A simulation model of PCB dynamics in the Lake Ontario pelagic food web

ELSEVIER

Ecological Modelling 93 (1996) 43-56

A simulation model of PCB dynamics in the Lake Ontario pelagic food web Leland J. Jackson

*

CenterforLimnology, Unirersiv of Wisconsin. 680 North Park Street, Madison. Wisconsin. 53706. USA Received 3 1 May 1995; accepted 16 November 1995

Abstract Salmonid stocking and reduced phosphorous loading have led to major changes in the Lake Ontario ecosystem over the last 25 years and during the last 15-20 years PCB concentrations in age 4 + lake trout have dropped from ca. 9 to 3 mg kg- ‘. Here, I present a simulation model that examines PCB dynamics by coupling a dynamic prey supply to a multi-species functional response model of predation. At 1994 target levels of stocking, the model predicts that chinook salmon will increase, but alewife, lake trout and steelhead will decrease in biomass over the next 10 years. PCB concentrations of the sport fish are predicted to be 1.28, 1.77 and 3.06 mg kg -’ for steelhead, chinook salmon and lake trout, respectively. If alewife growth rates are not reduced farther, the model predicts that alewife could recover from a severe winter mortality. Decreases in the growth rates of sport fishes were accompanied by higher PCB concentrations and lake trout always had higher PCB concentrations for a given age class than chinook salmon and steelhead. Therefore if minimal sport fish PCB concentrations are a principal goal, managers should stock less lake trout, and stock more of the faster growing and less contaminated chinook salmon and steelhead. Keywords:

Ecosystem model; Contaminants: Food web: Lake Ontario; Lake management

1. Introduction

Over the last 25 years decreased phosphorous (P) loading and stocking of exotic Pacific salmonids and native lake trout have led to major changes in the Lake Ontario pelagic food web. For example, total P loading to Lake Ontario decreased by about 80% between 1970 and the late 1980’s (International Joint Commission, 1987), and annual stocking of sport

* Present address: University of Illinois at Urbana-Champaign. Department of Natural Resources and Environmental Sciences. S-506 Turner Hall, 1102 South Goodwin Avenue, Urbana, IL 61801, USA. 0304.3800/96/$15,00 Copyright SSDI 0304.3800(95)00210-3

fishes increased from 0 (1968) to over 8 million (1987). The responses of intermediate steps of the pelagic food web were not immediate as no major changes in water clarity (as measured by Secchi disc transparency) or the abundance or composition of the zooplankton community had been detected by 1985 (Johannsson, 1987; Hartig et al., 1991). By 1991 high levels of salmonid stocking and decreasing P loading appeared as lowered zooplankton production and decreased biomass indices of alewife ( Alosa pseudoharengus) and rainbow smelt (Osmerus mordax). the lake’s two major planktivores (Great Lakes Fishery Commission, 1992). Signs of stress in salmonids, such as an increase in

0 1996 Elsevier Science B.V. All rights reserved

44

L.J. Jackson/Ecological

the number of dead or dying salmon in bottom trawl surveys, declines in the size of salmonids returning to spawn and reduced angler harvest rates (Great Lakes Fishery Commission, 1992) had also appeared. Because coho salmon (Oncorhynchus kisutch), chinook salmon (0. rshawytscha), lake trout (Salvelinus namaycush), steelhead ( = rainbow trout; 0. mykiss), and brown trout (Salmo trutta) prey heavily on alewife, an alewife collapse would be disastrous for the billion dollar sport fish industry (Talhelm, 1987). In response to concerns of a potential forage fish collapse, stocking of all salmon and trout in Lake Ontario was reduced from 1991 levels by 12% in 1992, and by an additional 28% in 1993. Target stocking levels for 1994 were set at 54% of 199 1 levels (Great Lakes Fishery Commission, 1994). Salmonid species composition is determined primarily by stocking as successful natural reproduction is low (Jones et al., 1993). Changes in stocking affect forage fish contaminant dynamics because predation influences forage fish size structure and growth rates. Because trophic transfer is the major pathway for organochlorine contaminant accumulation in fishes (Thomann and Connolly, 1984; Rasmussen et al., 1990; Rowan and Rasmussen, 1992; Gobas, 1993; Madenjian and Carpenter, 1993) responses of the forage fish feed back to affect salmonid diets, growth rates, and their contaminant concentrations too. Contaminant concentrations vary widely among species, even when corrected for size and age (Jensen et al., 1982; Masnado, 1987; Thomann, 1989). Such differences are due in part to variation in growth rates and diets (Weininger, 1978; Jude et al., 1987; Madenjian et al., 1994). Further P load reductions or decreases in salmonid stocking could alter growth and PCB dynamics of Lake Ontario fishes. However, the potential effects of such changes remain unexplored. There are currently no PCB data collection programs that survey the entire Lake Ontario pelagic fish community, or include steelhead and chinook salmon, yet synthesis papers (e.g., Spangler et al., 1987; Steedman and Regier, 1987; Stewart and Ibarra, 1991; Jones et al., 1993) have emphasized the importance of managing the Great lakes with an ecosystem perspective. In fact. a lack of whole-system analyses poses a substantial hurdle to reaching an ecosystem management goal (Lewis et al., 1987; Leach et al., 1987). Lake

Modelling

93 (1996) 43-56

Ontario is one case where only a simulation model can perform analyses at spatial and temporal scales relevant to questions at the ecosystem level. The most recent model of the Lake Ontario pelagic food web (Jones et al., 1993) did not reflect changes in sport fish stocking or alewife growth rates that occurred between 1988 and 1993, and did not consider PCB levels in sport fish. Such factors should be considered in an ecosystem evaluation of predatorprey dynamics and the future of Lake Ontario. This paper presents a simulation model designed to evaluate PCB dynamics in the Lake Ontario pelagic food web and to predict sport fish PCB concentrations given current levels of lake productivity and stocking. Because alewife have been subject to massive overwinter mortalities, I also consider the effect of such a mortality event on sport fish populations and their PCB dynamics.

