in

A fine-scale multi-step approach to understand fish recruitment variability

[adace-ad id="91168"]

To investigate the pathway from adult population characteristics to spawning behaviour, egg production, and ultimately to recruitment (Fig. 1), we used three data sources; an egg survey (for estimates of egg distribution, total egg production, and environmental variables), biological samples of the commercial fishery (for estimates of spawning duration and peak, and maternal body condition), and stock assessment outputs (for estimates of age-1 recruits, spawning stock biomass and age structure).

Figure 1

Conceptual framework of the pathway from spawners to recruits and the underlying mechanisms investigated (stock demographic structure and environmental conditions in red and green, respectively).

Full size image

Egg survey data

Sampling

Mackerel enter the southern Gulf of St. Lawrence (sGSL, Eastern Canada) in early June each year to spawn, after overwintering along the north-eastern US continental shelf (from Sable Island to the Mid-Atlantic Bight29,30). Each year, Fisheries and Oceans Canada (DFO) conducts a 2-week long mackerel egg survey in the sGSL (a 65-station fixed grid 20 nautical miles apart spanning the dominant mackerel spawning area) around the average mackerel peak spawning date of June 21st. Over this period, a large fraction of spawning occurs and the survey is therefore believed to reflect appropriately spawning intensity and spatio-temporal properties. Stations consist of double oblique tows using 61-cm Bongo nets with 333 µm mesh size and flowmeters carried out on board a research vessel at a speed of 2.5 knots from 0 to 50 m depth to estimate daily and total egg production while also measuring physical and biological oceanographic variables (see further details in SI Appendix A). This survey has been carried out consistently since 1982, except for no surveys in 1995 and 1997. Several indices are derived from this mackerel egg survey: total egg production, egg distribution, water temperature, and zooplankton biomass, species composition, abundance, and distribution.

Total egg production and distribution

Annual total egg production was calculated according to a standard DFO protocol based on the Daily Egg Production Method31. Stage 1 (spawned less than 24 h ago) and 5 (i.e., damaged stage 1 eggs) egg counts were standardized by the volume of filtered water and the depth of the sampled water column to provide egg densities per station (number m−2). These numbers were then adjusted for incubation time32 to obtain daily egg production point estimates. Spatial interpolation was done across a grid of 3320 coordinates using ordinary kriging to calculate a mean daily egg production estimate per grid cell, which was extrapolated to the surface area sampled. Annual egg production estimates were obtained by dividing by the proportion of reproductively active fish at the median date of the survey. This latter value, along with peak spawning date and spawning duration was calculated using a logistic model describing the daily evolution of the gonadosomatic index, based on corresponding biological data (see further details in Doniol-Valcroze et al.31, and in “Commercial fishery sampling”).

To examine the potential inter-annual spatial mismatch between spawning location and the optimal habitat for larvae, we calculated the spatial extent (spawning area) and the position of the centre of gravity (spawning longitude and latitude) of spawning for each year in the time series. The spatial extent of egg production was determined using an α-convex hull on stations where eggs were present33. The centre of gravity of total egg production was calculated by taking the arithmetic mean of the coordinates of each station weighted by their individual observed egg production.

Environmental indices

Sea surface temperature (SST, °C) directly affects early life stage growth and survival7, but might also have an indirect effect on recruitment through adult spawning behaviour, as mackerel generally spawn between 8 and 15 °C34. Therefore, we produced an SST index by averaging June CTD-measured mean water temperatures in the first 10 m over stations, where the majority of mackerel eggs and larvae occur35.

We hypothesized that the main adult mackerel prey (i.e., C. hyperboreus and capelin, Mallotus villosus36) might be influential as well, as they may affect spawning location and therefore be an indirect driver of recruitment. Capelin is despite its importance as prey in terms of weight36 not considered as a potential driver of spawning location, because its consumption by mackerel is infrequent, only important to the larger mackerel and likely opportunistic. As such, habitat selection is most likely to be related to copepod abundance and we developed spatial, biomass, and composition indices in June in the sGSL only for C. hyperboreus. As a proxy of adult mackerel prey location, we computed the annual centre of gravity of C. hyperboreus biomass (latitude and longitude) with the same methodology used for total egg production. Also, we estimated the total C. hyperboreus biomass (mg m−2) in the sGSL37. The percentage of C. hyperboreus biomass relative to the total Calanus spp. biomass (% C. hyp.) was calculated as we hypothesized that changes in C. hyperboreus proportion may have influenced adult mackerel feeding behaviour and thus spawning locations.

