1 Disclaimer

The following work is preliminary and intended only as tool for eliciting feedback on data, modelling and other aspects of these fisheries.

None of these results are final.

These analyses do not necessarily reflect the point of view of IMAS or other funders and in no way anticipate future policy in this area.

2 Objective

Develop an MSE framework for the Tasmanian Sand Flathead fishery that can inform management decision making including research prioritization, assessment methodology, specification of fishing regulations and enforcement.

3 Project details

‘Development of a draft operating model in openMSE for the assessment and management strategy evaluation of Southern Sand Flathead in Tasmania.’

Term 15/03/2024 - 1/7/2024
Funding body University of Tasmania
Funding stream Subcontract
Project No. T0030292
Project Partners IMAS, Blue Matter Science Ltd.
Blue Matter Team Drs. Tom Carruthers & Adrian Hordyk
IMAS Principal Investigators Dr. Sean Tracey

 

4 Current Issues

The straw-dog operating models have been revised following feedback from the group.

Tabulated below are a list of current issues / assumptions that should be addressed at the current stage of framework development (Table 1). Also tabulated is the current ‘to-do’ list for the Blue Matter team (Table 2)

Table 1. Current issues / assumptions / items for discussion

Issue Notes
Length at 50% maturity, and logistic slope From an online report by NRE (‘around 27 cm’) - but actually there is GonadState, Gonadweightand Statge Mature 3-7, in the historical data (how to interpret these?). StageMature 3-7 is 50% at length 31cm in the data (aggregated)
Selectivity of largest / oldest fish Currently assumed to be flat-topped, asymptotic.
Natural mortality rate Using value of 0.28 from Kruek et al. 2023 but what is the origin of this?
Minimum size limit of 32cm Is this correct, for all areas?
Background rate of discarding Assumed to be zero - but is this accurate? We can do bag limit modelling but need CPUE vs release rate by trip (baglimit)
Observation error a placeholder to get demo MPs working Later these observation processes can be characterized statistically
Observation biases assumed to be nil For now, observed catches, indices etc are assumed to be unbiased and not hyperstable or hyperdeplete
Implementation assumed to be perfect For now, for demonstration purposes, any management advice is assumed to be followed exactly
Model arbitarily started in 1974 (50 historical years) What is a suitable time to start fishing?
Equilibrium catches assumed to be negligible before 1974 (again arbitrarily) of those observed C_eq = 0.00001 * mean historical catch but can bring forward the initial model year and specify a differing C_eq
OM fleet and survey structure: 2 fleets (rec / commercial), 4 surveys (rec CPUE, commercial CPUE, historial length/age comp, rec survey length/age comp)
Fleet seleectivities are not informed by length / age data we assume that the rec survey reflects the recreational fishery
Nominal Effort could be improved as an index of exploitation rate Can we derive effort / habitat area. There is the potential to borrow information on catchability among areas/models - priors, metaanalysis, EM.
Catches are expanded to totals using expansion factor - no uncertainty How can we get observation error in total catches? How are expansion factors calculated - can we do bootstrapping etc?
Discard mortality rate assumed to be 6% as the max from deep water estimates of Lyle et al 2006 Lyle et al. 2006. This is used to include discard mortality in total catch data (in model conditioning [Catch = ExpFac x (Kept + Rel * DiscMort)] and used in projections that would affect any kind of regulation affecting discarding such as size limits, bag limits etc.
Total recreational effort currently calculated by Duration_hrs x Npersons x ExpWt (what is the ‘expansion factor’??)
rec_suvey_data.xls sheet 2017-18 has DurationHrs formatted as a date I manually changed this to ‘number’ format.
Recreational survey index by large region is standardized as a linear model log(CPUE) ~ Yr + Quarter + Region + WaterBody + Type (Large region is something like SEC, region is something like Tasman, Derwent esturary etc)
Commerical catch data not reconstructed by Large Region Commerical catch data are reconstructed (I think for all areas) from 95/96 to 06/06 (…total_catch_rel_effort…csv). These catches are not reconstructed by large region (EC, SEC etc).
Commercial cpue and effort units not clear ?
Not clear how to assign calendar year to commercial year Currently this is assumed to occur mostly in the second half, ie Nov 1 - Sep 1, so 2022/23 would be assigned the year 2023.
Imputation / extrapolation of catches The RCM model requires either complete effort or complete catches. For now I just linearly interpolated / constantly extrapolated catches. Maybe move to an effort model next?
Commercial fishery selectivity / retention is unknown For now I’m assuming it follows the survey comp data
Large region (LRegion) Region
SEC Derwent Estuary, Tasman, Frederick Henry/Norfolk Bay, South-eastern coast, D’entrecasteaux Channel, South, Northwest Bay, SECest, SEC
EC Great Oyster Bay, Central-eastern coast, Eastern coast, Coles Bay, Georges Bay, EC
NWC North-western coast, King Island, rocky cape, NWC
NEC Tamar River, North-eastern coast, Flinders Island, Spring Bay, Flinders/Eastcoast, NC, EC, Deal island, Hogan group, NEC, FI
WC Central-western coast, Western coast, South-western coast
unknown EAT, ECS, ET, SET, CBS, no sample

 

Table 2. Blue Matter to-do list.

High Meet to discuss straw-dog fits
Medium Data weighting profiling
Medium Parameter profiling

 

5 Study Area

Currently the study area was divided into the five large regions (top middle Figure 1): SEC, EC, NEC, NWC, WC. Limited data were available for WC in this round of operating model development.