2. The model I constructed a model of the Lake Ontario pelagic fish community based on (1) a synthesis of published data on PCB concentrations and biomasses of fish, and (2) the sustainability of intensively managed populations of large ecosystems model (SIMPLE; Jones et al., 1993). The SIMPLE model summarizes an assessment of carbon, but not PCB, dynamics in major fish populations of Lake Ontario. I modified the SIMPLE model to include PCB dynamics, calibrated it to available data, and then simulated selected scenarios of PCB and biomass dynamics.

/.-4-&--L>* 4,

Lake tr&t

Slimy sculpin

J

Chinook salmon

Alewife

Steelhead

Rainbow smelt

Fig. 1. Conceptual model illustrating the trophic relationships of a simplified Lake Ontario pelagic food web. Arrows represent the flow of carbon and PCBs.

L. J. Jackson /Ecological

My model represents the Lake Ontario pelagic food web with coupled ordinary differential equations for 6 fish species. The trophic interactions and the flow of carbon and PCBs are illustrated in Fig. 1. The model couples a dynamic simulation of prey supply to a multiple-species functional response model of predation (DeAngelis et al., 1975) driven by predator demand calculated from bioenergetics (Kitchell et al., 1977; Stewart et al., 1981; Rand et al., 1993). Spatial and temporal patterns of fish distribution have been included. The ordinary differential equations were solved annually with a fourth order Runge-Kutta algorithm (Press et al., 19921 with a step size of 10. This step size was chosen by comparing predicted deviations of biomass from a nominal run as a function of step size and choosing a small number of steps that produced a minimum deviation. 2.1. Population

dynamics

2.1.1. Preyfish The change in alewife or rainbow smelt biomass (Table la; all model variables summarized in Table 2) was determined as the difference between growth, and losses due to natural mortality and predation by sport fish. Growth was made density dependent using a logistic formulation (Table lb). I assumed that Lake Ontario alewife and smelt were near their carrying capacities (K) during 1990. and set K equal to 250 and 70 kt, respectively (O’Gorman et al., 1991). I calculated r, the intrinsic rate of growth, from biomass and production estimates for 1990 (O’Gorman et al., 1991). Natural mortalities (Table 1c) were obtained from Jones et al. (1993). An asymptotic (type II) functional response (DeAngelis et al., 1975) described predation. This ratio-dependent formulation was chosen because it provides greater stability (Berryman, 1992) and requires fewer parameters than the prey-dependent formulation of Jones et al. (1993). Because maximum consumption (C,,,,,) depends on fish weight and water temperature, I calculated a fish mass and annual water temperature for the average fish for each predator species. I grouped together age-classes of each species that fed on a common resource (e.g., all age-classes of chinook salmon feeding on alewives). I assumed that the lake was at 4°C for 5

Modelling 93 (I 994) 43-56

45

months, 8°C for 3 months, and during summer stratification adult lake trout, chinook salmon and steelhead would locate at depths corresponding to 10°C (Stewart, 19801, 11°C (Stewart and Ibarra, 1991) and 15°C (Rand et al., 1993). respectively. I adjusted sport fish biomass to fit production that led to measured biomasses from 1971 to 1990. I used the habitat overlap matrix of Jones et al. (1993) derived from knowledge of the life history, ecology and the seasonal patterns of distribution and habitat preference of the pelagic fish in Lake Ontario (Christie et al., 1987; O’Gorman et al., 1987; O’Gorman et al., 199 1) as the selectivity coefficients (Vanderploeg and Scavia, 1979) of the functional response. These values are consistent with predator and prey overlap determined hydroacoustically during 199 1 and 1992 (Goyke and Brandt, 1993). Habitat overlap values were assumed to be independent of age for the predators (Jones et al., 1993) and I used values that were typical of the overlap between predators and adult prey (Table 3). Recruitment of yearling alewife and rainbow smelt was calculated with a Shepherd (1982) style stockrecruitment function (Jones et al., 1993). Young-ofyear (YOY) alewife and rainbow smelt were determined by back-calculating their abundance from yearling equivalents with an assumed YOY-yearling mortality of 99% for alewife (Jones et al., 1993) and 69% for smelt (Lantry and Stewart, 19931. For the initial year of the simulation YOY biomass and numbers were included as input data. Slimy sculpins are present in the model as fixed numbers and biomass because insufficient data exist to determine a stock-recruitment function, and bottom trawl catches are not thought to be reliable indices of abundance due to low sampling effectiveness (Jones et al., 1993). Slimy sculpins are important prey for young lake trout only (Elrod and O’Gorman, 1991). Inclusion of sculpins in the model provides a fixed pool of alternative forage for the lake trout, and I used biomass estimates of Jones et al. (1993) that were chosen to produce the expected contribution of sculpin to the lake trout diets. 2.12. Sport fish Sport fish recruitment is primarily from stocking (Jones et al., 1993). Stocking records from the New York Department of Environmental Conservation and

46

L.J. Jackson/

Table 1 Equations

and parameters

Ecological Modelling 93 (1996) 43-56

of the model

(a) Basic differential equations for the slate uariables Forage fish (prey) populations v, - m, F,, i = 1,2,3 Sport fish (predator)

!$&_

populations Z,(m, + H, + L,), j = 1.2.3

(b) Rate equarions of the model

P,=r,(l-g Prey loss to predator

‘j’

Cmar,Z,F,S,, v, = 5 + &St,) i= I Predator growth due to predation

on prey ‘i

4,Cmm,ZjFi’t, Y, = Z,+

?(F,S,,) ,= I Prey recruitment %A,

R, =

I + (G, A,)P’ Predator recruitment R, = NR, + (SF&l - PSM,) Predator harvest due to sport fishing H, = i B,~.sFM,FE, c=o Predator loss due to sea lamprey attacks d L, = c B,LM, C=O

Predator gain/loss

of PCBs

PCB, = &FC,[PCB],) i=

- m,[PCB],

I

Cc) Additional model parameter Parameter

(symbol)

dues.

