ICES WKMSYREF5 REPORT 2017

Transcription

1 ICES WKMSYREF5 REPORT 2017 ICES ADVISORY COMMITTEE ICES CM 2017/ACOM:46 A Report of the Workshop to review the ICES advisory framework for short lived species, including detailed exploration of the use of escapement strategies and forecast methods (WKMSYREF5) September 2017 Capo Granitola, Sicily

2 International Council for the Exploration of the Sea Conseil International pour l Exploration de la Mer H. C. Andersens Boulevard DK-1553 Copenhagen V Denmark Telephone (+45) Telefax (+45) info@ices.dk Recommended format for purposes of citation: ICES Report of the Workshop to review the ICES advisory framework for short lived species, including detailed exploration of the use of escapement strategies and forecast methods (WKMSYREF5), September 2017, Capo Granitola, Sicily. ICES CM 2017/ACOM:46 A. 63 pp. For permission to reproduce material from this publication, please apply to the General Secretary. The document is a report of an Expert Group under the auspices of the International Council for the Exploration of the Sea and does not necessarily represent the views of the Council International Council for the Exploration of the Sea

3 ICES WKMSYREF5 REPORT 2017 i Contents Executive Summary Introduction Terms of Reference Conduct of Meeting Structure of the report TOR b and c Forecasting methods for short-lived species Presentations Compilation of Forecast Methods Discussion TOR a and d Current ICES stochastic forecast procedure and reference points for short-lived species Presentations Discussion TOR e Evaluating the long-term performance of forecast methods for short-lived species using Management Strategy Evaluation (MSE) Presentations Discussion TOR f Guidelines for conducting short-term forecasts for short-lived species TOR g Criterion 1 for MSFD Descriptor 3 for short-lived species Conclusions and Recommendations Acknowledgments References Annex 01 List of Participants Annex 02 Agenda Annex 03 Recommendations Annex 04 Working Documents presented... 63

4

5 ICES WKMSYREF5 REPORT Executive Summary The meeting was held at the IAMC-CNR Detached Unit of Capo Granitola in Sicily, Italy, over the period September The meeting was co-chaired by Knut Korsbrekke (Norway) and José De Oliveira (UK), and included 14 other participants (three remotely). TORs were fully covered, including a compilation of the features of various forecasting methods currently used for a selected number of short-lived stocks. However, work was not sufficiently advance to provide definitive guidelines on the statistical merits of the various forecasting methods, which would, inter alia, include testing these methods in a full Management Strategy Evaluation framework. Recommendations are made for the amendment and addition to current guidelines for Category 1 and 2 short-lived stocks. The workshop recommends the use of Fcap for both deterministic and stochastic forecasts unless simulation testing in a full MSE demonstrates that it is not needed to ensure precautionarity. There are also recommendations with regard to defining B lim for short-lived spasmodic stocks, the formulation of Bpa based on Blim, and requirements for conducting shortcut MSEs following best prac-tice.

6 2 ICES WKMSYREF5 REPORT Introduction 1.1 Terms of Reference 2016/2/ACOM46 A Workshop to review the ICES advisory framework for short lived species, including detailed exploration of the use of escapement strategies and forecast methods [WKMSYREF5] will be established (Co-Chairs: Knut Korsbrekke, Norway, José De Oliveira, UK) and will meet September in Capo Granitola, Sicily. to: a) Review the population dynamic characteristics of short lived species and consider whether ICES MSY approach regarding stock size aimed at achieving a high probability (>95%) of having a minimum biomass (Blim) required to produce MSY left to spawn the following year, is appropriate for short lived species and propose a different framework if necessary. b) Compile a list of available forecasting methods for short-lived species, including information on the stock specific characteristic of relevance for the choice of methods, e.g. timing of recruitment, fishery, assessment and advice. c) Review the statistical merits of the forecasting methods compiled under b), as well as the suitability of the methods as basis for advice on fishing opportunities for short-lived species. d) Evaluate appropriate methods to estimate MSY and PA reference points for short-lived species when no apparent stock-recruitment relationship is found. e) Compare the long term performance of escapement strategies and F based strategies in terms delivering MSY, inter-annual variability in yield and risk of SSB falling below Blim. The application of escapement strategies should include both deterministic and stochastic approaches to the short term forecasts. The framework of these comparisons should be made within the ICES MSE framework. Some work may have already been completed in this area using this approach and so the workshop should include a review of existing simulation work. f) Draft guidelines for benchmark and assessment working groups providing background for decisions on short-term forecasting methodologies given the biology, data, and assessment approaches for short-lived species. These guidelines should acknowledge the statistical merits of each approach. g) Consider how to handle the assessment of criterion 1 (level of fishing pressure) of the MSFD Descriptor 3 for stocks with an escapement strategy where, by definition, no FMSY is defined. This workshop should mainly consider stocks with an analytical assessment and forecast (category 1 & 2 stock), e.g. North Sea sandeel stocks (SA 1, 2, 3 and 4), North Sea sprat, North Sea Norway pout, Barents Sea capelin, Capelin in the Iceland-Greenland- Jan Mayen area, and Bay of Biscay anchovy. WKMSYREF5 will report by 27 September 2017 for the attention of ACOM. 1.2 Conduct of Meeting The meeting was held at the Institute for Marine and Coastal Environment (IAMC- CNR), based in Capo Granitola, Sicily in Italy over the period September There were 13 participants physically present, and three that participated remotely at various times (including one that gave two of the presentations). There were 11 presen-

7 ICES WKMSYREF5 REPORT tations in total, covering various aspects of TORs a-e. A compilation of forecast methodologies for the stocks listed in the TORs (North Sea sandeel stocks: SA 1, 2, 3 and 4, North Sea sprat, North Sea Norway pout, Barents Sea capelin, Capelin in the Iceland- Greenland-Jan Mayen area, and Bay of Biscay anchovy) was conducted during the meeting, along with a description of the characteristics that influence the forecast methodologies used. TORs f and g were covered during the meeting through discussions and the material presented for the other TORs. 1.3 Structure of the report The report has been structured along the lines of the TORs. Because they were closely aligned, we decided to combine TORs b and c (forecasting methods for short-lived species; Section 2) and TORs a and d (current ICES stochastic forecast procedure and reference points for short-lived species; Section 3). TOR e (evaluating the long-term performance of forecast methods for short-lived species using MSE) was covered in Section 4, TOR f (guidelines for conducting short-term forecasts for short-lived species) in Section 5, and TOR g (Criterion 1 for MSFD Descriptor 3 for short-lived species) in Section 6. Presentations (11 in total) were offered for TORs a-e (Sections 2-4), and these are presented as P1-P11 across the relevant sections with a short description of the presentation (Sections 2.1, 3.1 and 4.1). The one substantial piece of work that was conducted during the meeting was the compilation of a description of forecast methods and their features for the stocks highlighted in the TORs (Section 2.2). A summary of the discussion held following presentations is given in each of the sections (Sections 2.3, 3.2 and 4.2). Section 7 provides concluding remarks and summarises findings/recommendations from Sections 1-6 of the report.

8 4 ICES WKMSYREF5 REPORT TOR b and c Forecasting methods for short-lived species 2.1 Presentations P1: Stochastic forecast methods for SMS Mollie Brooks A review of forecast methods available for SMS models was presented to the group. The sources of this information were a working document and a presentation by Morten Vinther for WKSand (ICES 2017a). Some of these stochastic forecast methods are not published in peer-reviewed literature and have not been assessed in a management strategy evaluation framework. In general, deterministic forecasts set the TAC deterministically based on reference points while stochastic forecasts set the TAC using stochastic predictions of the probability distribution of escaped SSB (SSB remaining after the TAC was taken). Stochastic predictions of the escaped SSB can be made in one of three ways described below. A TAC can be chosen such that the 5th percentile of the distribution of escaped SSB is located at Blim, thus predicting a 5% probability of the escaped SSB dropping below Blim. Searching for this precise TAC requires re-estimating the distribution of escaped SSB for each TAC considered. An estimate of the distribution of the escaped SSB could come from a likelihood profile, but that would be computationally intensive. SMS estimates the mean and variance of escaped SSB expected after the TAC has been taken. Assuming a log-normal distribution on the escaped SSB gives a close approximation to the profile likelihood. The stochastic forecast can be done using uncertainty of the joint distribution of the number of individuals in each age class at the beginning of the TAC year (N) and the exploitation pattern, i.e. selectivity (E). SMS provides estimates of the joint distribution of N and E, defined by the mean vector, vector of standard deviations, and correlation matrix. The assumptions of SMS that go into estimating the joint distribution are described in the assessment section of Table Essentially for sandeel, N for ages one and over is based on estimated N and Z (fishing mortality plus natural mortality); N at age zero is generally the geometric mean of estimated recruitment over all past years except the most recent; fishing mortality at age is estimated in the last year of the assessment; the assessment has no intermediate year; natural mortality at age is a fixed input of the model. N and E are the only stochastic variables used in the forecast. In this forecast method, 10,000 samples from the joint distribution of N and E are taken. For a range of TACs, the samples are used to predict escaped SSB. This prediction uses the assumptions described in the Forecast Biology & Ecology section of Table Then, the TAC that gives a 5% probability of SSB < Blim is chosen. Similar to the method of getting samples of N and E from their estimated joint distribution, these samples can be taken from the SMS model s posterior distribution by MCMC sampling. In the posterior, N and E are varying with all estimated parameters, rather than just with each other; so it may include more uncertainty than other stochastic forecast methods. This method estimates lower TAC than the other forecast methods when uncertainty is high. This method requires checking that a sufficient number of independent samples have been chosen so that the tails of the distribution are adequately characterized. Collecting adequate independent samples is sensitive to features of the posterior distribution and requires expert knowledge of MCMC sampling methods.

