andré e. punt 1 school of aquatic and fishery sciences, uw 2 csiro marine and atmospheric research...
TRANSCRIPT
André E. Punt1School of Aquatic and Fishery Sciences,
UW2CSIRO Marine and Atmospheric
Research
How has Strategic Advice Been Usedin a Global LMR Context: A GlobalPerspective on How to Deal With
Ecosystem Model Uncertainty
Outline Some definitions (to provide context). A process for strategic evaluation.
Assigning plausibility weights Case studies I & II (environmental drivers of
recruitment) Case studies III & IV (trophic interactions & MRMs) Case study V (whole of ecosystem models)
Examples International Whaling Commission:
Aboriginal and commercial whaling. Australia:
Management of the SESSF. South Africa:
Penguins and anchovy. USA:
Evaluation of the GOA pollock harvest control rule.
Definitions andcontext
Tactical Advice -> What is next year’s catch limit for pollock?
Strategic Advice -> How well does the approach we use for determining next year’s catch limit for pollock perform relative to a set of agreed management objectives
Tactical Advice -> A number.
Strategic Advice -> A set of trade-offs.
Strategic Evaluation
For this talk a “strategic evaluation” asks the question:
How well does a set of tactics (monitoring, assessment,decision making) achieve a set of (agreed) managementgoals.
Strategic evaluation is not:
• How to determine what the goals should be?• Perfect knowledge analyses• Constant F projections
An “Ecosystem model”:
Anything which is NOT a single-species, single-area,population dynamics model driven by random perturbationsin recruitment and fishery selection.
Standardmodel
Environmentaldrivers
Trophicinteractions
Spatialstructure
Non-fisheriesdrivers
Elements of a Strategic Evaluation (aka MSE):• A set of management goals (appropriately quantified).• A set of candidate strategies to evaluate.• A set of operating models which span the space of possible realities.
Uncertainty arises because of:• model uncertainty (is our model right?).• process uncertainty (are the parameters constant?). • parameter uncertainty (given a model, can we estimate its parameters?).• implementation uncertainty (given a management decision, can we implement it as anticipated?).
The Indirect Approach
Time
Ca
rrying
capa
city(o
r natura
l mo
rtality
)
Simulation trialswere run for changes
in climate in anindirect way
[IWC testing of its revised management procedure]
A Process for Strategic Evaluation
Qualitative managementobjectives (aka the M-S Act)
Quantitative performancemeasures
Hypotheses for systembehaviour
Models of systembehaviour
Data andpriors
Models weights
Candidatestrategies
Strategyranks
Systemsimulation
A model weighting scheme
How strong is the basis for the hypothesis; in the actual data for the system under
consideration; in the actual data for a similar system; for any system; or in theory.
After Butterworth et al. (1996); Rep. Int. Whal. Comm 46: 637-40.
An IWC interpretation-I Step 1 of the previous scheme requires a
belief in the objective function (aka AIC, DIC, etc.); this is rarely possible.
The IWC approach: Assign each hypothesis (model) a rank of ‘high’,
‘medium’, ‘low’ or ‘no agreement’ using a “Delphi” approach.
Each rank is associated with an agreed (conservation) performance standard.
An IWC interpretation-II
What makes a hypothesis “low” plausibility? Obvious conflict with actual data. Obvious conflict with auxiliary information.
Quantitative Tools for Model Weighting
In order of relative ease: Fit diagnostics (observed versus predicted
data; residual plots, q-q plots, etc). Sensitivity tests Variance estimates
Bayesian; Bootstrap; delta method
Case Studies I & II
Environmental Drivers of Recruitment
Incorporating climate forcing(An empirical approach)
1960 1970 1980 1990 2000
-2-1
01
23
Pacific Decadal Oscillation
Year
Link torecruitment
Climate indices
Age-structured operating model
Management Strategy
TACData
“Climate”Decision rule??
Gulf of AlaskaPollock
A’mar et al. (2009); IJMS 66: 1614-32
Data from surveys andthe fishery
Stock assessment model
Target and limitreference points
Stock size, productivity
Fis
hing
mor
talit
y re
lativ
e to
F35
%
Stock size relative to SB47%
Acceptable BiologicalCatch (ABC)
The performance of this approach to setting TACcan be quantified in terms of:
• high stable catches;• low probability of reducing stock size to undesirable (low) levels; and• accurate and precise estimates of biomass (and status relative to target biomass levels). [essentially hindcast skill]
0 2000 4000 6000 8000 10000
020
0040
0060
0080
0010
000
R2 0.36
What drives pollock recruitment?
