vol. 167, no. 2
the american naturalist
february 2006
Plugging Space into Predator-Prey Models: An Empirical Approach
Ulf Bergstro¨m,* Go¨ran Englund,† and Kjell Leonardsson‡
Department of Ecology and Environmental Science and Umea˚ Marine Sciences Centre, Umea˚ University, SE-901 87 Umea˚, Sweden Submitted October 28, 2004; Accepted September 29, 2005; Electronically published January 9, 2006 Online enhancements: appendixes.
abstract: Extrapolating ecological processes from small-scale experimental systems to scales of natural populations usually entails a considerable increase in spatial heterogeneity, which may affect process rates and, ultimately, population dynamics. We demonstrate how information on the heterogeneity of natural populations can be taken into account when scaling up laboratory-derived process functions, using the technique of moment approximation. We apply moment approximation to a benthic crustacean predator-prey system, where a laboratory-derived functional response is made spatial by including correction terms for the variance in prey density and the covariance between prey and predator densities observed in the field. We also show how moment approximation may be used to incorporate spatial information into a dynamic model of the system. While the nonspatial model predicts stable dynamics, its spatial equivalent also produces bounded fluctuations, in agreement with observed dynamics. A detailed analysis shows that predator-prey covariance, but not prey variance, destabilizes the dynamics. We conclude that secondorder moment approximation may provide a useful technique for including spatial information in population models. The main advantage of the method is its conceptual value: by providing explicit estimates of variance and covariance effects, it offers the possibility of understanding how heterogeneity affects ecological processes. Keywords: functional response, moment approximation, extrapolation, spatial heterogeneity, variance, covariance.
* Author names are in alphabetic order. Present address: Swedish Board of ¨ regrund, Sweden; Fisheries, Institute of Coastal Research, Box 109, SE-740 71 O e-mail:
[email protected]. †
Corresponding author; e-mail:
[email protected].
‡
E-mail:
[email protected].
Am. Nat. 2006. Vol. 167, pp. 246–259. 䉷 2006 by The University of Chicago. 0003-0147/2006/16702-40707$15.00. All rights reserved.
Since the pioneering work of Gause (1934) and Huffaker (1958), in which they elegantly demonstrated how heterogeneity may affect predator-prey dynamics, questions of scale and heterogeneity have slowly grown to become a primary focus in ecology (Wiens 1989; Levin 1992; Tilman and Kareiva 1997; Dieckmann et al. 2000; Kemp et al. 2001; Schneider 2001; Englund and Cooper 2003). The potential consequences of scale and spatial variation for population dynamical outcomes, such as stability and persistence, mediated through predatory or competitive interactions have been thoroughly explored theoretically (Hassell et al. 1991; Matsuda et al. 1992; Goldwasser et al. 1994; Pacala and Tilman 1994; Bolker and Pacala 1997; Chesson 1998, 2000; de Roos et al. 1998; Nisbet et al. 1998; Donalson and Nisbet 1999; Pascual et al. 2001; Keeling et al. 2002; Hosseini 2003). Empirical examples are less abundant and primarily come from small-scale conceptual experiments (Huffaker et al. 1963; Luckinbill 1974; Holyoak and Lawler 1996; Ellner et al. 2001). Conducting experiments aiming at coupling scale and/or heterogeneity to population dynamics is for logistical reasons seldom possible in natural systems. One way to approach this problem for large-scale natural populations is instead to develop system-specific spatial population models, in which the effects of space on population processes may be explored. These models combine small-scale observations or experiments for measuring process rates with field measurements of spatial variability. This approach has a long history in the study of host-parasitoid interactions, where models originally developed by Bailey et al. (1962) and May (1978) have been used to examine the dynamic consequences of heterogeneity in the distribution of parasitoid attacks (e.g., Hassell 1980; Jones et al. 1993; Liljesthro¨m and Rabinovich 2004). A similar approach has been used in studies of host-parasite interactions (Dobson and Hudson 1992; Barlow 2000). An important empirical finding is that measures of heterogeneity vary with the spatial scale of observation (Schneider and Duffy 1985; Boulinier et al. 1996; Roland and Taylor 1997; O’Driscoll et al. 2000; Cronin 2003). Consequently, predictions about the
