Center for Independent Experts (CIE) Independent Peer Review Report on the 2016 South East Data, Assessment and Review (SEDAR 41) Workshop to Review the Assessments of South Atlantic Red Snapper and Gray Triggerfish Michael J Armstrong1 Centre for Fisheries, Environment & Aquaculture Science (Cefas) Pakefield Road Lowestoft Suffolk NR33 0HT United Kingdom [email protected] May 2016 1 Representing the Center for Independent Experts SEDAR 41 CIE review 1. Executive Summary The 2016 South East Data, Assessment and Review (SEDAR 41) Workshop to review the assessment of the status of South Atlantic Red Snapper and Gray triggerfish took place on 15-18th March 2016 in Charleston, South Carolina. The material presented for each stock was derived from a separate Data Workshop and an Assessment Workshop. The primary assessment model was the Beaufort Assessment Model (BAM), which is a length and age structured statistical model using fleet-specific landings, discards, length and age compositions and relative abundance indices. The data series for red snapper and gray triggerfish were based largely on the same data collection programmes and data processing methods, but differed in the exact mix of data inputs and the year ranges used. The BAM is an evolution of the one developed in SEDAR-15 and SEDAR-21 for red snapper. To examine sensitivity of stock status evaluations and benchmarks to the type of model used, a surplus production model (ASPIC) was also run using the same fishery-dependent and fishery-independent data sources as used for the BAM, but excluding any length or age data. Some additional analyses using catch-curves were carried out for red snapper to examine mortality rates using only the landings-at-age data. The main conclusion arising from the SEDAR-41 assessment of South Atlantic red snapper using BAM, and the sensitivity runs and uncertainty analysis of the assessment, is that there is a high probability that the stock is overfished and is experiencing overfishing. The perception of overfishing is driven by the addition of 2014 data, which revises all the 2010-2013 estimates upwards from values from runs of the same model terminating in those years. In other words, if the assessment had been done only to 2013, overfishing may not have been evident. The recent increase in fishing mortality estimates is evident primarily at ages 4 and over. It is uncertain how selectivity is changing since the moratorium in 2010 and the mini-seasons since 2012, and fishery data quality must be deteriorating due to the resultant increase in discarding, which is estimated with sampling error. There are also changes in recreational data collection involving intensive State surveys since 2012 to estimate private and charter recreational catches and discards, which is essentially a major change in design from just using the Marine Recreational Information Program (MRIP) in previous years. Additional sources of uncertainty for the model are the cessation of the fishery-dependent landings indices after 2009, concerns over altered fishing behaviour on catchability in the headboat discards index during the moratorium, and the short duration of the CVID index series. All these factors are likely to degrade the ability of the model to detect changes in fishing mortality, and the sudden change in perception with addition of 2014 data should be investigated further before concluding that overfishing is occurring and is increasing between 2010 and 2014. The conclusion that the stock is overfished and very slowly rebuilding in terms of egg production, however, appears robust. Some strong year classes formed in the 1980s, and since the mid 2000s have resulted in the stock being very abundant in numbers of young fish, which is corroborated by fishermen’s observations, but the numbers of mature fish have increased only very slowly over the last decade and the total egg production remains very low. Testimony from recreational and commercial fishermen at the review meeting strongly disputed the model estimates that large red snappers remain comparatively rare, and considered that selectivity of commercial lines and the trap survey is not asymptotic as assumed for the model. Independent, empirical observations of fish behavior in relation to the gears are needed to resolve this dispute. The BAM conclusions on stock status of red snapper are not supported by the comparative ASPIC model that indicates that the stock is currently no longer overfished (using biomass rather than egg production) and that overfishing is no longer occurring. However, the assessment working group and the review panel agreed that the BAM was more appropriate as it explicitly models recruitment, age-based dynamics and fishery selectivity, which is important because the stock is strongly driven by widely-varying recruitment. An error in age compositions for the chevron trap survey was detected during the meeting, due to not correcting to true 1 SEDAR 41 CIE review age, requiring a re-run of the assessment to provide a new base model. Changes to the results were relatively small, and the sensitivity analyses, Monte Carlo bootstrap uncertainty analysis and retrospective analysis of the original base model are used for the present review. The SEDAR-41 assessment of South Atlantic gray triggerfish using BAM had been fitted using both length and age compositions from the same trips, and the chevron trap series of relative abundance indices had been up-weighted by a factor of six to allow the model to fit the survey more closely than the age compositions, following some general guidelines by Chris Francis (New Zealand) who had commented