Exploring the ABC approach for IOTC Albacore OM conditioning

For the current suite of IOTC MSE work, the general approach to conditioning the required
set of Operating Models (OMs) has been to use the species-specific stock assessment model
structure as the basis for the OMs. A grid of model runs, formulated using a set of alternative
assumptions and inputs, is constructed based on the base case assessment model. In [1] an
alternate, complementary approach was outlined where, instead of the assessment being the
basis for conditioning, a suite of possible prior states of historical dynamics and current status are
defined. The available, but mostly the more contemporary, data are included within an estimation
scheme built on emerging Approximate Bayesian Computation (ABC) and Synthetic Likelihood
(SL) concepts [2, 3]. The aim is to generate a distribution of current abundance, mortality and
status that is consistent with both the available data and the suite of possible prior states of
nature defined beforehand. This can then be used to initialise the OMs used to project the stock
into the future and test the candidate MPs.
A stock assessment, in this context, can be viewed as our attempt to do both of these things
at once. Ideally, this is arguably a sensible option; however, it is not always successful. The
ongoing struggles with the Yellowfin tuna stock assessment, and the conditioning of OMs based
upon it, outline this problem: what if you cannot adequately reconcile the data, assessment
model structures, and the resultant estimates of current status and future projected dynamics?
In [1] we proposed an alternate approach arguing that using a stable, agreed and robust stock
assessment was a natural first option, but that the ABC approach was a potentially viable - and
scientifically pragmatic - alternative approach if the assessment route was unsuccessful.
In this paper we parameterise a real world example using Indian Ocean Albacore tuna that mirrors
(biologically and structurally) the most recent stock assessment, utilises length composition
and longline CPUE data, and is able to explore a wide range of stock status prior hypotheses,
many of them built on information from the results of the stock assessment.