|Home | About | Journals | Submit | Contact Us | Français|
In 1976 the North Pacific climate shifted, resulting in an average increase of the water temperature. In the Gulf of Alaska the climate shift was followed (i.e. early 1980s) by a gradual but dramatic increase in the abundance of groundfish species that typically prey on pre-recruitment stages of walleye pollock. In the present study we used a previously parameterized model to investigate the effect of these climate and biological changes on the recruitment dynamics of walleye pollock in the Gulf of Alaska. Simulations covered the 1970–2000 time frame and emphasized the medium-to-long temporal scale (i.e. about 5–10 years) of environmental variability. Results showed that during periods characterized by high sea surface temperature and high predation on juvenile pollock stages, recruitment variability and magnitude were below average, and recruitment control was delayed to stages older than the 0-group. Opposite dynamics (i.e. high abundance and variability, and early recruitment control) occurred during periods characterized by low temperature and predation. These results are in general agreement with empirical observations, and allowed us to formulate causal explanations for their occurrence. We interpreted the delay of recruitment control and the reduction of variability as an effect of increased constraint on the abundance of post age-0 stages, in turn imposed by high density dependence and predation mortality. On the other hand, low density-dependence and predation favoured post age-0 survival, and allowed for an unconstrained link between larval and recruitment abundance. Our findings demonstrate that the dominant mechanisms of pollock survival change over contrasting climate regimes. Such changes may in turn cause a phase transition of recruitment dynamics with profound implications for the management of the entire stock.
Phase transitions are low frequency and high amplitude changes of a system attribute in relation to slow changes of forcing variables (Duffy-Anderson et al. in press). Examples of population or community transitions or re-organizations are fairly common in the marine ecology literature, ranging from anchovy/sardine shifts in dominance (Chavez et al. 2003) and changes of cod recruitment (O'Brien et al. 2000), to entire community restructuring (Anderson & Piatt 1999; Cury & Shannon 2004). However, the variability and the shortness of many of these time-series make it difficult to conclude whether the observed changes are occasional events due to stochasticity or actual attractions toward new states of the system (Rothschild & Shannon 2004). The distinction between a stochastic event and a change of state is not just semantics, as the dynamic properties of a system may radically change over contrasting states (Collie et al. 2004).
Scheffer & Carpenter (2003) argued that the positive identification of a system transition cannot be solely based on observational time-series, but it is also necessary to possess a mechanistic understanding of how internal (e.g. density dependence) and external (e.g. environmental forcing) variables affect key properties of the system. Within this context, dynamic models are very useful tools, as they are easily manipulated for the exploration of new system properties (Steele 2004). If the models are based on empirical statistical analysis they become very reflective of the actual dynamics of the system. In the present study we investigate, through simulations of an earlier parameterized model (Ciannelli et al. 2004), changes of the Gulf of Alaska (GOA) walleye pollock (Theragra chalcogramma) recruitment dynamics in relation to environmental variables linked with climate change. Our overall intent is to establish a causal link between recruitment changes and climate variability, a link that may reveal whether pollock recruitment dynamics have undergone phase transitions and that may ultimately improve the management of pollock in the face of incumbent climate change.
This paper refers extensively to the terms ‘phase transition’ and ‘regime shift’—words sometimes being used interchangeably in the ecological literature but with quite different meanings. We adopt the definitions presented in Duffy-Anderson et al. (in press) that are summarized below.
Regime shift: refers to changes in the forcing variables of a system. In mathematical terms a change of regime corresponds to a change of the model parameters. Unlike phases, regimes refer to a homogeneous set of ‘controlling variables’. Climate, for example, represents a set of controlling variables acting on fish populations and entire biological communities, and thus a climate change is a ‘regime shift’ when referring to population and community dynamics.
Phase transition: describes the change of dynamic attributes and mechanistic properties of a system, potentially caused by a regime shift. The mechanistic properties of a system indicate how its primary attributes react to external (i.e. environmental) or internal (i.e. density-dependence) forcing. Examples of system attributes include diversity in a community or recruitment in a population. From these definitions, it follows that phase transition refers to a homogeneous set of ‘observational patterns’. Examples include: recruitment phase transitions, community phase transitions, etc.
We realize that our definitions of ‘phase’ and ‘regime’ may differ from the more common usage in marine ecology. For example, ‘regime shift’ is typically used to describe large scale and abrupt changes of one or more ecosystem attributes in relation to climate change (e.g. Beamish et al. 1999; Beaugrand 2004). We believe that this usage generates confusion as to what has actually changed in a system, since it makes no clear distinction between changes of forcing (e.g. climate) and forced (e.g. recruitment) variables. In our proposed terminology we attempt to reconcile these two facets referring to existing definitions from other scientific fields (Duffy-Anderson et al. in press).