Plugging Space into Predator-Prey Models effects of observed heterogeneity on population dynamics may vary with the extent and resolution of the sampling program (Horne and Schneider 1997; Chesson 1998), an issue that has received little theoretical attention (but see Perry et al. 2000). Another important empirical finding is that measures of spatial variation are often dynamic, which may have significant effects on predicted density dynamics (Hassell 1980; Chesson and Murdoch 1986; Schneider and Piatt 1986; Jones et al. 1993; Van Veen et al. 2002; Tobin and Bjørnstad 2003). Usually a phenomenological approach has been used when incorporating heterogeneity dynamics into models; that is, variances and covariances are modeled as functions of host or parasitoid densities (Hassell 1980; Chesson and Murdoch 1986; Murdoch and Stewart-Oaten 1989; Jones et al. 1993). More recently, the technique of moment closure has been used to derive dynamic equations for statistical moments based on assumptions about dispersal and local interactions (e.g., Bolker and Pacala 1997; Bolker et al. 2000; Keeling et al. 2002). This approach, combining theoretical models with observed heterogeneity, has not been used to explore the effects of spatial variation on the dynamics of predatorprey systems. Thus, there is a need for development of a conceptual framework for how to measure relevant heterogeneity in such systems and how to add this information to models. Such a framework should allow measures of heterogeneity to be included in models as dynamic entities (e.g., Hassell 1980; Chesson 1998), and it should deal with complications posed by the scale dependence of heterogeneity (Chesson 1998). Our objective in this article is to present and apply such a framework. We build on earlier work demonstrating how moment approximation can be used when investigating the effects of scale-dependent heterogeneity on process rates (Chesson 1998; Melbourne 2000). Moment approximation is a mathematical method that can be used to describe the effects of spatial variation on process rates in terms of familiar statistical moments (variance, covariance, skew, etc.). We start by explaining the concepts behind this method and how to estimate spatial variation at scales relevant to specific ecological processes. We then exemplify the use of moment approximation in an empirical study of a predator-prey system by showing how a functional response estimated in a small-scale experimental system can be adjusted to account for the heterogeneity observed in the natural system of interest. Thus, moment approximation can also be used for dealing with the problem of extrapolation from small-scale studies to larger, more heterogeneous systems (Englund and Cooper 2003; Melbourne et al. 2005). We also use the technique to incorporate spatial information into a simple predator-prey model of the system and explore how this spatial model reproduces the dynamics of the system compared to a nonspatial model.
247
Accounting for Spatial Variation Using Moment Approximation A general mathematical result states that heterogeneity in a driving variable affects the response if there is a nonlinear relationship between the driving variable and the response. It is often referred to as nonlinear averaging (Ruel and Ayres 1999; Wilson and Harder 2003), and ecologists dealing with the specific problem of scaling up have used the terms aggregation error (Bartell et al. 1988; Rastetter et al. 1992; Cale 1995), transmutation (O’Neill 1979), and scale transition (Chesson 1998). Several techniques can be used for incorporating effects of heterogeneity in driving variables when scaling up ecological functions, for example, extrapolation by expected value, explicit integration, partitioning, calibration, and moment approximation (King 1992; Rastetter et al. 1992; Chesson 1998; Englund and Cooper 2003). Partitioning has been widely applied in ecological research (see Rastetter et al. 1992), while other methods have found relatively little use. A strength of moment approximation, compared to other methods, is that it is computationally simple and offers a means of exploring how different aspects of heterogeneity affect process rates. Moment approximation is based on a Taylor series expansion, which approximates a function close to a known point. In the moment approximation, Taylor expansions are averaged over all observed variable values. The resulting polynomial for the sum of the Taylor expansions contains expressions for statistical moments around the mean, such as variance, skewness, and kurtosis. These statistical moments measure the heterogeneity of the observed variables, while the other part of the terms, the partial derivatives, measures the nonlinearity of the function (e.g., eq. [1] and app. A in the online edition of the American Naturalist). Thus, in a moment approximation, we can estimate the effect of spatial variation using a single equation, where the input data (mean, variance, etc.) can be determined from field data. In ecological applications, the series of statistical moments is often closed at second order (e.g., Bolker and Pacala 1997; Melbourne et al. 2005). It is thereby assumed that higher-order moments or derivatives are negligible, which is a reasonable assumption when either distributions are close to normal or the nonlinear function is close to quadratic over the range of densities observed (see “Evaluation of the Second-Order Approximation”). To demonstrate how the moment approximation can be used, we apply it to the functional response. The functional response is a central concept within population ecology, because it describes how the consumption rate of predators is related to prey density. Processes that may cause the functional response to be nonlinear with respect to prey densities include predator satiation, which sets an upper ceiling on the function at high prey densities, and