Kendall et al. Fish Ocean (1996)
Predicted recruitment(with environment)
Est
imat
ed r
ecru
itmen
t(f
rom
ass
essm
ent)
0.0
0.1
0.2
0.3
0.4
2010 2020 2030 2040 2050
Spa
wni
ng b
iom
ass
Year
0.0
0.5
1.0
1.5
2010 2020 2030 2040 2050
Spa
wni
ng b
iom
ass
(rel
ativ
e to
tar
get)
Year
050
100
150
200
250
2010 2020 2030 2040 2050
Cat
ch (
'000
t)
Year
Performance when:• the assessment is (almost) correct• recruitment varies about a mean
• the stock is left above the target and the average catch is ~ 150,000t.
0.1
0.2
0.3
0.0
0.4
0.8
1.2
2010 2020 2030 2040 2050
Spawning biomass:• Generally downward• Depends on model for forecasting future climate (two of eight IPCC models)
Year
Spa
wni
ng b
iom
ass
5010
015
020
00
250
500
750
2010 2020 2030 2040 2050
Year
Cat
ch
Spawning biomass:• Generally downward• Depends on model for forecasting future climate (two of eight IPCC models)
Catches:• React faster than abundance, especially for a declining resource.
Uncertainty Model uncertainty
Choice of IPCC model Relationship between environmental indices and
recruitment Process uncertainty
Variation in recruitment about the assumed relationship
Estimation uncertainty Parameter uncertainty (Bayesian analysis)
Eastern North Pacific Gray Whales
Brandon and Punt (2009): IWC Document SC/61/AWMP2
Eastern North Pacific Gray Whale
Ice conditions in the Bering Sea have beenpostulated to impact calf production.
Objectives and Strategies Objectives
Satisfy aboriginal need (Russia and the US) Achieve stock conservation objectives
Management strategy (default) Surveys (of absolute abundance) every 5-10
years. Strike limits based on the IWC’s “Gray whale
SLA”.
Previous Assessment
With climate
Performance Evaluation
Model uncertainty:• Sea-ice impacts calf production• Future catastrophic events are:
• random• related to population density.
Process uncertainty:• Random variation in calf production.
Estimation uncertainty:• Parameters are based on Bayesian estimation.
Other Studies Rock Lobsters off Southern Australia Pacific Sardine off the west coast of the US
Cases StudiesIII and IV
Trophic Interactions
(MRMs)
MRM Types Biological interactions
Competition, predation, etc.
Technical interactions Interactions through bycatch.
Anchovy and Penguins
How does penguin breeding success and adult survival depend on the abundance of
pelagic fish?
How does penguin breeding success and adult survival depend on the abundance of
pelagic fish?
Penguins as output statistics
0
1000
2000
3000
4000
5000
6000
1984 1989 1994 1999 2004
tho
usa
nd
to
ns
Cape Columbine
Robben
Island
Dassen
Island
Dyer Island
Cape Agulhas
20°E18°E
33°S
Stony Point
Seal Is.
Cape Town
Malgas, Marcus,
Vondeling &
Jutten Islands
Boulders
Geyser Is.
Cape Columbine
Robben
Island
Dassen
Island
Dyer Island
Cape Agulhas
20°E18°E
33°S
Stony Point
Seal Is.
Cape Town
Malgas, Marcus,
Vondeling &
Jutten Islands
Boulders
Geyser Is.
Anchovy and sardinecontrol rule
Uncertainty (sardine and pilchard) Model uncertainty
Stock-recruitment relationships
Process uncertainty Variation in recruitment Variation in bycatch rates
Estimation uncertainty Quantified using bootstrapping
Gulf of AlaskaPollock
A’mar et al. Fish. Res. (Submitted)
GOApollock
Arrowtoothflounder
PacificHalibut
Pacificcod
, , max
, ,,max max1
i a i i y yi y a
i i y
V B B BM
B B B
{Predator functionalrelationship
Pollock harvest policy
Predatorharvestpolicy
(const F)
M really isn’t constant it seems…
Age
Year
24
68
1012
14
1960 1970 1980 1990 2000
Age
Year
24
68
1012
14
1960 1970 1980 1990 2000
Age
Year
24
68
1012
14
1960 1970 1980 1990 2000
Type I
Type II
Type III
Uncertainty Model uncertainty
With / without predation mortality Predator feeding relationship Fishing mortality on the predators
Process uncertainty Variation in recruitment
Estimation uncertainty Parameter uncertainty (Bayesian analysis)
Other Studies Predator-prey interactions:
SSLA for krill management (CCAMLR) Cod and minke whales in the Barents Sea
Technical interactions: Hake off South Africa. Coral trout and red throat emperor off the Great
Barrier Reef, Australia. Prawns off Northern Australia.