Walleye pollock currently constitutes the second largest single species fishery in the world (FAO 2002). Most of the landings come from the Bering sea, but a large pollock stock, at present estimated at about 200000 metric tonnes (spawning biomass), is also present in the GOA (Dorn et al. 2002). Historically, the bulk of GOA pollock spawning has occurred in early April in Shelikof Strait, between Kodiak Island and the Alaska peninsula (Kendall et al. 1996; figure 1). During their maturation, eggs and larvae are advected southwestward by the Alaska coastal current. The egg stage lasts approximately two weeks, after which non-feeding yolk-sac larvae are hatched (duration about one week), followed by feeding larval (duration 40–60 days) and then 0-age juvenile stages. At age 1–2 (1 and 2 years old, respectively) pollock are detected by the acoustic and trawl surveys, and at age 4–5 (4 and 5 years old, respectively) they enter the fishery.
Information on pollock year-class strength is available as number of age-2 recruits and dates back to 1959. Year-class strength is estimated from an age-structured stock assessment model tuned statistically to the commercial harvests and acoustic and bottom trawl surveys (Dorn et al. 2002). In the present study, to ensure the data quality, we focus on the recruitment dynamics during the time frame 1970–2000 (figure 2). Early estimates may be biased due to inaccuracy of fishing and natural mortality rates during the start of the commercial fishery, non-random sampling of the commercial catch during the early years of the pollock fishery (K. Bailey, personal observation), and the use of a common age-length key over aggregated years (Megrey et al. 1995). The most recent estimates may also be unreliable because the cohort has been observed only a few times in the catches. From here, we refer to the recruitment time-series estimated from the stock assessment model as ‘observed’ to differentiate from the recruitment values estimated by our model simulations, referred to as ‘simulated’ or ‘predicted’.
In the presence of sudden changes or jumps in time-series, which may be indicative of a phase transition, the frequency distribution of key population variables should be multimodal (Scheffer & Carpenter 2003). In this study, we tested whether the analysed time series of pollock recruitment followed a multimodal distribution using an extended application of the Silverman test (Chan & Tong 2004). To increase the power of the test we adopted the Hall & York (2001) adjustment under the null hypothesis of unimodality (i.e. method 1 in Hall & York). The significance level was set at 5%.
In an earlier study (Ciannelli et al. 2004) we explored a variety of model formulations to reproduce pollock recruitment in the GOA. Using non-parametric regression techniques (i.e. generalized additive models) we modelled recruitment as a series of survival events, going from egg to age-0, from age-0 to age-1 and from age-1 to age-2. The statistically based best models for the egg to the age-0 and for the age-0 to the age-1 stages assumed a threshold non-additive effect of environmental and demographic variables on survival (table 1). In particular, it was found that the intensity of negative density-dependence (i.e. compensation) from the egg to the age-0 stage increased in the presence of high wind speed and low spring seawater temperature. Similarly, from the age-0 to the age-1 stage, compensation increased in the presence of high predator abundance and warm fall (autumn) water temperature (figure 3). The model selection criterion was based on the minimization of the generalized cross-validation (GCV). Briefly, the GCV is a measure of the predictive squared error of the model (Wahba 1990; Green & Silverman 1994; Wood 2000). Low values indicate the best compromise between model complexities (i.e. number of parameters) and fit to the observed data. In the case of parametric models it can be shown that the GCV and the more frequently used Akaike information criteria (AIC) are asymptotically equivalent (Stone 1977). However, the AIC cannot be directly applied to non-parametric models, like the one adopted in this study, because of the use of non-maximum likelihood estimation schemes.
In the current analysis we used the previously developed model to simulate critical pollock recruitment metrics under contrasting environmental scenarios. The simulated metrics include average magnitude, average coefficient of variation (CV=standard deviation/mean) and relative stage of recruitment control. Here, recruitment control is assumed to provide an indication of how early (late) recruitment is set, and is therefore representative of the ‘critical period’ (sensu Hjort 1914) of pollock life history. We inferred recruitment control from the coefficient of determination (R2) between the abundance of the pre-recruitment (i.e. eggs, age-0 and age-1) and the recruitment (i.e. age-2) stages; the higher the R2 between two stages, the earlier the recruitment control.
