Assessment of North Sea herring

2.6.1 Data exploration and preliminary results

North Sea herring is on the AFCM observation list, but was also classed as an update assess-ment in 2005 by ACFM. With this in mind limited exploration was carried out into the fit of the assessment. The full choice of assessment model, catch and survey weightings and the length of separable period where not explored in detail in 2005. It is proposed to carry out a benchmark assessment for North Sea herring in 2006.

2.6.1.1 Selection of weighting of indices in the assessment of North Sea herring

The usual assessment tool for the assessment of North Sea herring is ICA. The settings were the same as last year. Acoustic, Bottom trawl (IBTS), MIK and Larvae (MLAI) surveys are available for the assessment of North Sea autumn spawning herring. The surveys and the years for which they are available are given in Table 2.6.1.1.

In 2002 the HAWG moved from arbitrary index weighting as used for the previous 6 years (1996-2001) to a more objective method. This new method was developed from the work of the ACFM study group SGEHAP (ICES CM 2001/ACFM:22) which had one of its objectives to try to rationalise the survey index weighting in the assessment. The analysis is described in the 2003 report (ICES 2003/ ACFM:17). The weighting values are given in Table 2.6.1.2. The weightings applied account for sampling error of the surveys. The WG in 2002 selected index weighting which both minimised the variability in the assessment output but also reduced the retrospective revision of management parameters (F, SSB and recruitment). However, they could not find a method that minimised the revision of all of these parameters but selected the

one that performed best for two out of three. This was done by down-weighting the influence of catch of 0wr and 1wr in the assessment (Table 2.6.1.2).

The WG in 2003 made an extensive review covering both inverse variance and structural er-rors, and it considered that the inverse variance method provided the better method. This proc-ess meant that the weighting of surveys and catch is fixed as the sensitivity of the assproc-essment to these weighting values has been greatly studied in recent years. The weights express the WG view that the young herring are best estimated with MIK and IBTS surveys, the older herring are best evaluated through the acoustic survey and the SSB should be estimated through the MLAI.

2.6.1.2 Period of separable constraint

Changes in the regulations in 1996 have affected the various components of the fishery differ-ently. Recent meetings of this WG split the separable period into two different periods: 1992-1996 and 1997 onwards. In the WG 2001 it was considered that the number of years after the change in selection was long enough to use only a single separable period of four years. Dur-ing 2002-2004 a separable period of five years was used. The WG in 2002 found that year on year adaptation of the separable period did not improve the performance of the assessment model and that a fixed selection period gave more stable assessments, even with changing management. Last years WG explored a 4, 5 and 6 year separable period. No important differ-ences in the model fit or outputs were detected. The estimation of F at reference age (4wr) was not significantly different and differences in mean F2-6 and SSB where found negligible. So the 5 year separable period was maintained in the current assessment.

2.6.1.3 Model fit and residuals

The influence of the catch and the surveys was explored on the estimation of reference F and the model fit. ICA was run using all catch and survey data with the same procedure as last year (SPALY). The patterns in catch and survey residuals (Figure 2.6.1.1, upper panels) are similar to the assessments in 2003 and 2004 but greater in magnitude. In the assessments in 2003 and 2004, positive residuals in the catch of 2+wr fish against negative residuals in the acoustic survey and IBTS were detected. Using the same procedure as last year, the catch re-siduals in this year’s assessment showed positive rere-siduals for older ages in recent years, while at 2wr in the terminal year there is a large negative residual. Different trends in the residual patterns between catch and surveys indicate conflicting signals in the information.

To explore the contribution of the catch and the survey data to the specific patterns in the re-siduals, runs with modification to the data were explored:

i. Setting the weights for the catch in the separable period to 10% of the original values for the separable period to explore the influence of the catch data on the model.

ii. Setting the weights for the catch of 1wr and 2wr in 2004 at 0.01 to explore the in-fluence of the large negative residual for 2wr in the terminal year

iii. Using the acoustic survey as the only tuning fleet, iv. Using the IBTS as the only tuning fleet,

v. Using the MIK survey as the only tuning fleet,

vi. Using the MLAI as the only tuning fleet to explore the influence of the survey data on the model

Setting the weights for the catch in the separable period to 10% of the original values reduces the negative residuals in the survey data in recent years slightly (Figure 2.6.1.1, middle pan-els). Setting the weights for the catch of 1wr and 2wr in 2004 at 0.01 does hardly change the

ICES HAWG Report 2005 52

patterns in the group of negative residuals in the survey data in recent years (Figure 2.6.1.1, lower panels).

Using each individual survey as the only tuning fleet did not resolve the pattern of negative survey residuals in recent years for the acoustic and the IBTS survey (Figure 2.6.1.2). The MIK and MLAI survey showed random distributions in the residuals with no apparent trend.