on the SEDAR-24 assessment of red snapper. The review panel considered that this resulted in fitting too closely to the point estimates from the survey, including an unusually low value in the first year of the series (1990). A sensitivity run using only age data where both length and age were available resulted in poorer fits to the age composition data. The review panel requested that the assessment be put back to the assessment working group for a full review of how these different data sets should be weighted. Hence there is no final assessment run for gray trigger fish from SEDAR-41. All BAM sensitivity runs and uncertainty analysis presented at the review workshop indicated that the stock is not overfished, and that overfishing is not occurring, with declining fishing mortality and increasing spawning stock biomass in recent years. The comparative ASPIC model run fitted poorly due to lack of contrast in the data, but the conclusions on stock status were the same as from BAM. There does not therefore appear to be any current risk to stock status but further work on the BAM by the assessment working group is needed to confirm this finding. 2. Background South East Data, Assessment, and Review (SEDAR) is a joint process for conducting stock assessments, and peer-reviewing their outcomes, for stocks of interest to the South Atlantic, Gulf of Mexico and Caribbean Fishery Management Councils, NOAA Fisheries, SEFC, SERO and the Atlantic and Gulf States Marine Fisheries Commissions. SEDAR is organized around separate data, assessment and review workshops. The previous assessment of South Atlantic red snapper was conducted in 2010 (SEDAR-21), using a similar form of model to the one implemented at SEDAR-41. South Atlantic Gray triggerfish has not previously been benchmarked by SEDAR. Input data for the SEDAR-41 assessment were compiled during the Data Workshop (DW), and population models were developed during the subsequent Assessment Workshop (AW), taking into account recommendations from the SEDAR-21 independent peer review of the data and assessment models for red snapper and other subsequent developments. 3. Description of review activities The SEDAR 41 Review Workshop (RW) took place at the Crowne Plaza Convention Centre at Charleston from 8:30 am Tuesday 15 March 2016 to around midday Friday 18 March. The assessment results and background were clearly presented by the experts at the meeting. Time was set aside each day to allow observers and Council representatives to provide testimony, and for the review panel to seek clarification from the assessment team and to request additional model runs, which were done very quickly and led to fruitful discussion that helped to clarify a number of important issues. All proceedings were on record, and the coordinator maintained notes each day that were circulated. The provisional agenda for the meeting is given in Annex 3 of Appendix 2. The Review Panel itself comprised the Chair and two reviewers from SAFMC SSC, and three reviewers appointed by the CIE (Appendix 3). The assessment results were presented by two US technical experts from SEFSC Beaufort who were involved in the AW, with support from five assessment workshop 2 SEDAR 41 CIE review participants from the same laboratory. The RW was also attended by the SEDAR coordinator, Council and Agency staff, Council Representatives and some appointed observers. All documentation, including background documentation provided to earlier DW and AW meetings, was provided to the Review Panel in advance of the review workshop, and was comprehensive for the job in hand. The Review Panel developed a Summary Report during and after the meeting. The following report presents my personal evaluation of the review process together with more extended observations on the data and assessment models that are not necessarily shared with the other panel members. I accept all responsibility for any errors in my report due to misinterpretations of the data or analyses. Due to the need to revise the red snapper assessment at the end of the RW, the Review Panel report could not be completed during or immediately after the RW. Due to other commitments, the present CIE personal review had to be submitted before completion of the Panel Report in order to meet the CIE deadline. It is therefore possible that there are different views expressed here compared with the Panel report for which the reasons are not explained. 4. Summary of findings by Term of Reference 4.1 South Atlantic Red Snapper ToR 1. Evaluate the data used in the assessment, including discussion of the strengths and weaknesses of data sources and decisions, and consider the following: a) Are data decisions made by the DW and AW sound and robust? b) Are data uncertainties acknowledged, reported, and within the normal or expected levels? c) Are data properly applied within the assessment model? d) Are data input series reliable and sufficient to support the assessment approach and findings? A general statement on (a) – (d) is given below, followed by some supporting observations. (a) Data decisions made by the DW and AW were sound and robust. The development of input data and parameters for the BAM and ASPIC models required an extremely thorough compilation and evaluation of all available data at the DW. Modifications made subsequently by the AW were fully explained. (b) Data uncertainties were acknowledged, reported, and were within the normal or expected levels, as far as could be ascertained from information provided to the RW. The data varied widely in coverage and quality, and were often heavily manipulated and standardized to try