Natural mortality augments population fluctuations of forage fish

1 INTRODUCTION

Forage fish, small pelagic fish species such as sardines, anchovies and herring, are a primary food item for many other species in the oceans, such as predatory fish, sea birds and marine mammals (Cury et al., 2011; Engelhard et al., 2014). Besides serving as the main link between higher trophic levels, they are also significant to the world's fisheries where they constitute approximately one-third of the global catch (Pikitch et al., 2014). Their use is both direct human consumption and reduction fisheries, where they are processed to fish oil and meal that has a variety of applications in both agriculture and aquaculture (Alder, Campbell, Karpouzi, Kaschner, & Pauly, 2008).

Forage population dynamics are characterized by extreme boom and bust cycles, where they fluctuate wildly in abundance (Maccall, 2009; Essington et al., 2015; Figure S1). These fluctuations can be explained partly by changes in reproductive success (recruitment) owing to oceanographic conditions (Chavez, Ryan, Lluch-Cota, & Niquen, 2003; Lindegren, Checkley, Rouyer, Maccall, & Chr, 2013). Fishing also plays a pivotal part in the crashes of forage fish worldwide (Beverton, 1990; Essington et al., 2015). Specifically, fishing increases the duration and frequency of population crashes by increasing fishing pressure when productivity is already decreasing (Essington et al., 2015). Yet for most species, mortality due to predators is a dominant source of total mortality. Consequently, if this mortality rate changes through time, owing to changes in predator abundance or feeding habits, it follows that it can have large effects on population dynamics and their ability to recover (Swain & Benoît, 2015). Apart from a small number of case-studies that implicate release of top-down control in population changes of marine prey species, to date, there has been no assessment of the contribution of variable predation mortality on forage fish populations.

Here, we ask how large the contribution of predation mortality is as a driver of forage fish dynamics, to understand the role of top-down control on forage fish variability. Top-down control has previously been shown to influence marine prey populations (Worm & Myers, 2003), and models predict that decreasing predation mortality by fishing predators can lead to an increase in forage fish catches (Andersen, Brander, & Ravn-Jonsen, 2015; Jacobsen, Burgess, & Andersen, 2017; Szuwalski, Burgess, Costello, & Gaines, 2017). We specifically asked whether changes in predation mortality are depensatory (increasing with decreasing abundance and vice versa), as is expected from standard functional response models, or whether they are compensatory (increasing with increasing abundance) as is expected if predators shift feeding towards abundant prey (Anderson, 2001). Depensatory mortality is destabilizing and would therefore act to amplify population fluctuations.

We analysed time series of biomass, natural mortality and fishing mortality for 10 forage fish stocks and examined the evidence for top-down regulation on forage fish populations. We evaluate the contribution of natural mortality to forage fish fluctuations and compare it with the contribution of fishing mortality and productivity.

2 METHODS

We identified forage fish stocks that had independent sources of data to provide estimates of annual natural mortality, and where these estimates were used in the statistical population models used to estimate population abundance. This ensured that the estimates of biomass were based on fisheries-independent abundance data and rigorous statistical models (known formally as stock assessment). The forage fish stocks included in the analysis covered four different large marine ecosystems, eight different species and a total of 10 stocks (Table S1). Common for all stocks included in the analysis was that the time-varying mortalities were based on predator abundance and food preference, with most natural mortality estimates coming from the SMS model (eight of the 10 stocks; Lewy & Vinther, 2004). The mortality of the two remaining stocks is calculated from a “MSVPA-X” model (Atlantic Menhaden, [Brevoortia tyrannus, Clupeidae]) and from Barents Sea Cod (Gadus morhua, Gadidae) consumption (Capelin [Mallotus villosus, Osmeridae]), respectively (Table S1). For all years with time-varying natural mortality, we used the estimated numbers at age (N_a,t), weight at age (w_a,t), fishing mortality rate at age (F_a,t), and natural mortality rate at age (M_a,t) from the stock assessment. The natural mortality rate is the sum of predation mortality and background mortality, but for simplicity, we use it as a proxy for predation (background mortality rate was constant for all stocks included).