Pollock recruitment metrics were predicted from a set of pre-specified initial conditions and from the average of multiple model runs. Initial conditions included egg density, sampled from within the available time series, and the environmental variables typically associated with the inspected climate regime (see the section ‘Climate regimes’ for a description of the initial conditions associated with each simulated regime). We first predict age-0 from the initial egg abundance and relevant environmental variables, then age-1 from the predicted age-0 and finally age-2 (the age of recruitment) from the predicted age-1 abundance. The modelling results were the average from 1000 realizations of 15 years each. Each realization was derived by adding random error (sampled from the model residuals) to a single prediction, and by sampling the environmental variables from a normal distribution with mean and standard deviation determined from the observed data.
Climate regimes used to force our pollock recruitment model were characterized by a unique combination of environmental variables that are known to affect pollock early-life survival. Specifically, for the survival from egg to age-0, the environmental variables included mean May wind speed cubed, and mean April–June sea surface temperature (referred to as spring water temperature). For the survival from age-0 to age-1, and from age-1 to age-2, the environmental variables included mean October–November sea surface temperature (referred to as fall water temperature), and groundfish predation intensity on the age-0 and age-1 stages. A more complete description of the variables used to characterize climate regimes can be found in Bailey (2000) and Ciannelli et al. (2004).
For each of the environmental variables we assumed two magnitude and variability levels: low if it was below the median and high otherwise. This type of partition was justified by the fact that climate events in the North Pacific do not appear to have a specific periodicity (Bond et al. 2004), thus complicating the selection of time-explicit environmental phases. Furthermore, a time-explicit selection of the environmental variables would introduce a high degree of dependence between the time series used to calibrate the models and those used in the current simulations.
The distribution parameters (i.e. mean and standard deviation (s.d.)) of all environmental variables, except predation, were estimated from the observed time-series. Low or high mean and s.d. referred to the available time series below or above the median value, respectively. The predation index was fixed at 25.0 when low or 26.0 when high, an interval approximately covering the range of the available time-series. Such selection was justified by a long life cycle (15–20 years) of the species included in the predation metrics, suggesting that this index is unlikely to significantly change within the time frame of a simulated climatic regime, but is more likely to change over different regimes. For each simulated regime, the level of wind speed (i.e. high or low) was always the opposite of the spring temperature level; as in the GOA these two variables are to some degree negatively correlated: high wind induces deeper water mixing and cooling of surface temperature. The distributions of the environmental variables used in the model simulations are shown in figure 4.
The environmental scenarios simulated in our analysis mirrored climate and biological regimes having occurred in the GOA during the last 40 years (table 2). The pre-1976 regime was characterized by low sea surface temperature (SST) and biologically by low abundance of groundfish predators. Reversed conditions applied to the post-1976 regime. The post-1989 regime was a transitional state, characterized by opposite trends of spring and fall SST (former low, latter high; Hare & Mantua 2000) and by high groundfish biomass (Mueter & Norcross 2002). We also defined two additional transitional state which may have occurred after the 1976 regime shift: (i) ‘early-transitional’, characterized by high spring temperature, and low fall temperature and predation, and (ii) ‘late-transitional’, characterized by high spring and fall temperature, and low predation. Finally, we defined a post-1998 regime characterized by low fall and spring temperature, but still high predation. Although still debatable, such a transitional state was presumably indicative of a recent return to pre-1976 conditions of the North Pacific climate (McFarlane et al. 2000; Mantua & Hare 2002; Bond et al. 2004).
The environmental variables, associated with climate regimes, greatly influenced the stage of pollock recruitment control. In general, there was a delay of recruitment control with the increasing of both fall temperature and groundfish predation, typical features of the post-1976 regime (table 2). Increasing only the water temperature (late-transitional) shifted the recruitment control from the age-0 to the age-1 stage, and increasing both fall temperature and predation (post-1976) delayed recruitment control even further, as inferred by the reduction of correlation between the age-1 and age-2 abundances. In contrast to all of the post-1976 regimes, in the pre-1976 regime (including the early-transitional) we estimated a high correlation between the age-0 and the age-2 stages (table 2), indicating that recruitment was already set by age-0 juvenile stage, or even earlier (larval stage).
The magnitude of pollock recruitment, as predicted by our model simulation, was highest during the pre-1976 and immediately after the 1976 regime shift (early-transitional), when spring temperature was high and both fall temperature and predation were low. It was lowest during the post-1976 regime when fall temperature and predation were high, and went up again during the post-1998 regime, when spring and fall temperatures were low, but predation was still high (figure 5). A sample of 2000 realizations of simulated pollock recruitment values, with sample probabilities weighted by the approximate regime duration (in years), had a pronounced bimodal distribution. (The weight coefficients for each climate regime were: pre-1976=7, early- and late-transitional =5, post-1976=8, post-1989=3, post-1998=2.)