and develop coherent time series from diverse data sources of differing designs, coverage and accuracy. The combined data will have biases that in some cases are poorly understood, especially in earlier years of the time series. All decisions made by the DW and AW in compiling data were explained and justified in detail. Data quality metrics were provided by the DW in terms of numbers of samples, CVs, or alternative plausible data series or biological parameter values. These were used by the AW to weight data series in the assessment model, estimate the uncertainty in the assessment results using the Monte Carlo / 3 SEDAR 41 CIE review bootstrap method, or to explore the sensitivity of the assessment to data decisions and uncertainty. (c) The data were properly applied within the assessment model. Any issues with application of the data such as time periods for fitting, use of length and age data from the same sampling schemes, or weighting of data according to data quality metrics, were explored at the SEDAR-41 RW, if not previously evaluated by the DW and AW. (d) Data input series were applied if considered reliable and sufficient to support the assessment approach and findings. Reliability and sufficiency was evaluated based on a-priori criteria where possible, supported by data quality metrics such as numbers of samples or CVs and by model fits. The assessment is supported primarily by a wide range of fishery-dependent data covering landings and discards, and therefore is heavily driven by these data and assumptions related to their reliability and use. An additional fishery-independent trap and video survey data set unfortunately covers only the period since 2010 due to very low incidence of red snapper catches prior to the recent increase in abundance due to strong year classes. Additional supporting comments Life history parameters Overall, the DW and AW made sound and robust decisions on life history parameters, and the uncertainty in these, and the data were used appropriately in the assessment. Natural mortality (M) and stock-recruit steepness are key parameters related to stock productivity and biological reference points, but there are no direct estimates for red snapper and they have to be inferred from a meta-analysis approach for M-at-age using growth parameters and maximum observed age, and from the ability of the BAM to estimate steepness (which always tended to 1.0 due to high recruitment at the lower spawning stocks). These are conventional approaches in statistical assessment modelling. Sensitivity analyses of the BAM were conducted for different time-invariant values for M, but not for any plausible trends over time in M as there is no information to guide this. The stock-recruit relationship uses total annual egg production estimated based on maturity, length, number of batches and batch fecundity, allowing the effect of age structure on reproductive output to be reflected in setting SSB reference points and stock status. Trends or more random variations in fecundity in red snapper, considered an indeterminate batch spawner, are possible sources of uncertainty that could not be fully explored as historical information was not available. The low estimate of age at first maturity in females (43% at age 1) was considered by the review panel to be unusual for snappers, and it was speculated if it has declined as a compensatory response to heavy exploitation. Annual maturity data from the SERFS chevron trap survey could not be used to test this because sample collections have been from different areas in different time periods. Fishery removals The DW and AW made sound and robust decisions on fishery removals data, and the uncertainty in these (which is largest for discards estimates), and the data were used appropriately in the assessment, using only those data considered reliable enough for this purpose. Historical commercial and recreational fishery removals – landings and dead discards – were constructed back to 1950 to allow a sufficient burn-in period for the BAM with assumed stable age structure and low fishing mortality in the early period. The DW made a large number of decisions to infer historical values from more recent data or to calibrate data series where the survey design has changed, including adjusting NMFS Marine Recreational Fisheries Statistics Survey (MRFSS) 4 SEDAR 41 CIE review estimates from 1981 to 2003 to be consistent with catches from the Marine Recreational Information Program (MRIP: 2004 to present), and developing combined recreational landings back to 1955 using effort data from the National Survey of Fishing, Hunting, and Wildlife-Associated Recreation Survey (FHWAR: SEDAR41-DW17) combined with average MRFSS and SRHS CPUE data for 1981-83. All such calibrations and extrapolations may introduce additional biases in the data series. Recreational landings of headboats are estimated from the Southeast Region Headboat Survey (SRHS) log book scheme which has improved in quality over time due to introduction of mandatory reporting in 1996 and improved logbook supply from 2008 onwards. A lengthy review of the headboat survey that was carried out to address concerns of fishermen about the quality of the data (SEDAR41_DW46) found little evidence of problems that would preclude use of the data. The DW proposed a declining CV in three blocks from 1981 onwards to reflect improved quality of the data, which is reasonable in relative terms, but will be biased to some extent in absolute terms as it is not based on actual sampling errors. For earlier years in the BAM Monte Carlo bootstrap (MCB) uncertainty analysis, deviations were