9 ICES WKMSYREF5 REPORT P2: Stochastic forecast using N and E distribution Mollie Brooks One method of stochastic forecast as briefly described above uses the joint distribution of N and E to simulate the uncertainty in these values. The distribution is assumed to be lognormal. Means, variances, and a correlation matrix for log(n) and log(e) are extracted from the output file sms.cor. The mean of log(n) of age zero individuals in the mean vector is reset to the log of the geometric mean of estimated N for all years except the most recent. Then, these distributions are used to simulate values of log(n) and log(e); then exp(log(n)) and exp(log(e)) are taken as samples of N and E. This transformation ensures that the median (not the mean nor the mode) of N and E with lognormal distribution is the same as the best estimate of N in the SMS model; the choice of the median has not been evaluated to our knowledge. Different means, variances, and correlation matrices for log(n) and log(e) can be extracted for each year by doing a retrospective analysis. We investigated the choice of which correlation matrix is used in the forecast, from years 2011 to It is important to be aware that when using a retrospective analysis to generate distributions for all years, some years are not using all of the information that would be available in a true forecast. This is because SMS assumes that the exploitation pattern is separable with one component estimated as constant for a defined number of years. A breakpoint for the most recent constant period is set at 2010 in SAN-area-1r and 2r. Consequently, parameter estimates in 2011 have high uncertainty. A simulation study was conducted to investigate the effect of uncertainty (year-to-year variation in the estimated correlation matrices). In SAN-area-1r and 2r, the area-specific means and standard deviations as well as the reference point (Blim) and forecast inputs (median numbers at age, weight at age, natural mortality, and proportion mature) for 2016 were used for all stochastic forecasts. For each year from 2011 to 2016, the correlation matrix from that year was used to do five stochastic forecasts. Seeds for the random number generator were set to one through five for the five replicates, in case identical simulations need to be reproduced and examined in more detail. All stochastic forecasts used samples of N and E from the specified joint distribution (double the number of samples that ICES demonstrates will provide an estimate of the probability of SSB<Blim with high precision; Section 3.3 of ICES 2013a [WKGMSE]). In spite of the high number of samples, the five replicated stochastic forecasts per specified distribution still showed considerable variation (Figure 2.1.1). For a given distribution, the TACs from the five stochastic forecasts had a range of 5000 tonnes. In both areas, using the correlation matrix from 2011 lead to a lower TAC; other years were slightly higher and nearly equivalent, but with a slight upward trend. The most recent correlation matrix will contain the most accurate estimates as it has the most number of years after the 2010 breakpoint. Thus we recommend using the most recent correlation matrix for the forecast.

10 6 ICES WKMSYREF5 REPORT 2017 Figure Using correlation matrices of N and E for different years, five replicated stochastic forecasts were performed. Each forecast had the same mean, standard deviation, forecast inputs, and reference points respective of assessment area. P3: Multispecies assessments of M from WGSAM; maturity Anna Rindorf Estimates of natural mortality The WGSAM implementation of the North Sea SMS model (ICES 2013b) estimates natural mortality of sprat, Norway pout and sandeel based on the stock assessment data and stomach content information for a large range of predators. Due to the limited coverage of stomach samples, sandeel are included as two stocks: northern sandeel and southern sandeel. This spatial split is similar to the stock definition applied in the single species assessment before 1996 (ICES 1995; 1996). Specific details on the model can be found in ICES (2013b [WGSAM]) and ICES (2017a [WKSand]). The SMS estimate of predation mortality (M2) of all three short-lived stocks is high, especially for the younger ages (Figures 2.1.2a-d). For the southern sandeel stock (Figure 2.1.2a) mackerel, whiting and seabirds are the main predators, while haddock, saithe, whiting and grey seals are the main predators for the northern stock (Figure 2.1.2b). For sprat, the main predators are whiting, mackerel and seabirds (Figure 2.1.2c), while for Norway pout, the main predators are saithe, whiting and cod and lately also hake (Figure 2.1.2d). Mackerel (combined North Sea and Western components) is the major predator, with rather high partial mortalities in the second half of the year for sandeel. Even in quarter 1 and quarter 4 where age 1+ sandeel are in the sediment the majority of the time, M2 from mackerel is high (ICES 2017a). The group considered that WGSAM should look into the possible causes of this pattern. Predation rates are estimated by WGSAM every three years, and on these occasions, the general settings of the model are also updated if deemed necessary. For example,

11 ICES WKMSYREF5 REPORT in 2011, sprat was added as prey in the model, and in 2014, hake was added as a predator. As a result, the estimates of natural mortality of each species may change somewhat back in time. However, the temporal patterns tend to be relatively stable between updates (Figure 2.1.3, Table 2.1.1). WGSAM recommends using a smoothed version, for example 3-year averages, before including natural mortalities in annual stock assessments. They also recommend not using trends to extrapolate the time series, but instead using the terminal year value for subsequent years. Furthermore, they recommend considering the effects of new key runs on stock-recruitment relationships before updating time series outside benchmarks. If the effect on the stock recruitment plot (shape rather than level) is minor, the time series can be updated to use the new time series even outside a benchmark. Finally, to be used in half-yearly assessment models, the quarterly values of M must be combined to provide M by half-year. Table Correlations between time series of natural mortality based on the 2008, 2011 and 2015 key runs (ICES 2008, 2011 and 2015a [WGSAM reports]). KEY RUNS COMPARED AGE 1 AGE vs vs vs S. sandeel age: 2 Whiting Cod Harbour porpoise Mackerel G.gurnards R.radiata Birds S. sandeel age: 0 S. sandeel age: S. sandeel age: S. sandeel age: Figure 2.1.2a. Partial annual predation mortality (M2) of southern sandeel.

12 8 ICES WKMSYREF5 REPORT 2017 N. sandeel age: 2 Saithe Haddock Whiting Cod Harbour porpoise Grey seal H. mackerel Mackerel G.gurnards R.radiata Birds N. sandeel age: 0 N. sandeel age: N. sandeel age: N. sandeel age: Figure 2.1.2b. Partial annual predation mortality (M2) of Northern sandeel. Figure 2.1.2c. Partial annual predation mortality (M2) of sprat.

13 ICES WKMSYREF5 REPORT Figure 2.1.2d. Partial annual predation mortality (M2) of Norway pout.

14 10 ICES WKMSYREF5 REPORT 2017 Annual natural mortality Year Age Age Age Annual natural mortality Year Age Age Age Figure Estimates of annual natural mortality based on 2008, 2011 and 2015 key runs of the multispecies model SMS (ICES 2008, 2011, 2014a and 2015a [WGSAM reports]). The values of the 2015 key run are derived as the average of northern and southern areas, weighted by the abundance of the age group in the beginning of the year. Natural mortalities used in assessments The annual natural mortalities estimated from SMS are used directly in assessments of North Sea sprat, sandeel in area 1r and sandeel in area 3r. Norway pout and the remaining sandeel stocks use a constant natural mortality. For Bay of Biscay anchovy natural mortality is fixed at 0.8 for age 1 and 1.2 for older individuals (age 2+).

15 ICES WKMSYREF5 REPORT Maturity at age Maturity at age is highly variable between years for at least some of the short lived species. However, as there seems to be no link between recruitment and SSB, and there was no trend in the observed values, the sandeel assessments do not apply this variable maturity in the assessment, but instead uses an average value. In contrast, North Sea sprat uses highly variable annual maturities at age directly in the assessment (Figure 2.1.4). Figure Proportion North Sea sprat mature by age. From ICES (2017b [HAWG]). Comparison of scale of variability in natural mortality, weight at age and maturity For North Sea sprat, variable maturity, natural mortality and mean weight at age were all available from the assessment. Using the same method as described for P4 below (MARE) the variation in the different sources can be compared (Table 2.1.2). Table Interannual variability (MARE) of natural mortality, weight at age and maturity for North Sea sprat. SOURCE AGE 1 AGE 2 AGE 3+ Natural mortality Weight at age Maturity P4: Variation in catch weight-at-age Casper Berg [with additions from Anna Rindorf] Mean absolute relative error (MARE) on forecast weight-at-age data from various stock assessments were compared (WD1). The weight-at-age data were derived from stockassessment.org, except for sprat, sandeel, and Norway pout in the North Sea, where data from the ICES assessments were used. Only ages 1 8 were considered for all stocks (if present). Constant values of weight-at-age in the beginning of some of the time-series were removed from the data, because these are likely assumed values rather than real data. For sandeel only data from the first half year is used, for sprat only

16 12 ICES WKMSYREF5 REPORT 2017 quarter 3, and for Norway pout only quarter 4. The stockassessment.org data only contains annual means. Next year s weight at age were predicted using a 5-year running average by age group and stock. The predictions were compared to the observed values using mean absolute relative error (mean over years), observed predicted MARE a = mean ( ) predicted Finally, the average MARE over all age groups were computed for easier comparison over stocks. The median of the age-averaged MARE by stock is 0.09 and the mean is Capelin, sprat, sandeel, and Norway pout in the North Sea all have values above those values (0.12, 0.12, 0.13, and 0.16 respectively, Table , Figure 2.1.2). However, within the particular age groups, the variation of the short-lived species is slightly below the average across all stocks (short lived=0.14, other stocks=0.16) while for age 2, the value for short-lived species is slightly above that of other stocks (0.10 vs 0.086) (Figure 2.1.3). The group concluded that there was little evidence that variability of short lived stocks was greater than for other stocks. However, whereas this variation will not matter to short term forecasts using FMSY or a deterministic forecast aiming at Bescapement, it will matter to methods using the probability distribution of future SSBs. This includes both stochastic short term forecasts and all long term forecasts where the estimated probability of falling below Blim is having an effect on the agreed FMSY or the FMSY range.