248
The American Naturalist Table 1: Second-order moment approximations for Type I–III functional response functions Type
Functional response (Ne)
I II III
aPN aPN/(1 ⫹ ahN) aPN2/(1 ⫹ ahN 2)
Second-order moment approximation (Ne(N, P)) ≈ Ne(N, P) ⫹ ajN, P ≈ Ne(N, P) ⫹ [⫺a2hP/(1 ⫹ ahN)3]jN2 ⫹ [a/(1 ⫹ ahN)2]jN, P ≈ Ne(N, P) ⫹ [aP(1 ⫺ 3ahN 2)/(1 ⫹ ahN 2)3]jN2 ⫹ [(2aN)/(1 ⫹ ahN 2)2]jN, P
Note: Ne is the density of prey eaten per unit time, a is the attack rate, P is the predator density, N is the prey density, and h is the prey-handling time. The symbols of the moment approximation are as described for equation (1).
predator switching to other resources, which causes the function to accelerate at low prey densities. Nonlinearity with respect to predator density may occur because of interference or facilitation between predators. Functional responses are usually determined in controlled small-scale experiments. When functions derived in such experimental studies are used to estimate consumption rates in natural systems, heterogeneity in the distribution of predator and prey both within the experimental system (Bergstro¨m and Englund 2002, 2004) and in the natural habitat (Ruxton and Gurney 1994; Keeling et al. 2000) may distort the estimates. We focus on the latter of the two, the change in spatial variation when extrapolating from small-scale experiments to whole populations. In many cases, heterogeneity is ignored when scaling up functions, because only population-scale mean densities are used in the calculations. This method of (not) dealing with heterogeneity is usually referred to as the mean field approximation (MFA). We assume that the functional response is linear in P, that is, no interference or facilitation between predators, and nonlinear in N, which may reflect predator satiation, switching, and so forth. The general form of the secondorder moment approximation is then
\
⭸ 2Ne(N, P) 1 j2 ⭸N 2 2 N \
MFA
variance effect
Ne(N, P) ≈ Ne(N, P) ⫹
⫹
⭸ 2Ne(N, P)
j
.
N, P ⭸N⭸P \
(1)
covariance effect
The overbars in the equation are notations for spatial averages. The mean function Ne(N, P) is the regional-scale consumption rate, which is a function of prey (N) and predator density (P), while jN2 is the spatial variance in prey density and jN, P is the spatial covariance between prey and predator densities. The second-order moment approximation thus consists of the function for average densities of predator and prey, that is, the MFA, plus two terms that correct for the effects of spatial heterogeneity (see app. A for derivation). We refer to these terms as the “variance effect” and the “covariance effect.” The variance effect is caused by spatial variation in prey density in com-
bination with a nonlinear function, here the functional response. The covariance effect arises as an effect of spatial correlation in the distribution of predator and prey, which affects the number of prey encounters for the predator. In table 1, we provide the approximations for three common functional response models (Types I–III). Note that for a linear functional response (Type I), consumption is affected by the covariance but not by the variance. The nature of the variance and covariance effects can be illustrated with a conceptual example. Consider two equal-sized patches with different prey densities (Nlow and Nhigh) but similar predator densities (fig. 1A). Using the MFA, and thereby assuming perfect mixing, the consumption rate within a defined habitat and amount of time can be calculated for the average prey density (Ne(N ); fig. 1A, crosses). If there is spatial variation in the driving variable (N) and the functional response is nonlinear, the MFA produces biased estimates. The unbiased average consumption rate for the heterogeneous habitat (Ne(N )) should then be calculated as the mean of the two rates (fig. 1A, squares). For a Type II functional response, spatial variation in prey densities decreases the average consumption rate, compared to a uniform prey distribution, and the difference between Ne(N ) and Ne(N ) gives a measure of the error caused by using the MFA, that is, the variance effect. The sign and magnitude of the variance effect depends on the curvature of the function. The effect is negative if the function is concave down, as illustrated in fig. 1A. It is positive if the function is concave up and negligible if the function is nearly linear. The process causing the Type II response to level off is predator satiation. Thus, an intuitive interpretation of the variance effect is that predators at low prey densities have a low feeding rate because the encounter rate is low and that this effect cannot be fully compensated for by predators at high densities because of satiation. Next, we examine the effect of covariance between predator and prey densities. This effect can be understood as a result of altered neighborhood densities and encounter rates. A positive covariance means that predators experience higher prey densities and therefore have higher encounter rates than would be predicted from mean densities. Conversely, a negative covariance reduces encounter rates because predators and prey tend to be found in dif-