Model variable units are dejined in Table 2

Prey Alewife

Age-classes (c) 6 Intrinsic growth rate (r) 1.46 Carrying capacity ( K 1 250.0 Natural mortality (m) 0.04 Natural reproduction (NR) Post-stocking mortality (PSM) Weighted annual temperature CT)

Predators Rainbow smelt

Chinook salmon

Rainbow trout (Steelhead)

Lake trout

6 1.15 70.0 0.2

5

6

11

0.2 54 000 0.65 8.0

0.25 39000 0.80 10.0

0.15 0 0.60 6.0

L.J. Jackson/ Ecological Modelling 93 (1996) 43-56 Table

1 (continued)

Proportion

vulnerable

to the sport fishery at age (V) 0.7 1 1 1

I

2 3 4

0 0.3 1 1 1

>4 Fitted parameters

4.05 0.01973

4.228 0.4361

G

0.00001982

0.0003 1767

Table 2 Parameters

used in the Lake Ontario simulation

model

Symbol or variable

Description

A B

Biomass of spawning adults (prey) Biomass of an age-class Age-class indices Catchability of an age class Maximum consumption Biomass of forage species Predator species specific prey consumption Fishing effort Fitted parameter of the Shepherd model Harvest rate due to angling Prey fish indices Sport fish indices Prey fish carrying capacity Annual loss rate due to lamprey Mortality rate due to lamprey Instantaneous natural mortality rate Natural reproduction (sport fish) Total PCB pool PCB concentration Post-stocking mortality Prey fish intrinsic rate of growth Yearling abundance Selectivity of sport fish for forage fish Numbers of stocked fish Sport fishing mortality Temperature Age-specific vulnerability Index for years Sport fish biomass Fitted parameters of the Shepherd model Growth rate (prey species) Sport fish growth Prey consumption Gross conversion efficiency Sport fish PCB assimilation efficiency

j, 1 K L LM m NR PCB [PCBI PSM r R

s SF SFM T V .v Z @%P CL Y ” @ i

0 0 0 0 1

of the Shepherd model

;

c. d c c m.lx F FC FE G H i, k

41

Reference

Units ton kg none none ktkt-’ y-’ kt kt y-’ none ton -1 Y none none kt -I Y -1 Y -1 Y millions y -

1

1 1 1

2 3

’

kg mg kg-’ none kt y-’ millions y _ ’ none millions y - ’ none degrees Celcius none none kt tonnes, none m/ Y ktkt-’ y-’ ktkt-’ y-’ none none

1 1

1 2 1 1,4 1 1

1 2

5 6, 7

Refs.: (1) Jones et al., 1993; (2) Great Lakes Fishery Commission, 1992; (3) Koonce et al., 1993; (4) Great Lakes Fishery 1994; (5) Hewett and Johnson, 1992: (6) Madenjian et al., 1993: (7) Stow and Carpenter, 1994.

Commission,

48

L.J. Jmkson / Ecologicul Modelling 93 (1996) 43-56

Table 3 Assumed selectivity coefficients for stocked sport fish and their prey in Lake Ontario, derived from knowledge of spatial and temporal overlap between species (from Jones et al.. 1993 and Goyke and Brandt, 1993) Predator

Prey

Selectivity coefficient

Chinook salmon

Alewife Rainbow smelt

0.8 0.6

Lake trout

Alewife Rainbow smelt Slimy sculpin

0.7 I .o 0.7

Steelhead

Alewife Rainbow smelt

0.8 0.6

(S;;)

the Ontario Ministry of Natural Resources were used to describe stocking from 1971 to 1990. Stocking summaries from the Great Lakes Fisheries Commission 1994 annual report were used to describe stocking from 1991 to 1993, and 1994 stocking target levels were used as the baseline stocking for 1994 to 2005. Estimates of natural reproduction, based on the incidence of non-fin-clipped fish in creel surveys, were added to the stocking record to determine wild annual recruitment (Jones et al., 1993). Changes in sport fish biomass were the net balance between growth due to predation on forage fish, and losses due to natural mortality, harvest by anglers, and lamprey (Perromyzon marinus) attacks. Age classes of each species that fed on a common resource (e.g., all chinook salmon feeding on alewives) were aggregated for the functional response. Prey consumption was corrected for the predator’s gross growth efficiency to determine predator growth. Gross growth efficiencies were calculated from bioenergetic subroutines (Hewett and Johnson, 1992). Natural mortality rates of post-yearling sport fish were taken from Jones et al. (1993). After each year the aggregated population biomass for each species was allocated to next year’s postyearling age classes. (The adults (forage fish) or stocking plus natural recruitment (sport fish) determined next year’s yearling biomass.) The allocated biomass (Fig. 2) was a function of the biomass and maximum growth potential of each age class (Jones et al., 1993) as a fraction of the biomass and growth

potential summed across all post-yearling age classes. Changes in age-class proportions over time would therefore be a function of recruitment variability and population growth or decline. In addition to natural mortality sport fish experience mortality from angler harvest and sea lamprey attacks. I calculated age-specific angler harvest rates as the product of estimates of catchability, vulnerability and sport fishing effort, that Jones et al. (1993) determined from creel survey estimates. I also averaged, over 10 year periods (e.g., 1971-1979. 19801989, etc.) age-class specific mortality rates from the 1980 IMSL sea lamprey model for Lake Ontario (Koonce et al., 1993). Losses due to sport fishing and sea lamprey attacks were applied to the biomass allocated to the following year’s age classes. This reallocation procedure allowed application of age specific mortalities consistent with creel survey estimates.

,y,

Calculate next year’s yearling equivalents

I Aggregate population variables (except YOY’S) I Calculate population parameters (e.g.. average weight, total PCBs etc.) Runge-Kutta Algorithm

Ratio-dependent Functional Response

i Determine changes in population biomass

Reallocate biomass to next year’s *I

I Final population statistics (biomass, PCB concentration etc.) Fig. 2. Flow diagram for a simulation model of carbon and PC3 flow in a simplified Lake Ontario pelagic food web.