17 ICES WKMSYREF5 REPORT Table MAREa for all stocks. STOCK/AGE AVG Blue whiting North Sea Plaice Western Baltic Plaice Eastern Baltic American plaice NAFO Div 3LNO Atlantic cod Western Baltic Atlantic cod Kattegat Atlantic cod Faroe plateau Atlantic cod North Sea Atlantic herring Celic Sea & WOScot Baltic herring Bothnian sea Atlantic herring North Sea Baltic herring Gulf of Finland Baltic herring Western Baltic Baltic herring Central Baltic Atlantic herring US Gulf of Maine/Georges Bank Saithe Barents Sea Saithe North Sea Saithe Iceland Haddock North East Arctic Haddock Northern Shelf (North Sea) Sole Kattegat and Skagerrak Sole North Sea Whiting North Sea Capelin Iceland Greenland halibut Barents Sea Sprat North Sea Sandeel North Sea Norway Pout North Sea Figure MARE averaged by stock.

18 14 ICES WKMSYREF5 REPORT 2017 Figure MARE by age group for each stock. Shortlived stocks are in shades of pink. P5: Assessment and management of Bay of Biscay anchovy ( ) Leire Ibaibarriaga et al. This work summarises how the assessment and management of the Bay of Biscay anchovy has evolved since 1987 to handle recruitment uncertainty. It covers issues like the most adequate management calendar depending on the available information, methods to predict incoming recruitment, the fishery collapse and subsequent closure ( ), development of a recruitment index from an autumn acoustic survey on juveniles, and the implementation of a management strategy and its subsequent revision. Currently ICES advice for Bay of Biscay anchovy is based on a management strategy (STECF 2013, 2014). This consists of a harvest control rule that sets the TAC for January to December according to the expected spawning stock biomass in that year. The expected spawning stock biomass is computed from a deterministic projection of the population using the medians of the parameters estimated in the assessment (recruitment in the management year, remaining population at the end of the last assessment year, intrinsic growth and selection at age). In addition, natural mortality at age is assumed fixed (set at the same values as in the assessment) and the percentage of the catches taken in the first semester is set at 60% (as specified in the management strategy). The alternative catch options (annual TAC from the management strategy and other catch levels) are evaluated in terms of the probability of spawning stock biomass in the

19 ICES WKMSYREF5 REPORT management year being below Blim, median spawning stock biomass in the management year, % of TAC change from previous year and harvest rate (ratio of catch to spawning stock biomass). These are computed from a stochastic projection of the population (ICES 2013c [WKPELA], ICES 2013d [WGHANSA]) based on the assessment model equations. The next incoming recruitment, remaining population at the end of the last assessment year, intrinsic growth and selection at age are sampled from their joint posterior distribution, whereas natural mortality at age is fixed and constant over time. The posterior distribution of the next incoming recruitment distribution is mainly obtained from the latest JUVENA juvenile abundance index and the parameters of the JUVENA observation equations estimated from the assessment model (Figure 2.1.4). In the past, when the JUVENA survey was not included in the assessment and the management calendar was from July to June, alternative recruitment scenarios were constructed as a mixture of the past series of posterior distributions of recruitments as: w y p(r y. ) y where p(r y. ) denotes the posterior distribution of recruitment in year y and w y are the weights of the mixture distribution, such that y w y = 1. When no information about incoming recruitment was available, all the years were equally weighted, resulting in an undetermined recruitment scenario (Figure 2.1.5). Figure Distribution of next incoming recruitment for 2017 as assessed in the December assessment after the inclusion of the latest JUVENA index.

20 16 ICES WKMSYREF5 REPORT 2017 Figure Distribution of next incoming recruitment for 2014 as a mixture distribution of past recruitments. 2.2 Compilation of Forecast Methods Among the tasks entrusted to WKMSYREF5 was the preparation of a list that would include a representative sample of the different forecasting methods for short-lived species used within ICES, which would provide a broad perspective on the biological and fishing characteristics of relevance for the selection of the methods. The result of this review is shown in Table To develop this table, it was agreed to carry out a review of a representative number of short-lived stocks: Barents Sea capelin, Iceland capelin, 4 stocks of North Sea sandeel, Bay of Biscay anchovy, North Sea sprat and Norway pout. A large number of working papers describing benchmark processes, as well as advice sheets and some peer reviewed scientific publications were used to collect the necessary information. Finally, the stock-assessors and specialists in these stocks were asked to review this table to ensure that all the information included was correct and that details of relevance were not lacking. The table has been organized under six main headings, to cover the following topics for each stock: Scientific surveys: The existence or not of surveys for both adults and juveniles, as well as the date when surveys are usually carried out. Biological and fishery aspects considered in assessments: Among the elements analysed in this section, some are specific to the time and methodology used to carry out assessments, while others are more focused on biology and ecology, such as the way growth and maturation processes are modelled and/or if temporal variations are included. The inclusion of trophic interactions in the estimation of natural mortality over time was also included. Finally, the way in which the exploitation pattern is considered/estimated in the assessment is described. Biological aspects considered in the forecasting method: This section compiles all the relevant information for the development of the short-term projections from the ecological and biological point of view. Among the points considered of importance are the source of the initial population size and structure from which the population is projected, and the processes that influence the productivity of the stock during the simulated period, such as growth and natural mortality. The maturation process, the

21 ICES WKMSYREF5 REPORT spawning time, as well as the way in which recruitment prediction is made are also included in the table. Aspects of the fishery considered in the forecasting method: This section summarises elements such as the time during which the fishery usually takes place, as well as the origin of the information for the exploitation pattern used in the forecast. Finally, it was considered of relevance to highlight whether these methods included estimates of the percentages of recruits and immature fish that would be caught by the projected fleets. Additional information on forecast methods: This includes aspects such as whether the forecast is done using the assessment model or not, the forecast period and the treatment of the uncertainty that has been considered, if it has been included. Basis for the advice: This section describes the time of the year when the SSB in the projection is compared with the SSB reference points, the different criteria used to determine the value of Blim used for the advice, and finally the criterion used to define the Bescapement.

22 18 ICES WKMSYREF5 REPORT 2017 Table Forecast methods and characteristics of the data and assessments for a selection of short-lived stocks. Survey Assessment Stock Anchovy Bay of Biscay North Sea sandeel 1-4 North Sea sprat Forecasting method Adult Survey/Timing Juvenile Survey/Timing Assessment time Assess. framework Growth rate Maturity Natural Mortality Exploitation pattern Stochastic CBBM Bioman: DEPM Spring PELGAS: Acoustic Spring Deterministic Bescapement with Fcap (SMS model) Age 1/winter dredge survey Deterministic Bescapement with Fcap (SMS model) IBTS Winter and Autumn Summer Acoustic Survey North Sea Norway pout Stochastic SESAM IBTS Winter and Autumn JUVENA September Juveniles/winter dredge survey IBTS Winter Survey IBTS Autumn Barents Sea capelin Stochastic CapTool Barents Sea acoustic survey in August-September Barents Sea acoustic survey in August-September December January March October October CBBM SMS SMS SESAM Estimated in the assessment and kept constant along time but distinct across age groups Fixed all mature at age 1 Constant over time, variable by age. Uriarte et al 2016 Estimated in the assessment and kept constant along time but distinct across age groups Weight-at-age is estimated from commercial catches and varying from year to year Proportion mature by age is kept constant over time Depends on the management area. Varies from year to year in area1r and area 3r, but not in area 2r. Can be constant in the last years, since the SMS multispecies model producing the numbers is not updated annually Estimated in the assessment model and assumed to be constant within defined separability periods. Weight-at-age is estimated from commercial catches and varying from year to year Proportion mature by age varies from year to year Varies from year to year, but can be constant in the last years, since the SMS multispecies model producing the numbers is not updated annually Estimated in the assessment model and assumed to be constant within defined separability periods. Weight-at-age in the stock is estimated from survey and maintained constant over time Proportion mature by age constant over time Natural mortality at age/season derived from survey analyses and constant over time Variable by age and time Acoustic survey & Bifrost No increase mature and immature weight at age a by from October-March Bifrost: maturity as a function of length (cutoff at approximately 14cm) Bifrost. Predation from cod + Residual Natural mortality. Constant for all ages Fishing only on mature capelin in January- March (for info - minimum landing size is 11cm) Adult Icelandic capelin Simulation model starting with bootstrap replicas from acoustic surveys. September/October and January/February acoustic surveys September/October Acoustic survey October year i Revision February year i+1 Acoustic measurements in January assumed to represent the true stock size. Weight increase, change in q and reduction in numbers cancel each other approximately out from October-January Mature/immature proportions identified autumn and winter acoustic surveys. Total predation estimated for the period January-March taking into account uncertainty in feeding parameters and overlap between capelin and its predators. Fishing only conducted on the mature part of the stock, mostly in the last 2 months before spawning. Age range 1 to