In general, the recruitment CV decreased with the increasing predation. CV was lowest when both predation and fall temperature were high (i.e. post-1976 and post-1989 regimes) while it was highest when both spring and fall temperatures were high but predation was low (late-transitional; figure 5).
The metrics of pollock recruitment, predicted in the current analysis, were in general agreement with the empirical evidence gathered in the same study region and climate regimes covered by the simulations. With regard to stage of recruitment control, Bailey (2000) reported that during the decade following the 1976 North Pacific regime shift (i.e. our early- and late-transitional regimes), recruitment was already set at the larval stage, while in subsequent years (i.e. post-1976 regimes) it was delayed toward the juvenile stage. Similar patterns were predicted in our modelling analysis (table 2).
With regard to magnitude, the observed time-series of pollock recruitment shows a series of strong year classes during the decade 1970–80 followed by a decline of the average maximum recruitment from 1980 to the late 1990s (figure 2). During the decline there were other strong year classes, but these were not comparable in magnitude to those of the post-1976 decade. Similar patterns were predicted with our model. The management agency also estimated a strong year class in 1999 (Dorn et al. 2002), and so did our model in the post-1998 regime (figure 5). However, due to the partial nature of recruitment into the survey (2001), the actual magnitude of the 1999-year class is uncertain at the time of this writing. Finally, the observed time series of pollock recruitment was clearly multimodal up to the year 1995, while the applied test marginally failed to reject the null hypothesis of unimodality over the entire time series (figure 2).
Our simulation analysis shows that three critical metrics of pollock recruitment dynamics (i.e. magnitude, variability and control) can markedly change over contrasting climate regimes in a way that conforms to empirical observations. Bailey (2000) showed that during the mid- and late-1980s there was an increase of juvenile pollock mortality due to a gradual build-up of groundfish predators. Ciannelli et al. (2004) illustrated that density-dependent mortality is high when temperature and predation are also high. Here we show that the increase of fall temperature and predation, and relative increase of density-dependent survival, imposes a constraint on the abundance of the age-1 and age-2 stages (as shown by a decrease of the recruitment magnitude and CV). We also show that the constraint on the abundance of late juvenile stages can overcome the effect of the dynamics occurring during earlier stages, thereby delaying the stage of recruitment control. In contrast, during environmental regimes characterized by low water temperature and predation, juvenile pollock stages experience a reduction of density-dependent control, which allows for a more direct link between larval and recruitment dynamics. Consequently, the ‘critical period’ (Hjort 1914) is anticipated and the magnitude of year-class strength and variability are considerably increased.
The model parameters used in our analysis were estimated during an epoch (1975–98, Ciannelli et al. 2004) that is included within the periods covered by the current model simulations. However, the general agreement between the predicted and empirical recruitment patterns occurred independently from the model parameterization. First, the environmental variables used in the current simulations were drawn from statistical distributions, and for that reason were different in value from those used in the model parameterization. Second, and more importantly, with the exception of average recruitment magnitude, the patterns predicted here (i.e. stage of recruitment control and recruitment variability) are independent from the variables that were used to parameterize the model (i.e. abundance). In this respect, the newly simulated patterns should be considered as ‘emergent’ properties of pollock pre-recruitment dynamics. Finally, a fully additive model, parameterized on the same set of environmental variables, was not as efficient in reproducing the empirical observations (unreported simulations), pointing to the fact that it was the model structure (i.e. additive versus non-additive) rather than the model parameters that caused the agreement between data and predictions.
Due to the lack of suitable data, our study does not consider a number of factors that may also affect pollock recruitment. For example, concomitant with the decrease of the pollock stock after the mid-1980s, there has been a decrease in the abundance of the Shelikof Strait spawners, while other secondary spawning aggregations (e.g. Shumagin Islands or in various locations along the shelf edge) have remained relatively stable (Dorn et al. 2002). The decline of spawning in historically rich locations (e.g. Shelokof Strait) may affect pollock recruitment dynamics if the environmental conditions of new or secondary locations are less than ideal.
Competitive interactions with other species may also affect pollock year-class strength in the GOA. Up to the late 1960s there was a large population of Pacific ocean perch (POP, Sebastus alutus) in the GOA, which was decimated by the fishery in the early 1970s. Likewise capelin (Mallotus villosus) were abundant in the late 1960s and early 1970s, but have since nearly disappeared (Anderson & Piatt 1999). Both POP and capelin are zooplankton feeders and their disappearance may have released walleye pollock from competitive pressure. Finally, density-dependent egg production, egg quality (Rjinsdorp 1994) and spawner age-related quality of eggs and larvae (Berkeley et al. 2004) are also factors where improved understanding of mechanisms may lead to more refined models.
