SCRS 2015/113
Results of preliminary runs of the CMSYmethod against data limited ICCAT stocks
Rainer Froese^{1}
SUMMARY
CMSY is a new method for estimating maximum sustainable yield (MSY) and related fisheries reference points (Bmsy, Fmsy) from catch data and resilience, to help with preliminary stock assessment in datalimited stocks. CMSY was applied to 18 stocks of datalimited ICCAT species using default settings. The preliminary results are meant to be discussed by ICCAT experts and a rerun with more realistic priors is expected to provide results that may be deemed fitforuse for at least some of the species.
KEYWORDS
Datalimited stocks, Catch MSY, Reference points
1. Introduction
CMSY is a method for estimating maximum sustainable yield (MSY) and related fisheries reference points (B_{msy}, F_{msy}) from catch data and resilience. It is an advanced implementation of the CatchMSY method of Martell & Froese (2013), which was tested and found satisfactory at the WKLIVE IV workshop in Lisbon, October 2014 (ICES 2014). If managers, experts or stakeholders have a perception about the depletion history and the current status of a given stock, then CMSY can test such hypotheses against observed catches and the known resilience of the species. If combinations of productivity and stock size are found that are compatible with catches and resilience, then the stock status and exploitation rate are presented in an MSYframework. CMSY has been tested against simulated data, where the “true” parameter values were known, and against over one hundred fully assessed stocks, with good agreement between CMSY predictions and “true” or observed data. The respective publication (Froese et al. submitted) is under review as of this writing. Appendix I contains three examples of ICCAT stocks (ALB, BET, SWO) from this publication, assessed with CMSY as well as with a Bayesian statespace implementation of the Schaefer model analyzing catch and biomass data.
With the CMSY method, prior parameter ranges for the maximum intrinsic range of population increase (r) and for unexploited population size or carrying capacity (k) are filtered with a Monte Carlo approach to detect ‘viable’ rk pairs. A parameter pair is ‘viable’ if the corresponding biomass trajectories calculated with a Schaefer model are compatible with the observed catches, in the sense that predicted biomass does not overshoot carrying capacity nor crash the stock. Also, predicted biomass shall be compatible with prior estimates of relative biomass ranges for the beginning and the end of the respective time series. Optionally, a third intermediate prior biomass range can be provided to reflect extraordinary year classes or stock depletions. Also optionally, an indication whether the stock is likely to crash within three years if current catches continue can be given. This will improve the estimation of biomass in the final years.
A plot of viable rk pairs typically results in a triangularshaped cloud in log rk space. A special algorithm is applied to select the most probable rk pair from near the tip of the triangle and to establish approximate confidence limits.
2. Material and Methods
CMSY is written in R and the version used for this working paper was CMSY_44_ICCAT_5.R.
The CMSY method requires prior information about the range of possible rvalues for the considered species. As a proxy for rranges, the resilience of the species as stated in FishBase (www.fishbase.org) can be used as shown in Table 1. In a real CMSY application for stock assessment, experts are of course encouraged to use more specific prior ranges for r.
The CMSY method requires prior estimates of relative biomass at the beginning and end of the time series, and optional also in the middle. For example, the relative biomass ranges shown in Table 2 can be used for low or strong biomass depletion. Experts are of course free to use more suitable prior ranges for relative biomass.
CMSY input data are contained in two files, here SmallTuna_Catch.csv and SmallTuna_ID.csv. The first file contains time series of catch and, if available, total biomass or CPUE, with mandatory headers for the stock ID “stock” (e.g. “FRI”), a column for the years with available data “yr” (e.g. 1950…2013), a column for catches “ct” (e.g. 581760…511416) and an optional column for total (=exploited) biomass or CPUE “TB” (e.g. 7053207…3937277). The second file contains information about the stock and the priors to be used for r, k, initial relative biomass and final relative biomass, and the “FutureCrash” indicator with options “Possible” or “No”. A column with header “Btype” classifies available total biomass data as “observed”, “simulated”, “CPUE”, or “None”, i.e., CMSY can also be used if no biomass or CPUE data are available.
If total biomass or CPUE data are available, CMSY also fits a full Schaefer model (BSM) using a statespace Bayesian approach for estimating the most probable rk pair from biomass and catch trends. The statespace approach explicitly assigns process and observation errors to improve posterior distributions of the wanted parameters. The full Schaefer model is used to obtain “similar” types of reference points with which the results of CMSY can be compared (rather than using MSY, F_{msy} and B_{msy} estimated with other methods). Examples for three ICCAT stocks from Froese et al. (submitted) are shown in Appendix I. In addition BSM is applied to catch and biomass and catch and CPUE data for Bluefin tuna (BFT) (Appendix III).
3. Results
CMSY was applied to altogether 18 stocks of datalimited ICCAT species. The results are detailed in the Appendix II. For every analyzed stock, CMSY produces a screen printout describing the analyzed data, the priors, the results of the full Schaefer analysis (if any), and the results of CMSY. For visual examination, CMSY also produces standardized graphs. Figure 1 shows such graph for West African Spanish mackerel Scomberomorus tritor (MAW).
