Fishing policy exploration

Parameters

Discount rate: What is it worth for you today to get a million Euro next year, compared to getting it today? Economists discount the present value of future income by using a discount rate. The optimization routine uses a default a discount rate of 4%, which means that the current value of getting a Euro in one year is €0.96 (= (1+Discount rate)^-1). (Use default if you don’t have a prescribed discount rate to use)

Number of runs: If you set the Initialize using option to “Random F”, the optimization routine will use a random starting point. As the routine may get stuck on local minima, it makes sense to try several times. (perhaps 10). In nonlinear optimization practice, this is called “multiple shooting”.

Initialize using: There are three options,

1. Ecopath base F: Starts with the Ecopath base fishing effort, (which typically will be 1)
2. Current F: when used with a base year that’s different from year 1, selecting this option will make the policy search start with the effort of the base year
3. Random F: selects random effort for each fleet. This option is important to evaluate if there are local minima. The policy search can get stuck on local minima, and the best way to evaluate the risk of that is to run a number of trials, starting each with random effort.
   Note that if the Number of runs is set to more than 1, the Initialize using will change to using Random F.

Max. no of evals: The optimization routine does a large number of Ecosim runs during an optimization, often several hundred. The parameter sets a limit for how many. (Use default setting unless your model won’t converge with this).

Base year: By default this is year 1, and if so, your economic parameters should be for the Ecopath base year (= year 1). Assuming your model is fitted to time series data, e.g., for years 1 to n, it may make sense to do the optimization looking forward to the post-fitting period (year n+1 on). In that case, add, e.g., 20-30 year to the run time (Ecosim > Input > Ecosim parameters > Duration of simulation (years)), and change the Base year accordingly. If you increase the base year, the interface will block out (in black) the years prior to the base year, i.e. Ecosim will run with stored effort for those years, not try to estimate any.

To minimize the risk that the optimization will result in overexploitation towards the end of the simulation, the simulation routine will run for an additional 20 years beyond the last year, and the results for that last block of years will be included in the output.

Max effort change: Provides a limit for how much effort is allowed to change from one year to the next. The parameter should be entered as a factor 𝛼 ≥ 1. With a value of, e.g., 𝛼 = 1.1, the routine will at most accept a change in effort for the following year in the range [0.909, 1.1]. With the default value of 0, there is no limit for how much effort can change.

The changes are only for between-blocks years, for instance where there are several blocks for a given fleet, perhaps to have some additional years for rebuilding a stock followed by a different management patterns for later years. In that case, the max. effort change will be evaluated only when changing from one block of years to the next.

Search using: Decides which search algorithm to use for the minimization, either DFPmin or Fletch,

1. DFPmin: The Davidson-Fletcher-Powell ^[1] optimization procedure is a nonlinear optimization procedure to iteratively improve an objective function by changing relative fishing rates, where each colour-coded “year/fleet block” defines one parameter to be varied by the procedure, (e.g., setting four colour code blocks means a 4-parameter nonlinear search).  DFP runs the Ecosim model repeatedly while varying these parameters; in the search output display, each simulation trial is labelled an “eval” or function evaluation.  So if you are running a large model for many years, where each simulation takes several seconds to do, the search may take quite a long time to do enough function evaluations to find a maximum for the objective function.
   The parameter variation scheme used by DFP is known as a ‘conjugate-gradient’ method, which involves testing alternative parameter values so as to locally approximate the objective function as a quadratic function of the parameter values, and using this approximation to make parameter update steps.  It is one of the more efficient algorithms for complex and highly nonlinear optimization problems like the one of finding a best fishing pattern over time for a nonlinear dynamic model.
2. Fletch: An efficient non-linear search routine, older and less-well known than DFP, but faster and more efficient for many problems. Used by default.

Use plug-in economic data: When defining a fleet in the Ecopath base model, there are input values for costs as a fraction of income (Ecopath > Input > Fishery > Fleets) based on which the baseline costs are obtained (default is that cost is 80% of revenue). The revenue is obtained from the Ecopath baseline landings times off-vessel price.  In Ecosim, revenue for each time step will be calculated the same way, while cost will change proportionally to effort.

The checkbox will be enabled if a value chain has been defined for the model, and the “Run with searches” option has been checked (at Ecopath > Output > Tools > Value chain > Parameters). If the Use plug-in economic data is checked, the fishing policy search routine will run the value chain for each run and obtain revenue and costs by fleet and time from there.

Limit cost > earnings: When maximizing for jobs (landed value) the search routine can create rather extreme policies, often driving the system towards mono-culture system configuration or fishing at a loss (implied subsidy to increase jobs). For Anchovy Bay, as an example, this may result in higher landed value, comprised almost exclusively of shrimp and anchovies. This corresponds to a “sole owner” or societal approach (see textbox in Fishing policy exploration chapter), and it is implemented if you do not check the Limit cost > earnings option (default option).