Plugging Space into Predator-Prey Models
249
Figure 1: Effects of spatial variation in prey and predator distributions on estimates of consumption rates. In this example, we have a habitat consisting of two equal-sized patches with differing prey densities. The curve describes a Type II functional response. A, The variance effect is caused by spatial variation in prey density; N is the average prey density in the habitat, and Nlow and Nhigh are the densities in the two patches. The consumption rate obtained if prey variance is not considered (mean field approximation) is indicated with crosses, while the exact consumption rate, which is given by the mean of the consumption rate in the two patches, is represented by filled squares. B, The covariance effect is caused by spatial variation in both predator and prey densities, giving rise to either a negative or a positive correlation between the two. Here P is the predator density when predators distribute evenly between the two patches. The covariance is then 0, and the consumption rate for the whole habitat (filled squares) is given by the mean consumption in the two patches (open squares). A positive covariance between predator and prey densities means that the prey densities perceived by most predators are larger than the overall mean and thus that the consumption rate per predator is increased (filled triangles). A negative covariance decreases the consumption rate because most predators are found in the patch with low prey density (diamonds).
ferent areas. A graphical illustration of covariance effects is given in figure 1B. Consider again a habitat consisting of two patches with different prey densities. If predators distribute equally between the two patches at a density P , the covariance is 0. The consumption rate can be found by averaging the consumption in each habitat (open squares), as illustrated in figure 1A. A positive covariance— for example, if most predators aggregate in the patch with high prey density (open triangles)—will increase the encounter rate and therefore the average consumption rate (filled triangles). The opposite holds true when the covariance is negative; experienced prey density, encounter rate, and therefore average consumption rate will be lower than expected from overall mean prey densities. This relationship between perceived prey density and the covariance jN, P, measured at a scale corresponding to the perceptual range of the predator, can be quantified using Nperceived p jN, P /P ⫹ N (Schneider et al. 1987). In the example above, the net effect of heterogeneity on the consumption rate is a function of the nonlinearity of the functional response, the average prey density, the variance of the prey distribution, and the covariance of prey
and predator distributions (and also, in cases when distributions deviate from normal, higher moments). Here moment approximation provides an instrument for partitioning the effect of spatial variation between well-known statistical descriptors, that is, the statistical central moments, thereby effectively clarifying the somewhat diffuse concept of heterogeneity. A Framework for Operationalizing the Moment Approximation The moment method outlined above can be used for making nonspatial population models spatial. Instead of coupling the model to a map, as in spatially explicit models, the method treats space implicitly by describing variation statistically. Such a framework involves the following steps: first, derive the nonspatial model and make it spatial by performing moment expansion for each driving variable that is nonlinearly related to the response; second, measure the nonlinear functions in a small-scale system; third, measure spatial variation in the form of variances and covariances in the natural system of interest; and finally, find
250
The American Naturalist
functions that relate the variance and covariance to densities of prey and predator (the variables tracked by the model; Chesson 1998; Melbourne et al. 2005). In order to accomplish this last step, data that cover a range of mean densities (e.g., from different areas or time periods) are needed. The spatial variance of a population can usually be modeled as a power function of its mean density (Taylor et al. 1980; Hanski 1987; Downing 1991; Chesson 1998). Finding a model for the covariance is more challenging, as empirical studies of its density dependence appear to be lacking. Keeling et al. (2000) analyzed a metapopulation model and found that a delayed density dependence was manifested in the covariance between predator and prey, which suggests that the covariance may be modeled as a function of both past and present densities. Estimating Relevant Heterogeneity An important issue when using moment approximation for extrapolation is how to determine the scale at which to measure spatial variation in the natural system, as this choice may have a large effect