We calculated the total predatory consumption in a year t, C_t, on forage fish as,

(1)

which is a variation on the Baranov catch equation, with the change that it calculates total consumption instead of total catch by interchanging M and F in the first fraction. The equation assumes that fishing and natural mortality work on the same time scale. Similarly, we calculated total catch, Y, as,

(2)

Total biomass is calculated as,

(3)

We defined the fraction of the total biomass taken each year as consumption as,

(4)

For simplicity, we refer to this as “natural mortality” for the remainder of the paper. The exploitation rate was calculated as,

(5)

To simplify the calculation of the stock productivity, we collapse the dynamics into a biomass model where we calculate the surplus production, P,

(6)

To compare among stocks, we normalized this number by dividing by the average biomass, thus, getting a “productivity rate.” For each stock, this calculates the remaining productivity not included in the two mortality terms (e.g., annual changes in recruitment and growth). We compare the productivity rate with the natural mortality rate and the exploitation rate. The model formulation assumes that all changes in biomass can be attributed to yield, losses due to natural mortality or the remaining productivity.

We classified time-series blocks as biomass dynamics leading up to a trough or a peak based on over five consecutive years of decline or increase, respectively. The time series leading up to peaks or troughs were identified by an algorithm that detected consecutive negative or positive gradients in the biomass dynamics and then saved consumption rate, productivity rate and exploitation rate from those years. To remove interannual variability (white noise) that would mask trends in some of these time series, we used the “loess” smoother in R on the biomass time series (using α = 0.3) prior to classifying time-series blocks (Figure S1).

Finally, we calculated the intensity of the fluctuations leading up to troughs or peaks by projecting the stocks forward in time using Equation 6, while assuming that the consumption was equal to the average consumption in the time period leading up to the trough or peak.

3 RESULTS

We identified 15 time-series blocks with over 5 years of consecutive increase in biomass (leading up to peaks) and 16 time-series blocks of over 5 years decline in biomass (leading towards troughs) and compared them with their trends in natural mortality, U (Figure 1). Generally, natural mortality decreased before a peak and increased before a trough, but with large variation among stocks (Figure 1, Figure S2). In the terminal year at the peak, 67% of the time series had below average mortality, whereas 56% of the time series had above average mortality in the trough. However, 2 years prior to the terminal year leading to troughs, mortality peaked where 81% had an above average U (Figure 2).

We calculate the average trajectory of mortality the years before a peak or a trough as ∑(U_i−mean(U_i))/n, where n is the number of biomass time series and i is the stock. The mortality is scaled by the long-term average (mean(U_i)) to compare the mortality trajectory among species with large differences in productivity. The relative average decline in natural mortality before a peak is moderate at −0.02 year⁻¹ from 7 years before the peak (Figure 2). On the contrary, the relative increase in U is approximately +0.09 year⁻¹ until it is highest 2 years prior to the valley (Figure 2). To ensure that individual stocks were not driving this trend, we examined the sensitivity by removing the highest and the lowest block for each of the two time series, but that only slightly damped the result (−0.01 and +0.07 year⁻¹ change before peaks and troughs, respectively).

We investigated the total contribution of mortality to fluctuations by calculating the mortality-adjusted production, P (Equation 6) required to fully explain the observed variability and compared it to the exploitation rate and consumption rate. Here, we present three representative stocks (Figure 3), the remaining ones are showed in Supporting information (Figure S3). If mortality (natural plus fishing) explained all variability, the mortality-adjusted production productivity would be constant before a peak or trough occurred, as the two mortalities would be responsible for all the biomass removals. In contrast, if there is depensatory mortality, total mortality and productivity will fluctuate out of phase, as high productivity leads to higher abundance and thereby lower mortality, and vice versa.

We find evidence suggesting that mortality fluctuations and productivity fluctuations jointly contributed to population fluctuations, but the patterns varied widely across species. Baltic sprat (Sprattus sprattus, Clupeidae) experienced a long-term decline in mortality (down approximately 55% from 1982 to 1992), observed at the same time as a general increase in productivity occurred (Figure 3a), consistent with depensatory mortality. Barents Sea capelin showed several periods of high productivity with lagged responses of natural mortality, such that increasing mortality was followed by strong declines in productivity and thus biomass (Figure 3b, Figure S1). Conversely, in years where productivity was increasing rapidly, mortality was low. For Barents Sea capelin, the years with lowest and highest abundance had low and high natural mortality, respectively, whereas the years leading up to throughs/peaks showed depensatory trend in natural mortality, with low increasing mortality leading up to a trough and a reduction in mortality leading up to a peak. Finally, North Sea herring (Clupea harengus, Clupeidae) showed different patterns than the other species in the analysis; natural mortality rate often followed the productivity rate. Rather, the fishing mortality rate went up when productivity went down, and fishing intensity decreased when productivity increased, emphasizing a tendency to compensatory dynamics in this species.