A change of recruitment metrics and survival mechanisms may have profound ecological and management implications. For example, recruitment control has long been considered in fisheries oceanography (Houde 1987; Sundby et al. 1989; Bailey & Spring 1992; Leggett & DeBolis 1994; Ellersten et al. 1995; Helle et al. 2000). Current knowledge assumes that recruitment control is partly a species-specific function of the life cycle. Pollock, for example, fall in the category of the ‘cod-like’ group, for which Houde (1987) hypothesized a control during the juvenile stage. Analogously, Myers & Caddigan (1993) suggested that recruitment control of the Atlantic cod (Gadus morhua) was set at the juvenile stage. Species with a short life cycle and high larval growth and mortality (e.g. anchovies) may be controlled at earlier stages, such as the larval stage (Houde 1987). Here, however, we show that environmental variables, related to climate change, may also anticipate or delay the stage of recruitment control.
Our modelling results point to the possibility that pollock recruitment dynamics have undergone a phase transition in the last 30 years. Essentially, we can divide the period 1970–2000 into two recruitment phases: the pre- and post-1976, each characterized by distinct dynamic properties, and recruitment mechanisms. The passage between the two phases may not have happened abruptly (i.e. within a year), but rather through an inter-annual transitional period with dynamic properties representative of both the pre- and post-1976 phases. In the pre-1976 climate regime, recruitment magnitude and variability were both high and occasionally low year-classes may have simply occurred due to chance. Within this regime pollock recruitment was already set at the age-0 stage, indicating that larval dynamics were central in setting the year-class strength. In the post-1976 climate regimes, pollock recruitment magnitude and variability were low and recruitment control was delayed to the juvenile stage. Predation by groundfish became an important forcing variable, as did density dependence during the juvenile stages.
Our finding of a phase transition of pollock recruitment dynamics is in agreement with ecological theory and empirical observations of other systems. For example, Collie et al. (2004) illustrated that different stock-recruitment functions over different environmental regimes may, in a changing environment, give rise to abrupt changes of demographic variable (e.g. population size, recruitment). The stock-recruitment function of pollock changes over contrasting environmental regimes (Ciannelli et al. 2004) and here, in line with the observations of Collie et al. (2004), we demonstrate that the dynamic implications of these compensatory changes may give rise to recruitment phase transitions. The existence of a bimodal distribution of recruitment magnitude is also indicative of sudden changes of system properties, and therefore a phase transition (Scheffer & Carpenter 2003). Our simulated recruitment values clearly follow a bimodal distribution, while the observed time series of pollock recruitment supports a bimodal distribution up to the year 1995, but not afterwards (figure 2). However, the latest estimates of pollock recruitment are the least reliable and therefore noisier, as they are based on fewer observations of year class strength. Moreover, the latest years may correspond to a transitional phase toward a pre-1976 regime. Thus, we believe that there is support for a bimodal distribution also in the observed values of pollock year-class strength.
In the presence of phase transition driven by climate regime shifts, it becomes important to alter the reference and target metrics typically adopted in fisheries management (e.g. Hilborn & Walters 1992). For example, during periods characterized by early recruitment control, there can be a more sensitive link between the harvest and the strength of the ensuing year-class (through spawner-recruit functions). Furthermore, as we show in our analysis, recruitment becomes more responsive to short-scale environmental fluctuations and consequently more variable, when controlled during early-life stages. Lastly, a better understanding of the mechanisms regulating the stage of recruitment control can be used for monitoring purposes as the first stage that is significantly correlated with recruitment abundance also provides the first reliable estimate of the year-class strength. The emphasis of current fisheries management practices on the forecasting of the population abundance from one year to the next obscures the significance of longer time-scales, typically related to climate regimes. Hence, there is a need to formulate a new theoretical framework in fisheries science that balances short-term biocomplexity with adaptive management over medium to long time-scales in the presence of changing climate (Steele 1996; Rothschild & Shannon 2004; Duffy-Anderson et al. in press).
Comments provided by Geir Ottersen, Jeff Napp, Janet Duffy-Anderson and two anonymous reviewers improved earlier versions of this paper. LC was funded by the NOAA Steller Sea Lion Program and by the Pollock Conservation Cooperative Research Center. KSC was partially supported by the University of Iowa BSI programme. NCS acknowledges support from the Norwegian Science Council to the EcoClim-project as well as from the University of Oslo. AB acknowledges the support of an NSF Biocomplexity grant (DEB-0083422). This research is contribution FOCI-0493 to NOAA's Fisheries–Oceanography Coordinated Investigations.