Tutorial: Policy exploration procedure

The EwE policy exploration module is a complex but capable beast, designed for policy exploration of trade-offs, not for providing management advice for direct implementation. The policy advice it can produce is strategic rather than tactical (i.e. broad policy advice rather than specific management advice). It can thus be at the table where policy discussions take place, in particular about options for and trade-offs in ecosystem-based management. In this tutorial, we’ll go through and explain details for how the module may be used for serious policy exploration.

As part of this, we will outline, step by step, a procedure we find useful for conducting a more complete policy exploration that can be published and potentially can contribute to policy development.

Model scope and behaviour

We assume that your model is indeed to be used for actual policy exploration, and advise that the model, while being predictive (rather than descriptive, see the Defining the ecosystem chapter) should include the fleets among which the policy module will seek to balance trade-offs – this may well be all fleets operating in the given ecosystem as the tradeoffs often are through food web interactions and bycatch.  Further, the target species for the fishery should be included in the model, along valued species that they take as bycatch, along with their prey and competitors, and where applicable top predators such as marine mammals and species of conservation concern.

It is important for the model behaviour in response to proposed fisheries changes that the model is fitted to time series data – that is that density dependence related to carrying capacity has been considered (see Density dependence and carrying capacity chapter), and the vulnerability multipliers that affect population resilience have been modified accordingly, (see Vulnerability and vulnerability multiplier chapter).

What fleets to consider?

If you have fleets in your system that it does not make sense to optimize for, e.g., optimizing for profit for a recreational fleet or a “catch-all” IUU fleet, they can be considered in optimizations without being varied in optimization searches.  For this, on the Blocks form in the interface select the first (black) block, then click the name of the fleet in the spreadsheet, and all years will be blackened out. When this is done, the Ecosim effort will remain as entered for that group, but the calculations of objectives will still include impacts of the blocked fleet(s).

As an example, the recreational fleet may be relying on a species that is also a target for a commercial fleet that is considered in the optimizations. Abundance changes caused by changing the commercial fleet effort will then impact the catch rates of the recreational fleet, which in turn will impact optimization measures that are affected by the recreational fleet.

Exploratory analysis

Objective ranges

Policy explorations are often intended to explore less extreme, more balanced solutions for fleet tradeoffs. That calls for using weights on several policy objectives (see textbook Policy exploration chapter) – but what weights are needed to make the resulting fishing efforts “balanced”? Using the same weight (e.g., 1) for all objectives is not likely to results in a reasonable balance of performance measures. What then?

When you are ready to explore the policy optimization, the first step is to evaluate the range of optimized fishing efforts that result for each objective. Open your model (or Anchovy Bay from this link), load the Ecosim scenario you want to use, but do not load time series. Then run four policy searches with default settings varying only the objective weights. In the first search give the Net economic value a relative weight of 1, and leave the relative weight on all other objectives at 0. In the second run, give the Social value (employment) a value of 1, and all others 0.  Then third and fourth runs are with only weights on Ecosystem structure and Biodiversity, respectively.

There is no need to include Mandated rebuilding in the range of tested optimizations as this objective differs in behaviour. When invoked it is a forced rebuilding, which is intended to take precedence over all other objectives, i.e. it should have a weight that trump all other objectives, (so very high, e.g., 100). The Mandated rebuilding is functional group specific, and only impacts the optimization when the biomass of a specified group falls below a user-defined reference level (B_lim). When the biomass is at or above B_lim, the objective has no impact on the optimization.

For Anchovy Bay, we may get results as in Table 1 from the single objectives optimization runs.

Table 1. Policy optimization objective ranges for four runs, each with weight on only one objective at the time.  

Table 1 shows the outcome from the four objective-by-objective runs. The range of objective values are indicated (ignoring negative Net economic values) and make it clear that the two economic objectives have the largest range, followed by Ecosystem structure. The Biodiversity objective has a much more narrow range.  There is no truth or absolute values coming out of this exploratory analysis, but it serves to illustrate that using equal weight on all objectives is unlikely to lead to a balanced solution. Instead, as a first estimate for weights to use across fleets one may be able to use the inverse of the ranges, see Table 1.

When running the policy optimization for Anchovy Bay with objective weights = the inverse of ranges from Table 1, the results in the table below are obtained.

Table 2. Objective function values and fleet effort for Anchovy Bay with weights set to inverse range of objectives

Local maxima

As part of the exploratory analysis, it is important to check whether the maximization search is impacted by the start point, i.e. whether the optimization solutions are unique. By default the optimization routine will start with the fishing rates defined by the Ecopath baseline. It’s possible, however, to instead using random fleet effort (Random F’s in the policy interface) to check if the optimization routine is likely to get stuck at local maxima.  All optimization routines are impacted by this, the ones in EwE being no exceptions.