Figure 1. Study area. Area definitions (top left), areas of high recreational effort (top middle), areas of research focus (top right), commercial effort distribution (bottom left), commercial catch per unit effort (bottom right).

 

6 Time Series Data

6.1 Recreational Fishery

Figure 2. Recreational catches by large region.

 

Figure 3. Recreational effort by large region.

 

Figure 4. Standardized Recreational CPUE indices by large region [log(CPUE) ~ Year + quarter + small_region + Method + Waterbody].

 

6.2 Commercial Fishery

Figure 5. Commercial catch by large region.

 

Figure 6. Commercial effort by large region.

 

Figure 7. Commercial effort by large region.

 

6.3 Historial Composition Data

Figure 8. Aggregated historical length frequencies by large region.

 

Figure 9. Aggregated historical age frequencies by large region.

 

6.4 Fishery Independent Composition Data

Figure 10. Aggregated length frequencies from the fishery independent survey by large region.

 

Figure 11. Aggregated age frequencies from the fishery independent survey by large region.

 

7 Life history chracteristics

In order to include plausible uncertainty in the life-history dynamics for sand flathead, frequentist models of somatic growth and the length-weight relationship were fitted to data and parameter values draw from the variance covariance matrix arising from those fits (Figures 12 and 13)

Figure 12. Generation of stochastic life history parameters (females) for a preliminary operating model for Flinders Island. Top left is the fit of a preliminary von Bertalanffy somatic growth model to observed age-length data. Top right is the correlation among simulated asympotic length (Linf) and maximum growth rate (K) parameters drawn from the variance-covariance matrix of the model fit. Bottom left is the simualted natual mortality rate (M) given a fixed ratio of M/K and a lognormal error with CV of 10%. Bottom right is the simulated length at 50% maturity (L50) given a fixed ratio of L50/Linf and a lognormal error with CV of 10%.

 

Figure 13. Generation of stochastic weight-length parameters (females) for a preliminary operating model for Flinders Island. Top left is the fit of a preliminary weight length (W=aL^b) growth model to observed length-weight data. Top right is the correlation among the slope (a) and power (b) parameters drawn from the variance-covariance matrix of the model fit.

 

The mean ‘maturity’ (fraction of individuals mature) ogive was calculated from the historical data (all regions combined, Figure 4a) and was the basis for simulating correlated uncertainty in L50 (length at 50% maturity) in Figure 12. A very different maturity ogive is provided by the more recent fishery independent survey (all regional combined, Figure 14b).

Figure 14a. The logistic relationship between length and fraction mature (fraction ‘stage gonad development 3-7’) of the historical data set.

 

Figure 14b. The logistic relationship between length and fraction mature (fraction ‘stage gonad development’ > 2) of the fishery independent survey data set.

 

8 Operating Models

Operating models were fitted using the Rapid Conditioning Model of the openMSE R package. The model is structured to have two fisheries (recreational and commercial) and four surveys (Recreational CPUE, commercial CPUE, historical age/length compositions, recreational survey age/length compositions). Currently fishery (Rec, Com) selectivities are specified by the user but could be linked to the length/age composition of the surveys if that is appropriate.

Current the model is ‘conditional on catch’ which means that a complete catch history for both fleets had to be imputed. This was done by linear ramping from a starting year (1974 - picked arbitrarily) to the first observation, linearly interpolating between observations and then constant extrapolation of the last observation. An alternative would be to start the model later and assume an ‘equilibrium average catch’ prior to the first year, or to instead interpolate effort data and make the model ‘conditional on effort’.

The models were fitted to length an age composition data assuming a multinomial likelihood function and an average effective sample size of 50 observations per year for each composition data stream (hist / rec x age / length).

The full operating model fitting reports are available to download from Table 3.

 

Table 3. RCM Operating Model Fitting Reports

9 Supporting Information

10 Acknowledgements

Many thanks to Sean Tracey, Nils Krueck, Kate Stark, Alyssa Marshall, Peter Coulson, Barrett Wolfe, Katie Cresswell, Ruth Sharples.

 

11 Appendix 1: MSE terminology

11.1 Operating models

An operating model is a theoretical description of fishery and population dynamics used for the testing of management strategies that could include, for example, data collection protocols, stock assessment methods, harvest control rules, enforcement policies and reference points. In fisheries, operating models are used in closed-loop simulation to test management procedures (aka. harvest strategy) accounting for feedbacks between the system, data, management procedure and implementation. A management procedure is any codifable rule that calculates management advice from data. Management Strategy Evaluation uses closed-loop simulation of management procedures as a core technical component but is a wider process of stakeholder and manager engagement that identifies system uncertainties, performance metrics, viable management procedures, ultimately aiming to adopt an MP for the provision of management advice for an established time period.

 

11.2 Reference Case Operating Models

The reference case operating model is used as the single ‘base’ operating model from which reference set and robustness set operating models are specified. Reference and robustness tests are typically 1-factor departures from the reference case OM, however sometimes reference set OMs are organized in a factorial grid across primary axes of uncertainty.

 

11.3 Reference Set Operating Models

Reference set operating models span a plausible range of the core uncertainties for states of nature. These are often the types of alternative parameterizations or assumptions that would be included in a stock assessment sensitivity analysis.

The role of the reference set operating models is to provide the central basis for evaluating the performance of candidate management procedures, for example rejecting badly performing harvest strategies.

 

11.4 Robustness Set Operating Models

Robustness set operating models are intended to include additional sources of uncertainty for providing further discrimination among management procedures that perform comparably among reference set operating models.

Robustness operating models often represent system states of nature that are not empirically informed or are hypotheses of a subset of stakeholders.