Large interannual variability in biomass is caused by changes in productivity rather than mortality, something that is common among the species explored here (Figure 3, Figure S3). Interestingly, the amount of biomass removed by natural mortality was higher than the biomass removed by fishing for all species included in the analysis, underscoring the role of natural mortality for forage fish population dynamics.

We find that changing mortality amplified peaks and deepened troughs. When we simulate population dynamics with constant mortality, on average biomass dynamics before a peak would have been 25% lower than was observed (Figure 4a). Similarly, on average biomass before a trough would have been 37% higher (Figure 4b). The analysis also showed that fluctuations would have occurred even in the absence of temporally changing natural mortality (Figure 4), indicating that productivity is the main driver behind long-scale fluctuations.

4 DISCUSSION

Here, we quantified the role of natural mortality fluctuations on forage fish population dynamics, illustrating a depensatory trend in mortality in periods of population cycles leading to population peaks and troughs. Namely, we found that density-dependent mortality worked in two ways: on average mortality increased before a trough and decreased before a peak, albeit with a stronger signal before trough. The analysis also showed that the peaks and troughs would have occurred if mortality was constant, but in smaller magnitudes (Figure 4). This result does not implicate that mortality will always be depensatory during population cycles, but that depensatory mortality is frequently occurring in forage fish populations. As fishing amplifies crashes in forage fish populations (Beverton, 1990; Essington et al., 2015), the effects of natural and fishing mortality are likely synergistic in contributing to the large-scale fluctuations often observed in forage fish populations. This work contributes to specific case-studies illustrating Allee effects and depensation (Gascoigne & Lipcius, 2004; Wood, 1987), but is the first to demonstrate natural mortality as a widespread depensatory mechanism across multiple marine fish populations.

Our results do not imply that natural mortality is the main mechanism driving forage fish population fluctuations, but rather they magnify the effects caused by changes in residual productivity. There is significant recruitment variation in forage fish populations, some which can be explained by a range of factors such as climate change (Checkley, Alheit, Oozeki, & Roy, 2009; Lindegren et al., 2013), temperature, zooplankton abundance and spawning biomass (Cardinale et al., 2009), or indices of upwelling (Cury & Roy, 1989). Pre-recruitment mortality can also be a factor controlling recruitment but is not explicitly considered in the stocks presented here. Growth changing among years could also influence biomass fluctuations, but it is regularly assumed that high abundance causes density-dependent decrease in growth, and thus a compensatory effect (Lorenzen & Enberg, 2002).

The biomass dynamics used here are output obtained from fitting population models to data. Biomass estimates have their own associated uncertainty, depend on the assumptions in the models used, as well as the quality of the data used to fit the model. The models that estimate biomass dynamics combine a mechanistic population model that describes the underlying biological mechanisms, and a statistical model that confronts the biological model with data. The models are all age-structured, but they differ in the way they estimate recruitment, fishing mortality at age and types of underlying statistical models. The stocks included in this analysis come from a range of different model frameworks (Table S1 for details). The natural mortality estimates used in this study are calculated externally and used as input to the population models rather than being estimated within the models. Recent studies have shown how time-varying natural mortality additionally can be estimated within the models (Fu & Quinn, 2000; Jiao, Smith, Reilly, & Orth, 2012). Swain and Benoît (2015) estimated natural mortality for a range of demersal species within the assessment models and found that their natural mortality had increased prior to crashes and contributed to the continued decline of biomass even after fishing stopped.

As the majority of the temporal mortality estimates used in stock assessments globally are from the SMS model (ICES, 2017; Lewy & Vinther, 2004), the mortality estimates are subject to the assumptions and constraints in that model. The model calculates natural mortality based on a combination of size-based predator prey preferences, prey abundance (target species and other) and predator selectivity (for details Supporting information). Multispecies models often deploy a functional response to promote stability and coexistence (e.g. Christensen & Walters, 2004). While there is not a direct functional response in the SMS model, the total prey abundance is included in the mortality calculation. In cases where a predator only prefers a small suite of prey items, the model formulation can cause depensatory dynamics when the target prey is changing abundance substantially, while predator abundance and all other prey remain constant. This formulation assumes that predator intake is directly correlated with abundance.

