User menu
This paper summarizes progress on the development of Operating Models (OMs) for IOTC
yellowfin (YFT) tuna, highlighting priorities for technical feedback. A short stand-alone summary
document describing the most recent reference set Operating Model (OM) is included at
attachment 1. This paper focuses on OM developments since the IOTC MSE Task Force meeting in
March 2019 (Kolody and Jumppanen 2019a,b). MP evaluation updates for yellowfin and bigeye
tunas are described in Kolody and Jumppanen (2019c). The latest version of the MSE software and
technical documentation is publicly available from github https://github.com/pjumppanen/niMSEIO-
BET-YFT/. Key developments include:
• The reference set YFT OM (OMref19.4.500) was updated from a reference case Stock
Synthesis stock assessment presented to the 2018 WPTT (Fu et al. 2018), and expanded to
represent uncertainty in 11 dimensions. The result is considerably more pessimistic than
the reference set OM considered by the 2018 WPTT/WPM, however, the model is also
more stable, and did not require the “bi-variate grid sampling” approach used to select a
plausible model ensemble in 2018.
• The reference set OM proposed at this time was more comprehensively investigated than
the OM which was used for the 2019 TCMP results and corrected a couple minor
specification errors, but resulted in very similar MP evaluation performance.
• OM development requests from the various IOTC working groups have led to a potentially
unwieldy number of yellowfin model specifications (4608, if all assumption interactions are
evaluated). The computational burden is further exaggerated by numerical instability and
sensitivity to initial parameter values. A few alternative approaches for grid specification
were explored, which increase confidence that a relatively small number of models (e.g.
50-100) can probably provide similar MP evaluation advice to the full grid. A subset of 49
models (filtered from a main-effects fractional factorial design of 144) was able to produce
very similar results to a much larger ensemble (2-way interaction fractional factorial design
of 1152, filtered to 420). Similarly, the large grid was subject to a jitter analysis, which
demonstrated substantial variability among many individual “converged” models.
However, comparison of MP performance based on OM ensembles comprised of the best
and worst fit of the converged models (i.e. selected from within the 3 replicates of each
configuration) were very similar.
• The majority (~2/3) of the yellowfin models in the ensemble grids explored were rejected
with numerical problems, identified by a relatively large SS3 catch penalty. We interpret
this to mean that these models struggle to remove the observed catch and presumably
explains much of the retrospective patterns that were noted in the 2018 WPTT. This is an
issue in both the assessment and OMs, and suggests that a substantial rethinking of
assumptions might be warranted. The yellowfin OM will need to take consider the findings
of the international yellowfin assessment review project. As part of this broader process,
we explored a few new OM configurations and observed that:
o The model did not provide compelling evidence for strong non-linearity in the LL
CPUE – abundance relationship post-1972.
o If the LL CPUE are aggregated into a single (regionally-weighted) index, the LL
vulnerable biomass appears relatively stable for the past decade. When this is fit in
a spatially-aggregated model (mimicked by forcing high movement rates), the result
is somewhat more optimistic than the current assessment. However, it is unclear
the extent to which this is primarily due to the simplification of regional processes
or the removal of the tags (poor tag mixing renders them inappropriate for very
large spatial regions).
o If one assumes that i) northwest LL CPUE is reliable from 1972-2006 (pre- Somalian
piracy period), ii) LL CPUE is reliable from 1972-2017 in the other regions, and iii)
(standardized) PSFS catchability has increased at a continuous rate from 1986-2017,
then a PSFS catchability increase of ~ 0.71% per year appears to largely reconcile
the LL and PSFS CPUE series. When the PSFS CPUE series is included in this manner,
the recent depletion level is estimated more optimistically than the assessment, but
the productivity (MSY) is more pessimistic. At this time, we would be hesitant to
include the PSFS series in this manner, because it requires an unexplained
continuous catchability increase assumption, and the catch size composition of this
fleet suggests that two modes of operation have been pooled (the rapid change
between small and large fish observed in the quarterly catch statistics does not
appear to be consistent with estimated recruitment patterns).
o We briefly explored the Ricker stock recruit function, (requested as an OM
robustness test). We would question whether it is worth pursuing further, since the
estimates qualitatively resembled Beverton-Holt functions (possibly with steepness
lower than was considered plausible for this population).
• We encourage further feedback from the WPTT and WPM on all elements of the Operating
Model, especially the approach used to create the reference set ensemble, methods for
dealing with numerical instability, process for evaluating plausibility (especially in an
automated context), and specific requests for reference set and robustness tests. Some
specific options are proposed for discussion, but it is recognized that the YFT review
process may fundamentally shift the direction of the OM.