on the estimates (Chesson 1998). Determining the extent of the sampling program is straightforward, as it should match the extent of the population(s) of interest. Finding the appropriate sampling resolution, that is, the area covered by an individual sample, is more complicated. Here we need to consider the resolution of the nonlinear process of interest. We define process resolution as the smallest scale of spatial variation in a driving variable that affects the outcome of the process (Englund and Cooper 2003). Variation at scales smaller than the resolution scale is thus not “detected” by the process. To ensure that such small-scale variation is not detected by the sampling program, the area covered by an individual sample should be tuned to match the resolution scale. An exception is when the process resolution is smaller than or equal to the area of the experimental arena where the process function was determined. Then the area of an individual sample should be adjusted to match the size of the experimental arena rather than the process resolution scale. This is to avoid detecting small-scale variation that was present in the experimental system and thus already is integrated in the function. The application of the concept of process resolution is hampered by a lack of empirical methods for its quantification. However, if the biological mechanism causing nonlinearity is understood, it is often possible to use intuitive arguments to identify a scale domain within which the resolution scale can be found (Melbourne et al. 2005). When heterogeneity has been quantified at multiple scales using hierarchical ANOVA or ANCOVA, it is also possible to evaluate whether the estimates obtained by moment approximation are sensitive to changes in sample scale. In
this way, processes where the definition of resolution scale is critical for the outcome of the moment approximation may be identified, which may necessitate a more careful identification of process resolution. The method we propose for quantifying spatial variation at relevant scales assumes that the heterogeneity in the experimental system is similar to the heterogeneity at the same scale in the field. In many cases, however, the walls of the containers affect the distribution of animals, which may have substantial effects on the consumption rate observed in predator-prey experiments (Bergstro¨m and Englund 2002, 2004). If the magnitude of the unwanted heterogeneity can be estimated, this caging artifact may be extracted from the functional response (e.g., Bergstro¨m and Englund 2004). The corrected functional response may thereafter be used in the moment approximation. The application of this method for estimating heterogeneity is complicated by the fact that the nonlinear behavior of a model may be the result of several processes that may have different resolution scales. For example, the covariance between predator and prey affects the rate of consumption of prey by affecting the encounter rate and thereby the perceived prey density. Here heterogeneity, even at small scales compared to the mobility of the predator, will affect the number of prey encountered and thus the consumption rate (fig. 1B). On the other hand, the variance effect on a Type II functional response (fig. 1A) is caused by a higher degree of predator satiation in patches with high prey density. Heterogeneity on scales that are small compared to the mobility of the predator will not matter, because there will be no difference in the prey densities experienced by hungry and satiated predators. Thus, if moment approximation is used in population models including several processes, it may be desirable to quantify spatial variation at scales matching the resolution scale of each of the processes. If a hierarchical sampling scheme is adopted in which variance and covariance components are statistically partitioned between several scales, spatial variation relevant to different processes can be measured simultaneously (Chesson 1998; Melbourne 2000). An Empirical Example The Saduria-Monoporeia System The isopod Saduria entomon (L.) is a predator and scavenger, common on soft bottoms of all depths throughout the Baltic Sea (Green 1957; Haahtela 1990). Its main prey in the northern Baltic Sea is the amphipod Monoporeia affinis (Lindstro¨m), although the diet may also include other macro- and meiobenthos, dead fish, and conspecifics (Haahtela 1990; Leonardsson 1991a). Monoporeia is a deposit feeder (Lopez and Elmgren 1989) and can be very