The “[stock] catch” graph in the upper left indicates the acronym used for the stock by the assessment working group (here: MAW), and shows the time series of catch data used by CMSY. The red circles indicated the highest and the lowest catch, respectively. If the user does not provide prior information on biomass ranges, simple prior rules are applied to catch relative to maximum and minimum catch, and are used to establish likely relative biomass ranges for the beginning and the end of the time series, as well as for an intermediate year (blue vertical lines in in the “Pred. biomass vs observed” graph).
The “Finding viable rk” graph shows the filtered logrkspace, with viable rk pairs in grey. While CMSY is executed, this graphs shows progress by adding grey dots as viable rk pairs are found. This search for viable rk pairs is the most timedemanding part of CMSY.
The “Analysis of viable rk” graph shows the result of the CMSYanalysis, with viable pairs in grey and the predicted most probable rk pair in blue, with 95% confidence limits. rk pairs to the left of the vertical dashed line are excluded from the analysis, as this section of the viable rk space is not expected to contain the maximum intrinsic rate of population increase.
The “Pred. biomass vs None” graph shows in bold the median relative biomass trajectory predicted by CMSY, with 2.5th and 97.5th percentiles. The dashed horizontal line at 0.5 k indicates B_{msy} and the horizontal dotted line at 0.25 k indicates half B_{msy} and thus the border to stock sizes that may result in reduced recruitment. The blue vertical lines show the prior biomass ranges set by the user or by prior rules applied to the catch pattern. The purple point in the final year indicates the 25th percentile of predicted biomass, which could be used as precautionary starting point for harvest control rules.
The “Equilibrium curve” graph shows the Schaefer parabola with catch expressed relative to MSY on the Yaxis and biomass relative to k on the Xaxis. Grey dots are catch over biomass predicted by the CMSY method. Dots falling on the parabola indicate catches that will maintain the corresponding biomass. Dots above the parabola will shrink the biomass; dots below the parabola allow the biomass to increase.
The “Exploitation rate” graph shows the time series of the catch/biomass ratio (u) relative to the ratio corresponding to MSY. It can be understood as depicting F/F_{msy}. The black curve is the exploitation rate resulting from catch relative to biomass predicted by CMSY. The dashed horizontal line indicates the maximum sustainable exploitation rate.
4. Discussion
4.1. Warning about reduced recruitment at low stock
Productivity models such as used by CMSY assume average recruitment across all stock sizes, including stock sizes below half of B_{msy}, where fisheries textbooks predict an increased risk of reduced recruitment. In other words, if recruitment is indeed reduced, then production models such as CMSY will overestimate production of new biomass and will underestimate exploitation rates. Thus, if the final biomass predicted by CMSY is close to half of B_{msy}, then extra precaution should be applied if CMSY is used for management. For example, instead of the median a lower percentile of predicted biomass could be used, such as the 25th percentile or even less. Stock recovery predicted by CMSY from low biomass should always be confirmed by independent data, such as CPUE.
4.2. Impact of using landings instead of catch
Whenever possible, stock assessment is based on true removals from the stock, i.e., including discards and other unallocated removals. But for datalimited stocks, estimates of discard are typically not available and only the reported landings can be used as indicator of removals. The effect of using landings instead of catch for CMSY assessment was explored previously for cases where the landings underestimated true catches by about 30%. As a result, the estimate of r remained practically unaffected, but the estimates of MSY, k and biomass were reduced by also about 30%. However, the relative estimates of B/k and c/B remained unchanged. Thus, the CMSY method seems capable of providing reliable relative assessments for stocks for which only landing data are available.
4.3. Preliminary CMSY evaluation of datalimited ICCAT stocks
CMSY was successfully fitted to all datalimited ICCAT stocks. Results differ greatly in their confidence limits for the desired reference points. Not surprisingly, long timeseries with contrast in the catch data gave much narrower confidence limits than short or monotonic ones. For several stocks it cannot be excluded that biomass is below half of B_{msy} and thus recruitment may be impaired and predicted recovery of biomass based on average recruitment may be too optimistic. In these cases (e.g. BOP) it would be prudent to use not the median but the 25th or 5th percentile of predicted biomass for precautionary management purposes.
4.4. Improving CMSY evaluation of datalimited ICCAT stocks
This preliminary investigation used default settings for prior ranges of r, k, and relative biomass. Serious application would start with gathering best available expert knowledge on these priors. Also, any independent indication of abundance, which can even be annual catches in a single fishers logbook (assuming same gears and similar effort), should be added, to build confidence in the CMSY analysis of available catch data. Questions to be put to experts are exemplified in Table 3. Preliminary results of using more realistic priors and CPUE data for some of the stocks are shown in Appendix IV.
References
Froese, R., Demirel, D., Coro, G., Kleisner, K.M., Winker, H. Estimating fisheries reference points from catch and resilience. Submitted to Fish and Fisheries on 28 February 2015
ICES, 2014. Report of the Workshop on the Development of Quantitative Assessment Methodologies based on LIFEhistory traits, exploitation characteristics, and other relevant parameters for datalimited stocks (WKLIFE IV), 27–31 October 2014, Lisbon, Portugal. ICES CM 2014/ACOM:54. 241 pp.
Martell, S. and R. Froese, 2013. A simple method for estimating MSY from catch and resilience. Fish and Fisheries 14: 504514
Table 1. Prior ranges for parameter r, based on classification of resilience.
Resilience