applied equally to all years and not independently by year. In practice, errors in subsequent years, where only half of the logbooks were returned (Fig.4.1.1 from SEDAR41_DW46), may also be correlated between years if non-response is related to size of catch; but no information was available on this and the assumptions made about independence of errors in the MCB analysis were adequate although potentially underestimating uncertainty. Fig. 4.1.1. Compliance rates in submission of headboat survey logbooks over time (from SEDAR41_DW46_NMFS-SEFC_HBDataEval_7.20.2015.pdf). Private boat and charter boat landings since the early 1980s were estimated from MRFSS/MRIP, which has a robust and peer-reviewed statistical design that has substantially reduced bias and improved precision over time, and for which CVs are estimated directly based on efficient estimators. Discards estimates from commercial and recreational fleets, where available, are subject to sampling error if based on observation, and potential for bias also exists where discards are recalled, as in the MRFSS/MRIP intercept surveys. Additional time-series error is introduced by extrapolating recent data to historical periods with no data. For example, commercial handline discard observations for 2002-2009 were extrapolated back to 1992, with zero discards assumed prior to that, due to low minimum landing size. Similarly, head boat discards estimates from log books and some at-sea observation since 2004 were extrapolated back in time based on changes in length frequencies recorded by dockside sampling before and after changes in minimum landing sizes, with zero discards 5 SEDAR 41 CIE review assumed pre-1984. All these data manipulations can introduce additional error in the time series. Discards estimates from MRFSS/MRIP are self-reported by anglers intercepted at landing sites and are not verified. The MRIP surveys provided too few estimates of red snapper landings or discards for the very brief mini-seasons since 2012, and additional State surveys using a variety of off-site and on-site methods were also used for these periods, based on collaboration between MRIP staff and State laboratories, which the Review Panel was advised is continuing in order to develop options for future sampling. Continued collaboration is encouraged. Whilst these surveys may be providing much more data than MRIP, it is not clear if this has resulted in a discontinuity due to different designs which may have different bias characteristics. The period covered by the mini-season could be treated as a separate post-stratum in MRIP for red snapper, albeit with low sampling frequency, and a statistically robust procedure then adopted to estimate discards within this stratum using additional samples collected by State laboratories, ensuring that sampling is representative and compatible with MRIP in terms of bias characteristics. This would avoid the more complex decision making processes on which data set to use. A draft manuscript by Sauls et al. provided to the Panel after the RW describes the relatively intensive on-site State survey in Florida. Discarding of red snapper has increased over time due to changes in minimum landing size to 20 inches in 1992, increases in abundance of young fish from above-average year classes in some recent years, and the introduction of the moratorium in 2010 and 2011, and the small commercial catch limits and recreational bag limits in the mini seasons for 2012 onwards. Most of the catch is now discarded, and the inherent errors in discards estimation based on limited observer trips and fisher recalls means that the quality of total fishery removals estimates may therefore have deteriorated significantly, which will impact estimation of stock size and fishing mortality. Any initiatives to improve the quality of discards estimates would likely be beneficial, particularly as the BAM requires these and any landings estimates to be treated as precise. Overall, the DW and AW made the best possible evaluation of fishery removals data and their uncertainties. Length and age compositions The DW and AW made sound and robust decisions on length and age composition data, and the uncertainty in these, and the data were used appropriately in the assessment, using only those data considered reliable enough for this purpose. Length and age composition data for fisheries were collected by port-side and in some cases by at-sea sampling, and by sampling of fish in the trap surveys. The AW used age composition data in preference to length composition data in BAM where both data exist, except for discards where selectivity is fitted separately for discards and landings, but age data are available only for landings. Length composition data were fitted only for commercial handlines from 1984 – 1992, commercial discards in 2009 and 2013, and headboat discards from 2005 to 2014. Age compositions were fitted for commercial handlines landings from 1990 onwards, for headboat landings in two widely separated blocks in the 1980s and 2000s, for general recreational landings since 2001, and for the CVID survey from 2010. The CVID age data were found towards the end of the review meeting to have not been converted to calendar ages, and revised data were provided along with some preliminary assessment results which indicated some relatively small changes to the overall assessment results and stock status. Samples below a specified number of trips and fish were excluded from the BAM, but there is still a widely varying number of trips sampled, which can be considered as the primary sampling units. The 6 SEDAR 41 CIE review patchy nature of