Variations in life-history traits in exploited and unexploited populations of Perth herring (Nematalosa vlaminghi): evidence for density-dependent growth?

Description of estuaries

The Swan–Canning Estuary (SCE) (32.0°S, 115.8°E) has a catchment of ~126 000 km². The estuary surface area is ~55 km² (Brearley 2005), including a narrow entrance channel, a series of basins and the tidal parts of the Swan and Canning rivers (these riverine sections are collectively referred to as the ‘upper estuary’) (Fig. 1). The upper estuary is shallow (<3 m in both rivers). The Swan River contributes the majority (85%) of flow into the estuary. Under low-flow conditions, saline waters can extend 60 km upstream from the sea in the Swan River (Spencer 1956), but only ~25 km upstream in the Canning River, where a weir (built in 1926) is closed in low-flow months (typically October–May) to prevent intrusion of saline water.

Fig. 1.

Location of the Swan–Canning Estuary (SCE) and Peel–Harvey Estuary (PHE) on the south-western coast of Australia. Insets: estuary entrances, basins and main tributaries of SCE and PHE. Tidal reaches of the tributaries collectively comprise the ‘upper estuary’.

The Peel–Harvey Estuary (PHE) (32.6°S, 115.7°E) is ~50 km south of the SCE. The middle part of the PHE consists of two connected basins (‘Peel Inlet’ and ‘Harvey Estuary’), with a combined surface area of ~135 km² (Fig. 1). The basins are connected to the sea at the northern end via a natural channel and, on the western side via an artificial entrance (the Dawesville Channel) built in 1994 to increase flushing and improve water quality in the basins. Three shallow (<5 m) rivers, the Serpentine, Murray and Harvey rivers (collectively the ‘upper estuary’), contribute ~95% of total flow into the estuary (Ruibal-Conti et al. 2013). On the Murray River, a permanent weir located ~35 km upstream from the ocean prevents saline-water intrusion in summer. Salinity in the upper estuary ranges from almost fresh to hypersaline during low-flow periods.

The region has a Mediterranean-type climate, with over 85% of the rainfall occurring in the cooler months (May–October). As rainfall and river flow decline, a salt wedge extends into the upper estuaries in spring and typically remains upstream until late autumn, or until winter in dry years. Under the stratified conditions of the salt wedge, the saline (bottom) layer often becomes deoxygenated (Kurup and Hamilton 2002).

The SCE and PHE catchments have been extensively cleared for agricultural, urban and industrial purposes (Riggert 1978; Bradby 1997; Brearley 2005). The upper reaches of both estuaries are now eutrophic or hypereutrophic because of runoff of anthropogenic nutrients from the naturally oligotrophic, sandy soils of the catchment (Bradby 1997; Chan et al. 2002; Brearley 2005; Cloern et al. 2014; Thomson 2019). These microtidal estuaries experience limited flushing and thus tend to accumulate pollutants and nutrients, especially during low-flow periods (Warwick et al. 2018). Superimposed on these processes are the effects of a drying climate, which has resulted in streamflow reductions of ~80% since the mid-1970s (Silberstein et al. 2012; Ruibal-Conti et al. 2013; Smith and Power 2014; see https://www.watercorporation.com.au/Our-water/Rainfall-and-dams/Streamflow). Reduced flow has allowed greater upstream penetration of the salt wedge, more extensive stratification, and increased hypoxia (DO < 2 mg L⁻¹) and anoxia (0 mg L⁻¹) in the upper estuaries (Cottingham et al. 2014, 2018). Reduced flushing and sediment nutrient releases under low-flow conditions have contributed to the increasingly eutrophic state of the upper estuaries (Douglas et al. 1997; Twomey and John 2001; Kurup and Hamilton 2002; Hamilton et al. 2006). In the PHE, construction of the artificial entrance increased saltwater intrusion, resulting in further upstream penetration of the salt wedge and stratification in the rivers.

In the PHE, dense macroalgal blooms occurred in the estuary basins in the 1960s and 1970s, which were partly replaced by extensive cyanobacterial blooms in the warmer months in the 1980s and early 1990s (Lukatelich and McComb 1989; McComb and Lukatelich 1995). These blooms caused hypoxia and fish kills, and clogged fishing nets (Potter et al. 1983; Lenanton et al. 1985). Under these conditions, commercial fishing for N. vlaminghi effectively ceased in 1984 and did not resume until 1996 after the artificial entrance was opened (Fig. 2b).

