Example

• Allow stakeholders to agree management objectives and the trade-offs between them as the first step when conducting Management Strategy Evaluation.
• Explore multiple conflicting objectives and the trade-offs beyween them.
• Develop a risk based framework for evaluation management strategies, where is risk is an uncertainty that matters and what matters are management objectives.

Management Strategy Evaluation

When conducting MSE an Operating Model (OM) is used to represent alternative hypotheses about the dynamics of the system.

This is then used to evaluate the robustness of different management strategies, i.e. do they still work despite what we do not know and can not control.

Advice depends on the combination of data, estimation method, reference points and the harvest control rule used.

These are therefore modelled as a Management Procedure (MP).

The control actions from the MP are fed back into the OM so that its influence on the stock and hence on future fisheries data is propagated through the stock and fishery dynamics.

Management Strategy Evaluation

To conduct MSE requires six steps; namely

1. Identification of management objectives and agreeing performance measures to quantify how well they are achieved;
2. Selection of hypotheses about system dynamics for building Simulation Models (i.e. Operating Models);
3. Conditioning the Operating Models on data and knowledge, and rejecting and weighting different hypotheses;
4. Identifying alternative management strategies, (i.e.the combination of pre-defined data, stock assessment methods, reference points and control rules;
5. Running simulations using the management strategies as feedback controllers; and
6. Agreeing the Management Strategies that best meet management objectives.

Objectives

• Safety
  • Make sure the stock does not fall below a level were productivity is impaired
• Status
  • Stock should be capable of providing the Maximum Sustainable Yield
• Yield
  • High yields
• Variability
  • Yield should not vary wildly between years

Management Strategy Evaluation

Operating Model:

Simulation

Simulation

Simulation

Perfect Management

Perfect Management

Perfect Management

Kobe

Kobe

Kobe

Kobe

Kobe

Management Procedures

MPs can either be model based, where a stock assessment is used to estimate quantities such as biomass and exploitation rate relative to reference points, or model free based on data alone.

Using model-based MPs does not necessarily ensure a more robust management if the model used in the MP is misspecified. Model-based strategies are attractive because they may be linked to the stock assessment results and generally have a greater capacity to learn about stock productivity.

Alternatively MP may be based on empirical indicators, e.g. that follow trends in catch rate, size composition, tag recovery rate, survey estimates of abundance or species composition. Such rules have an advantage of being more easily understood by managers and stakeholders.

Management Procedures

A model free MP based on data has several parameters that require tuning; i.e. the “best” parameters are found by choosing the values that best meet the objectives of asset and stakeholders. Deciding which is a “best” MP requires an iterative process involving between managers, stakeholders and scientists. The MP selected here uses an index of abundance where catches are increased when the trend is positive, alternatively catches are decreased if the trend is negative , namely

Empirical MP

Empirical HCR catches are increased when the trend in an index is positive, and decreased if the trend is negative

\[TAC^1_{y+1}=TAC_y\times \left\{\begin{array}{rcl} {1-k_1|\lambda|^{\gamma}} & \mbox{for} & \lambda<0\\[0.35cm] {1+k_2\lambda} & \mbox{for} & \lambda\geq 0 \end{array}\right. \]

where \(\lambda\) is the slope in the regression of \(\ln I_y\) for the most recent \(n\) years, \(k_1\) and \(k_2\) are parameters and \(\gamma\) actions asymmetry so that decreases in the index do not result in the same relative change as as an increase.

The TAC is the average of the last TAC and the value output by the HCR.

\[ TAC_{y+1} = 0.5\times\left(TAC_y+C^{\rm targ}_y\right)\\ \]

Never perfect

Optimising energy and overhead for large parameterspace simulations

[Kell, Forshaw, and McGough (2019)(https://arxiv.org/pdf/1910.02516.pdf)

Pareto Frontiers

Sustainability

Yield and Variability

+ Use Support Vector Regression for calibration.

Trade-offs

Run MSE