sampling of the fisheries means that the mix of data used in the model differs in the three time blocks where landings selectivity is fitted (1950-91; 1992-2009; 2010 onwards). Sampling for age was particularly intensive for headboats in the mid-1980s, commercial handlines and headboats in 2009 and for general recreational in 2014. The true effective sample sizes were not computed, but this would be a useful exercise to determine the relationship between numbers of trips, numbers of fish and the effective sample sizes, and hence evaluate the relative weighting factors for years. Relative abundance indices Overall, the DW and AW made sound and robust decisions on most of the relative abundance indices and the uncertainty in these, and the data were used appropriately in the assessment, using only those data considered reliable enough for this purpose. Three fishery-dependent indices were used in the assessment: commercial handline landings per unit effort (1993-2009), headboat landings per unit effort (1976-2009), headboat discards per unit effort from observer trips (2005-2014), and one fishery-independent survey, the chevron trap abundance index combined with video records from same survey (SERFS combined video and trap - CVID - 2010-2014). The general approaches to analysis of the data appear appropriate and are widely used for such data in assessments in the USA. Changes in red snapper management will affect targeting and catch-release behavior, and therefore any CPUE data from 2010 onwards are likely to be incompatible with earlier years, which is of concern for use of the headboat discards index. The CVID index was used only from 2010 due to increased spatial coverage, integration with video data, and sufficient operations with red snapper catches to allow a reliable index calculation. A zero- inflated negative binomial model with trap soak minutes as an offset was used for trap and video, with stepwise removal of covariates (SEDAR41-DW54 for trap survey and DW45 for video survey). The covariates included in both the trap and video index were depth, latitude and bottom temperature, with some additional covariates in the video index. Subsequent to the DW, a combined analysis was conducted as the two data sets are not independent. The statistical methods adopted appear sound, but the series is unfortunately still very short and it is therefore difficult to evaluate if it provides abundance indices directly proportional to stock abundance in the longer term. The DW adopted the widely used practices in the USA of filtering fishery CPUE data using the Stephens-MacCall (2004) method to remove fishing operations from areas and habitats with low probability of occurrence of red snapper based on species compositions observed, and then applying a delta-GLM with predictor variables to explain CPUE patterns accounting for spatial and temporal coverage, and trip type, among other factors. There was some discussion at the RW that, as these fisheries cover different depth ranges, a combined analysis might be beneficial external to the BAM, and this should be considered for future benchmark assessments. Possibly of more concern is the assumption of constant catchability in the fishery CPUE series, which if violated would bias the BAM (and ASPIC) model estimates. The short CVID and the longer handline index series show random residuals (AW report Figs 10 & 11) whilst the headboat fleet shows strongly serially correlated residuals with substantial under-fitting of the large 2008 and 2009 indices, with closest fit to the series of very low indices from 1992 – 2007. It is notable that the 2008 and 2009 indices correspond to a period of substantially increased compliance in logbook returns from less than 50% to almost 100% (Fig. 4.1.1 above). The estimation of total headboat landings assumes that vessels not supplying logbooks have the same CPUE on average as those supplying logbooks, and can therefore be applied to the headboat activity records (HAR) of fishing effort to generate catches where these are not supplied – this assumption should be investigated further by comparing the activities of vessels that now supply logbooks but persistently did not before 2008. 7 SEDAR 41 CIE review Further work is also warranted to better understand factors that could cause trends in catchability in commercial and recreational hook-and-line catch rates, such as changes in species targeting or effects of hook competition and handling time for other species on deck due to changes in their abundance relative to red snapper. It would be useful to consider trends in species compositions, particularly given the difficulty of BAM and ASPIC to fit the headboat indices for the early part of the series. Other factors affecting CPUE trends include technology creep in catchability due to improvements in fishing gear, positioning (GPS) and communication systems, and also by rising fuel costs in recent years. SEDAR24 attempted account for improvements in technology (notably, GPS systems) by linearly increasing catchability of South Atlantic red snapper by 2% per year, beginning in 1976 for headboats, and 1993 for commercial lines, until 2003 and then holding it constant thereafter. This approach was not adopted in SEDAR41. The Review Panel proposed that changes in management actions such as the moratorium, mini-season and reductions in bag limits are expected to alter fishing behavior and hence catchability in fishery- dependent indices, and should inform decisions on inclusion of data or periods of data in assessments. A member of the SAFMC stated on record at the SEDAR41 RW meeting that the behavior of