(million
s)
N/k
m2
t/km2
Jul
31.8
5.8
1,159
320
0.010
600
Aug
39.5
27.5
945
261
0.010
400
Sep
58.1
25.5
770
213
0.012
Oct
138.7
40.7
628
173
0.024
Nov
145.7
38.5
512
141
0.021
Dec
172.9
81.4
418
115
0.020
Jan
154.9
36.7
340
94
0.015
Feb
318.5
93
278
77
0.024
Mar
408.3
153.7
226
62
0.026
Apr
584.5
184.8
185
51
0.030
May
777.6
232.3
150
42
0.032
Jun
919.0
252.3
123
34
0.031
Jul
1128.9
322.5
Aug
1320.6
336.6
Sep
1523.1
397.6
1200
Month
1000
800
200
0
Jul
Aug
Sep
Oct
Nov
Dec
(Ishida et al. 1998)
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Additional data
16
14
OSP temperature
SST (°C)
12
10
8
6
4
2
Month
Prey group
cal/g wet
weight
species and range
Copepods
700
Neocalanus cristatus, 627-748
Euphausiids
1000
Thysanoessa spp., Euphausia spp., 840-1050
Pteropods
650
Limacina helicina; 624-940
Amphipods
800
Parathemisto pacifica; 852-1010
Ctenophores
50
Beroe sp., 47
Salps
36
Salpa sp., 36
Chaetognaths
450
Sagitta elegans; 455-488
pelagic forage
fish
1200
mesopelagic
forage fish
2000
Stenobrachius leucopsarus; Tarletonbeania crenularis;
Leuroglossus schmidti; 2041-2365 (Bering Sea)
Micron. squid
1500
Berryteuthis anonychus; 1307-1737 (increases with
increasing mantle length).
Gasterosteus aculeatus; 1166-1533
Prey quality
Dec
Nov
Oct
Sep
Aug
Jul
Jun
May
Apr
Mar
Feb
Jan
0
Results
2.5
Euphausiids, amphipods and pteropods
Copepod
Micron. squid
biomass/reference biomass
2
1.5
1
0.5
biomass/reference biomass
0
1.2
A
S
O
N
D
J
F
M
A
M
J
J
A
S
O
N
D
J
F
M
A
M
J
J
A
month
1
0.8
0.6
0.4
Pink salmon biomass
0.2
0
pink salmon body weight (g)
1400
A
S Empirical
O
N
1200
D
J
F
M
A
M
J
J
A
S
O
N
D
J
F
M
A
M
J
J
A
O
N
D
J
F
M
A
M
J
J
A
month
Modelled
1000
800
600
400
200
0
A
S
O
N
D
J
F
M
A
M
J
J
A
month
S
Matching growth rates
Run calibrated for “fast growth” (fits
base growth rate to later data points)
Original run
pink salmon body weight (g)
1400
Empirical
1200
Modelled
1000
800
600
400
200
0
A
S
O
N
D
J
F
M
A
M
J
J
A
month
S
O
N
D
J
F
M
A
M
J
J
A
Sensitivity analysis
Original run
pink salmon body weight (g)
1400
Empirical
1200
Modelled
1000
800
600
400
200
0
A
S
O
N
D
J
F
M
A
M
J
J
A
S
O
N
D
J
F
M
month
% change in input variable
Parameter
Base
+10%
-10%
Salmon entry day
228 (Julian day)
-0.8%
+1.1%
Starting body weight
39.5 g
+3.8%
-4.1%
Average water temperature
8.68 °C
-2.8%
+1.1%
Seasonal temperature
amplitude
3.16 °C
-1.2%
+1.1%
Prey cal/gram wet weight
Varies by diet
item
+24.3%
-22.1%
% change in final pink
salmon body weight
A
M
J
Diet switching?
100%
Diet % by weight
90%
80%
Other
70%
60%
Fish
50%
Squid
Amphipods
40%
30%
Pteropods
Euphausiids
20%
Copepods
10%
0%
625
(n=3)
875
1125
1375
1625
1875
2125
(n=65) (n=92) (n=125) (n=137) (n=66) (n=26)
2375
2500
(n=10)
(n=5)
Body weight category
1
Proportion squid in diet
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
500
1000
1500
2000
Body weight (by category)
2500
3000
Finally: mixed layer depth
Original run
pink salmon body weight (g)
1400
Mixed-layer depth
Empirical
1200
Fast growth
Modelled
1000
800
600
400
200
0
A
S
O
N
D
J
F dependence
M
A
M
J
Density
J
A
S
O
N
D
J
F
M
A
M
J
J
month
Concentration of prey density inversely proportional to mixed-layer depth.
A
Model
WHAT NEXT?
Biogeography: N/S or E/W??
Circulation connecting gyres (squid and
other species).
More detailed salmon models with more
detailed density-dependence: MULTIPLE
BOTTLENECKS model.
More realistic climate scenarios (eg.
changes in mixed-layer depth -> NEMURO
-> food webs in coupled models).
What made this successful: Key feature:
Can swap out components
Some specific modeling data needs
(example)
• Movement (float tracking, etc) for development of spatial models
(Stockhausen in progress).
• Local scale studies (hotspots, especially for pollock and other
forage).
• Interaction matrices (diets, bycatch, economics).
• The rest are issue specific (analyses can be provided).
• Ecosystem Considerations needs, especially climate/fish links
(e.g. flatfish) (come back/talk to Boldt).
Collaboration Lesson
•
Explicit between management/decision-bodies as well as scientists: REFM,
NMML, AK, Fish and Wildlife
•
Data SYSTEM
•
RESULT Framework for nesting and feedback of nesting (getting together
data within the center takes a long time).
•
Scientists from multiple fields
•
Workshop-driven construction processes are a good thing, multiple people
at table.
PSEIS mandate has driven GOOD integration and team-building within
groundfish community, models have helped focus that quite a bit. But
it was a painful process (time). Next level, integration with marine
mammal management is limited by different mandates and cultures.
Framework needs
• Predictive framework
• Results exchange/feedback
• Data framework