L.J. Jacbon/Ecologicrrl

I did not include coho salmon and brown trout in my model, yet did not want to completely ignore their impact on forage fish stocks. Estimated forage fish consumption by these piscivores for 1990 (Great Lakes Fishery Commission, 1992) was assumed to represent their long-term average, was fixed at 7.5 kt annually from 1971 to 1990, and alewife and smelt consumed was proportional to their biomass for a given year. Following 1994, I reduced the brown trout and coho salmon consumption by 46% to reflect changes in combined stocking of these two species from 1 563 050 to 850000 individuals between 1991 and 1993 (Great Lakes Fishery Commission, 1994). 2.2. PCB dynamics Exceedingly detailed mechanistic models of PCB accumulation in fish exist (e.g., Barber et al., 1991) but provide no better fits to empirical data than single, aggregated differential equation models (e.g., Thomann and Connolly, 1984; Thomann, 19891. I have considered dietary exposure of PCBs to be the principal route of accumulation, although gill exchange may be important for light weight congeners. Predator consumption summed across prey determined the flow of PCBs from prey to predators. The change in PCB pool of an age-class was the difference between new PCB inputs due to consumption of prey, corrected for net dietary PCB assimilation efficiencies (steelhead: 0.50 (Madenjian et al., 1994); chinook salmon: 0.53 (Stow and Carpenter, 1994) lake trout: 0.75 (Niimi and Oliver, 19831, and PCB losses due to natural mortality, sport fishing and lamprey attacks. PCB loss terms were the product of their respective biomass and PCB concentration. PCB concentration was computed by dividing the resulting PCB pool by the total fish biomass at the end of the year, for each age-class. Total PCB concentrations in 15 g rainbow smelt and 5 g slimy sculpin exhibited no significant change between 1977 and 1988 (Borgmann and Whittle, 1992). I used the average concentration from this period as smelt (1 .Ol mg kg- ’) and sculpin (1.16 mg kg- ’) concentrations for the model. There are no data that illustrate how PCB concentrations in alewife may have changed prior to 1991, or how concentrations vary with size for any of the Lake Ontario prey

Modelling 93 (1996) 43-56

49

fish. Total PCB concentrations in composite alewife samples collected near Grimsby and Cobourg, Ontario, during 1991 and 1992 (D.M. Whittle, Fisheries and Oceans Canada, Burlington, Ontario, unpublished data) failed to reveal a significant trend between PCB concentration and alewife size (mean value 0.43 mg kg ’ for 9 samples of 5 similarly sized fish). Therefore, I assumed that the total PCB concentrations of alewife, smelt and sculpin were invariant with size. There are no data for chinook salmon or steelhead PCB concentrations at the beginning of the simulation. Concentrations for 1971 were estimated by assuming that both species had similar diets to the lake trout and that their body burdens would differ by a factor proportional to the differences in their PCB assimilation efficiencies. Therefore, I gave comparatively aged steelhead and chinook salmon 67% and 71%, respectively, of the lake trout’s PCB concentration. These values were not estimated based on relative total lipid content as Borgmann and Whittle (199 1) demonstrated empirically (for lake trout at least) that there is no advantage to lipid correcting contaminant concentrations for those contaminants that are accumulated primarily through dietary exposure or for which bioenergetics control final concentrations.

3. Results 3. I. Model calibration I first calibrated the model with adult alewife biomass estimates from spring bottom trawl surveys conducted between 1978 and 1992. Whole-lake alewife biomass estimates (Fig. 3) were calculated from data from New York waters by dividing the

Fig. 3. Measured (estimated from spring bottom trawls) and model predictions of adult alewife biomass in Lake Ontario, 1978-1993. The model assumes a 50% winter mortality of alewife in 1976.

50

L.J. Jackson/Ecological

Mode&g

Table 4 Comparison of SIMPLE and this model predicted stocked sport fish biomass (kt) for various years for the simulation period. Both scenarios assume stocking levels from 1991 onward at SIMPLE levels Species

Year

This model

SIMPLE

Chinook salmon

1971 1980 1990 2005

0.001 0.52 4.45 4.36

0.00 1 0.52 4.45 4.37

Lake trout

1971 1980 1990 2005

0 0.52 1.72 1.76

0 0.54

1971 1980 1990 2005

0.016 0.42 1.16 1.66

0.016 0.42 1.19 1.67

Steelhead

1.72 1.75

lake into depth strata (O’Gorman and Schneider, 1986) calculating the total lake area in each strata, and multiplying by the alewife density measured in each corresponding strata (Jones et al., 1993). The general pattern of alewife biomass is captured by the model (Fig. 3). The model does a good job at predicting the recovery of alewife after the 1976 die-off, but does not predict the magnitude of the recovery nor the ca. 2 year cycle in alewife biomass between 198 1 and 1992. The predicted biomass for the sport fish species from my model is in good

Table 5 Model output of biomass target stocking levels

and PCB concentrations

93 f 1996) 43-56

agreement with that of the SIMPLE model (Table 4) from the initial year (1971) through the end of the simulation (2005) given identical stocking levels. The maximum consumption estimates calculated by bioenergetics for the sport fish populations based on the average weight fish had to be reduced to have my model predicted growth resemble the SIMPLE model predicted growth. Once my model had been calibrated to the SIMPLE model, I decreased alewife growth rates as a surrogate for declining energy densities, and adjusted sport fish stocking to reflect changes that occurred in Lake Ontario between the late 1970’s and early 1990’s. Such changes lead to a lower alewife biomass (ca. 100 kt versus 119 kt at 2005) and also change the predicted sport fish biomass (compare the biomasses at 2005 in Table 5A with Table 4). The increased in chinook salmon biomass results from target stocking levels of 1.45 X lo6 compared to 9.65 X lo5 individuals annually, while the lower lake trout biomass results from roughly half the numbers stocked annually (1.02 X lo6 compared to 2.03 X 106). Steelhead are roughly the same for both the SIMPLE model and 1994 GLFC targets (9.43 X lo5 vs. 8.09 x 105). The only published time series describing PCB concentrations in Lake Ontario sport fish is that of Borgmann and Whittle (1991) for lake trout (Fig. 4A). They showed that age 4 + lake trout dropped in concentration from ca. 8.5 mg kg-’ to ca. 3.3 mg

for Lake Ontario sport fish by year 2005 given early 1990’s lake productivity

(A) Baseline conditions Species

Biomass (kt)

Alewife Chinook Lake trout Steelhead

100.81 5.07 1.07 1.75

PCB (mg kg-‘) 0.43 1.77 3.06 1.28

(Bi 50% Alewife winter mortality Species Alewife Chinook salmon Lake trout Steelhead (A) Baseline conditions.

