General information
Description of the stock and management units is provided in the stock annex (ICES, 2019b).
The stock was benchmarked, and the management plan evaluated, in March 2019 (ICES, 2019a). The result was no change in assessment setup. A minor change in the management plan was introduced as MGT Btrigger was decreased from 65 to 61 thous. tonnes to be in line with ICES MSY Btrigger. Other reference points were unchanged except HRlim and HRpa. were introduced to replace Flim and Fpa.
Fisheries-dependent data
Landings of saithe in Icelandic waters in 2023 are estimated to have been 41 717 t (Figure 1), decreasing from 61 881 t in 2022 and much less than the TAC that was 70 300 tons for the fishing year 2022/2023 and 66 533 tons for the fishing year 2023/2024 (Figure 4).
Figure 1. Saithe. Above: Total landings from Icelandic waters. Below: Total landings and percent by gear in Icelandic catches. Catches taken by foreigners before 1978 are nearly all taken by bottom trawl.
Information on landings of saithe exist since 1905 (Figure 1). From 1905-1938 most of the catch was taken by foreigners, and also from 1950-1975 when foreigners, mostly Germans, accounted for 60% of the saithe landings. Mean annual catch of saithe has been 65 thous. tons since 1955, 73 thous. tons before 1980 but 60 thous. tons after 1980. In the last five years, the catch by foreigners has always been less than 300 tons and 0.5%, of which nearly all the catch has been taken by Faroese vessels.
Of the landings in 2023, 35 924 t were caught by trawl, 1333 t by gillnets, and the rest caught by other fishing gear (Figure 2). In the last two decades, most of the catch was taken by bottom trawl (83% in 2010–2017, 90% in 2018–2023,with gillnet and jiggers taking the majority of the rest, 5% each fleet. The share of the gillnet fleet was larger in the past, 26% in 1987–1996 compared to 8% in 1998–2020 (fig-landingsbygear). Reduction in the gillnet fisheries is caused by general reduction in gillnet boats that are mostly targeting cod and increased mesh size in gillnet fisheries targeting cod. The catch by foreigners before 1978 was nearly all taken by bottom trawl.
The reduction in the gillnet fleet was driven by boats changing from gillnets to longlines, a change driven by cod and haddock fisheries. The price of large gillnet cod sold for bacalao reduced compared to “normal size”, so it became more economical to operate longliners that supply fish evenly through the year. An increase in the haddock stock in the early 2000s and progress in automatic baiting were also important factors. For saithe fisheries it is important to note that saithe is rarely caught by longliners, so the fleet has become much less directed toward saithe fishing than before. The share of longlines increased gradually from 0.8% before 2000 to 2.2% in 2013–2016 reducing to less than 1% in 2021 and 2022. The share of longlines increased again in 2023 but the amount caught was similar. Faroese vessels take 35% of their catch by longlines.
The fleet using demersal trawl can be divided in two parts, those that freeze the catch and those that land it fresh. The trend in the last decade has been that the proportion of the trawler fleet that land the catch fresh has increased. Freezing trawlers have taken large proportion of the catch of saithe and redfish but much less of cod and haddock (Figure 3). The main reason for this is relative price of frozen vs fresh fish for each species. Mixed fisheries issues, like avoiding redfish when catching fish to be landed fresh, can be a factor, as redfish scratches the catch. The same vessels are catching redfish and saithe in the same area but not in the same hauls. Redfish is mainly caught during daytime and saithe during night.
Most of the saithe is caught by bottom trawl at 150-250 m depth. Other gear include gillnets that catch saithe at 50-200 m depth and Danish seine and handline that catch saithe at depth less than 150 m.
Spatial distribution of the saithe fisheries changed much from 2002–2014 (Figure 5). Before 2002 most of the saithe was caught south and west of Iceland but since 2012, 40–50% of the catch has been taken northwest of Iceland. Comparable percentages before 2002 ranged 3–8%. Similar increases can be seen for golden redfish, but redfish and saithe have for a long time been caught by the same vessels, not necessarily in the same hauls, rather as night versus day fishery. The area where saithe is caught now, Hali (Figure 6), has since the beginning of the 20th century been the most important cod fishing ground for trawlers.