Mackerel larvae mainly feed on the early life stages (eggs, nauplii, and young copepodites) of C. finmarchicus, Pseudocalanus spp. and Temora longicornis25. The copepod daily egg production (CEDP, µg egg carbon L−1 d−1) of these three copepod taxa, calculated based on adult female abundance and species-specific per capita daily egg production (see details in the SI Appendix A), was previously recognized as a good predictor of mackerel recruitment23,24,25. High larval prey abundance might, however, be irrelevant when there is a temporal or spatial mismatch with larval distribution. An annual (y) index of a temporal match was therefore calculated in June in the spawning area as the proportion of older stage 6 female C. finmarchicus, producing prey for mackerel early life stages, with respect to the number of younger immature copepodite stages 4 and 526 (Eq. 1).

$${Temporal match}_{y}=100%times {N}_{C. fin female}/{N}_{C. fin stages 4-5}$$

(1)

Higher percentages of stage 6 female copepodites during mackerel spawning (i.e., a later development of the plankton community) should improve the temporal match between hatching and the availability of prey for emerging larvae26. This same index could not include Pseudocalanus spp. and Temora longicornis as only data for stage 6 adults were available. C. finmarchicus is, however, considered to be a good indicator of the overall zooplankton phenology in spring and early summer in the sGSL and should also reflect Pseudocalanus spp. and Temora longicornis phenology27. An annual index of a spatial match between mackerel egg distribution and their near-future prey was determined as the sum of mackerel daily egg production (DEP) at stations (s) with sufficient prey (i.e., copepod daily egg production above a threshold value) divided by the daily egg production of mackerel over all stations (Eq. 2).

$${Spatial match}_{y}=100%times {sum }_{s=1}^{S>threshold}{DEP}_{s,y}/{sum }_{s=1}^{S}{DEP}_{s,y}$$

(2)

The threshold copepod daily egg production value was determined as the 25th quantile of values measured for all years and stations, which excludes zero and near-zero prey availabilities unlikely to be able to support larval survival. This index of spatial match captures a combined effect of the abundance and distribution of the prey in relation to the distribution of the fish eggs. Note that due to the availability of taxonomic zooplankton data, Pseudocalanus spp., Temora spp., C. finmarchicus and C. hyperboreus data and hence all indices derived from it were available for only 21 years (but covering the entire span of the time series; 1982, 1985, 1987, 1990, 1993, 1996, 1999, 2000, 2003 and 2006 to 2017). Spatial and temporal match–mismatch proxies were based on a match with the mackerel eggs rather than the early larval phase. We expect this to introduce little noise as the development time of mackerel eggs is typically less than 6 days and mackerel larval development is fast (about 20 days32). All the environmental variables used and the associated hypotheses are summarized in Table 1.

Table 1 Summary of all the hypotheses tested along the pathway from spawners to recruits and associated references.

Full size table

Commercial fishery sampling

Adult mackerel samples are collected annually by DFO from the commercial fishery. The sampling covers the entire spawning area and period (thrice a week) and on average 4998 (range 421–14,858) individual fish are analysed each year. We used this data to calculate the annual peak spawning date (spawn. peak), spawning duration (spawn. duration), and maternal body condition.

Peak spawning date and duration were calculated each year based on the fit of a logistic model of the daily evolution of the gonadosomatic index. The mean value of the derived symmetrical probability density function was defined as the peak spawning day and the time between the 2.5% and 97.5% quantiles was estimated to represent the spawning duration in days.

As relatively fatter individuals might spawn more and higher quality eggs38, mature females (i.e., reproductive stages 3–839) sampled between their arrival in the sGSL and June 21st (the average peak spawning date) were selected to investigate the potential influence of pre-spawning fat reserves on total egg production and recruitment with the relative body condition index (Kn40, Eq. 3):

$${K}_{n}=frac{W}{{W}_{r}}$$

(3)

where W is the observed somatic weight (g) of an individual and Wr the predicted weight of an individual of a given fork length (FL, cm) calculated with Wr = αFLβ (α and β are nonlinear least-squares regression parameters).

Mackerel SSB, recruitment and age structure

Annual mackerel SSB, recruitment residuals and an index of age structure were derived from an age-structured state-space stock assessment model applied to the period 1968–201828. Note that the model was calibrated using an SSB index directly calculated from total egg production. In the assessment model, a two-parameter Beverton-Holt stock-recruitment relationship was used to estimate annual recruitment (abundance at age 1), and the residuals of this relationship were used in subsequent analyses (Rres). An indicator of the annual age structure was considered as bigger, older mackerel spawners (> age 5) are known to have a greater fecundity, and spawn in different spatial and temporal niches than younger females35,41. Mean biomass-weighted age (MA) was calculated using mature biomass-at-age (({SSB}_{a})) as follow in the Eq. (4):

$$MA=frac{sum_{a=1}^{a=10}(a{SSB}_{a})}{sum_{a=1}^{a=10}{SSB}_{a}}$$

(4)

MA was based on biomass rather than abundance to better reflect the stock’s reproductive potential42.