Plugging Space into Predator-Prey Models abundant, with densities frequently reaching 10,000 individuals per square meter (Sparrevik and Leonardsson 1998). In the low-saline Gulf of Bothnia, these two species totally dominate the macrofaunal community at depths larger than 40–50 m. As other predators on these species, such as benthivorous fish, are scarce at these depths (Leonardsson 1991b), Saduria and Monoporeia make up a tightly coupled predator-prey system, where predation from Saduria appears to have a strong influence on the dynamics of Monoporeia (Sparrevik and Leonardsson 1999). The functional response of Saduria preying on Monoporeia is sigmoid and well described by the Type III function in table 1 (Aljetlawi et al. 2004). The field data on the dynamics and the spatial distributions of Monoporeia and Saduria come from an environmental monitoring program in the Gulf of Bothnia, at 63⬚20⬘N, 20⬚20⬘E. Here 11 stations within a 300-km2 area were sampled in May each year during 1983–2002. The depth at theses stations was 46–129 m. At each station, three replicate samples within a radius of approximately 30 m from the center of the station were collected with a van Veen grab (0.1-m2 area). This data set thus provided information on spatial variation at two scales, the sample scale and the station scale. Quantifying the Moment Approximation The experimentally obtained functional response was extrapolated to the natural Saduria-Monoporeia system using second-order moment approximation. To parameterize the moment approximation for the Type III functional response (derived in app. A), we used experimental data on attack rate and handling time (Aljetlawi et al. 2004; app. B in the online edition of the American Naturalist), together with field estimates of Monoporeia and Saduria densities, Monoporeia variance, and Saduria-Monoporeia covariance in the study area from 1983–2002. The consumption rate calculated using the moment approximation was compared to the rate predicted by a nonspatial model based on mean densities of predators and prey. This latter method, which is the MFA, can also be described as a first-order moment approximation, consisting of only the first term of equation (1). To evaluate the performance of the moment approximation, we compared estimates generated with this method to estimates obtained using partitioning (sensu Rastetter et al. 1992). In partitioning, the number of prey consumed was calculated for each sample separately, and the total consumption was found by summing over samples and stations. As the number of terms retained in the moment approximation increases, the moment approximation estimate approaches the partitioning estimate. Thus, by comparing the estimates obtained by partitioning and
251
second-order moment approximation, we can evaluate the effect of ignoring higher-order statistical moments. This comparison will, however, not be exact, since the partitioning estimate is calculated as if the whole population was known, while the variance and covariance components are adjusted for being estimated by a sample of a larger population. In our case, this source of error is negligible because the sample size is large. The results of the moment approximation showed that the heterogeneous distribution of Saduria and Monoporeia in the studied area had a large effect on Saduria’s consumption rate. The effect of the variance, averaged across years, was to increase the consumption rate by 47%, compared to the mean field estimate, while the effect of the covariance was to decrease consumption rate to 65% of the mean field estimate (fig. 2, top). The variance effect was positive in all years, because densities were in the range of the functional response where it curves up. The covariance effect, on the other hand, was negative in most years. Because of the opposite influence of these two mechanisms, the resulting effect of heterogeneity on the consumption rate was relatively small in general, as demonstrated by the difference between the MFA and moment approximation estimates. However, in the last years, both the covariance and the variance effects were highly positive, which together raised the consumption rate to almost five times the mean field estimate (fig. 2, bottom). The second-order moment approximation seemed to capture the effect of spatial variation quite well, as the difference between the moment approximation and partitioning was on average 13%. The difference was largest in years with low prey density. We also evaluated the performance of a third-order moment approximation (app. A), where the effect of skewness of distributions is also taken into account. The third-order term further improved the accuracy of the moment approximation, as the difference compared to partitioning was then on average 4%. The precision of the moment approximation also depends on the match between the resolution of the sampling program and the size of the experimental system and/or the resolution of the process that generates nonlinearity. The size of the experimental system and a single sample matched well, these being 0.2 and 0.1 m2, respectively. Following the scheme for how to measure relevant heterogeneity (outlined in “Estimating Relevant Heterogeneity”), we should also examine whether the resolution scale of the process is larger than the experimental/sample scale. As discussed above, this procedure is more complicated if the variance and covariance effects may have different resolution scales. Based on our knowledge of Saduria behavior when preying on Monoporeia, we believe that the resolution scale of the covariance effect is smaller than the size of the sample. The resolution scale of the
252
The American Naturalist olution scale was not critical for the outcome of the moment approximation in most years.