23 ICES WKMSYREF5 REPORT Forecast Biology & Ecology Forecast Fishery Initial Stock in forecast Recruitment prediction Growth rate Maturity Natural Mortality Timing spawning F &M before spawning Timing of fishery Exploitation pattern Percentage immature Posterior distribution of SSB in the last assessment year Posterior distribution of recruitment in the last assessment year Posterior distribution of growth rate by age from the assessment Fixed all mature at age 1 Constant over time, variable by age. Uriarte et al 2016 Numbers at age projected by stock assessment model Long-term geometric mean recruitment, unless regime shifts is indicated then geometric mean of the latest regime is used. Constant weight at age in stock and catch (average of last 3 years) Assessment value/constant at age over time Taken from the assessment model. Multispecies SMS 3 years average Numbers at age projected by stock assessment model Long-term geometric recruitment mean for the entire time-series Constant weight at age in stock and catch (average of last 3 years) Average last 10 years Taken from the assessment model. Multispecies SMS 3 years average Posterior distribution of SSB in the last assessment year Posterior distribution of recruitment in all years in the assessment Constant weight at age in stock and predicted weight in the catch. Taken from the assessment Constant at age over time Natural mortality at age/season derived from survey analyses and constant over time Based on numbers and weight at age, September acoustic survey None as age 0 (some predictions of age 1 abundance based on estimates of age 0 abundance from 0-group trawl survey) None Taken from the Arctic Fisheries WG assessment. Bifrost Taken from the Arctic Fisheries WG assessment. Bifrost Bootstrap replicas of biomass from January and Autumn surveys. October surveys get lower weight as the measurements are earlier in time. From acoustic measurements of age 1. Usually 1-2 acoustic measurements on the adult stock are added before first fisheries on the yearclass take place. Increase stock by 13% from October-January Indices from acoustic measurements are split into immature and mature part. From predation model in January March. Cancels out with weight increase and different q from October-January. Spring January Summer Winter April March. Proportional to the elapsed time until spawning All year Fishing mortality by age/semester from the posterior distribution of the assessment 0% (Age 0 not in the assessment) Zero April-June Taken from the last year in the assessment model Zero Predominantly autumn and winter. For Q1-Q2 catches are assumed as the average % relative to Q3_Q4 of last three years (%of TAC) Taken from the last year in the assessment model Predominantly autumn and winter Recent exploitation pattern according to changes between seasons and ages. --- Yes, estimated by Bifrost January-March Monthly pattern set according to monthly pattern observed in recent years, fishery only on mature capelin None (observations of actual catch indicate < 5%) M estimated by predation model and all fisheries are conducted before spawning. July March but usually only January March. Mature part of the stock is fished. ~0%

24 20 ICES WKMSYREF5 REPORT 2017 Forecast Additional details Basis for the advice Percentage incoming recruitment Forecast within Assessment model Forecast period Assumptions/Uncertainty Date Comparison SSB and reference points Blim 57% (Average percentage of age 1 in the catch) 0% N/A ~0% Yes Yes Yes Yes CapTool The model is only a forecast model, taking bootstrap replicas from acoustic surveys as true stock size. January-December January -December July -June November-October October-April October-January Stochastic MCMC all parameters from joint posterior distribution Deterministic Bescapement with Fcap Deterministic Bescapement with Fcap Stochastic projections from joint posterior distributions of N and F in terminal year and recruitment for all years No uncertainty in catches Uncertainty in survey catches with CV of 0.2 based on the historic replicates Uncertainty in the acoustic measurements is estimated by bootstrapping Stochasticity in some of the predation model parameters and spatial overlap with predators. May January July October January Median of SSB estimates in the years 1987 and 2009, the minimum estimated biomass that produced substantial recruitment Sandeel 1: The lowest SSB at which a high recruitment is observed Sandeel 2: Average SSB of the two lowest SSB estimates providing high recruitment (2001, 2009) Sandeel 3: The lowest SSB at which a high recruitment is observed Sandeel 4: Average SSB of two lowest SSB estimates providing high recruitment (2003, 2009) Average of different approaches to determining Blim Bescapement No Bpa Bpa Fcap No Yes (estimated in MSE to ensure that P(SSB<Blim) does not exceed 5% in the long term) Yes (estimated in MSE to ensure that P(SSB<Blim) does not exceed 5% in the long term) Blim = Bloss, the lowest observed biomass in 2005 Quota set so that P(SSB>Blim) =95% Above SSB1989, the lowest SSB at spawning time that has produced a good year class Quota set so that P(SSB>Blim) =95% Blim is the average of the 3 lowest values of SSB from the years 1990, 1981 and 1982, all leading to average yearclasses. The value is 150 thous. Tonnes Earlier rule was based on Bescapement=400 th tonnes. The current rule is more precautionary it includes more predation and No No No.

25 ICES WKMSYREF5 REPORT Discussion Forecast methods The diversity of forecast methods currently applied for the six selected stocks was surprisingly high (Bay of Biscay anchovy, North Sea sandeel 1-4, North Sea sprat, North Sea Norway pout, Barents Sea capelin, and adult Icelandic capelin). Only North Sea sandeel and North Sea sprat relied on the same forecast method (Deterministic Bescapement forecast using Bpa as target). Three stocks utilize a type of stochastic forecast, but based on very different methods. One stock, Icelandic capelin, uses a survey based forecast approach (as opposed to forecasts based on stock assessment outputs). There is no clear pattern in respect to the choice of forecast methodology and biology, assessment input (configurations), etc. It appears that the choice of forecast methods relates to the stock assessment model applied and/or reflects non-generic tools developed to fit special requirements and local expertise. For example, when the SMS stock assessment model is used (sandeel and sprat), the built-in forecasting tool is applied, and the same can be said for Norway pout that runs in the SESAM stock assessment model. The pending question is whether it is advisable to write guidelines on how to select forecast methods from a generic tool box. Another observation is that only some of the stocks apply forecast methods/strategies that have been evaluated in an MSE (anchovy, sandeel and sprat). The anchovy case is unique in the sense that it applies a deterministic harvest control rule approach to set the TAC, but uses an MCMC based stochastic forecast to calculate the risk associated with the advised TAC (which does not affect the harvest control rule, but is used as supplementary information for managers). None of the case study stocks implement patterns of auto correlation or environmental indicators of recruitment strength or other inputs to the forecast. Recruitment Recruitment of short lived stocks is often very poorly related to SSB (see Section 3.1, P7 on Blim) and the variation in recruitment should always be included in stochastic forecasts (short as well as long term). In contrast, recruitment often shows a high degree of correlation with environmental variables, density of predators on juvenile stages (such as herring predating on capelin) and density dependence (such as is seen for North Sea sandeel, where a large year-class is never followed by another large year-class in the subsequent year). In short term forecasts for short-lived pelagic stocks, recruitment is generally forecast without accounting for these other sources of variation, even in cases where they have been demonstrated repeatedly in scientific studies. In some cases, environmental variables or surveys of eggs and larvae show a high correlation with the recruitment estimated in the stock assessment or survey time series. In these cases, it should be investigated whether the time series in question significantly improves the prediction of recruitment. If this is the case, these parameters could potentially be included to improve the CV of stock numbers in the short term forecast if the considerations on availability of data and causality of the relationships given below are accounted for. A lack of updated information on environmental variables occurs both when the environmental variable cannot be observed before the time of the assessment (e.g. when the recruitment of 0-group in the TAC year is determined by temperature during the

26 22 ICES WKMSYREF5 REPORT 2017 TAC year), or when the processing of environmental variables is not integral to the ICES system. In the latter case, an effort should be made to ensure that the information can be made available to the assessment WG before the time of the assessment to which it may contribute added information. A lack of knowledge of the causality of the relationship between recruitment and a particular variable also accounts for their lack of use in assessments and forecasts. There are numerous examples of correlations between different variables breaking down over time or turning out to be spurious at closer examination (Myers et al. 1998, Punt et al. 2014). Since our correlations are based on observations being on a time scale (time series), this may result in relationships initially inferred from these correlations later being discovered to be of a spurious nature (time being the common and confounding effect). Causal mechanisms should be well understood before this additional information is used in forecasting (De Oliveira and Butterworth 2005, De Oliveira et al. 2006, Bogstad et al. 2013). The causal link should be formulated as a clear hypothesis and be confirmed by other available data before being introduced in short term forecasts. It should also be considered whether relationships are nonlinear. For example, a large year-class of sandeel in area 1 seems to be consistently followed by a small yearclass. However, a small year-class can be followed by either a large or a small yearclass. Hence, the relationship cannot be included in short term forecast by introducing a simple autocorrelation as this would lead to the prediction of a high incoming yearclass following a low year-class, though this is not supported by data. Incorporating variability in recruitment, natural mortality, weight at age and maturity in short and long term forecasts To provide a scientific basis for a decision on whether variation in recruitment, natural mortality, weight at age or maturity should be accounted for in short term forecasts, this section has described analyses of the variation in natural mortality, maturity (Section 2.1, P3) and mean weight at age (Section 2.1, P4) in short lived species and other fish stocks. In many cases, one or all of the variables recruitment, natural mortality, weight at age and maturity will exhibit long term temporal trends or other types of persistent long term changes. In these cases, the most recent level should be used in short term forecasts. This is currently the case for some of the short lived species, but not all. In one stock (North Sea sprat), natural mortality, weight at age and maturity all varied between years, whereas for the remaining stocks, one or more of these was kept constant. Adding this variability to stochastic forecasts would tend to increase the variation around the estimated SSB, except if there are correlations between these parameters which counteract this effect. For example, the greatest variation for any of these variables in North Sea sprat was for maturity at age 1 (Table 2.1.2), which varied between 15 and 70%. Including this in short term forecasts would add substantial variation that is not related to fisheries management. Furthermore, as recruitment shows no relationship with SSB, the actual maturity level does not seem to have any effect on the probability of experiencing a low recruitment. A similar case can be made for weight at age. Hence, if an objective of the management target is a close link to the management measures suggested, it may be preferable to exclude the variability in weight at age and maturity from the short term forecast and from evaluations of the historical performance of management.