Biomass (kt) 99.12 5.06 1.07 1.75

% change from baseline

PCB (mg kg-‘)

% change from baseline

- 1.7 -0.2 0.0 0.0

1.77 3.07 1.28

0.0 -0.3 0.0

(B) As ‘A’ but with a severe alewife winter mortality

imposed in 1994.

and 1994

L.J. Jackson/Ecological

Modelling 93 (I 996143-56

51

kg-’ between 1977 and 1988, but also experienced increased concentrations during 1982- 1984. My model captures the general pattern (Fig. 4A), but does not reproduce the 1982-1984 increase. I assumed that the PCB concentration of 4 year old lake trout was highest in 1975 and declined thereafter, and calculated a line of best fit to the lake trout PCB time series. I then calculated the decrease in alewife PCB concentration (Fig. 4B) necessary to lead to the measured lake trout concentration decline. If diet PCB concentration was the only factor leading to the observed lake trout PCB decline (Fig. 4A), then alewife PCB concentrations would have to have dropped from ca. 3 mg kg-’ in 1971 to the 1991/2 measured value of 0.43 mg kg-‘(Fig. 4B). Fig. 4. Concentrations of PCBs in Measured PCBs in age 4+ lake 1991) and concentrations predicted in PCB concentration of alewives served lake trout PCB concentration sition and PCB concentration were trout PCB concentration decline.

Table 6 Ranks of model sensitivity

Lake Ontario pelagic fish. (A) trout (Borgmann and Whittle, by the model. (B) The change necessary to lead to the obdecline assuming diet composolely responsible for the lake

to individual

Parameter

parameter

3.2. Sensitiuity

analysis

The model’s parameters can be organized into four groups: (1) initial conditions: biomass, numbers

perturbation

Response variables Biomass Alewife

Population parameters

PCB concentration Smelt

Lake trout

Steelhead

Chinook salmon

Lake trout

Steelhead

(15)

Chinook Post-stocking mortality Lake trout post-stocking mortality Steelhead post-stocking mortality Functional response parameters

3 3 1 (13)

1

Chinook annual temperature Lake trout annual temperature Steelhead annual temperature Chinook maximum consumption Lake trout maximum consumption Steelhead maximum consumption Bioemyqetics

Chinook salmon

parameters

2 2 3

1

1

2

1 1 2

(31

Chinook PCB assimilation efficiency Lake trout PCB assimilation efficiency Steelhead PCB assimilation efficiency

3

I 1

Parameters ranked 1 elicited the greatest change in the response variable. Only those parameters to which the model was sensitive ranked. The total number of parameters tested for each parameter group is indicated in parentheses.

were

52

L.J. Jackson/Ecological

and PCB concentrations of various age classes; (2) population parameters: growth rate (prey fish), natural mortality, post-stocking mortality, sport fishing and lamprey mortality (predators); (3) functional response parameters: annual temperature, maximum consumption and predator selectivities for prey and, (4) PCB assimilation efficiencies. To explore the sensitivity of model parameters and initial conditions I varied parameters, one at a time, and compared the resulting changes in response variables. Response variables were biomass of all species except sculpin and PCB concentrations of age 4 + chinook salmon, lake trout and steelhead. The sensitivity analysis was performed on the model calibrated as described above, and run from 197 1 to 1990. Each coefficient value was changed by f 10% and the model was considered to be sensitive to a coefficient if any response variable changed > 10% in response to this increase in the coefficient value (Table 61. The model always recovered from perturbations of initial conditions by simulation year 1990. The model was most sensitive to functional response parameters and PCB assimilation efficiencies, and less sensitive to population parameters. Variables affecting chinook salmon growth (temperature and maximum consumption) were also important to alewife and smelt biomass, as chinook salmon was responsible for the largest amount of alewife and smelt consumption during most of the stocking period until 1990. None of the response variables were sensitive to selectivities of predators for prey used in the functional response equations.

YEAR Fig. 5. Predicted biomass of alewives in Lake Ontario under various assumptions: (A) Lake productivity and stocking rates from the SIMPLE model (circles); (B) 1994 lake productivity and GLFC target stocking rates (squares); (C) ‘B’ with a severe winter mortality event in 1994 (triangles).

Modeiling 93 (1996) 43-56

Fig. 6. Comparison of age 4+ lake trout PCB concentrations given two scenarios of alewife growth. (A) 1994 lake productivity and GLFC target stocking. (B) As for A but with a 50% alewife winter mortality in 1994.

3.3. A severe alewife winter mortality To simulate a severe alewife overwinter mortality event I applied a one time 50% mortality to all alewife age classes during 1994 (Fig. 50. Alewife were predicted to recover from this mortality event and had returned to the same population biomass as the baseline conditions by 2005 (Table 5B). The effect on piscivore PCB concentrations is exemplified by age class 4 + lake trout whose growth rates declined following the alewife mortality; their PCB concentrations increased concomitant with their decrease in growth rates (Fig. 6).