Discarding is not considered to be a problem in the Icelandic saithe fisheries, with an estimated discard proportion of 0.1% (annual reports by Palsson et al., 2003 and later). Recently, the fleet does also seem to have difficulty in catching the set TAC, making discards more unlikely.
In 1999–2005, substantial fisheries of blue whiting were conducted within Icelandic and Faroese jurisdictions. From 2003-2005, bycatch of saithe in those fisheries was estimated to be 1500-4000 tons per year, less than half of that within Icelandic juridiction. Since 2007, blue whiting fisheries in Icelandic jurisdiction have decreased significantly compared to 2000–2005.
Catch per unit effort
Catch per unit effort (CPUE) from the bottom trawl fleet shows considerable variability, and has decreased considerably from its peak in 2018. CPUE in 2022 and 2023 is the lowest it has been since 2011 (Figure 7). Unit effort is here hours trawled and the CPUE index for each year is the median of the CPUE for the selected hauls.
When compiling CPUE indices, deciding which hauls to base the analysis on is not straightforward. All hauls inside a particular area, all hauls with saithe recorded, or all hauls with saithe accounting for more than a defined proportion of total catch could be chosen. The larger the stated fraction, the greater the variability in the CPUE index (Figure 7).
CPUE in 2022 and 2023 is not low as compared to earlier years, especially if the index is compiled based on all hauls where saithe has been registered. The question is then if data 15 years ago are comparable to modern data, owing to technological advances. However, CPUE indices show considerable similarity to total biomass index from SMB (Figure 20).
Landings, advice and TAC
For all Icelandic stocks that are managed by a TAC system the TAC is given for fishing year where fishing year y/y+1 is from September 1st in the year y to August 31st in year y+1. Assessment done in the spring of year y, is used to give advice for the fishing year starting September 1st the same year. For most stocks the survey conducted in March is the most influential data source in the assessment and the most recent survey from March in the assessment year is used for the advice given in June the same year.
The management plan and assessment for Icelandic saithe have been unchanged since 2010 and advice based on the same 20% harvest control rule as used for cod. Since 2014/2015 the set TAC has not been caught (Figure 8) but in the period 1997/1998 to 2013/2014 the TAC was caught in all years except 2007/2008 and 2008/2009. The catch in the fishing year 2022/2023 was 45.8 kt while the TAC was 70.3 kt so only 65% of the TAC was caught. Trends in landings indicate similar result, that less than 70% of the TAC will be caught in the fishing year 2023/2024 (Figure 10).
The Icelandic Fisheries management system allows some transfer between species based on cod-equivalence factors that are supposed to reflect the price of the species compared to cod (see ICES, 2021). Transfer to cod is though not allowed in the system that is quite limited. In recent years saithe has been converted to other species (Figure 9) that are probably more economical to catch than saithe. Possibilities of species transfer were recently restricted when transfer to shared stocks like redfish were banned.
Even though some part of the saithe quota has been transferred to other species, a considerable part of the TAC has not been used at all. This is an indication that the saithe fisheries are not very economical, either due to smaller than estimated stock, or difficulty in catching. Historical assessment shows that fishing mortality of Icelandic saithe was never high, even in periods were fisheries were not limited (ICES 2002).
Sampling from catches
Compilation of catch in numbers is based on age and length distributions from the catches where the number aged is usually considerably less than number length measured.
Assessment
The samples used to derive catch in numbers are both taken by observers at sea and from shore samples. The trawlers that freeze the catch account for majority of sea samples, while all shore samples are from fresh fish trawlers. In addition, relatively few fishes from sea samples are sampled for otoliths but the age-length keys are similar between sea and shore samplesm even though the length distributions differ. Few sea samples were taken in 2020 and 2022 (Figure 11).
Sampling from commercial catch has been revised in recent decades, the number of samples has reduced and also the number of otoliths per sample (Figure 13) (ICES, 2019b). Sampling in 2020 was much less than in the years before, the number sea samples and number of age samples was especially low. The main explanation seems to be the COVID-19 epidemic. In 2021 the sampling was back to the level in 2017-2019 but has reduced again in 2022 and 2023.