27 ICES WKMSYREF5 REPORT Natural mortality poses a special problem as often annual data for this are unavailable or are only updated at 3-year intervals. The structural uncertainty in the estimates can be substantial in the last years of the estimation, in particular if the predators exhibit strong retrospective patterns as was the case for eastern Baltic cod. This uncertainty is not easily quantified. However, natural mortality estimates seem to change slowly over time due to the slow development in predator stock size, at least when derived from models such as those presented in WGSAM. Finally, variation in natural mortality and weight at age can be highly correlated and this correlation structure needs to be taken into account if the variability is to be included in stochastic short or long term forecasts. In the investigation, there was no evidence to support that this issue was greater for short lived stocks than long lived stocks, and hence, this should be a focus in evaluations of both types of stocks.

28 24 ICES WKMSYREF5 REPORT TOR a and d Current ICES stochastic forecast procedure and reference points for short-lived species 3.1 Presentations P6: SESAM - Seasonal State-space Assessment Model Applied to Norway Pout in the North Sea Casper Berg and Anders Nielsen [with additions from the WKPOUT report (ICES 2016) Norway Pout advice sheet] SESAM (Seasonal SAM) is a seasonal extension of the SAM model (Nielsen and Berg, 2014). The model preserves useful properties of the SAM model, namely that the fishing mortality is specified via an unobserved stochastic process that allows for gradual changes in both fishing pressure and selectivity, and catches are treated as observations with noise (ICES 2016a). This presentation was specific to the application of SESAM to Norway Pout in the North Sea. The former Fcap and MSY Bescapement reference points for Norway pout are no longer used because the forecast is now stochastic and uncertainties in the assessment and forecast are directly taken into account to ensure the SSB immediately following the fishing season stays above Blim with 95% probability (ICES 2016a). The forecast provides a TAC advice according to a calculated yield in the forecast year where the probability of SSB being below Blim by 1 st October in the forecast year is less than 5%, i.e. the forecast estimates the yield according to SSB that meets the 5% criterion at the Blim date (1 st October). Accordingly, the TAC is used for the management year 1 st November 31 st October. This is an approximation and will be sustainable unless radical changes occur in the seasonal fishing pattern used in the forecast. Forecasting in SESAM is performed as follows: 1. Assume values for M, weight-at-age in the catches and in the stock, and maturityat-age for the projection period. Since all of those quantities except weight-at-age in the catches are assumed constant over time, only weight-at-age requires special treatment (see below). 2. Draw K samples from the joint posterior distribution of the states (log N and log F) in the last year with data, and the recruitment in all years. 3. Assume that log Ft = log Ft-S + log ψt, for all future values of t where ψt is some chosen vector of multipliers of the F-process. If ψt = 1 for all t this corresponds to assuming the same level and quarterly pattern in F for all future time-steps as in the last data year. 4. Create K forecasting trajectories starting from the samples of joint posterior distribution of the states. The is done by sampling K recruitments directly from the random walk recruitment process estimated by the model, or from the vector of historic recruitments obtained in step 2, and then projecting the states forward in time using the stock equation with randomly sampled process errors from their estimated distribution. 5. Find ψt such that the fifth (or any other) percentile of the catches (total mass) in the projections equal some desired level (optional). Blim = Bloss, the lowest SSB estimated in quarter 4, was selected because this is closest to the beginning of the fishing season (1 st November), and is considered the most appropriate to use as a Blim reference point because the probability of SSB being below Blim can then be evaluated immediately after the fishing season for which a TAC is being calculated. It was argued that the quarter 4 SSB (an existing output of the SESAM

29 ICES WKMSYREF5 REPORT model) was adequate for this purpose because any attempt to calculate an SSB corresponding to 1 st November would require further assumptions and would effectively only be an interpolation between the quarter 4 and subsequent quarter 1 SSBs, thus unnecessarily complicating the calculation of the SSB. Forecasting weight-at-age in the catches There is substantial variation in weight-at-age in the commercial catches from year to year, which means that usual methods of using running averages will be quite sensitive to the bandwidth of the running average. This is important, since TAC estimates calculated in step 5 above depend directly on the catch weight-at-age. The following models is used: E ( CW a,q,t ) = μ a,q + s(cohort, a) + U t where μ a,q is a mean for each combination of quarter and age, s( ) is tensor product smoothing spline, and U t are normal distributed random effects. There square root transform is used to achieve variance homogeneity in the residuals. P7: Estimating the reference point Blim Mollie Brooks According to the ICES Advice Technical Guidelines, the biomass limit reference point (Blim) for short-lived species is evaluated in the same manner as for long-lived species (ICES 2017c). The procedure consists of the following steps: (1) identifying appropriate stock-recruitment data from the most recent assessment, (2) identifying stock type, and (3) estimation of Blim following the procedure for each stock type. Aspects of estimating Blim using Type 1 and 2 stock types described in ICES Advice Technical Guidelines (ICES 2017c) were investigated in an initial simulation study. Type 1 for spasmodic stocks states that "Blim is based on the lowest SSB where large recruitment is observed unless F has been low throughout the observed history, in which case Bloss = Bpa". This guideline has led to subjectivity in how Blim is estimated in practice for these stocks. The potential to define "large recruitment" using some upper quantile was investigated. Stock recruitment curves were derived from a hockey stick model with a known true value of Blim. We also simulated stock recruitment relationships from Ricker and Beverton-Holt curves which were similar to the hockey stick but have no true value of Blim. The results are preliminary since the parameter values used in the simulations are first approximations. Assuming that the underlying curve is hockey-stick shaped, we found that if the quantile method is used, then the quantile should be set so that several points fall into the "large recruitment" category. Using a longer time series provides more accurate estimates of Blim. Shorter time series and hence a low number of observations resulted in estimates of Blim which were substantially higher than the true value from which the data were derived (Figure 3.1.1). Statistical properties will cause the minimum observed value (Bloss) to decrease with increasing number of samples. In general, the sample minimum is the least robust statistic from a sample and is maximally sensitive to outliers. The methods seemed to result in a downward trend in the estimated Blim over time as more data became available and random variation combined with a weak stock-recruitment relationship increased the probability of observing above average recruitment at low SSB.

30 26 ICES WKMSYREF5 REPORT 2017 The stock recruitment plots for short lived stocks available to the group are shown in Figures 3.1.2a-g below. Based on these plots and the result of the simulation exercise, the group concluded that the relationships were generally erratic and could not be clearly assigned to a single type in the ICES guidelines (Figure 3.1.2a-g). The type used differed between stocks and between benchmarks of the same stock. In general, using the existing guidelines, which sets Blim as the lowest observed biomass, directly resulted in very low estimates of Blim, which might not be appropriate. The issue was further confounded with the strong link between Blim and the management target, which in effect means that setting a very low Blim is automatically resulting in a default management target (Bescapement or Blim for deterministic and stochastic methods, respectively) at a very low level. However, a low level for Blim is not necessarily problematic for some stocks. For example, an arguably low value for Blim in the case of Bay of Biscay anchovy is considered appropriate, and is informed by the fact that the stock experienced the red zone of impaired recruitment in the mid- to late-2000s, and the fishery collapsed as a result. Other methods to define limit reference levels for short-lived species include the Lenfest recommendations to maintain forage fish biomass above 80% (low information level) and 40% (intermediate information level) to 30% (high information level) of the virgin biomass, and fishing at 50 to 75% of the estimated FMSY (intermediate and high information level, respectively) (Pikitch et al. 2012), and the seabird-derived notion to avoid reducing biomass below 30% of the virgin biomass (Cury et al. 2011). All of these approaches require a method to estimate virgin biomass. The group discussed whether a possible solution would be to develop specific guidelines for defining Blim for short-lived species in the ICES Advice Technical Guidelines. In practice, points near the minimum were often averaged (Table 3.1.1) and it may be worthwhile to modify the guidelines accordingly. It may be preferable to use a lower quantile or other summary statistic (other than the minimum) of the SSB observed to give large recruitment for Type 1 SR models. Furthermore, it may be preferable to decouple the definition of Bescapment and Bpa altogether. Table lists the Blim and Bpa values for selected short-lived stocks, along with the basis for calculating them.

31 ICES WKMSYREF5 REPORT Table Blim and Bpa values with the basis for these values, from the 2016/2017 advice sheets. BLIM BPA Stock value basis value Basis Bay of Biscay anchovy Barents Sea capelin t Blim: median of SSB estimates in the years 1987 and 2009, the minimum estimated biomass that produced substantial recruitment t Above SSB1989, the lowest SSB that has produced a good year class (estimated at spawning time) Not defined Not defined Icelandic capelin t Bloss Not defined Norway pout t, 4th quarter North Sea sandeel 1r North Sea sandeel 2r North Sea sandeel 3r North Sea sandeel 4r Blim = Bloss, the lowest observed biomass in t The lowest SSB at which a high recruitment is observed t Average SSB of the two lowest SSB estimates providing high recruitment (2001, 2009) t The lowest SSB at which a high recruitment is observed t Average SSB of two lowest SSB estimates providing high recruitment (2003, 2009) North Sea sprat t Blim was set to ensure that years of very good recruitment mainly occurred when the stock was above Blim, and years of very low recruitment only occurred when the stock was below Blim t, 4th quarter = Blim exp( ) t Bpa = Blim exp(σ 1.645), with σ = 0.17 estimated from assessment uncertainty in the terminal year t Bpa = Blim exp(σ 1.645), with σ = 0.25 estimated from assessment uncertainty in the terminal year t Bpa = Blim exp(σ 1.645), with σ = 0.29 estimated from the assessment uncertainty in the terminal year t Bpa = Blim exp(σ 1.645), with σ = 0.46 estimated from assessment uncertainty in the terminal year t Bpa = Blim exp (σ 1.645), with σ = 0.28 estimated from assessment uncertainty in the terminal year

32 28 ICES WKMSYREF5 REPORT 2017 Figure Each panel shows the estimated Blim versus the length of the time series used for estimating Blim. We used the estimation method specified for spasmodic stocks (i.e. Type 1). The solid lines represent the mean of 10,000 simulations; the grey ribbons encompass the 2.75 to 97.5 percentiles. We assumed that the underlying stock recruitment curve was hockey-stick shaped with a breakpoint at 100 (i.e. the true Blim). We assumed that SSB was observed in a range that included the part of the curve where recruitment is impaired due to low SSB. The amount of variation around the stock recruitment curve varies with each row of panels. The quantile used for defining good (i.e. large) recruitment varies across the columns of panels.