Modeling Predator-Prey Dynamics with and without Spatial Variation To study how spatial variation may affect the SaduriaMonoporeia system over timescales longer than the snapshot picture of the functional response, we developed a simple continuous time model of the predator-prey system. We have simplified the model by ignoring the size structure of the species, as well as the cannibalistic behavior of Saduria (Leonardsson 1991a). In the model, we used a logistic prey growth function and the experimentally determined functional response (Aljetlawi et al. 2004):
Figure 2: Quantitative effects of the moment approximation on the consumption rate estimates for the Saduria-Monoporeia system in 1983– 2002. Top, influence of the variance and covariance effects on the mean field estimate. Bottom, rate of consumption of prey calculated by partitioning, mean field approximation, and moment approximation.
variance effect, on the other hand, may be larger than the size of the sample but still most likely smaller than the intersample distance (∼30 m). This means that heterogeneity at the station scale will certainly matter for both the variance and the covariance effects, while variation at the sample scale possibly acts only on the covariance effect. To examine whether this was a problem, we estimated variances and covariances at both the sample and the station scale and calculated the moment approximation for each scale. The variance and covariance components were partitioned between scales by performing one-way ANOVAs and ANCOVAs (Sokal and Rohlf 1995; Underwood 1997). The analysis showed that using station, instead of sample, as the resolution scale had little influence on the result of the moment approximation, as the mean difference between the estimates was only 3.4%. Most of the variance in Monoporeia density (mean 89%, range 66%–99%), as well as of the covariance between Monoporeia and Saduria (mean 82%, range 4%–99%), came from the station scale. Thus, since most of the spatial variation came from the station scale, the choice of res-
dN N aPN 2 p rN 1 ⫺ ⫺ , dt K 1 ⫹ ahN 2
( )
(2)
dP aPN 2 pf ⫺ qP ⫹ b, dt 1 ⫹ ahN 2
(3)
where r and K are the growth rate and the carrying capacity of the prey, respectively, and N and P are prey and predator densities, respectively, measured as biomass per unit area, a is attack rate, h is the handling time of the functional response, f is the conversion efficiency, and q is predator mortality rate. Predator growth due to feeding on an alternative food source, dead fish, was represented by adding a small constant b. This formulation was chosen because this type of resource is detected from long distances and rapidly consumed; that is, we assume that a fixed amount of biomass enters the predator population via this route. The parameterization of the model is described in appendix B. This nonspatial model was made implicitly spatial using moment approximation. To accomplish this, we express the variance and covariance as functions of prey and predator densities. These functions were estimated from the time series data available; that is, yearly measurements of densities, variances, and covariances were used as observations. The variance in prey distribution was expressed as a power function of prey biomass (jN2 p 1.12Nt1.56; r 2 p 0.94, P ! .001). The covariance was modeled by a linear function of prey biomass at times t and t ⫺ 1 and predator biomass at t (jN, P p ⫺3.53Nt ⫹ 3.58Nt⫺1 ⫺ 3.10Pt, r 2 p 0.60, P ! .02 for all terms; details about model selection in app. B). The significant effect of the prey density from the previous year (Nt⫺1) shows that there is a delayed density dependence that is manifested in the covariance dynamics. The covariance model worked well except when negative covariances occurred at increasing prey densities. Then, the
Plugging Space into Predator-Prey Models
253
model could produce negative consumption. To correct for this, negative consumption was set to 0. This implies that predators relied on alternative resources (b in the model), which in the model occurred in two years per cycle. The spatial model, with a second-order moment approximation of both the prey growth function and the functional response, was given by N aPN 2 dN r p rN 1 ⫺ ⫺ jN2 ⫺ dt K K 1 ⫹ ahN 2
( )
⫹
[
aP(1 ⫺ 3ahN 2) 2 2aN jN ⫹ jN, P , (1 ⫹ ahN 2)3 (1 ⫹ ahN 2)2
]
(4)
aPN 2 aP(1 ⫺ 3ahN 2) 2 dP pf jN 2⫹ dt 1 ⫹ ahN (1 ⫹ ahN 2)3
[
⫹
]
2aN jN, P ⫺ qP ⫹ b. (1 ⫹ ahN 2)2
Figure 3: Classification of the dynamics produced by a predator-prey
(5)