Around 90% of the length samples are taken from trawl that accounts for ~90% of the catches.
Length and age distributions from catches.
Catch in numbers are compiled based on 2 fleets, bottom trawl and gillnets, 1 region and 1 season. Bottom trawl accounts for 90% of the landings and other fleets than bottom trawl and gillnet are included with the bottom trawl. The same length-weight relationship (\(W= 0.02498 \times L^{2.75674}\)) is applied to length distributions from both fleets.
Foreign catcges that were 147 tonnes in 2023 are included in the landings above. They were caught by longlines (75 tonnes), handlines (42 tonnes) and demersal seine (31 tonnes). Nearly all the foreign catches have in recent years been taken by the Faroese fleet.
Length distributions from sea and shore samples show some difference, where the shore samples show usually more of large fish (Figure 14). This difference might be reflecting the difference in composition of the catch of the trawlers that freeze the catch and those that land the catch fresh.
Length distributions from bottom trawl (Figure 15) show a tendency to catch smaller fish from 2003–2017 but again larger fish in 2018–2020 (Figure 10). In 2020 the +110 cm group was especially abundant, but proportion of 60-69 cm fish was above average in 2022. In 2023 proportion of 50-60 cm saithe was above average but the proportion of 60 cm and larger below average.
Catch in numbers at age shows more of ages 3 and 4 than predicted last year (Figure 16). This is in line with that which can be seen in length distribution were more of small fish is seen (Figure 15).
Mean weight and maturity at age
Weights of ages 3–6 have been low in recent years, but older ages are close to average weight (figures 12–14). The large 2012 cohort has the lowest mean weight of all year classes, both in catches and in the survey. This is in line with density dependent growth that has been observed in this stock and can for example be seen for year classes 1984 and 2000 that are both large. Year classes 2013 and 2014 that seem to be above average have higher mean weight at age than the 2012 cohort. The long-term trend since 1980 has been a gradual decline in the mean weight of all ages. Mean weight at age in catches was close to average in 2023 for most age groups, similar to 2022.
Weights at age in the landings are used to compile the reference biomass (B4+) that is the basis for the catch advice. Catch weights are also used to compile the spawning stock. Catch weights for the assessment year are predicted by applying a linear model using survey weights in the assessment year and the weight of the same year class in catches in the previous year as predictors (Magnusson, 2012; ICES, 2019b).
Maturity at ages 4–9 has decreased in recent years and is currently below average since 1985 (Figure 19). A model using maturity at age from the Icelandic groundfish spring survey is used to derive smoothed trends in maturity by age and year (ICES, 2019b).
Scientific surveys
In the benchmarked assessments from 2010 and 2019, only spring survey data are used to calibrate the assessment. Compared to the autumn survey the spring survey has larger number of stations (lower CV) and longer time series. Saithe is among the most difficult demersal fishes to get reliable information from bottom trawl surveys. In the spring survey, which has 500–600 stations, a large proportion of the saithe is often caught in relatively few hauls and there seems to be considerable inter-annual variability in the number of these hauls.
The biomass indices from the spring survey (Figure 20) fluctuated greatly from 1985–1995 but were consistently low from 1995–2001. Since 1995 the indices have been variable but compared to the period 1985–1995 the variability seems “real” rather than noise. This difference is also seen by the estimated confidence intervals of the indices that are smaller after 1995. In 2018 the indices were the highest in the series and had tripled since 2014. Most of the increase was caused by year class 2012 that was strong in the surveys 2015–2018 (Figure 21). The index reduced much from 2018-2020. It has been variable in last 4 years, was lowest in 2022 but increased again in 2023 and 2024. The trend 2021-2024 can though as stable with considerable noise.
The high index in March 1986 (Figure 20) is mostly the result of one large haul that is scaled down to the second largest haul when compiling indices for tuning. The scaling is from 16 tonnes to 1 ton.