33 ICES WKMSYREF5 REPORT Figure 3.1.2a. Stock recruitment plot of North Sea sprat (ICES 2017b [HAWG]). The hockey stick has a forced breakpoint at the agreed Blim. Figure 3.1.2b. Stock recruitment plot of Area 1r sandeel (ICES 2017a [WKSand]). The hockey stick has a forced breakpoint at the agreed Blim.

34 30 ICES WKMSYREF5 REPORT 2017 Figure 3.1.2c. Stock recruitment plot of Area 2r sandeel (ICES 2017a [WKSand]). The hockey stick has a forced breakpoint at the agreed Blim. Figure 3.1.2d. Stock recruitment plot of Area 3r sandeel (ICES 2017a [WKSand]). The hockey stick has a forced breakpoint at the agreed Blim.

35 ICES WKMSYREF5 REPORT Figure 3.1.2e. Stock recruitment plot of Area 4 sandeel (ICES 2017a [WKSand]). The hockey stick has a forced breakpoint at the agreed Blim. Figure 3.1.2f. Stock recruitment plot of Norway pout (ICES 2016a [WKPout]).

36 32 ICES WKMSYREF5 REPORT 2017 Figure 3.1.2g. Stock recruitment plot of Bay of Biscay anchovy. The vertical dash line represents Blim at 21000t (ICES 2013c [WKPELA]). P8: The project Ecosystem Based FMSY Values in Fisheries Management (ECOSYSTEM- FMSY) Søren Anker Pedersen The objective of the project (duration: ) is to derive FMSY values based on ecosystem functioning for each of ICES data rich fish stocks. These FMSY values could potentially be applied directly by ICES in its routine fisheries advice. The approach suggested in this project is not a full multispecies one, but focusses on adding mainly density dependent growth, maturity and (mainly for cod) cannibalism, to the current single-species approach for estimating FMSY. Thus, managers need not consider the balance between species when using the proposed set of FMSY values. Alternative sets of FMSY values will be based on a meta-analysis of published values for FMSY in multi-species and ecosystem models, including population dynamic characteristics of each stock in terms of growth, maturity and longevity parameters. Surplus production models (SPM) will be applied to some selected data-rich stocks, using estimated F time series as effort. The selection will be based on the dynamic range of F, so that only the best-suited stocks for surplus production models will be analysed. Surplus production models will give FMSY (and BMSY) as an integral part of the model. These are based on actual realisations over the past history of the stock and therefore implicitly include ecosystem functioning in their FMSY (and BMSY) calculations. Further information on the project can be found in (

37 ICES WKMSYREF5 REPORT Discussion Fcap for short-lived stocks Most of the fisheries for short lived species are highly seasonal and all of the assessments address seasonality in some way. For all stocks, the 1-year-olds contribute substantially to the fishery and hence, most models start with 0-groups as the youngest age. Generally, the observations have a temporal resolution which is finer than the annual scale. The assessments are all conducted immediately after the first reliable survey of the incoming year-class. This means that there is in general only one data point to base a large fraction of the catch prediction on. While this is preferable to having no index of the incoming year-class, it is still a method which is inherently sensitive to the survey estimate. Hence, survey estimates should be investigated in detail at benchmarks to ensure that the survey is still unbiased and variability remains the same, as this is generally assumed in assessment models. The reliability on one survey index makes the models highly sensitive to values outside the previously observed range. While a very low value generally does not change the forecast very much even if multiplied by a factor 2 or more, this is not the case for very high recruitment estimates, where error in the observed recruitment is transferred directly to error in the estimated TAC. This error has been addressed in some forecasts by introducing a cap on F. It is unclear if the introduction of a stochastic forecast is sufficient to eliminate the need for an Fcap value as no large scale MSE has been conducted based on the stochastic forecast. In WKSand (ICES 2017a), a series of evaluations were conducted based on selected years, and here it is clear that the stochastic forecasted F can exceed the current Fcap, but it is not clear whether this will affect the long term performance of the management strategy, or even whether this can be investigated in an MSE framework which is relying on a range of assumptions regarding the distribution of recruitment and observation error. Where the Fcap has been estimated for North Sea sprat and sandeel, the value is not related to natural mortality as has been suggested for other types of stocks (Table 3.2.1) and hence, using natural mortality as a proxy for Fcap is inappropriate. Table Fcap and average natural mortality of ages 1 and 2 (consistent with the age range for which F-bar is estimated), and σ standard deviation of ln(ssb) in the terminal year of the assessment. STOCK AVERAGE NATURAL MORTALITY (AGES 1 AND 2) FCAP North Sea sprat Sandeel area Sandeel area Sandeel area Sandeel area Σ Bescapement for short-lived stocks Under a deterministic forecast, Bescapement is in many cases defined as Bpa, which in turn is defined as Blim exp(1.645σ), where σ is the standard deviation of ln(ssb) in the terminal year of the assessment. Bpa is generally defined at benchmarks and not updated between benchmarks. It was agreed that in a stochastic assessment, a stock can

38 34 ICES WKMSYREF5 REPORT 2017 be considered to be below Blim if the probability that SSB<Blim was greater than or equal to 50%. The stock can be considered to be between Blim and Bescapement if the probability of SSB<Blim is greater than 5% but less than 50%. The use of Bpa as the management target links the estimated Blim and the estimated CV in the terminal year directly to catch opportunities. The group discussed how biomass reference points such as Bescapement could be defined for a short-lived stock that uses a stochastic forecast. Several options were suggested (Table 3.2.2). In connection with these, some general observations were made. Firstly, the CV of the estimated SSB is highest in the terminal year, lowest in the middle of the time series and again somewhat higher in the beginning of the time series (Figures 3.2.1a-f). Assuming that Blim is the same through the period and that the stock size remains constant, this means that the status allocated to the stock based on the annual risk of being below Blim varies over the time period (Figure 3.2.2). Following these general observations, the pros and cons of the different approaches are listed in Table In discussions, it was pointed out that in some cases, the CV in the terminal year is fairly constant when assessments are rerun each year and hence a Bescapement, if estimated, would remain approximately constant over time. In other stocks where recruitment contributes substantially to the subsequent SSB, the CV of the terminal year SSB can be more variable. It was also pointed out that the σ for selected short-lived stocks in the North Sea varied between 0.17 (sandeel 1) and 0.46 (sandeel 4) (Table 3.2.1). ICES uses Bpa = 1.4 Blim for many of the longer-lived stocks (1.4 is an upward rounding of e 1.645, where =0.2), and it was suggested that a candidate default value for short-lived stocks could be σ 0.3 (approximately the average of the values in Table 3.2.1) corresponding to Bpa = 1.64 Blim.

39 ICES WKMSYREF5 REPORT Figure 3.2.1a. CV of estimated SSB, recruitment and average F estimated for sandeel in area 1r (ICES 2017b [HAWG]).

40 36 ICES WKMSYREF5 REPORT 2017 Figure 3.2.1b. CV of estimated SSB, recruitment and average F estimated for sandeel in area 2r (ICES 2017b [HAWG]).

41 ICES WKMSYREF5 REPORT Figure 3.2.1c. CV of estimated SSB, recruitment and average F estimated for sandeel in area 3r (ICES 2017b [HAWG]). Figure 3.2.1d. CV of estimated SSB, recruitment and average F estimated for sandeel in area 4 (ICES 2017b [HAWG]). This assessment has very low effort and very few samples in the middle of the period, hence the high CVs in this period.

42 38 ICES WKMSYREF5 REPORT 2017 Figure 3.2.1e. CV of estimated SSB, recruitment and average F estimated for North Sea sprat in 2015 (left) and 2016 (right) (ICES 2015b, 2016b [HAWG reports]).

43 ICES WKMSYREF5 REPORT Figure Effect of a changing CV on the assigned status of a stock with a Blim of t (grey line). The blue line shows the development in CV over time, the red line the resulting lower 95% CI with a constant mean stock size of t and assuming that the confidence interval is symmetrical at the log scale. The indicative yellow and green horizontal line shows the status of the stock (risk of being below Blim being respectively above 5%, or at/below 5%).