4. Discussion 4. I. Model evaluation My model and SIMPLE predict similar biomasses for alewife and their stocked predators for similar stocking levels (Table 4). Quantitatively similar trends may be expected given the degree of overlap in the parameters and initial conditions of the models. However, the similarity of model predictions demonstrates that the functional response I used can be parameterized to provide quantitatively similar population dynamics. My functional response has fewer parameters than that of the SIMPLE model and is therefore a more parsimonious approach to estimating the predator-prey interactions. The ability of a parsimonious approach to provide an adequate model (Ludwig and Walters, 1985) of the predatorprey interactions should not be too surprising given

L.J. Jackso~~ / Ecological

the available data. The type of functional response I used allows interference competition between predators (DeAngelis et al.. 197.51, and selectivity by predators from multiple prey categories (O’Neill, 1969, 1976; Bartell et al., 1988; DeAngelis et al., 1989). Maximum consumption predicted from bioenergetics had to be reduced by about 60-70% to have predicted growth of the predators match observed biomasses of alewife and predicted predator biomass by SIMPLE. This indicates that stocked salmonids are growing at rates below their physiological maxima. The necessary reduction was greater for chinook salmon and steelhead than for lake trout, and is consistent with much lower intrinsic growth rates for the lake trout. The correction applied to the C max estimate is also an adjustment for the assumption that consumption estimated from the average temperature and fish mass is indicative of the annual average for each population. The predicted decrease in alewife PCB concentration (Fig. 3B) does not account for the increase in age 4 + lake trout PCB concentration from 19821984 (Borgmann and Whittle, 19911. Slimy sculpin and rainbow smelt PCB concentrations remained unchanged from 1977 to 1988 (Borgmann and Whittle. 19921 possibly because they consume primarily a benthic diet and PCBs in sediment pools may have remained nearly constant through the measured time period. Therefore, it seems likely that a change in alewife PCB concentration (or a change in sport fish diet composition) must have occurred to account for the observed increase in lake trout PCB concentration during the early 1980’s. A change in alewife PCB concentration might have been possible if alewife growth and survival increased at this time. Unfortunately, alewife were not included in contaminant survey programs between 1977 and 1988 and my model does not include an increased PCB concentration with alewife body mass. Therefore, the model cannot adequately address the potential role that size selective predation may have played in the increased PCB concentration of the lake trout. Size selective predation has been shown to be important in Lake Michigan (and presumably Lake Ontario) as salmonids prefer the largest alewife available (Stewart and Ibarra, 1991). The bottom line is that my model fits the limited available data and the SIMPLE model reasonably

Modelling

93 (1996) 43-56

53

well. However, it is important to recognize that the model includes information available subsequent to the development of SIMPLE (e.g., decreased growth rates of age 1 + alewives and actual numbers of sport fish stocked from 199 l- 1993 plus revised 1994 target levels) in addition to considering PCB dynamics, and thus evaluates the future of the Lake Ontario pelagic food web under a different set of conditions than those of the SIMPLE model. Lower alewife growth rates and higher stocking of chinook salmon have important implications to the dynamics of the food web to 2005. For example, stocking more chinook salmon effectively increases the total demand on the forage base, which is reflected in the lower alewife biomass. This result emphasizes the interaction between alewife productivity and salmonid stocking on alewife population dynamics. Alewife population dynamics are currently the key to dynamics of the other fish populations in the Lake Ontario pelagic food web. Growth of age 1 + alewife has been poor since 1986 and was the lowest measured in 1992 (R. O’Gorman, National Biological Survey, Oswego, NY, personal communication). The average weight of alewives of age class 2 + and older dropped from 41 g in 1978 to 19 g in 1989 and was accompanied by a 25% reduction in energy density between 1979 and 1985 (Rand et al., 1994). The ability of the alewife population in Lake Ontario to sustain itself has been the subject of considerable concern over the last few years (Jones et al., 19931. The model predicts that the decline in alewife growth rates and energy densities, coupled with 1994 target salmonid stocking levels, may still result in a considerable challenge ahead for the long-term management and sustainability of Lake Ontario, despite the declines in stocking that have occurred since 1990. The predicted biomass of alewife in 2005 is 100 kt, a level that is near the top of a 70-90 kt range that Jones et al. (1993) suggest is a critical crossover point below which alewife are unlikely to be able to recruit sufficient numbers to supply 1990 stocking levels of salmonids. This modelling analysis predicts that by 2005, adult alewife biomass will be approaching a level below which recruitment may not be sufficient to provide a sustaining population of alewife (Jones et al., 1993). Alewife and chinook salmon are clearly the key species to long term management of the Lake Ontario pelagic food web

54

L.J. Jackson/

because these species represent the majority forage supply and predator demand. 4.2. Directions for future

Ecological

of the

research and monitoring

Alewife have not been included in Lake Ontario contaminant surveys until recently despite the dependency of stocked sport fish on the alewife. Future monitoring clearly needs to measure the PCB concentrations of all age classes of alewife. Future modelling could then derive a bioenergetics model of PCB accumulation for the alewife as has been done for Lake Michigan (Jackson and Carpenter, 1995). and then explore the potential interaction between alewife growth and PCB accumulation, and size selective predation by stocked sport fish in determining contaminant concentrations in the sport fish. Such a model would allow a better, and more critical examination of the feed back between population dynamics, and the potential importance of those dynamics to PCB flow between the species considered. Because alewife play a key role in the dynamics of the Lake Ontario pelagic food web, a better understanding of the factors driving their dynamics would aid managers in predicting alewife population dynamics. Population dynamic models, coupled with validated PCB accumulation models would provide fisheries managers with tools that would allow them to evaluate the degree to which PCBs are a problem in Lake Ontario, and prioritize their management actions accordingly.

Acknowledgements I am extremely grateful to M.L. Jones (Ontario Ministry of Natural Resources) for providing SIMPLE model input data and D.M. Whittle (Fisheries and Oceans Canada) for providing unpublished data on PCB concentrations of L. Ontario alewife. R. O’Gorman (US Fish and Wildlife) provided unpublished data on alewife growth rates and condition indices. I thank M.L. Jones and D.M. Whittle for engaging in numerous discussions on L. Ontario population and PCB dynamics. B. Ropers-Huilman provided invaluable assistance with the programming. L. Eby, S.R. Carpenter, and A. Trebitz provided advice and comments on model structure and

Modelling

93 (1996)

43-56

procedures. I am most grateful to M.C. Barber, S.R. Carpenter, T. Legovic P.S. Rand, D.E. Schindler and A. Trebitz for improving this manuscript through their constructive reviews and comments. This project was funded primarily by a Natural Sciences and Engineering Research Council of Canada post-doctoral fellowship, and also by the University of Wisconsin Sea Grant Institute under grants from the National Sea Grant College Program, National Oceanic and Atmospheric Administration, US Department of Commerce, and from the State of Wisconsin. Federal grant NA90AA-D-SG469, project R/MW-41.