anglers has changed substantially since the moratorium, to avoid catches and discarding of red snapper. The panel therefore considered that all the fishery CPUE series for red snapper should be applicable only to 2009, the year before the moratorium, and earlier years, and I agree with this conclusion. ToR 2. Evaluate and discuss the strengths and weaknesses of the methods used to assess the stock, taking into account the available data, and considering the following: a) Are methods scientifically sound and robust? b) Are assessment models configured properly and used consistent with standard practices? c) Are the methods appropriate for the available data? A general statement on (a) – (c) is given below, followed by some supporting observations. a) The BAM stock assessment method used as the primary assessment tool for South Atlantic Red Snapper is scientifically sound and robust, and is based on well tried and tested integrated assessment models such as Stock Synthesis, and has been applied to a number of stocks in the USA. The ASPIC production model is also well tested worldwide, but the conditions under which it can give reliable results are more limited as it does not explicitly represent year class dynamics and fishery selectivity, and requires sufficient contrast in data to reliably estimate model parameters. Approaches to investigate uncertainty, including retrospective analysis, sensitivity analysis and Monte Carlo bootstrap analysis were all scientifically sound and robust. b) The BAM and ASPIC assessment models are configured properly and their use is consistent with standard practices. c) The methods used are appropriate given the nature of the available data. In particular, the patchy distribution of length and age sampling across time and fleets, and the variable sampling rates where sampling occurs, require a statistical modeling framework such as BAM to fit the underlying population model to the available observations in a statistically sound way. A downside is that patchy occurrence of short fishery composition series with very 8 SEDAR 41 CIE review variable sampling rates can cause the model to over-weight and over-fit some series and years within series. Additional comments are given below: BAM vs ASPIC The BAM is a superior model to the ASPIC production model as it makes direct use of fishery and survey length and age compositions to estimate annual recruitments (which have varied widely and are the most dynamic component of the intrinsic rate of increase (R) parameter in ASPIC for this stock), and can also account for changes in selectivity. In recent years, the total stock biomass has been dominated by young snappers, and therefore is heavily recruitment driven. The ASPIC model also proved difficult to fit. The BAM has many assumptions and many estimated parameters, but the base model configuration appears to have reasonable assumptions and parameter estimates. As indicated by sensitivity analyses reported by the Assessment Workshop and during the Review Workshop, the most important data and modeling decisions were around choice and weighting of relative abundance indices, and modeling the form and time variation in selectivity. Catch curves of age composition data were provided as exploratory information at the RW, but are not a valid basis for status determination as the apparent mortality rates do not reflect selectivity. Configuration of the BAM base assessment and weighting of input data The BAM fits to a wide range of fishery landings and discards data, multinomial length and age composition data, and relative abundance data. Where data are from a sampling design (rather than census), such as length compositions, recreational catches and trap surveys, the input data have associated estimates of sampling error. For catch estimates and surveys a CV is provided, whilst for length and age compositions an input sample size is included as a proxy for effective sample size – in this case the number of fishing trips sampled. The input CVs and sample sizes inform the model about the relative precision of input data. A penalized likelihood approach is used for fitting the model, using robust multinomial likelihoods for length and age compositions, and lognormal likelihoods for removals and index values. For parameters defining selectivities, CV of size at age, and recruitment variability (σ ), normal priors were applied. An age error matrix is included to allow R for this additional source of error when fitting age composition data. The assessment follows the common practice of weighting compositional catch data and abundance indices in two stages. The input data are first assigned relative weights before the model is run using input CVs and sample sizes. Assuming the model converges correctly, the data weights are iteratively adjusted upwards or downwards during a series of model runs to improve model fit. The method of doing this for red snapper was to adjust the weights until the standard deviations of normalized residuals were near 1.0 (Francis 2011). The relative input weights between years in a data set should reflect information about sample sizes, e.g. annual numbers of trips sampled for age as a proxy for effective sample size, or CVs calculated for annual abundance indices. For red snapper, this is done for length and age compositions, and for the headboat discards relative abundance index and the CVID index, but not for the commercial lines and headboat landings abundance indices, which were set to 0.2 for all years (for headboats this is at odds with the CVs used for the landings of this fleet in MCB runs of the BAM, where precision was 9