Checking the Limit cost > earnings option, leads to an optimization where the policy search will seek to make each fleet profitable, corresponding to a “multiple fishing rights” approach, (see textbox in Fishing policy exploration chapter).

Maximize portfolio utility: As described in the Fishing policy exploration chapter, there are two alternative objective functions in the policy search. Checking the Maximize portfolio utility option invokes the alternative objective function that invokes risk-averse utility measures that favour a balanced ‘investment portfolio’ of fishing activities. See the Risk-averse portfolio utility chapter for details.

Blocks

This section of the interface is designed to set up how many fishing effort fleet-time blocks that the fishing policy search should try to optimize. A block is represented by a colour, and by default the routine searches for one fishing effort per fleet (one time block). The base year (by default year 1) and any preceding year are by default coloured black, that means that the routine will not search for those years but instead retain whatever the fishing effort variable has in memory.

If you as an example colour the whole spreadsheet with one colour, the policy search will find the one (relative) fishing effort across all fleets that optimizes the objective function

No of blocks: Use this to get more colours = blocks

Selected: With many colours (blocks) it can be difficult to see which one is selected. It can be easier to change the number of Selected to selectively set blocks. The horizontal slider under the colours indicate which one is selected.

If the optimizations starts at a point where the system is degraded/overfished, it can make sense to have an initial rebuilding period followed by sustainable use. This can be set with by setting two colours blocks per fleet, one for the rebuilding period and one for the sustainable use period.

Set block year and sequence

This, somewhat complex part of the interface, can be used to set blocks of years. The default is that there is one block per fleet spanning from year 2 to the end, the optimization thus starting in Year 2 and ending in the last year of simulation.

Year: Gives number of years per block. The default is number of years in simulation. Changing this, for instance to 10 years,  will signal that there should be one block for each 10 years simulated for each fleet

Set: Clicking will set the number of blocks by fleet based on the Year setting.

Start: Increasing the start year will result in the years prior be blocked (indicated in black) and not included in simulations

End: Decreasing the end year will cause the following years to be blocked and not included in simulations.

Objectives, iteration results and plot results

The lower panel has three tabs, each explained next.

Objectives

This tab will be displayed while setting up the policy scenario.  The objectives are described above, this section deals with how they are implemented.  The objectives are described in the Fishing policy exploration chapter of the EwE textbook.

Search objective, relative weight: List the five objectives (or three if the Maximize portfolio utility option has been checked) with relative weights. Be aware that putting the same weight on two or more objectives does not translate into these objectives actually being balanced (see the Policy exploration procedure tutorial in the EwE textbook. This is because changing for instance how much Ecosystem structure may change will be much harder than changing Net economic value.  To get an overview of the range for each objective it may be advisable to conduct a series of optimizations placing weight on one objective at the time.

If the Maximize portfolio utility option is checked, there will be three objectives listed: Net economic value, Prediction variance and Existence value.  The details of these are explained in the Risk-averse portfolio utility chapter in the EwE textbook.

Jobs/catch value: This parameter implements the Social value (employment) objective, and can be obtained from economic analysis, e.g., through value chain analysis. If such data are not available, consider the assumption that the number of jobs is proportional to the value of the catch. The default parameter (1) is based on this assumption.

A tweak: If instead of entering landed value in your Ecopath model, you enter the trophic level of each group in the Ecopath > Input > Fishery > Off-vessel price form, the optimization can be for Trophic level of the catch instead of for Social value.

Mandated relative biomass:  The default parameter for this parameter is 0 for all groups, which means that by default, the Mandated rebuilding will not be considered in the optimization even if there’s a wight entered in the Search objective spreadsheet.  If you want to invoke the objective, change the Mandated relative biomass for groups where you want to invoke the Mandated rebuilding objective to a minimum acceptable biomass (often called B_lim) and give considerable weight to the Mandated rebuilding objective to make sure that this objective is considered fully.

Structure relative weight: This objective is used to implement the Ecosystem structure objective, and is by default parameterized with the inverse production/biomass (unit: year) for group with a P/B (year^-1) lower than ~5 year^-1.  The objective thus implements one of the ecosystem maturity parameters.

The default values can be replaced, e.g., by other network analysis indicators. An example could be the trophic level of the group, or it could be indicators such as the number of species in a functional group to bring in more biodiversity aspects.

Maximum fishing mortality: This parameter is not linked to any of the objectives, but if a Max F is entered on this spreadsheet, the search routine will penalize runs where the Max F is exceeded. This parameter can thus be used to reduce the risk of collapse of fished groups.

The default value is set so high (1000 year⁼¹)  that it does not have any impact.

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.