prior r range

High

0.6 – 1.5

Medium

0.2 – 0.8

Low

0.05 – 0.5

Very low

0.015 – 0.1

Table 2. Prior relative biomass ranges B/k used by CMSY for analyzing the simulated data.
Point in time series

Strong depletion

Low depletion

Beginning

0.1 – 0.5

0.5 – 0.9

Intermediate

0.01 – 0.4

0.3 – 0.9

End

0.01 – 0.4

0.4 – 0.8

Table 3. Example of questions to be put to experts to establish priors for CMSY analysis.
Prior

Question to experts

Start year for catch time series

From what year onward are catch data deemed reliable?

Relative start and end biomass
B/B_{0}

What was the most likely exploitation level at the beginning and end of that time series: light, full, or overfishing?

Relative intermediate biomass
B/B_{0}

Is there an intermediate year where, e.g., exploitation changed from light to full, or where an extraordinary large year class entered the fishery?

2 M ≈ r

What is your best guess for the range of values including natural mortality of adults (M)?

2 F_{msy} ≈ r

What is your best guess for the range of values including maximum sustainable fishing mortality (F_{msy})?

B/B_{0} < 0.2 : Possible / No

If current catches continue, is it likely that the stock will be outside of safe biological limits within the next 3 years?

Figure 1. Graphical output of CMSY applied to West African Spanish mackerel Scomberomorus tritor (MAW).
Appendix I
A.I.1. Results of applying CMSY to 3 fully assessed ICCAT stocks
The following three pages contain a comparison of CMSY analysis of catch and resilience with a Bayesian statespace analysis (BSM) of catch and biomass data for three ICCAT stocks (ALB, BET and SWO). These three pages were copied from the Appendix of the submitted CMSY paper (Froese et al. submitted). The interpretation of the Figures for CMSY is as explained above. In addition, the upper middle panel shows a cloud of black points, which are viable rk pairs as estimated by BSM, with the green cross indicating the most probable rk pair with its confidence limits. The upper right panel shows as black dots the observed catch and biomass data relative to MSY and k as estimated by BSM. The lower middle panel shows as red curve the time series of observed biomass relative to k as estimated by BSM, with dotted indications of the 2.5th and 97.5th percentile. The lower right panel shows as red curve the observed exploitation rate, relative to the catch/biomass ratio that would be compatible with MSY as estimated by BSM.
Species: Thunnus alalunga , stock: Albacore_NA, ALB
Name and region: Albacore , Albacore  North Atlantic
Catch data used from years 1930  2011 , biomass = observed
Prior initial relative biomass = 0.5  0.9
Prior intermediate rel. biomass= 0.3  0.9 in year 1963
Prior final relative biomass = 0.01  0.4
If current catches continue, is the stock likely to crash within 3 years? No