Fig. 2.

Annual commercial catches of Nematalosa vlaminghi in the (a) Swan–Canning Estuary (SCE) and (b) Peel–Harvey Estuary (PHE), 1941–2020.

The upper parts of the SCE and PHE are characterised by high phytoplankton densities (Thompson and Hosja 1996; Thomson 2019). Since the 1990s, the upper estuaries have experienced an increasing frequency of blooms by potentially harmful phytoplankton species, including toxic dinoflagellates that, in concert with hypoxia, contribute to regular fish kills in summer–autumn (Twomey and John 2001; Chan et al. 2002; Adolf et al. 2015; Hallett et al. 2016; Thomson 2019). Significant fish kills occurred in 2003–2006, 2010 and 2012 in the SCE, and in 2003–2005 and 2017 in the PHE, with minor fish kills in most other years (Cosgrove et al. 2014; Thomson 2019).

Fish collection and measurement

Fish were sampled by fishery-independent methods during 2016–2018 in the SCE and 2017–2018 in the PHE. In addition, fish were sampled during 2016–2017 from commercial fishery landings in the PHE. Full details of fishery-independent sampling methods and locations are described by Valesini et al. (2017) and Hallett et al. (2019). Briefly, fish were collected at night in deep (>2 m) waters by a multi-panel gill net comprising 20 m long × 2 m high panels, each with a different mesh size (38, 51, 63, 76, 89, 102, 115 or 127 mm). In addition, small juveniles (Age 0+ years) were sampled in the day in shallow (<1.5 m) waters by a beach seine net with two 25-mm mesh wings (length 20 m) and a 9-mm mesh bunt. Together, these methods captured the full size range of N. vlaminghi (Smith et al. 2022). Commercial fishery catches were taken in deep waters by a haul net with a mesh size of 52 mm.

In the SCE, ~90% of fish were caught in the Swan River or estuary basin, with the remainder being caught in the Canning River. In the PHE, ~95% of fish were caught in the Murray River or estuary basin, with the remainder being caught in the Serpentine River.

Fish were frozen after capture. The effect of freezing on length was not investigated in this study. Shrinkage of 1–2% after freezing has been reported in other clupeids (e.g. Wessels et al. 2010). After defrosting, the total length (TL) and fork length (FL) of each fish were recorded to the nearest 1 mm. All lengths reported in the Results are TL. Any missing values of TL for individual fish were estimated using its relationship with FL determined from the other fish sampled in this study, as follows:

where r² = 0.98, n = 144, P < 0.001.

Sagittal otoliths were extracted, dried and stored in paper envelopes. Fish were aged by counting validated annual growth increments in thin sections of their sagittal otoliths and assuming an average birth date of 1 January for all fish (Smith et al. 2022).

Gonads were extracted and assigned to one of the following eight macroscopic developmental stages adapted from Laevastu (1965): I, immature; II, resting; III, developing; IV, maturing; V, prespawning; VI, spawning; VII, spent; and VIII, recovering.

Fish were collected under permits W1006/03, R3000/17 and RW2706/14 issued by the Murdoch University Animal Ethics Committee and research exemptions 2671, 2783 and 3032 issued by the Department of Primary Industries and Regional Development.

Growth

Because the sex of most small fish (Age 0+) could not be determined macroscopically, their lengths-at-age were assigned randomly, but in equal proportions, to female and male datasets. The inclusion of these small fish was necessary to estimate reliably the growth of young juveniles in the populations. Growth was described by fitting Schnute growth curves (Schnute 1981) to length-at-age data for each sex and also to the combined data for both sexes. The Schnute growth model has been shown to provide statistically better fits to length-at-age data for N. vlaminghi than has the commonly used von Bertalanffy growth model (Smith et al. 2022). The Schnute growth equation is as follows:

where L₁ and L₂ are the estimated total lengths at selected reference ages τ₁ and τ₂ (i.e. 0 and 10 years respectively) and a and b are constants. Growth curves were fitted (using nlminb) in R (ver. 4.3.2, R Foundation for Statistical Computing, Vienna, Austria, see https://www.R-project.org/) by minimising the negative log-likelihood (λ) calculated as follows (Haddon 2011):

In this equation, n is the number of fish and σ^ is the standard deviation, calculated as follows:

where L_{t, j} and L^t,j are the respective observed and expected lengths at age for the jth fish. Approximate 95% confidence limits for the estimated model parameters were calculated as 1.96 multiplied by their estimated standard error, derived from the variance–covariance matrix for estimated model parameters, calculated in R by using the optimHess function.