We illustrate this for Anchovy Bay by running five^[1] optimizations for each of the objective-by-objective optimizations in Table 1. The outcome of that exploratory analysis is presented in Table 3.

Table 3. Policy optimization objective ranges for five runs for each objective, all with random starting efforts.  The last five columns gives the effort by fleet for each optimization run. 

Table 2 has four sections, one for each of the objective-by-objective optimizations. As an example, the first sections shows the outcome of the five runs with random starting point effort when optimizing for Net economic value only. The objective function stopped at a very similar value (2.31) in all five runs, and the effort is very similar for each fleet for all runs, with one exception. The effort for Sealers is 0.95 in the first run, and 1.9 – 2.2 in the four others. This may seem like a big difference, but in the optimization, all five effort levels result in the same basic ecological outcome: a total collapse of the target species, seals.

Examining the entire Table 2, it is clear that that in this example, there is very little tendency for the optimization to get stuck on local maxima. The variation in the objective estimates and effort patterns are very similar across each optimization type, with only a few runs indicating presence of local maxima. The fourth run for Ecosystem structure is the only run that seems to differ in any substantial way, indicative of the optimization being unable to find any unique fishing efforts to optimize this objective (which should not surprise you since it is quite a vague objective in the first place).

The conclusion is that policy optimizations for Anchovy Bay are not very prone to get stuck on local maxima. This is also what we have found for many other ecosystem model optimizations, giving some comfort that the starting point isn’t very critical. Still, that needs to be checked for all models, so including a search with random Starting F’s should be included in all more serious policy explorations.

Fleet trade-off analysis

A next step of exploratory analysis is the fleet trade-off analysis described in the Fishing policy chapter (link to fleet trade-off). We refer to that section for description, including for code to produce plots.

We suggest that you perform the fleet trade-off analysis for your model and explore the outcome. For Anchovy Bay (see plot to the left), the plot indicates that a 10% reduction for the fleets mentioned to the left, causes a reduction in the value of the catch (red circles), and in some cases this leads to an increase for other fleets (blue circles) through food web or technical interactions.  The impacts are displayed so that the circle areas are proportional to the changes in value of the catch, and are thus comparable across fleets^[2].

For Anchovy Bay, note the blue circles in the figure. Reduction of Sealers effort has a positive impact on Trawlers catch value, while reduction in Trawlers effort has a positive impact on the last three fleets. These results indicate that trophic interactions are indeed important in creating catch value tradeoffs at least in the Anchovy Bay demonstration model, and in fact in most models for which such calculations have been done.

What are your questions?

Having explored the behaviour of your model, e.g., as described above, the next issue is to clearly define what questions you are asking for your model. An example of this is Alms et al. 2022, ^[3] who explored

If when you start running optimizations with the “limit cost > earnings” option, both the net economic value and the social indicator (jobs) are negative already at the two initial runs, which indicates that a penalty has been applied in the search routine where the penalty increases rapidly as the ratio cost/income increases toward and exceeds 1.0. Such a penalty is needed to make the optimization move away from fleet efforts that drive cost > earnings.

When it happens from the onset, it indicates that the baseline effort is unsustainable. In the case of Anchovy Bay, the culprit is the sealers fleet, which has a high, unsustainable effort in the base year. In Ecosim run, the fleet is shut down after a few years, but in the optimization, the high initial effort is maintained through the run. The optimization takes the cost and value at the baseline and sums the cost and value at the last year, and it multiplies that last year with a discounted value of what the last years catch would be worth if it were continued for an additional 20 years (i.e. there’s a high weight on the end state relative to the baseline).

In some cases, the optimization routine can find its way out of the unsustainable fleet effort range, but not always. If the routine keeps producing negative indicators for the first two objectives, try making a run where you flatline the effort (1 throughout), and see which fleets end up having negative profit in the last year. Then reduce the effort for those fleets, and run the optimization again

 

Acknowledgement

This chapter was developed for the EcoScope project to guide implementation of the EwE Policy Search for the project case studies. EcoScope is funded from the European Commission’s Horizon 2020 Research and Innovation programme under grant agreement No 101000302. Project coordinator: Aristotle University of Thessaloniki, Greece.  Parts of the text are from the unpublished EwE User Guide: Christensen V, C Walters, D Pauly, R Forrest. Ecopath with Ecosim. User Guide. November 2008.