Prior range for r = 0.2  0.8 , prior range for k = 75.8  910
Results from Bayesian Schaefer model using catch & observed biomass
r = 0.355 , 95% CL = 0.322  0.392 , k = 461 , 95% CL = 428  497
MSY = 40.9 , 95% CL = 37.8  44.3
Biomass in last year = 175 or 0.38 k
Results of CMSY analysis with altogether 1334 viable trajectories for 735 rk pairs
397 rk pairs above r = 0.242 and 494 trajectories within rk CLs were analyzed
r = 0.377 , 95% CL = 0.246  0.578 , k = 458 , 95% CL = 286  734
MSY = 43.2 , 95% CL = 39.7  47.1
Predicted biomass last year= 0.265 , 2.5th = 0.0742 , 25th = 0.167 , 97.5th = 0.394
Predicted biomass next year= 0.292 , 2.5th = 0.063 , 25th = 0.179 , 97.5th = 0.439

Species: Thunnus obesus , stock: BETuna_A , BET
Name and region: Bigeye Tuna , Bigeye tuna  Atlantic
Catch data used from years 1950  2009 , biomass = observed
Prior initial relative biomass = 0.5  0.9
Prior intermediate rel. biomass= 0.3  0.9 in year 1993
Prior final relative biomass = 0.01  0.4
If current catches continue, is the stock likely to crash within 3 years? No
Prior range for r = 0.2  0.8 , prior range for k = 160  1925
Results from Bayesian Schaefer model using catch & observed biomass
r = 0.467 , 95% CL = 0.418  0.521 , k = 847 , 95% CL = 809  886
MSY = 98.8 , 95% CL = 89.4  109
Biomass in last year = 406 or 0.48 k
Results of CMSY analysis with altogether 3484 viable trajectories for 540 rk pairs
263 rk pairs above r = 0.321 and 1334 trajectories within rk CLs were analyzed
r = 0.509 , 95% CL = 0.332  0.781 , k = 705 , 95% CL = 433  1149
MSY = 89.8 , 95% CL = 79.9  101
Predicted biomass last year= 0.309 , 2.5th = 0.173 , 25th = 0.246 , 97.5th = 0.397
Predicted biomass next year= 0.306 , 2.5th = 0.137 , 25th = 0.23 , 97.5th = 0.413

Species: Xiphias gladius , stock: Swordfish_NA , SWO
Name and region: Swordfish , Swordfish  North Atlantic
Catch data used from years 1950  2011 , biomass = observed
Prior initial relative biomass = 0.5  0.9
Prior intermediate rel. biomass= 0.01  0.4 in year 1995
Prior final relative biomass = 0.4  0.8
If current catches continue, is the stock likely to crash within 3 years? No
Prior range for r = 0.2  0.8 , prior range for k = 48.8  390
Results from Bayesian Schaefer model using catch & observed biomass
r = 0.462 , 95% CL = 0.415  0.514 , k = 127 , 95% CL = 120  134
MSY = 14.6 , 95% CL = 13.4  16
Biomass in last year = 72 or 0.569 k
Results of CMSY analysis with altogether 2189 viable trajectories for 844 rk pairs
337 rk pairs above r = 0.597 and 1128 trajectories within rk CLs were analyzed
r = 0.691 , 95% CL = 0.602  0.793 , k = 85.7 , 95% CL = 73.1  101
MSY = 14.8 , 95% CL = 14.2  15.4
Predicted biomass last year= 0.635 , 2.5th = 0.423 , 25th = 0.55 , 97.5th = 0.736
Predicted biomass next year= 0.652 , 2.5th = 0.449 , 25th = 0.582 , 97.5th = 0.742