Within each estuary, a likelihood-ratio test (Cerrato 1990) was used to test for differences in growth between sexes and also between estuaries. The test statistic was calculated as twice the difference between the negative log-likelihoods calculated when fitting a common growth curve to the data for the two sexes (or estuaries) combined, compared with fitting separate growth curves to the data for each sex (or estuary). The hypothesis that growth could be appropriately described by a combined growth curve was rejected at the α = 0.05 level of significance if the test statistic exceeded χa2(q), where q represents the degrees of freedom, determined as the difference between the numbers of parameters needed to fit separate curves v. a single curve (Cerrato 1990).

The average growth rate during the juvenile phase (G_juv) was calculated in each estuary for each sex as follows:

where L₅₀ is the total length (mm) and A₅₀ is the age (years) at which 50% of fish attain maturity.

Maturation

Logistic regression analysis was used to estimate the lengths at which 50% (L₅₀) and 95% (L₉₅) of fish attained maturity. An individual was considered ‘mature’ if its gonads were at stages III–VIII during the reproductive period (October–March; Chubb and Potter 1984). The probability (P) that a fish was mature was calculated as:

Using nlminb in R (R Foundation for Statistical Computing), logistic curves were fitted employing non-linear optimisation to minimise the negative log-likelihood (λ) calculated as follows:

where the jth fish was represented by X = 0 if immature and X = 1 if mature (e.g. French et al. 2014). Approximate 95% confidence limits were calculated for estimated parameters from their estimated standard errors. This logistic regression analysis was also used to estimate the ages at which 50% (A₅₀) and 95% (A₉₅) of fish attain maturity. For each maturity curve, a parametric resampling approach (assuming the estimated maturity parameters conform to a multivariate normal distribution) was used to produce 95% confidence limits for probabilities of maturity estimated for a range of specified lengths (and ages). This resampling analysis employed the MASS package (ver. 7.3-60.0.1, see https://CRAN.R-project.org/package=MASS; Venables and Ripley 2002) in R, together with the estimated variance–covariance matrix and maturity parameters to produce 500 estimates for probability of maturity at each length (or age), with the 95% confidence limits calculated as the 2.5 and 97.5 percentiles of these length values at each length (or age). Within each estuary, likelihood-ratio tests (Cerrato 1990), as described above for growth modelling, were used to test for a significant difference between sexes, and between estuaries, in the length (and then age) at which maturity was attained.

Mortality estimation

To estimate annual mortality, an age-structured model allowing for inter-annual variability in relative recruitment (Deriso et al. 1985; see also Smith et al. 2022) was fitted to age-composition data collected, each year, using gill nets in 2016–2018 in the SCE and in 2017–2018 in the PHE. The analysis assumed ‘knife-edge’ at Age 4 years, on the basis of selectivity analyses conducted by Smith et al. (2022), which showed N. vlaminghi becoming fully vulnerable to the gillnet sampling gear at ~180 mm. The selectivity S_a of fish of age a was thus set to 1 for all fish aged 4 or more years. As with many other forms of catch-curve analysis, fish were assumed to experience constant mortality.

Time was measured in ‘biological years’, defined as the 12-month period following the assumed annual birth date for N. vlaminghi of 1 January (and thus, for this species, with biological years equating to calendar years).

The expected number of fish N^a,t of Year class y that are of Age a years at the beginning of the first year of sampling, t₁, was as follows:

where Z is the instantaneous rate of total mortality, j is an index of age (where 0 ≤ j ≤ A), Rt1 is the number of 0-year-old fish recruiting to the population in Year t₁, R¯ is the average recruitment level across all years, and A is a plus group containing all fish of Age A years or older. For subsequent years:

R_y, the number of recruits of Age 0 years in Year class y, was calculated as:

where ε_y is the estimated ‘recruitment deviation’ for Year class y, with εy~N(0,σR2), where σR2 is the variance of the natural logarithms of the recruitment deviations. Recruitment deviation parameters were estimated for successive year classes for which there were substantive samples.

