Presenting Model Uncertainty - Gulf of Mexico Fishery Management

Tab G, No. 3
SEDAR Uncertainty Workshop
February 22-26, 2010
Charlotte, North Carolina
Some Major Sources of Uncertainty
For the Stock Assessment Process
• Sampling/Observation Error
• Input Parameter Uncertainty
 Model Uncertainty/Structural Complexity
 Projection Uncertainty
 Stock Vulnerability
 Management Implementation Uncertainty
Input Error
Age reader error
Ex: red porgy (SEDAR 1)
14
12
NC Age (jittered)
10
8
6
4
2
0
0
2
4
6
8
SC Age (jittered)
10
12
Input Parameter Uncertainty
Red Snapper
Z
M
study
age 0-1
age 1
age 0
age 1
Szedlmayer, age 0-1
2.1-3.2
0.54
~2
0.54
Nichols (2004)
Gazey et al 2008
2.3-3.7
?
?
Emigration could increase Z, no
trawling, so mostly M
Incomplete selectivity of small fish
could decrease Z; Emigration could
increase Z
inefficiency, no contrast in effort,
emigration
1.3
2
1.2
emigration to structure bias Z high
1.98
Brooks & Porch (2004)
2.2
problems
0.58
low q for age 0, emigration to
structure bias Z high, model mispec.
RE model est
3.3-3.7 1.6-2.25 3.3-3.7 0.76-1.4
RE model Dens Dep
2.6-3.5
0.6-1.3
low q for age 0, emigration to
structure bias Z high, model mispec.
neg. survival, bias from error
stucture, nonsensical regression
ratios, linear reg.
3.48
3.1
NS
2.96
SEDAR 7
1.5
1.2
0.98
0.6
based upon VPA, ratios of surveys
0.5
0.3
Substantial uncertainty
1999 assessment
Stock Assessment Model choice
Data
Indices Effort2
PAA1
Removal
M
Biology3
x
x
x
x
x
x
Delay-difference
x
x
x
x
x
Age Structured SP
x
x
x
x
x
Stochastic SRA
x
x
x
x
x
Catch-survey (stage)
x
x
x
x
x
x
Model type
Statistical CAA
Tuned VPA
x
x
Cohort analysis
x
x
Surplus production
x
PSA
x4
Data needs
High
x
x
x
Table 1. Some common stock assessment model types and their data requirements, from most complex to least.
1observed
proportion-at-age data are not needed in some age-structured models where age composition is inferred using input selectivities.
indices indirectly inform the analyses on effort
3some of the biological characteristics used to estimate spawning biomass for estimating spawner-recruit relations are not used in some model formulations
4Productivity-susceptibility analysis, as used in the Southeast U.S., include relative vulnerabilities to different fisheries
2fishery-dependent
Low
Presenting Model Uncertainty
• Monte Carlo/Bootstrap procedure
Ex: red
grouper
Parameters and Output
Projection Uncertainty
Note how confidence intervals quickly widen
With further projections
SSB/MxH_SSB
Trends of SSB ratios E-BFT
2.5
50th percentile
80th percentile
Trends of SSB ratios E-BFT
2
1.5
1
0.5
2020
2015
2010
2005
2000
1995
1990
1985
1980
1975
1970
0
Management Implementation
Uncertainty
•Fishermen’s behavior
•Enforcement
•Weather
•Economy
•Biological unknowns
(e.g., average weight
of fish)
2009 Red Snapper Recreational
Landings (mp)
Expected
Actual
4.27
2.45
Gulf of Mexico Regional PSA Results: Snapper
Gulf of Mexico Snapper
Susceptibility (low --- high)
3.00
2.50
2.00
1.50
1.00
1.00
1.50
2.00
2.50
Productivity (high --- low)
Silk Snapper
Mangrove Snapper
Mutton Snapper
Yellow tail Snapper
Lane Snapper
Vermilion Snapper
Red Snapper
3.00
Conclusions
•
Move beyond single ‘run’ when providing results and recommendations.
•
SEDAR should provide an OFL estimate and distribution around that estimate that
addresses uncertainty and enables the SSC to determine ABC in accordance with
its ABC control rules
•
SEDAR should better communicate uncertainties and the purpose of typical
techniques used to evaluate uncertainties
•
SEDAR should strive to improve consistency between assessments
•
SEDAR should strive to explicitly identify the primary and most influential
uncertainties at each step of the assessment process……and ensure these are
carried forward to subsequent steps?
•
Potential management actions should be linked with projections made through
population models
The End