Depending on the data source explored, the reference F (4wr) varied between 0.20 and 0.40 (Figure 2.6.1.3a). The runs with down weighted catch (i and ii) and using only the IBTS as tuning fleet gave similar perceptions of fishing mortality on the reference age (4wr) of the separable model with estimates within the 90% confidence intervals of the run with last years settings. The acoustic survey gave slightly higher estimates of F, while the MIK and the MLAI gave slightly lower estimates. The MIK and the MLAI gave higher estimated of SSB associ-ated with the lower reference F (Figure 2.6.1.3b). The MIK is not expected to have great power, since it is a recruitment index with information on only the 0wr.

2.6.1.4 Exploring other assessment models

The performance of ICA is explored against XSA and SURBA (ICES CM2003/D:03; Needle 2004), which is a survey-only based assessment model. ICA has been used for the assessment of North Sea herring during the last decade. Concern at WGMG was raised about the instabil-ity in the selection patterns at older ages impacting on the earlier part of the time-series (ICES CM2003/D:03). The WG in 2003 and 2004 explored the performance of ICA against another regularly used assessment model, XSA. The approach used was to choose XSA settings that reflect as many of the assumptions of the ICA model of North Sea herring.

The model settings for XSA in this years assessment are given in Table 2.6.1.3 and the sum-mary of the results in Table 2.6.1.4. The XSA assessment is consistent with the ICA assess-ment (Figure 2.6.1.4) for the recruitassess-ment and F2-6 and SSB in historic period. Only during the last 3 years does XSA show higher estimates of mean F2-6 and lower estimates of SSB. Higher estimate of F and lower estimate of SSB in the terminal year by XSA might be influenced by the exclusion of the MLAI in the XSA (which cannot use biomass indices) and the effect of the weak shrinkage during a period of declining F, maybe responsible for part of the increase.

A new version of SURBA that could assess North Sea herring was developed immediately prior to the HAWG in 2005. This version could combine multiple surveys, with weighting factors and incorporated the use of a biomass tuning fleet. Its use should be viewed as ex-ploratory and its results as preliminary as no major testing of the model or sensitivity analysis of the model settings has taken place as yet. The SURBA run had a higher mean F2-6 and lower SSB in the terminal year than ICA and XSA, and showed more between year variability in F in recent years than the other models.

2.6.1.5 Conclusions of exploration of the assessment

The formulation of the assessment was recently supported by an external and independent review of the North Sea herring assessment carried out for the North Sea Commission Fisher-ies Partnership, where the consistency, precision and quality of the assessment were judged as credible and fully acceptable as a tool for management advice. However patterns in the residu-als seen in previous years and current assessment indicate that catch and survey indices show different signals. Closer investigation of the model suggested a slight underestimation of F and an overestimation of SSB in current assessment (see section 2.10), and this should be moni-tored closely. Therefore WG strongly suggests a benchmark assessment of North Sea herring next year. Exploratory runs of ICA and XSA however showed similar trends in the develop-ment of mean F2-6, SSB and recruitment and are within the precision levels of the ICA model.

As exploration provides no simple solution to the apparent conflicts between catch and survey

data, it was concluded that this years assessment method is maintained as last year, with com-parable settings, tuning indices and weightings.

2.6.2 The stock assessment 2.6.2.1 The model used

The assessment of the stock was carried out by fitting the integrated catch-at-age model (ICA) including a separable constraint over a five-year period as explained above (Patterson, 1998, Needle 2000). The input data are shown in Table 2.6.2.1.

2.6.2.2 Results

The ICA output is presented in Tables 2.6.2.2 and 2.6.2.3, with model fit and parameter esti-mates in Table 2.6.2.4 and Figures 2.6.2.1 - 2.6.2.19. Uncertainty analysis of the final assess-ment is presented in Figure 2.6.2.20, although this only reflects the uncertainty in fitting the model and does not include uncertainty in the model specification. Estimates in 2004 of mean F2-6 vary in a similar way to last year, between 0.21 and 0.31 (25 and 75 percentile respec-tively), and for SSB between 1.73 and 2.14 million tonnes (Figure 2.6.2.20). There appears to be a relatively good agreement between the point estimates of the final assessment and the median values of the ICA bootstrap realisations. The estimation of mean F2-6 varies less than the SSB (Figure 2.6.2.21). Long-term trends in yield, fishing mortality, spawning stock bio-mass and recruitment are given in Figure 2.6.2.22.

The spawning stock at spawning time in 2004 is estimated at approximately 1.89 million ton-nes. The abundance of 0wr fish in 2005 (2004 year class) is low for the third consecutive year.

A low recruitment was also observed in the two previous years for the 2002 and 2003 year classes. The strong 1998 and 2000 year classes are still evident in the population, with the 2000 year class at 3wr in 2004 being the highest since 1964 and the 1998 year class at 5wr being the highest since 1962. Mean fishing mortality on 2-6wr herring in 2004 is estimated at around 0.25, and on 0-1wr herring at 0.05. The value of F for 2003 in this years assessment is in close agreement with last year’s assessment, which was 0.24.