The expected catch of fish of Age a in Year t was calculated as follows:

and the expected proportion of fish of Age a in the catch in Year t as follows:

Using n_a,t to denote the number of fish of Age a in the sample in Year t, the observed proportion at Age a for that year was as follows:

The overall log-likelihood for the catch curve model is as follows:

where λ_age, the log-likelihood of the age composition data, is as follows:

and λ_R, the log-likelihood associated with the recruitment deviations, is as follows:

This study followed several other assessments (e.g. Smith and Punt 1998; Maunder and Deriso 2003) that assume a prior probability distribution for the recruitment deviations, using the value 0.6 for the standard deviation of the natural logarithms of the recruitment deviations (Beddington and Cooke 1983; Mertz and Myers 1996). A penalty was also added to the objective function to prevent estimates of Z being confounded with trends in recruitment deviations (Fairclough et al. 2014).

Lifespan and reproductive lifespan

Typical lifespan (t_max) was the estimated age at which only 1% of individuals remain alive, assuming that total mortality (Z) is 0.32 year⁻¹ in the SCE and 0.84 year⁻¹ in the PHE (see Results). The reproductive lifespan (RL) was calculated as:

where A₅₀ is the age at which 50% of individuals attain maturity.

Evidence for density-dependent somatic growth

The differences in life-history traits between populations of N. vlaminghi in this study were consistent with the hypothesis that commercial fishing has resulted in higher total mortality in the fished population than that in the unfished population, leading to faster somatic growth because of density-dependent processes triggered by the harvesting of adult fish, and a younger mean age at maturity associated with the faster growth. Thus, results showed that during 2016–2018, the total mortality (Z) of adult fish was ~2.6 times higher in the PHE than in the SCE (Z = 0.84 v. 0.32 year⁻¹). At this time, the average juvenile growth rate (in terms of length) in the PHE was more than twice that in the SCE, which allowed females to attain maturity ~4 years earlier, but at a length similar to that of females in the SCE.

In the SCE, the mean age at maturity in 2016–2018 (6.2 years for females, 7.6 years for males) was substantially higher than that reported four decades earlier, when fishery catches of N. vlaminghi were at their peak in this estuary. In the late 1970s, most juveniles in the SCE were reported to mature at the age of 3 years, at an estimated total length of 176 mm (estimated from a growth curve for both sexes; Chubb and Potter 1984). Thus, the typical age at maturity was at least 3 years younger than that observed in 2016–2018. Whereas the historical mean length at maturity was smaller, the average juvenile growth rate (G_juv) was faster than in 2016–2018 (59 v. 33 mm year⁻¹). Adult growth was also greater in the late 1970s, as indicated by larger lengths at age (e.g. 293 v. 225 mm at age 8 years) and a substantially greater maximum observed length than in 2016–2018 (361 v. 291 mm; Chubb and Potter 1984; Smith et al. 2022).

Juvenile and adult N. vlaminghi co-occur in the upper parts of estuaries during the warmer months and so the removal of adults by fishing could potentially reduce competition for resources and result in faster growth of juveniles. Juvenile and adult gizzard shad are omnivorous filter-feeders that consume detritus, phytoplankton, benthic algae and zooplankton (Yako et al. 1996; Higgins et al. 2006; Pusey et al. 2021). Individuals that feed extensively on zooplankton have faster growth than do those that consume primarily detritus (Mundahl and Wissing 1987; Schaus et al. 2002). Detritus typically dominates the diet, but higher-quality food is preferentially consumed when available (Yako et al. 1996; Sternberg et al. 2008). At high densities, gizzard shad quickly deplete higher-quality zooplankton prey and shift to consuming abundant, but lower-quality, detritus. Hence, gizzard shad growth is dependent on food quality, which, in turn, can be influenced by its population density. Density-dependent regulation of growth has also been reported in Dorosoma cepedianum, a well-studied, anadromous gizzard shad in North America (DiCenzo et al. 1996; Kim and DeVries 2000; Schaus et al. 2002; Vanni et al. 2006).