References Berryman, A.A.. 1992. The origins and evolution of predator-prey theory. Ecology 73: 1530-1535. Barber, M.C.. Suarez, L.A. and Lassiter, R.R., 1991. Modelling bioaccumulation of organic pollutants in fish with an application of PCBs in Lake Ontario salmonids. Can. J. Fish. Aquat. Sci. 48: 318-337. Bartell, S.M.. Brenkert. A.L.. O’Neill. R.V. and Gardner, R.H.. 1988. Temporal variation in regulation of production in a pelagic food web model. In: ed. S.R. Carpenter, Complex Interactions in Lake Ecosystems. Springer-Verlag, New York, NY, pp. 101-118. Borgmann, U. and Whittle, D.M., 1991. Contaminant concentration trends in Lake Ontario lake trout (Sahelinus namapuh): 1977 to 1988. J. Great Lakes Res. 17: 368-381. Borgmann, U. and Whittle, D.M., 1992. DDE, PCB, and mercury concentration trends in Lake Ontario rainbow smelt (Osmerus mordax) and slimy sculpin (Cofrus cognatus): 1977 to 1988. J. Great Lakes Res. 18: 298-308. Christie, W.J.. Scott, K.A., Sly, PG. and S&us. R.H.. 1987. Recent changes in the aquatic food web of eastern Lake Ontario. Can. J. Fish. Aquat. Sci. 44 (Suppl. 2): 37-52. DeAngelis, D.L., Goldstein, R.A. and O’Neill, R.V.. 1975. A model for trophic interaction. Ecology 56: 881-892. DeAngelis. D.L., Bartell, S.M. and Brenkert, A.L., 1989. Effects of nutrient recycling and food-chain length on resilience. Am. Nat. 134: 778-805. Elrod, J.H. and O’Gorman, R., 1991. Diet of juvenile lake trout in southern Lake Ontario in relation to abundance and size of prey fishes. 1979-1987. Trans. Am. Fish. Sot. 120: 290-302. Gobas. F.A.P.C.. 1993. A model for predicting the bioaccumulation of hydrophobic organic chemicals in aquatic food-webs: application to Lake Ontario. Ecol. Modell. 69: 1-17. Goyke. A.P. and Brandt, S.B., 1993. Spatial models of salmonine growth rates in Lake Ontario. Trans. Am. Fish. Sot. 122: 870-883. Great Lakes Fishery Commission, 1992. Status of the Lake Ontario Offshore Pelagic Fish Community and Related Ecosystem in 1992. 33 pp.

L. J. Jacksorr /Ecological Great Lakes Fishery Commission, 1994. Lake Ontario Fisheries Unit 1993 Annual Report. 592 pp. Hartig. J.H.. Kitchell, J.F., Scavia, D. and Brandt. S.B.. 1991. Rehabilitation of Lake Ontario: the role of nutrient reduction and food web dynamics. Can. J. Fish. Aquat. Sci. 48: 15741580. Hewett, SW. and Johnson. B.L., 1992. Fish Bioenergetics Model 2. UW Sea Grant Technical Report No. WIS-SG-92-250. 79 PP. International Joint Commission, 1987. Great Lakes Water Quality Board, Appendix B - Surveillance. Vol III.. 20 pp. Jackson, L.J. and Carpenter, S.R., 1995. PCB contents of Lake Michigan invertebrates: reconstruction based on PCB contents of alewives and their bioenergetics. J. Great Lakes Res. 21: 112-120. Jensen, A.L.. Spigarelli, S.A. and Thommes. M.M.. 1982. PCB uptake by five species of fish in Lake Michigan, Green Bay, and Cayuga Lake. NY. Can. J. Fish. Aquat. Sci. 39: 700-709. Johannsson, O.E., 1987. Comparison of Lake Ontario zooplankton commumities between 1967 and 1985: before and after the implimentation of salmonid stocking and phosphorus control. J. Great Lakes Res. 13: 328-339. Jones. M.L.. Koonce. J.F. and O’Gorman. R., 1993. Sustainability of hatchery-dependent salmonine fisheries in Lake Ontario: the conflict between predator demand and prey supply. Trans. Am. Fish. Sot. 122: 1002-1018. Jude. D.J., Tesar, F.J.. DeBoe, F.S. and Miller. T.J.. 1987. Diet and selection of major prey species by Lake Michigan salmonines. 1973-1982. Trans. Am. Fish. Sot. 116: 677-691. Kitchell, J.F.. Stewart. D.J. and Weininger, D.. 1977. Applications of a bioenergetics model to yellow perch (ferca &cescens) and walleye (Srizostedion citreum). J. Fish. Res. Board Can. 34: 1922-1935. Koonce. J.F., Eshenroder. R.L. and Christie. G.C.. 1993. An economic injury level approach for establishment of the intensity of sea lamprey control in the Great Lakes. N. Am. J. Fish. Manage. 13: l-14. Lantry. B.F. and Stewart, D.J.. 1993. Ecological energetics of rainbow smelt in the Laurentian Great Lakes: an interlake comparison. Trans. Am. Fish. Sot. 122: 951-976. Leach, J.H.. Dickie, L.M., Shuter, B.J.. Borgmann, U., Hyman. J. and Lysak, W., 1987. A review of methods for prediction of potential fish production and its application to the Great Lakes and Lake Winnipeg. Can. J. Fish. Aquat. Sci. 44 (Suppl. 21: 47 I-485. Lewis, C.A., Schupp, D.H., Taylor, W.W., Collins, J.J. and Hatch. R.W., 1987. Predicting Great Lakes fish yields: tools and constraints. Can. J. Fish. Aquat. Sci. 44 (Suppl. 21: 41 l-416. Ludwig, D. and Walters. C.J.. 1985. Are age-structured models appropriate for catch-effort data? Can. J. Fish. Aquat. Sci. 42: 1066-1072. Madenjian, C.P. and Carpenter, S.R., 1993. Simulation of the effects of time and size at stocking on PCB accumulation in lake trout. Trans. Am. Fish. Sot. 122: 492-499. Madenjian, C.P.. Carpenter, S.R., Eck, G.W. and Miller, M.A., 1993. Accumulation of PCBs by lake trout (Sa/c~e/inus namav-