Estimated CV from the survey is often relatively high and many relatively low values appear in the survey matrix, both for the youngest and oldest age groups. The youngest age group (age 3–4 and younger) are considered to inhabit waters shallower than the survey covers and the older age groups are reducing in numbers and could also be pelagic. The high index in 2018 came from relatively large catches in many hauls so the estimated CV was around average.
The autumn survey shows similar trend as the spring survey and the index is at high level in 2017 (2004 and 2018 are outliers due to large CV). The values before 2000 might be underestimate due to stations added in 2000 where large schools of saithe are sometimes found. Excluding these stations leads to lower but more stable index. An index based only on the stations taken since 1996 shows much difference between 1996-2000 and 2001-2022.
One characteristic of high uncertainty indices is that the exact trends depend on the stratification used, for species like saithe direct mean (all station with the same weight) might be the best approach.
Internal consistency in the March survey measured by the correlation of the indices for the same year class in 2 adjacent surveys is relatively poor, with R2 close to 0.46 where it is highest (Figure 21).
Catch curves from the survey indicate that Z ~ 0.5 assuming similar q with age (Figure 22).
Indices from the gillnet survey (SMN) conducted south and west of Iceland since 1996 were at highest level in 2019 but have decreased since then (Figure 24). Compared to earlier years the 2024 index in SMN is below average, the lowest since 2014 and much lower than in 2023. SMN is mostly targeting large saithe (mean weight in 2024 was 6.3 kg). The indices from SMH and SMN that are not used for tuning in the assessment indicate average and below average stock.
Assessment method
In accordance with the recommendation from the benchmark (ICES, 2019a), a separable forward-projecting statistical catch-age model Muppet (Björnsson 2019), developed in AD Model Builder, is used to fit commercial catch at age (ages 3–14 from 1980 onwards) and survey indices at age (ages 2–10 from 1985 onwards). The selectivity pattern is constant within each of 3 periods (Figure 26). Natural mortality is set at 0.2 for all ages.
To take low values in the survey series into account the survey residuals are compiled as \(\frac{log(I + \epsilon)}{log(\widehat{I} + \epsilon)}\) where \(\epsilon\) is a number that should avoid giving low values too much weight as they do in log-log fit. Typical value of \(\epsilon\) is the value that 3–4 otoliths will give, that would be 0.15 for saithe. Higher values are used for saithe 0.3 for the older ages, 0.5 for ages 3–5 and 0.7 for age 2, a value giving age 2 very low weight except the index if very high. The survey residuals (\(\ \frac{log(I + \epsilon)}{log(\widehat{I} + \epsilon)}\) ) are modelled as multivariate normal distribution with the correlation estimated (one coefficient).
The assessment model is also used for short term forecast, the Muppet model can’t be run without prediction. Future weights, maturity, and selectivity for short term prognosis are assumed to be the same as in the assessment year, as described in the stock annex. Recruitment predictions are based on the segmented stock-recruitment function estimated in the assessment model which is essentially geometric mean when the stock is above estimated break point that is near Bloss.
Reference points and HCR
In April 2013, the Icelandic government adopted a management plan for managing the Icelandic saithe fishery (Ministry of Industries and Innovation, 2013). ICES evaluated this management plan and concluded that it was precautionary and in conformity with ICES MSY framework.
The management plan for the Icelandic saithe fishery, adopted for the first time in 2013 was re-evaluated by ICES in March 2019 and found to be precautionary and in conformity with ICES MSY approach (ICES, 2019a).
The TAC set in year t is for the upcoming fishing year, from 1 September in year t, to 31 August in year t+1. The TAC according to the management plan is calculated as follows.