An alternative hypothesis to explain the differences in life-history traits between N. vlaminghi populations is that environmental factors, such as temperature, salinity, productivity or dissolved oxygen, differed between estuaries and led to different rates of N. vlaminghi growth or mortality. However, there are no obvious large-scale environmental differences between the PHE and SCE that might explain the trait differences. These estuaries are closely located and thus share the same climate and have similar physical characteristics and histories of environmental degradation, including increasing hypoxia and eutrophication over the past four decades (see ‘Description of estuaries’ in Materials and methods). Previously, Smith et al. (2022) speculated that density-dependent processes associated with increases in primary production or the extent of hypoxia since the 1970s may have contributed to the slower growth of N. vlaminghi in the SCE. However, given the current observation of fast growth of N. vlaminghi in the PHE where there is currently high primary production and extensive hypoxia, this suggests that density-dependent processes associated with fishing, rather than changes in primary production or hypoxia, have had the greatest effect on the growth of this species in these two systems.

Environmental degradation may be contributing to total mortality of N. vlaminghi in each estuary. Over the past two decades, numerous ‘fish kills’ associated with low-oxygen events, often exacerbated by toxic dinoflagellate blooms, have occurred in summer–autumn in the upper parts of both estuaries (Smith 2006; Cosgrove et al. 2014; Thomson 2019). The largest recorded fish kills occurred in the SCE in 2003 (>200 000 fish) and in the PHE in 2017 (>30 000 fish) (Cosgrove et al. 2014; Thomson 2019). The majority of these recorded deaths are related to Acanthopagrus butcheri, a reasonably large-bodied demersal species that resides permanently in estuaries. However, major fish kill events are logistically difficult to survey because they cover large areas and last several days, and minor events are sometimes not recorded. Also, smaller fish decompose rapidly, and certain species tend to sink after death and are not observed. Thus, official records of fish kills in these estuaries probably underestimate the frequency of events, and the actual number of species and individuals affected. The limited available information for N. vlaminghi indicates that moderate numbers (>1000) of adults of this species have been killed in several low-oxygen events in each estuary (Smith et al. 2004; Smith 2006), but the extent to which N. vlaminghi total mortality has been altered by these events over the past two decades is unknown.

Environmental degradation, including declining freshwater flows and increasing extent of hypoxia, may also have reduced the carrying capacity for N. vlaminghi in each estuary. A reduced population size in the PHE would explain why recent fishery catches in this estuary, which are relatively small compared with those taken in the 1960s, are associated with high total mortality in this population.

Does faster growth really compensate for high mortality?

For N. vlaminghi, juvenile somatic growth (in terms of length) in the PHE population was 2.4 times faster for females and 3.3 times faster for males than that in the SCE population, which suggests high phenotypic growth plasticity in this species. Such plasticity is likely to be a useful adaptation for N. vlaminghi because it allows populations to quickly respond to the naturally variable conditions that occur in local estuaries. It also facilitates density-dependent processes that help buffer populations against the effects of high fishing mortality or environmental degradation. In recent years, the faster growth of juveniles in the PHE has allowed them to mature earlier and produce offspring more quickly than those in the SCE, potentially compensating for the reduced reproductive lifespan in the PHE.

Density-dependent recruitment may also be increasing productivity and compensating for high adult mortality in the PHE population (Rose et al. 2001). The larvae of gizzard shad are obligate zooplanktivores (Miranda and Gu 1998; Kanou et al. 2004; Medeiros and Arthington 2008; Uehara and Tachihara 2016), and lower adult densities may allow higher larval survival by reducing competition for zooplankton prey. Lower adult densities are also likely to reduce cannibalism of eggs and larvae. Density-dependent larval survival is reflected in the over-compensatory (Ricker-type) stock–recruitment relationship of the gizzard shad D. cepedianum (Miranda et al. 2020).