Cases StudiesV
Whole of System Models
South East AustraliaWhole of System
Review
Beth Fulton, pers. commn
SE Australian Atlantis-I
EEZ
Claimable shelf
Aim: To rethink management arrangements in the SESSF
Complications:1. Multi-everything2. Relatively data poor3. Many objectives
Atlantis:1. Physical component.2. Biological component.3. Assessment component.4. Management component.5. Social component.6. Non-fishing impacts.
SE Australian Atlantis-II
Advantages:• Considerable “realism”• Appeals to decision makers
Key difficulties:• Driven to an unknown extent by assumptions• Very difficult to calibrate• No variance estimates (ever) relative performance of high level policies (at a PEIS level; perhaps even beyond “strategic”).
Calibration Tests for Atlantis-I
Observed and predicted diet composition for gummy shark
Calibration Tests for Atlantis-II
Forecast basedon Atlantis
Uncertainty Model uncertainty
Productivity / susceptibility – alternative parameterizations.
Structural sensitivity (loop analysis; social network theory).
External forcing scenarios. Process uncertainty
Emergent property of the model. Estimation uncertainty
In a formal sense - N/A (ever?)`
Uncertainty of StrategicEvaluation
(Adoption, Uncertainty, and the State of the Art)
Strategic Evaluations(directly used!)
HakeAnchovySardine
Rock lobster
SardineMackerel
cod
Rock lobster
Toothfish
Minkewhales
Overall Summary(the State of the Art?) Sensitivity tests / model scenarios
IPCC data sets (pollock) Productivity scenarios (Atlantis) Predation functions (pollock)
Process uncertainty Climate-recruitment (pollock) Ice coverage – birth rate (gray whales)
Variance estimation Gray whales, pollock, etc.
How are strategies based ecosystem models used? USA
Pollock A requirement for (continuing) MSC certification. Presented to the NPFMC SSC (but validates current management
strategy). Pacific Sardine
Included in the PFMC CPS FMP
IWC Aboriginal subsistence whaling and commercial whaling
management schemes all tested accounting for “ecosystem changes”
The ENP gray whale analysis will form (part of) the basis for the review of the current Strike Limit Algorithm for gray whales in 2010.
How are strategies based ecosystem models used? South Africa (OMPs have legal status)
Penguin model currently “on hold” while it is being refined. Management strategies for sardine and pilchard have
taken technical interactions (and between sector-allocation) into account for over a decade.
Australia Used to “guide” decision making rules. Atlantis provided direction that helped set policy directions
in SESSF (gears, spatial, quota, etc.)
Pre-pre-Implementation Assessment (1)
First Intersessional
Workshop
Agree completed at an Annual Meeting
First Annual Meeting
Pre-Implementation
Assessment (2+)
Second Intersessional
Workshop
Second Annual Meeting
Option or options presented to the Option or options presented to the CommissionCommission
Catch limit?Catch limit?
Commission
2 years
What is actually necessary toprovide strategic advice? The Objective:
How robust are the current / alternative strategies (note that strategies which are “deterministically optimal” will not necessarily be given uncertainty).
Stakeholder Buy-in: Most successful applications are associated with strong
stakeholder involvement: Workshops to identify candidate strategies, hypotheses,
desired trade-offs. Stakeholder involvement is key when “implementation
uncertainty” is important.
What is actually necessary toprovide strategic advice? Think carefully about candidate strategies:
The default strategy should always be the current one. A TAC which is 20% of current biomass will always be
preferred to the outcome of complicated (e.g. ecosystem) model.
Look for a “good enough” solution which is easily explained rather than “complex perfection”.
Avoidance of “unrealistic” scenarios Avoid scenarios which “while interesting” are not strongly
supported by the data (IWC “rejects” all scenarios which are “low” plausibility).
What is actually necessary toprovide strategic advice? Capture the major uncertainties, but avoid 1,000
scenarios: Consider when to “integrate” (process error) and
“scenario”. A balance here is key. Always include assessment error (at realistic levels). Keep the scenarios “balanced” (e.g. high vs low
productivity) Combinations for factors are nice, but usually just add
confusion.
University of Washington Teresa A’mar John Brandon
CSIRO• Beth Fulton • Eva Plaganyi-Lloyd
UCTEva Plaganyi-Lloyd
KEYAcknowledgements