44 40 ICES WKMSYREF5 REPORT 2017 Table Options for defining Bpa for short-lived stocks that uses a stochastic forecast, along with pros and cons. OPTION PROS CONS Bpa set at latest benchmark from terminal year CV Bpa updated every year SSB relative to Bpa, defined as the SSB at which, for a given CV, there would be 5% risk of falling below Blim, updated annually (reference level: SSB/Bpa=1) Simple Consistent with methods for reference points for other stocks Provides a conservative estimate of a biomass where the risk of being below Blim is 5% or less, as the CV in the terminal year, which is used to define Bpa, is generally the highest CV of any SSB in the time series Provides the opportunity to estimate biomass quantitatively relative to this level to determine distance from this as an approximation of the target of management Simple to update Reflects that Bpa is not stable Provides the opportunity to estimate biomass quantitatively relative to this level to determine distance from this as an approximation of the target for management Simple to update Reflects that Bpa is not stable Provides the opportunity to estimate biomass quantitatively relative to this level to determine distance from this as an approximation of the target for management Does not reflect that Bpa is not stable Does not reflect that Bpa is not necessarily the management target More difficult to explain to the nonexpert Does not reflect that Bpa is not necessarily the management target Not consistent with methods for reference points for other stocks If using a retrospective analysis, it is not possible to estimate reference points for early parts of the assessment period If using the current estimates of CVs for all years, there will be a tendency for the stock to shift status along the time series even if retaining the same biomass (below Bpa at the beginning and end, above Bpa in the middle of the period) More difficult to explain to the nonexpert Not consistent with methods for reference points for other stocks Does not reflect that Bpa is not necessarily the management target If using a retrospective analysis, is not possible to estimate reference points for early parts of the assessment period If using the current estimates of CVs for all years, there will be a tendency for the stock to shift status along the time series even if retaining the same biomass (below Bpa at the beginning and end, above Bpa in the middle of the period)

45 ICES WKMSYREF5 REPORT OPTION PROS CONS Probability of SSB being below Blim listed annually No reference level for biomass used Simple to update Reflects that Bpa is not stable Reflects that Bpa is not necessarily the management target Provides the opportunity to estimate biomass quantitatively relative to this level to determine distance from this as an approximation of the target of management Reflects that Bpa is not necessarily the management target Likely difficult to explain to the nonexpert Not consistent with methods for reference points for other stocks If using a retrospective analysis, is not possible to estimate reference points for early parts of the assessment period If using the current estimates of CVs for all years, there will be a tendency for the stock to shift status along the time series even if retaining the same biomass (below Bpa at the beginning and end, above Bpa in the middle of the period) Eliminates the opportunity to estimate biomass quantitatively relative to this level to determine distance from this as an approximation of the target of management

46 42 ICES WKMSYREF5 REPORT TOR e Evaluating the long-term performance of forecast methods for short-lived species using Management Strategy Evaluation (MSE) 4.1 Presentations P9: MSE for Bay of Biscay anchovy Leire Ibaibarriaga et al. In the past, the Bay of Biscay anchovy TAC was set at 30 or 33 thousand tonnes regardless of the management advice. Furthermore, the actual catches were sometimes larger than the TAC. The stock and fishery collapse in 2005, due to successive low recruitments, brought to the forefront the need for a change in the management of this stock. In 2008 the European Commission defined the management objectives and launched the process of developing a management plan. During two STECF meetings (STECF 2008, 2009) various Harvest Control Rules (HCRs) were compared using Management Strategy Evaluation (MSE). The operating model was based on the same dynamics as in the stock assessment model (two-stage biomass dynamics) and recruitment was generated according to a Ricker-type stock-recruitment model. Observation and assessment error were considered jointly and estimated spawning stock biomass that forms the basis of the HCR was generated according to a log-normal distribution with a CV of The TAC was then set for a management calendar from July to June next year and no implementation error was considered (the TAC was assumed to be fully taken as long as the stock was large enough). The HCRs considered included an escapement strategy (harvest a proportion of the observed biomass above Blim) and a constant harvest strategy (harvest a proportion of the estimated spawning stock biomass). Dialogue with stakeholders, lead to proposals for new HCRs and suggestions for a maximum TAC and a minimum economically-viable yield below which the TAC should be set to zero. Main performance indicators were related to the stock status, the long-term yield and the TAC variability. The average catch associated with a 0.05 probability of being below Blim was compared among HCRs. The maximum TAC constraint reduced the average catch but stabilized catches and reduced the probability of SSB being below Blim. The minimum viable TAC increased the closure of the fishery but ensured its profitability. Results were very sensitive to a persistently low recruitment scenario. Alternative quota shares among countries did not affect the stock status and the long-term yield, but determined the rentability of the fishery of each country. All the results were presented to and discussed with the SWWAC. The stakeholders proposed an HCR that was finally agreed and adopted by the European Commission and the member states. The management plan was first used to set the TAC for July 2010 to June 2011, after the re-opening of the fishery. The management plan had a clause of revision after three years of application. In addition, in 2012 and 2013, ICES changed the stock assessment methodology and the natural mortality assumptions, and incorporated the juvenile abundance index from the JUVENA autumn acoustic surveys as an indicator of recruitment. Consequently, the management plan was revised in two STECF meetings (STECF 2013, 2014). The MSE was conducted using FLBEIA (Garcia et al., 2017). The initial population and the parameters of the operating model were taken from the latest stock assessment. As before, a Ricker-type stock-recruitment model was used, observation and assessment error were considered jointly and no implementation error was assumed. The management plan was assessed as precautionary in the light of the new assessment settings. New HCRs (trying to avoid discontinuities) and a management calendar change were also

47 ICES WKMSYREF5 REPORT evaluated. With the inclusion of the JUVENA juvenile abundance index, the management calendar for January-December halved the risk of being below Blim and of fishery closure, and gave larger and more stable catches in comparison to the July-June management calendar. Sensitivity to three consecutive low recruitments, to various CVs of observation and assessment error, and to the assumed proportion of catches by semester, were tested. All the results were discussed within the SWWAC. A new HCR with a new management calendar (January-December) was agreed by the different parties and implemented for the first time to set the TAC for A variation of the HCR with a larger maximum TAC of t was approved during P10: Description of deterministic Bescapement forecast/forecast using Fcap Mikael van Deurs The annual advice cycle for sandeel in management area 1r, the MSE basis for the current management strategy, and the management strategy itself (Bescapement with Bpa as target and an Fcap to ensure P(SSB<Blim) = 5%) was described. Fcap was estimated in an MSE model, reviewed in the context of WKMSYREF2 (ICES 2014b). The MSE model follows guidelines on how to carry out an MSE (Punt et al. 2016, ICES 2013a [WKGMSE]) and adheres to risk type Prob 1 (see ICES 2013a), with weight-at-age, maturity and M assumed to be constant. The MSE follows the shortcut approach, where rather than having the full assessment model in the simulation loop, the stock assessment model behaviour is approximated by generating assessment stock numbers-atage from a distribution with a mean equal to the numbers-at-age in the true stock (the operating model) and a standard deviation equal to the estimation errors on numbers-at-age (ignoring potential correlation between different age-groups) from the assessment model output. This approach is also described in ICES 2013a. Following recommendations since WKMSYREF2 (ICES 2014b), the MSE model was updated prior to this meeting, now incorporating the log-normally distributed co-variance structure of numbers-at-age and F when generating the assessment stock numbers-at-age. The updated model takes into account that estimated stock numbers-at-age are correlated. This update was made after the benchmark in 2016 and has therefore not yet been implemented in the advice. At this meeting, the need for thorough documentation that demonstrates the approximation of the stock assessment model behaviour in the MSE is valid, was pointed out. This can be achieved by comparing a few simulations with the actual assessment model and the approximation to check whether the approximation is adequate (Punt et al. 2016). Although this was done on an earlier occasion, this exercise should be repeated as the current parameterization of the model has changed since then. It was also discussed if implementation error on the TAC should be incorporated. For example, should very high TACs in the MSE be cut off at a level corresponding to the known maximum capacity of the fishmeal processing factories (see P11)? This was tested during the meeting, and the effect on Fcap was small. A remark on the importance of evaluating how well-defined Fcap is, was also made. Plotting Fcap on the x-axis and P(SSB<Blim) on y-axis (from MSE runs for a range of Fcap values) can be useful; for example, a steep slope of the P(SSB<Blim) vs. Fcap curve at the intersection with P(SSB<Blim) = 5% indicates that Fcap is well-defined. A comparison of the performance of Bescapement (with Fcap), FMSY, and HCR was presented for sandeel in area 1r (Table 4.1.1). Here the definition of FMSY is the maximum fixed F (i.e. not varying between years) that the stock can sustain while accepting a risk of 5% of SSB falling below Blim. HCR is conceptually similar to what is applied for herring in the North Sea: F equals 0 for spawning stock biomasses (SSB) between 0 and Blim,

48 44 ICES WKMSYREF5 REPORT 2017 whereas when SSB is between Blim and B-trigger (here B-trigger = Bpa), F increases linearly from zero to Ftgt (Ftgt is optimized in the MSE framework in the same way as Fcap), and for SSB above B-trigger F = Ftgt. Long-term average TAC was highest for the Bescapement approach, whereas FMSY and HCR showed less variation in TAC (indicated by lower P(TAC<50000) values). FMSY and HCR gave about the same long-term average TAC. An attempt to replace Fcap with a TACcap in the Bescapement strategy was also presented (i.e. where TACcap is optimized in the MSE framework instead of Fcap), and also here Bescapement with Fcap resulted in highest long-term average TAC, but reduced TAC stability. Table An MSE comparison of the performance of four harvest strategies, based on sandeel area 1r. (TACs and SSB in tons). Bescapement with Fcap AVERAGE TAC P(TAC<50000) AVERAGE F AVERAGE SSB ~0.05 FMSY ~0.05 HCR ~0.05 Bescapement with TACcap ~0.05 P(SSB<BLI M) At the sandeel benchmark in 2016 (ICES 2017a), stochastic and deterministic forecasts were discussed. However, the decision making was problematic because: (1) difficulties in demonstrating differences (i.e. differences in long-term average TAC, TAC stability, etc.) and (2) lack of agreement on best practice for stochastic forecasting. Some aspects of point (2) are addressed in Section 2 of this report. Regarding point (1), the optimal solution would be to test the stochastic forecast in the same MSE framework that is applied to evaluate the deterministic forecast, although this approach would probably require having the full assessment model in the loop, rather than applying the short cut approach. However, to the knowledge of the meeting participants, such an exercise, and associated methodology, has not been developed for short-lived stocks to date. Instead, an alternative approach to compare the performance and behaviour of the two types of forecast strategies was presented. This approach applied both forecast methods to a common range of stock sizes and incoming recruitments, and plotted the stock size (or incoming recruitment) against, for example, the TAC. Figure compares TACs from stochastic (method described for Norway pout in P6, Section 3.1) and deterministic Bescapement forecasts (the latter including the use of an Fcap) for different levels of initial SSB (from the terminal year of the assessment), which forms part of the input to both types of forecasts. [Note: a similar plot could be produced for the in-coming year class (i.e. number-at-age 1 in the terminal year of the assessment).] Ideally, the plotted values representing the stochastic forecast in Figure should have been derived from an MSE, as was done for the deterministic forecast, but due to time constraints, it was not possible to conduct an MSE for evaluations of stochastic forecasts before this meeting. The results indicate that the stochastic forecast is more profitable at high levels of stock size and/or when there is a large incoming year-class (recruiting to the fishery),