Density-dependent growth and recruitment can compensate for the effects of high mortality only if conditions are suitable for reproduction and recruitment. In the estuaries of south-western Australia, conditions for reproduction and recruitment can be unfavourable for extended periods because of environmental factors, including low flow and hypoxia (Smith et al. 2022). N. vlaminghi has a potential longevity of at least 20 years, which is one of the longest reported for any clupeid species (Smith et al. 2022). A long lifespan allows individuals to survive long enough to take advantage of favourable conditions when they occur (a ‘storage effect’) (Secor 2000). However, when lifespan is substantially reduced, the number of opportunities for successful reproduction decrease. As a result of the higher adult mortality in the PHE, the reproductive lifespan of females during 2016–2018 was estimated to be 44% lower in the PHE than in the SCE (4.5 v. 8.0 years). Thus, individual females in the PHE have almost half of the number of annual spawning opportunities that are available to females in the SCE throughout their life.

Age truncation may also be reducing the likelihood of reproductive success in the PHE population through various maternal effects. For example, age-related differences in spawning times and locations can increase the chance that offspring will encounter favourable conditions, and older females tend to produce a greater quantity and quality of eggs, which may also increase offspring survival (Wright and Trippel 2009; Hixon et al. 2014; Ohlberger et al. 2022; Yoneda et al. 2022). Furthermore, age truncation tends to destabilise fish populations, making them more variable and prone to collapse in the event of unfavourable conditions (Anderson et al. 2008; Hsieh et al. 2010; Rouyer et al. 2012; Hixon et al. 2014; Barnett et al. 2017; Ottersen and Holt 2023). Given these effects, it is likely that strong age truncation is decreasing the resilience of the PHE population to cope with exploitation and habitat degradation in the upper estuary, irrespective of any compensatory density-dependent responses that may be occurring as a consequence of high mortality.

Implications for the meta-population of N. vlaminghi

Consistent differences in the age composition and size at age of N. vlaminghi between the PHE and SCE across multiple years in this study indicate distinct patterns of recruitment, growth and mortality in each estuary, and imply low connectivity between populations despite their estuary entrances being located only ~50 km apart. These patterns suggest that most adults return to the same estuary to spawn (natal philopatry) and, therefore, the migratory behaviour of N. vlaminghi is similar to that of most other anadromous fish, which display a strong tendency to ‘home’ to their natal estuary (e.g. Grunwald et al. 2008; Jolly et al. 2012; Engstedt et al. 2014; Keefer and Caudill 2014; Palkovacs et al. 2014; Martin et al. 2015; Davies et al. 2020).

Most ‘homing’ anadromous fish species display some degree of straying, usually into nearby rivers (Keefer and Caudill 2014; Martin et al. 2015). Thus, the larger SCE population could still be an important source of recruitment to other estuaries, helping sustain small populations or replace extirpated populations. There are no estimates of N. vlaminghi absolute abundance in either estuary, but there is strong circumstantial evidence that the SCE has always had a greater carrying capacity for N. vlaminghi than has the PHE. Since 1940, commercial catches of N. vlaminghi have been consistently higher in the SCE than in the PHE, despite a larger fishing fleet and consistently higher effort levels in the PHE (and high market demand in both fisheries). During the peak periods of demand for this species, in the 1940s and the 1960s, the annual catch peaked at 80 and 178 tonnes (Mg) respectively, in the SCE but only 30 and 53 Mg respectively, in the PHE. During 1965–1995, N. vlaminghi comprised 40% by weight of total commercial landings (including all species of fish and invertebrates) in the SCE, but only 1% in the PHE (Smith and Lenanton 2021). Historical records also suggest that N. vlaminghi was relatively abundant in the SCE, compared with other estuaries. For example, the strong association of this species with the SCE, which flows through the city of Perth, gave rise to its common name of ‘Perth herring’ with this estuary historically being regarded by fishers as the ‘true home of Perth herring’ (Quain 1947).

Owing to the drying and warming climate in south-western Australia, the major breeding stocks of N. vlaminghi are likely to have been limited to just four estuaries that remain permanently open, including the SCE and PHE (Smith et al. 2022). With so few breeding stocks, the persistence of each one is important to the sustainability of the species. In anadromous species, the existence of a diverse ‘portfolio’ of spawning stocks is a strategy that spreads risk and buffers against local environmental fluctuations, promoting stability and resilience in the meta-population (Schindler et al. 2010; Waldman et al. 2016). Maximising resilience in each N. vlaminghi stock should therefore be a management priority. The current ‘low resilience’ status of N. vlaminghi in the PHE is of concern. Fishery and habitat management strategies that reduce mortality and preserve older age classes in this small population should be adopted to enhance its resilience.