Prior knowledge incorporating acoustic abundance indices into Bayesian
Prior knowledge: incorporating acoustic abundance indices into Bayesian assessments Richard L. O’Driscoll & Andy Mc. Kenzie
The New Zealand experience 1. 2. 3. 4. Importance of uncertainty Bayesian stock assessment Hoki example Incorporation of indices into assessment 5. Are our acoustic surveys doing a good job?
Why uncertainty is important Survey Acoustic best practice • • Target strength Target identification Sampling error Sound absorption • • Vessel motion Noise removal Calibration Deadzone Abundance 100, 000 t Assessment/Decision Rule e. g. Yield = 0. 2 x Biomass Yield 20, 000 t
Why uncertainty is important Survey Acoustic best practice • • Target strength Target identification Sampling error Sound absorption • • Vessel motion Noise removal Calibration Deadzone Abundance 50, 000 – 200, 000 t Assessment/Decision Rule e. g. Yield = 0. 2 x Biomass Yield 10, 000 – 40, 000 t
Survey uncertainty (my one equation!) Ii = q. Biεi Ii is the survey index in year i Bi is the population biomass in year i q is the systematic error (or bias or “survey catchability”) εi is the random error (or precision or CV) in year i If q varies between years variability becomes a component of εi
Estimating random error εi Further information: O’Driscoll (2004). ICES Journal of Marine Science 61: 84 -97
Estimating bias (or catchability) q? • Requires some independent estimate of surveyed population • Here we use a data-rich assessment model for New Zealand hoki
Hoki (Macruronus novaezelandiae) • NZ’s largest fishery • First large whitefishery to achieve MSC certification in 2001 • Current TACC 150, 000 t (value ~$800 million NZD)
Hoki fishery information • • • Two stocks Commercial catch-at-age data Biological parameters Acoustic surveys Trawl surveys
Bayesian assessment • Probabilistic approach to evaluate alternative hypotheses (e. g. , stock status) incorporating uncertainty • Two sources of information (with distributions): 1. Observations (data) 2. Prior knowledge (experience) • Information combined using Bayes Theorem • Output expressed as a posterior probability distribution for each hypotheses • Estimation approach - independent of population dynamics model Further information: Punt & Hilborn (1997). Reviews in Fish Biology and Fisheries 7: 35 -63
Hoki assessment no acoustic indices
Acoustic survey q prior
Hoki assessment with acoustic indices
Estimated q posteriors 0. 50 (95% CI 0. 34– 0. 69) 0. 39 (95% CI 0. 21– 0. 65)
NZ hoki surveys biased …. • Turnover (negative bias) • Target identification (positive bias? ) • Target strength (negative bias)
And uncertain …… Cook Strait
But still informative
Conclusions • We live in an uncertain world • Precise absolute acoustic abundance estimates may be unachievable • Bayesian methods provide framework for incorporating uncertainty • Long time-series and a variety of alternative abundance indices required to allow survey uncertainties to be estimated • Models can show where our observations &/or priors don’t fit • We can improve (tighten) priors by better understanding of acoustic methods
Acknowledgments • NZ Ministry for Primary Industries • Alistair Dunn, Chris Francis, Darcy Webber • Members of ICES WGFAST
- Slides: 21