49 ICES WKMSYREF5 REPORT whereas the opposite is the case for lower levels (Figure 4.1.1). Figure also illustrates the 2017 forecast for North Sea sandeel in management area 1r for both forecasting methods, and shows a 10-15% difference in the TAC. However, based on the actual forecast inputs used in 2017 (instead of the MSE results given in Figure 4.1.1), the deterministic Bescapement forecast (with Fcap) for sandeel in management area 1r, produced a higher TAC (around 13% higher) than the stochastic Bescapement forecast, whereas, the opposite was the case for sandeel in management area 2r. Spawning stock size was low in both areas and expectations of a large incoming year-class were high. However, for area 2r the incoming year-class was expected to be the largest in the time-series, possibly explaining why a stochastic forecast yielded a slightly higher TAC in this particular case. Blim Mean SSB 2017 forecast Highest SSB observed since 1990 Figure A comparison of TACs from a stochastic Bescapement forecast (green) and a deterministic Bescapement forecast (that includes the use of Fcap) (red) for different levels of initial SSB (i.e. SSB in the terminal year of the assessment), based on sandeel in management area 1r in the North Sea. The exercise initially used the exact same inputs to the forecast as used during the 2017 HAWG meeting (yielding a TAC of tonnes for the deterministic Bescapement strategy and tonnes for the stochastic Bescapement strategy; ICES 2017b). Stochastic forecasts (green dots) with different SSB starting points were obtained by increasing or decreasing SSB by multiplying all numbers-at-age from age 2 upwards by a factor greater or less than 1, while keeping numbers-at-age 1 constant and identical to the number used as input for the 2017 forecast. Dots relating to the deterministic forecast (red) were taken from an MSE with 500 by 20 years of simulations (i.e simulated forecasts); from these simulations, all years where numbers-at-age 1 going into the simulated forecast were approximately equal to the numbers-at-age 1 going into the actual forecast performed in 2017 were selected and the associated SSB values and TACs plotted.

50 46 ICES WKMSYREF5 REPORT 2017 P11: Capacity ceiling for sandeel MSEs Claus Sparrevohn, Søren Anker Pedersen, Henrik S. Lund This presentation (WD2) sets out to estimate the maximum annual catch of sandeel in the North Sea, called the catch-ceiling throughout. The premise is that the maximum catch taken during a fishing season will not exclusively depend on the total allowable catch (TAC) but also on the capacity of the fleet catching the fish, and the capacity of the industry processing the fish. The reason for estimating the catch ceiling is to allow for a management strategy evaluation wherein the maximum catch per year is explicitly assumed. At present, this is not done as it is assumed that the entire TAC advice is caught, independent of the magnitude of the advice. This can potentially lead to a bias in the MSE evaluation and outputs such as average yield, inter-annual variability in yield and Fcap. Similarly, the MSE should also assume that years with zero TAC most likely will initiate a real-time monitoring fishery, and hence a quota will be given and fished, despite the advice from ICES being zero catch. Figure Number of vessels that landing sandeel. Blue are vessel less than 30 m, and red larger. (From Slut rapport for EFF projektet: Stabilt Industrifiskeri) The catch ceiling depends on the capacity of the fleet as well as the capacity in the production lines (i.e. the fishmeal and oil factories). Since the capacity of the fleet involved in the sandeel fishery is volatile and can potentially change from year to year (Figure 4.1.2), the estimated maximum annual catch estimation will be based in this analysis on the fishmeal/oil factory capacities alone. The working document (WD2) is organized such that there is first a qualitative description of the sandeel fishing season 2017, which is used as an example to illustrate the mechanisms which kicks in when large catches are being taken. This part will also describe the seasonality of a sandeel season which typically starts around 1 st April and ends in July. Second part will estimate the total sandeel capacity of the EU fishmeal processing and suggest using this as a capacity ceiling.

51 ICES WKMSYREF5 REPORT Sandeel fishing season 2017 Figure Landings of industrial fish per week to Danish factories during the sandeel season The total TAC for sandeel in area 1-3 (EU zone) 2017 was the largest TAC since 2005 (see 2017 ICES advice for these stocks). Out of the total EU TAC, Denmark gets the majority (94.3%). Fishing for sandeels started 1 st April 2017 and it had high catch rates from the start of the fishing season. Already, about a week after the start of the fishery, there was a fishing vessel waiting queue at the factories to deliver sandeel catches. The sandeel catches at this time exceeded the production capacity of the plants. One of the reasons for the relatively fast-developing queue was that the Danish factories, in addition to sandeel, also received blue whiting early April (Figure 4.1.3). The Danish factory in Thyborøn tried to counteract the rising queue from late April by letting the vessels wait a period before they could sail from the harbour again, depending on the amount of sandeel they landed. This waiting period was maintained until the beginning of June, when quantities fell and the waiting period was again abolished. The other two factories in Denmark did not introduce the waiting period; however, vessels at these factories were also in a queue throughout April and May, but to a lesser extent. During June, there was also queue during periods where many vessels wanted to land fish over the same few days. The 2017 sandeel fishery season gives an indication for the maximum production capacity of the Danish fishmeal factories (Figure 4.1.4). For all three factories in Denmark, there were no technical problems or breakdowns. The Norwegian factories experienced a lockout, during which the factories did not take fish in. This lockout happened during May because of a conflict between Sildesalgslaget and Pelagia.

52 48 ICES WKMSYREF5 REPORT 2017 Figure Landings of sandeel per week to Danish factories during the seasons In Figure showing the weekly landings of sandeel to Danish factories, the maximum level is reached around t fish per week, which corresponds to the maximum capacity in the Danish factories. Capacity ceiling Since Denmark has 94,3% of the quota, the main question that should be asked, when deciding on a capacity ceiling, is to what extent Danish vessel land their fish in non- Danish harbours. The rest (5.7%) is assumed to be landed in other countries. In Figure 4.1.5, the relationship between the Danish TAC, Danish catches and the Danish catches landed in Denmark are shown. The overall picture is that the majority of Danishcaught sandeel are being processed in Denmark. When estimating the total capacity ceiling in the sandeel fishery, some assumptions are made. In WD2 the following approach was used. It is assumed that 94.3% of the sandeel are caught by Danish vessels and that these volumes are being processed in Denmark, whereas the remaining 5.7% will be processed outside. It is assumed that the factories can run with maximum capacity for 10 weeks. This leads to that the capacity ceiling for sandeel in SA area 1, 2 and 4 as follows: 6500 tons per day 7 days a week 10 weeks / 94.3% = tons. Based on this number and taking into account that some Danish catches (up to 5%) can be landed in other countries, we suggest to set the capacity ceiling for total sandeel catches in SA1, 2 and 4 to tons.

53 ICES WKMSYREF5 REPORT % 4.8% % -5.1% 1.6% 5.1% 2.3% 0.2% -5.9% -7.5% -10.7% % DK TAC DK catches DK landed in DK Figure Comparing the EU TAC with EU catches and how much of the Danish catches are eventually landing at Danish fishmeal factories. Percentage above bars are percentage Danish landings not landed in Denmark. 4.2 Discussion From the selected short-lived stocks ICES provides advice for, an escapement strategy is used for North Sea sandeels, Norway pout and sprat stocks, whereas there are management plans for the Barents Sea capelin and the Bay of Biscay anchovy stocks. ICES advice for capelin in the Iceland-Greenland-Jan Mayen follows the precautionary approach with an initial TAC revised within-year. The escapement strategy forms the basis of the ICES MSY approach for short-lived species. It consists of achieving a high probability (95%) of having the minimum amount of biomass required to produce MSY (Blim) left to spawn the following year. It is based on two reference points: MSY Bescapement and Fcap (for some stocks). ICES guidelines describe MSY Bescapement as: for short-lived species, a deterministic biomass limit below which a stock is considered to have reduced reproductive capacity, including any identified additional biomass need, and therefore it is defined: to be robust against low SSB and includes a biomass buffer to account for uncertainty in the assessment and catch advice (ICES 2017c). The basis for calculating MSY Bescapement is: Blim plus an additional biomass if the advice is based on a deterministic forecast (ICES 2017c). It is derived from Blim accounting for the uncertainty of the assessment in the terminal year, and is usually set at Bpa. However, this creates two issues: (1) MSY Bescapement will be underestimated because the forecast uncertainty is larger than the assessment uncertainty; (2) conceptually, MSY Bescapement, which is a target reference point, is confounded with the reference points related to stock status such as Bpa itself. The working group discussed whether MSY Bescapement would be better defined using a stochastic simulation approach (MSE) as done for Fcap. Fcap is a limit to F which is used when providing catch advice without directly estimating the probability of SSB > Bscapement (ICES 2017c). Some simulations for the North Sea sprat proved that the escapement strategy was not sustainable unless including a cap for the F (ICES 2014b [WKMSYREF2]). Fcap was also necessary for North Sea sandeel in management area 1r (presentation P10). Comparison between Fcap and maximum TAC for this stock showed that Fcap produced the maximum average yield but also more