Mackerel early life stages are prey for pelagic fish sharing the surface waters of the sGSL. Herring are, relative to other potential predators, dominant, widely distributed and known predators of mackerel eggs and larvae36. Hence, we used cumulated spring and fall herring model-derived annual biomass43 as a proxy of predation pressure on mackerel early life stages.

Statistical analyses

Recruitment variability driven by spawning aspects and environmental gradients

We analysed the relationships between the successive steps leading to recruitment (spawning aspects, egg production and recruitment) and both demographic and environmental effects using generalised linear models (GLMs). All model configurations (response and explanatory variables) are given in Supplementary Table S2. Explanatory variables were normalized (i.e., by subtracting the mean and dividing by the standard deviation for each variable) to facilitate comparison of their respective effects (i.e., through their coefficients). When the response variable was Rres (with a 1-year lag), residuals were assumed to follow a Gaussian distribution with an identity link function, whereas for the other response variables a Gamma distribution with a log link function was used (as they can only take positive values44). Before performing GLM computations, collinearity between explanatory variables was measured using variance inflation factors (VIFs), considering a VIF threshold of 344. Specifically, mackerel SSB and MA were highly correlated (Pearson correlation coefficient > 0.7, see Supplementary Fig S1), so distinct sets of GLMs testing SSB or MA on spawning aspects were used. A backwards model selection procedure was performed, choosing the model with the lowest Akaike’s information criterion corrected for small samples sizes (AICc). If independent models including either SSB or MA showed an AICc difference less than 2, both were reported. Assumptions of homoscedasticity and normality were checked using residual plots while assumptions of independence (to ensure no autocorrelation was present) were checked using correlograms. By replacing GLMs with generalized additive models, the same conclusions were reached and there were no indications of strong non-linear effects.

Variability in total egg production (TEP) could not be linked directly to SSB and MA using regression techniques, because of model circularity (a TEP derived SSB index was used to estimate SSB) and collinearity (SSB and MA are significantly correlated and difficult to disentangle). Although the relative effect size of both variables could not be measured, the positive link between them is well established in the literature (i.e., that larger, older fish produce more eggs41). We, therefore, focussed our efforts on the possible link between TEP per unit of biomass, thereby removing the effect of fish number- and weight-at-age, and maternal body condition. Furthermore, by working with stock–recruitment residuals, we removed in large part the intrinsically related process of TEP. That is, the stock–recruitment relationship is presumably created by the biological dependence of TEP on SSB, and subsequently of recruitment on TEP. This link was hence not explicitly considered, although being present. A Jackknife procedure was conducted to assess the consistency and robustness of the optimal models explaining recruitment residuals (see SI appendix A). Also, recruitment estimates are inherently dependent on the modelling choices45, and we verified that recruitment residuals obtained under different assumptions (i.e., through a Virtual Population Analysis, VPA46) were not differently explained by the considered variables (see SI appendix A for more details).

Stability of the recruitment-larval prey availability relationship

Since Castonguay et al.23, a different stock assessment model has been employed, resulting in new recruitment timeseries47. As a baseline for comparison, we, therefore, refitted the recruitment–CEDP relationship from Castonguay et al.23 with the updated estimates and including all years (1982–2017, linear modelling). We hypothesized that, with the addition of new years of data, potential changes in the performance of this quantitative food index (i.e., CEDP) in predicting recruitment would be driven by a temporal change in the relationship because of altering underlying mechanisms. The latter could manifest itself as changes in the spatial or temporal match between the CEDP and the spawning distribution (a proxy of larval distribution), i.e., the ‘effective’ prey availability. Thus, we examined whether changing larval prey availability in space and time, coupled with a changing mackerel larval quality (using adult Kn as a proxy), can explain residuals and the potential breakdown of the Rres-CEDP relationship. Then, the drivers behind the spatial match-mismatch between mackerel eggs and larval prey were investigated. We considered maternal body condition, SST, and C. hyperboreus longitude (i.e., spawner prey). We also retained the relative abundance of C. hyperboreus in the Calanus spp. community (% C. hyp.), as this species does not produce eggs and nauplii available to mackerel larvae in the summer in the sGSL37,48 and appears to reduce abundance of C. finmarchicus early life stages (i.e., mackerel larval prey) through predation49. Thus, years with a large proportion of C. hyperboreus in the plankton community may display a larger mismatch between mackerel eggs and CEDP. A beta regression model was used to study the spatial match (as it is a proportion). All statistical analyses were conducted with R (version 3.3.250).

Ethical approval

This study was approved by DFO Research Ethics Board and conducted with methods in accordance with the Canadian Council on Animal Care (ISBN: 0-919087-43-4).


Source: Ecology - nature.com

Validating the physics behind the new MIT-designed fusion experiment

Utilizing conductivity of seawater for bioelectric measurement of fish