If \(\text{SS}B_{y} \geq MGT B_{\text{trigger}}\)
\[\text{Ta}c_{y/y + 1} = \frac{\text{Ta}c_{y - 1/y} + 0.2 \times B_{4 + ,y}}{2}\]
If \(\text{SS}B_{y} \leq MGT B_{\text{trigger}}\)
\[\text{Ta}c_{y/y + 1} = \alpha \times Tac_{y - 1/y} + (1 - \alpha) \times \frac{\text{SS}B_{y}}{\text{MGT B}_{\text{trigger}}} \times 0.2 \times B_{4 + ,y}\]
\[\alpha = 0.5 \times \frac{\text{SS}B_{y}}{\text{MGT} B_{\text{trigger}}}\]
Where \(\text{Tac}_{y/y + 1}\) is the TAC for the fishing year starting 1 September in year \(y\) ending 31 August in year \(y + 1\). \(B_{4 + ,y}\) the biomass of age 4 and older in the beginning of the assessment year compiled from catch weights. The latter equation shows that the weight of the last years Tac does gradually reduce from \(0.5\) to \(0.0\) when estimated \(\text{SSB}\) changes from \(\text{MGT B}_{\text{trigger}}\) to \(0\).
Reference points were also re-evaluated at WKICEMSE 2019 (See table below and ICES, 2019a). Blim, Bpa, MSYBtrigger, HRMSY and HRMgt were unchanged, MGMTBtrigger changed from 65 to 61 thous. tonnes and HRlim and HRpa were defined but earlier Flim and Fpa had been defined.
| Item | Blim | Bpa | MSYBtrigger | MGTBtrigger | HRMSY | HRMgt | HRlim | HRpa |
|---|---|---|---|---|---|---|---|---|
| Value | 44 | 61 | 61/65 | 61 | 0.2 | 0.2 | 0.36 | 0.26/0.25 |
| Basis | Bloss/1.4 | Bloss | Bpa | Bpa | Stochastic simulations. |
The recipe to evaluate MSY Btrigger and HRpa has changed since 2019 so those reference points were evaluated based on the same simulations as in 2019, leading to MSY Btrigger = 65 thousand tonnes and HRpa = 0.25.
Definition of harvest rate was changed in the 2023 assessment from \(HR_{y} = \frac{C_{y/y + 1}}{B_{y}}\) to \(HR_{y} = \frac{C_{y}}{B_{y}}\). The former method is thought to indicate better how the manage plan works but the latter method was considered by an advice group as a better indicator of fishing pressure. This change did not affect the way that advice was calculated.
Analysis of the results of stock assessment
The model fit to the total biomass index of saithe (@marchsur-aggfit_plot) indicates that the model does not predict the survey biomass well, neither the high nor low value of the index. Looking at the index the variability is too much for any model to follow. It must also be kept in mind the the tuning is done by comparing logarithm of observed and predicted survey number by age (Figure 25), comparison of biomass index is some kind of summary of this comparison.
The model residuals by age are shown in (Figure 25). In SMB negative residuals of an age group for few years and positive for few years can be seen. Positive and negative blocks can also be seen in some years, indicating that the multivariate normal distributions for survey residuals in each year has not removed all correlation from the data. A prominent block of positive residuals can be seen for the 2012 yearclass in the years 2016-2020.
Catch residuals shows blocks for some age groups that can be related to the model having fixed selection pattern within each of 3 specified periods. The estimated selection pattern in the first 2 periods is similar but the fleet is targeting small saithe more in the third period (Figure 26).
Catch residuals in 2023 are positive for the youngest age groups (Figure 25); nonetheless, targeting of young fish can be seen in (Figure 16) and (Figure 15).
Results of stock assessment
The stock assessment indicates that in 2024 both the reference biomass and spawning stock are above average and the harvest rate in 2023 was very low (Figure 27). Recruitment is estimated above average in recent years, except for yearclass 2015. Yearclass 2012 is estimated large.
Retrospective pattern indicates considerable overestimation of stock size in recent years (Figure 28). Estimated 5 years Mohn’s \(\rho\) is outside limits (0.31 for reference biomass, 0.37 for spawning stock, -0.21 for harvest rate and -0.178 for recruitment). Five years is in relatively short time for this stock were harvest rate is most likely moderate.