While this study quantifies the degree of depensatory mortality, it does not identify the mechanisms that produce it. If predators have a saturating functional response such that per capita feeding is constant over a range of high prey densities, inversely density-dependent predation mortality would ensue. The same effect can occur in fisheries (and thus increases the overall effect), where constant catches while productivity is decreasing amplifies crashes (Essington et al., 2015). However, many piscivores have linear functional responses over broad time and space scales (Essington, Hodgson, & Kitchell, 2000) likely because they have evolved high stomach storage capacity to take advantage of high encounter rates with prey (Armstrong & Schindler, 2011). Another possible explanation is that prey avoidance behaviours such as schooling are diminished at low population sizes. That is, if school sizes are smaller, individuals may be more vulnerable to predators. One possible explanation of why mortality increases when population abundance is decreasing and decreases while abundance is increasing is that predators keep their consumption constant when productivity changes. This causes the fraction of consumed prey over total biomass to change in spite of total consumption of the predator being constant (only the denominator of Equation 4 changes).

Fishing mortality only marginally contributed to population biomass removals in comparison with natural mortality. This result ignored that natural mortality and fishing mortality target different sizes of fish; natural mortality is highest for small individuals and declines when they get larger (Gislason, Daan, Rice, & Pope, 2010; Lorenzen, 1996). Fishing often targets larger individuals that are the reproductive component of the population and, thus, has the potential to have more population-level impact than targeting abundant small individuals. Fish populations frequently have little correlation between reproductive biomass and recruitment (Szuwalski, Vert-Pre, Punt, Branch, & Hilborn, 2014), but several forage species also display low recruitment in years of low spawning biomass (Cardinale et al., 2009; Lindegren et al., 2013). Fishing, therefore, has the potential to have a significant impact on populations, even if the total biomass removed is smaller than the removals from mortality.

The forage stocks included in this analysis are all from temperate ecosystems with similar properties. The ecosystems have similar seasonality, are shelf ecosystems and include only a small to moderate amount of species. We expect the results presented here to be applicable to forage stocks from other systems, as forage stocks from different ecosystems often display similar properties; they are naturally fluctuating (Essington et al., 2015), they are important prey items for predators (Pikitch et al., 2014) (and thus susceptible to time-varying natural mortality), and they are often pelagic. Detecting natural mortality changes in (managed) forage fish stocks should be a research priority, as they are important for economies and fisheries in ecosystems worldwide (Alder et al., 2008).

All the stocks considered are exploited stocks and of significant management interest. This study, therefore, underscores the importance of the ecosystem approach to fisheries (Pikitch et al., 2004), which should be implemented on two different time scales (Plagányi et al., 2014): short-term tactical management and longer-term strategic management. Tactical management would require tools that detect increases in mortality while productivity was declining, and accordingly adjust exploitation (via catch restrictions). If management failed to observe large increases in mortality, there would be large risks of population and fisheries crashes (Essington et al., 2015; Siple et al., 2017) with slow or no recovery rates (Swain & Benoît, 2015). Currently, less than ten stocks on a global scale have temporal natural mortality included in their tactical management (Skern-Mauritzen et al., 2016). A possible framework for implementing natural mortality in tactical management is models of intermediate complexity that incorporate the minimum number of components in an ecosystem to capture the most crucial dynamics (Plagányi et al., 2014). Strategic management can use this information to develop harvest strategies that are robust to depensatory mortality. For instance, harvest strategies might be developed to reduced exploitation rates when populations are in decline and increase them when populations are increasing. These types of harvest strategies have been tested in food-web models (Pikitch et al., 2012; Smith et al., 2011) and are used in some forage fish fisheries (e.g., U.S. Pacific coast sardine). At the broader scale, strategic and tactical management is needed to address long-term management goals of exploited ecosystems such as total catch, biodiversity protection, the indirect effects of fishing (e.g., changes in mortality due to fishing predators). How fishing mortality should be distributed on an ecosystem scale to accommodate the indirect effect is still an open question with conflicting proposals: a decrease in forage fish exploitation to accommodate food for predators (Pikitch et al., 2014), an increase in catches of smaller fish due to their high productivity (Garcia et al., 2012; Jacobsen, Gislason, & Andersen, 2014) or the more pragmatic “efficiency frontiers” that choose the best strategy in an ecosystem based on economic or yield efficiency (Jacobsen et al., 2017). We recommend combining strategic and tactical approaches to obtain a better understanding of mechanisms underlying forage fish dynamics to guide management on long and short time scales.

We conclude that natural mortality plays a pivotal role in the dynamics of forage fish. In general, we observed a trend where mortality increased before a trough and decreased before a peak, but with ample variation among stocks. We recommend increased effort in estimating natural mortality of forage fish species to improve predictability of fluctuating populations to guide fisheries management and prevent long-term troughs of forage fish populations.