Groundfish Assessment Program
Performance of Modern Age-Structured Stock Assessments with High Survey Measurement Error
The objective of AFSC bottom trawl surveys in Alaska is to obtain biomass estimates for all major Alaska
groundfish species. However, many species of rockfish (Sebastes spp.) are patchily distributed and
therefore are imprecisely estimated in these multispecies surveys, which are based on a stratified random
design. This same general problem pervades the stock assessments of many fish species worldwide.
In our study, the stock assessment model for Pacific ocean perch (Sebastes alutus) in the Gulf of
Alaska was used to explore the consequences of survey imprecision and other uncertainties in components
of this type of model. The characteristics of the Pacific ocean perch assessment can be generalized to
other long-lived, iteroparous fish species with uncertain survey biomass estimates. The results of the
2003 stock assessment model served as the "true" values, and simulated data sets were
constructed in five experiments to answer the following questions:
- What is the effect of measurement error in survey biomass estimates on stock assessment results?
- What is the effect if the catchability coefficient changes over time from either gear changes or
- Does adding an additional biomass index increase model precision?
- How sensitive are model results to applying arbitrary weights to different data sources such as
fishery length distributions and survey ages?
- How sensitive are model results to prior distributions (a distribution representing prior knowledge
about a parameter that influences estimation) imposed on key parameters?
The five simulation experiments yielded the following general answers to these questions:
- High measurement error (coefficient of variation equal to 50%) rendered the stock assessment
inaccurate and imprecise.
- A catchability trend in the biomass index was undetectable and created a large bias in biomass
and parameter estimates.
- The addition of a second, more precise biomass index with a shorter time series improved performance of the
model. Examples of an additional index could be a hydroacoustic index or a dedicated rockfish survey.
- This type of assessment was robust to various data weightings, indicating that the stock assessment
scientist could merely give each data source equal weight.
- The prior distribution for natural mortality needed to be precisely specified, while the prior distributions
for the catchability coefficient and recruitment variability could be relatively uninformative.
Overall, the study showed that the high measurement error in the survey index (for species such as rockfish)
can render stock assessment intractable, data weighting was less important than expected, and prior
distributions on parameters (except natural mortality) could be uninformative.
By Dana Hanselman
2005 Sablefish Longline Survey
The AFSC has conducted an annual longline survey of sablefish and other groundfish off the coast of
Alaska since 1987. The survey is a joint effort involving the ABL and the Centerís Resource Assessment
Conservation Engineering (RACE) Division. In February 2005, a new longline survey contract was awarded
to two vessels over a period of 4 years. The Ocean Prowler will conduct the survey in 2005 and
2007, and the Alaskan Leader in 2006 and 2008. The schedule will be similar to previous years,
with sampling in the Gulf of Alaska annually and biennial sampling of the Bering Sea (2005, 2007) and
Aleutian Islands region (2006, 2008). The 2005 survey begins on 28 May and ends on 1 September.
By Chris Lunsford
AFSC Quarterly Research Reports Jan-Mar 2005