Uncertainty in the stock assessment
There is significant uncertainty in stock assessment of saithe due to variability in indices, inadequate recruitment estimates, and variable selection pattern. Estimated uncertainty in the reference biomass (\(B4+_{2024}\)) in the assessment model is 22% and analytical retrospective pattern for the assessment years 2001-2018 (years where the assessment has nearly converged (Figure 29)) gives a standard deviation of 0.24, autocorrelation of 0.5, and Mohns \(\rho\) 0.04 but Mohns \(\rho\) is a measure of bias in the time period investigated. In 2019, when the management plan for saithe was reevaluated, assessment error was based on retrospective analysis for the years 2001-2015. Based of assessment years 2001-2024 (Figure 29) the estimated value of Mohns \(\rho\) should be included in HCR evaluations.
Comparison with other models and settings indicates that the current estimate of the stock size is on the higher side (Figure 30). If the model is only based on age-disaggregated catch (SMB gets very low weight) the estimated stock is considerably small and with wide confidence intervals. The Muppet model cannot use age-disaggregated catches only, but the weight of surveys can be reduced substantially.
The problem in the stock assessment is that age-disaggregated catch indicates smaller stock than the surveys. If both SMB and SMH are used for tuning the surveys get more weight and the estimated stock size is larger. If one survey is used SMH indicates smaller stock as the survey time series is shorter and variability higher (less weight on survey).
The largest stock is obtained by using Winsorised indices, so that the two highest value in each year’s survey is scaled down to the third highest. This setup increases the weight of the survey relative to age-disaggregated (less noise more weight) catch which leads to a larger estimated stock.
Traditional VPA analysis (Adapt) estimates CV for each age group in the survey, leading to somewhat different patterns of CV compared to the other models that cannot estimate the CV for each age group independently. The Adapt model indicates similar stock size as the adopted model.
In most of the models except the SAM model, only the age composition of the catch is used in estimating the stock size, not the fact that the catch decreased much between 2022 and 2023. In the SAM model, fishing mortality is modelled as multivariate random walk. When catch changes as much as happened between 2022 and 2023, this term leads to a reduction in catches not immediately being been converted to a reduction in fishing effort, but rather a combination of reduced stock size and decreased fishing mortality. This is one of the reasons that the SAM model indicates the lowest biomass of the models and settings analysed (excluding the model where the survey is heavily downweighted). This random walk model is questionable when changes in fishing mortality are caused by management measures. In the case of Icelandic saithe where the TAC has not been caught for many years, the random walk term is appropriate.
Estimate of advice.
The state of the saithe stock is highly uncertain. The surveys SMB and SMH indicate average stock size but the gillnet survey (SMN) that the stock is small. The TAC has not been caught since 2024. In the fishing year 2022/2023, only 2/3rd of the TAC, indicating that the TAC might be too high.
CPUE was low in 2022 and 2023 compared to preceding years when it was high. The pattern is though somewhat similar as in SMB so it might already be accounted for in the assessment.
References
Bjornsson, Hoskuldur, Einar Hjorleifsson, and Bjarki Elvarsson. 2019. “Muppet: Program for Simulating Harvest Control Rules.” Reykjavik: Marine and Freshwater Research Institute. http://www.github.com/hoski/Muppet-HCR.
MRI, 2005. “Mælingar á brottkasti botnfiska og meðafli í kolmunnaveiðum 2004. Discard of demersal fishes 2004 and bycatch in blue whiting fishery 2004.” MRI Report. Reports of the Marine Research Institute. Vol. 117. MRI/117.
MRI, 2008. “Mælingar á brottkasti botnfiska 2007. Discard of demersal fishes 2007.” MRI Report. Reports of the Marine Research Institute. Vol. 142. MRI/142.
ICES. 2019a. “(Report of the workshop on the benchmark assessment and management plan evaluation for Icelandic haddock and saithe (WKICEMSE2019), 26-28 March 2019, Copenhagen, Denmark. ICES CM 2019.” International Council for the Exploration of the Seas; ICES publishing. http://doi.org/10.17895/ices.pub.5091.
ICES. 2019b. “Stock Annex: Saithe (Pollachius virens) in Division 5.a (Iceland grounds).” https://ices-library.figshare.com/articles/report/Stock_Annex_Saithe_Pollachius_virens_in_Division_5_a_Iceland_grounds_/18623102