Mode&g

93 C19961 43-56

55

tush): An individual-based model approach. Can. J. Fish. Aquat. Sci. 50: 97-109. Madenjian, C.P., Carpenter, S.R. and Rand. P.S.. 1994. Why are the PCB concentrations of salmonine individuals from the same lake so highly variable? Can. J. Fish. Aquat. Sci. 51: 800-807. Masnado, R.G.. 1987. Polychlorinated biphenyl concentrations of eight salmonid species from Wisconsin waters of Lake Michigan: 1985. Wisconsin Dept. Nat. Resources Fish Manage. Rep. 132, 55 pp. Niimi, A.J. and Oliver, B.G., 1983. Biological half-lives of polychlorinated biphenyl (PCB) congeners in whole fish and muscle of rainbow trout (S&no gnirdneri). Can. J. Fish. Aquat. Sci. 40: 1388-1394. O’Gorman. R. and Schneider, C.P.. 1986. Dynamics of alewives in Lake Ontario following a mass mortality. Trans. Am. Fish. sot. 115: l-14. O’Gorman, R., Bergstedt. R.A. and Eckert, T.H., 1987. Prey fish dynamics and salmonine predator growth in Lake Ontario, 1978-84. Can. J. Fish. Aquat. Sci. 44 (Suppl. 2): 390-403. O’Gorman. R., Mills, E.L. and DeGisi, J.. 1991. Use of zooplankton to assess the movement and distribution of alewife ( Alosa pseudoharmgus) in south-central Lake Ontario in spring. Can. J. Fish. Aquat. Sci. 48: 2250-2257. O’Neill, R.V., 1969. Indirect estimation of energy fluxes in animal food webs. J. Theor. Biol. 22: 284-290. O’Neill, R.V., 1976. Ecosystem persistence and heterotrophic regulation, Ecology 57: 1244-1253. Press. W.H., Teukolsky, S.A., Vettering, W.T. and Flannery. B.P.. 1992. Numerical Recipes in C: The Art of Scientific Computing. Cambridge University Press, New York, NY, 994 pp. Rand, P.S.. Lantry, B.F., O’Gorman. R., Owens, R.W. and Stewart, D.J., 1994. Energy density and size of pelagic prey fishes in Lake Ontario. 1978- 1990: implications for salmonine energetics. Trans. Am. Fish. Sot. 123: 519-534. Rand. P.S., Stewart, D.J.. Seelbach. P.W., Jones. M.L. and Wedge, L.R., 1993. Modeling steelhead population energetics in Lakes Michigan and Ontario. Trans. Am. Fish. Sot. 122: 977-1001, Rasmussen, J.B., Rowan, D.J. Lean. D.R.S. and Carey, J.H., 1990. Food chain structure in Ontario lakes determines PCB levels in lake trout (Sal~~elinus mma~cush) and other pelagic fish. Can. J. Fish. Aquat. Sci. 47: 2030-3038. Rowan, D.J. and Rasmussen, J.B.. 1992. Why don’t Great Lakes fish reflect environmental concentrations of organic contaminants? An analysis of between-lake variability in the ecological partitioning of PCB’s and DDT. J. Great Lakes Res. 18: 724-741. Shepherd, J.G.. 1982. A versatile new stock-recruitment relationship for fisheries, and the construction of sustainable yield curves. J. Conseil. Conseil Int. pour 1’Exploration de la Mer 40: 67-75. Spangler. G.R.. Loftus. K.H. and Christie, W.J.. 1987. Introduction to the International Symposium on Stock Assessment and Yield Prediction (ASPY). Can. J. Fish. Aquat. Sci. 44 (Suppl. 2): 7-9. Steedman. R.J. and Regier. H.A.. 1987. Ecosystem science for the

56

L.J. Jackson/ Ecological Modelling 93 (1996) 43-56

Great Lakes: perspectives on degradative and rehabilitative transformations. Can. J. Fish. Aquat. Sci. 44 (Suppl. 2): 95-103. Stewart, D.J., 1980. Salmonid Predators and Their Forage Base in Lake Michigan: A Bioenergetics-Modeling Synthesis. PhD thesis, University of Wisconsin, Madison, WI, 225 pp. Stewart, D.J. and Ibarra, M., 1991. Predation and production by salmonine fishes in Lake Michigan, 1978-88. Can. J. Fish. Aquat. Sci. 48: 909-922. Stewart. D.J.. Kitchell, J.F. and Crowder, L.B., 1981. Forage fish and their salmonid predators in Lake Michigan. Trans. Am. Fish. Sot. 110: 751-763. Stow. C.A. and Carpenter, S.R., 1994. PCB accumulation in Lake Michigan coho and chinook salmon: individual-based models using allometric relationships. Environ. Sci. Technol. 28: 1543-1549.

Talhelm, D.R.. 1987. Economics of Great Lakes Fisheries: A 1985 Assessment. Technical Report #54. Great Lakes Fisheries Commission. Ann Arbor, MI. Thomann, R.V.. 1989. Bioaccumulation model of organic chemical distribution in aquatic food chains. Environ. Sci. Technol. 23: 699-707. Thomann, R.V. and Connolly, J.P., 1984. Model of PCB in the Lake Michigan lake trout food chain. Environ. Sci. Technol. 18: 67-71. Vanderploeg, H.A. and Scavia, D.. 1979. The calculation and use of selectivity coefficients of feeding: zooplankton grazing. Ecol. Modell. 7: 135-149. Weininger, D.. 1978. Accumulation of PCBs by Lake Trout in Lake Michigan. PhD. thesis, University of Wisconsin, Madison, Wisconsin, 232 pp.