Note that the effect of prey variance on the logistic growth function is given by the second term in equation (4), whereas no correction term is required for the linear function describing the predator death rate in equation (5) (the second derivative equals 0). The behavior of the mean field model and the spatial model was explored for realistic values of predator attack and death rates. The models were not examined within the whole parameter space, as the aim of this modeling exercise was merely to capture qualitative differences between the models. The dynamics predicted by the nonspatial and the spatial model differed substantially. While the nonspatial model was stable within the explored parameter space, the spatial model also exhibited bounded oscillations (fig. 3). These results may be compared with the observed dynamics of the Saduria-Monoporeia system, where both our data (fig. 4) and longer time series of Andersin et al. (1978) and Leonardsson et al. (2002) for Monoporeia suggest that the dynamics are oscillatory. Our data cannot be used to determine the frequency of oscillations, but the 37-year record for the Bothnian Sea provided by Leonardsson et al. (2002) shows regular 6–8-year cycles, which is in good agreement with the 6-year cycles predicted by our spatial model. However, the spatial model predicts lower predator densities and more positive covariances than observed in the field (fig. 4). It can also be noted that the spatial model predicts that Saduria rely on alternative resources in years with strong negative covariance. When including the covariance effect but not the variance effect, the predicted dynamics were similar to those predicted for the fully adjusted model (fig. 3), while adjusting only for the effects of prey variance produced stable
model when using moment approximation (eqq. [4], [5]) to include effects of spatial heterogeneity in predator and prey densities. The model was adjusted either for both prey variance and predator-prey covariance (fully adjusted model) or only for the covariance. If the model was not adjusted for heterogeneity (mean field approximation) or adjusted only for prey variance, the predicted dynamics were stable throughout the investigated parameter space (not shown). The parameter values were K p 59.7 g m⫺2, r p 0.003 day⫺1, f p 0.7, h p 4.8 days, and d p 0.004 day⫺1. The plus sign denotes the values of attack rate and growth on alternative resources that were used in figure 4.
dynamics throughout the investigated parameter space (not shown). Thus, we conclude that the spatial model was destabilized by the covariance effect. The covariance affects the dynamics by altering contact rates. To demonstrate this intuitive interpretation, we used the potential contact measure proposed by Schneider et al. (1987) and O’Driscoll et al. (2000) to calculate the prey densities experienced by a predator at the station scale (fig. 5). Negative covariance during the buildup phase causes the prey density experienced by a predator to be lower that the regional mean density, whereas the two densities are more similar during the phase when prey densities decrease.
Discussion Evaluation of the Second-Order Approximation To evaluate the versatility of the second-order moment expansion as a tool for calculating effects of heterogeneity, we need to examine the accuracy of this approximation for different scenarios. We start this exercise by identifying situations where the second-order approximation will calculate regional-scale functions exactly. This is the case when the function to be approximated is quadratic (par-
254
The American Naturalist
Figure 4: Regional-scale Saduria and Monoporeia dynamics in the study area during 1983–2002 and the final dynamics predicted by a predatorprey model using either the mean field approximation (eqq. [2], [3]) or the moment approximation (eqq. [4], [5]). Attack rate a was 1.8 # 10⫺5 m4 g⫺2 day⫺1, and predator growth on alternative resources b was 0.005 g day⫺1 m⫺2.
abolic), as well as when the variables of the function are normally distributed, because then all higher-order moments (e.g., skew, kurtosis) will be 0 (Rastetter et al. 1992; Keeling et al. 2002). Only in cases when both of these conditions are violated will the second-order moment expansion be imprecise—if either condition is satisfied, a second-order equation will be exact. Small deviations from both these conditions will still allow for an exploration of how spatial heterogeneity affects the rate of an ecological process, as higher-order terms will be small. This situation may be exemplified by the Type III functional response of the Saduria-Monoporeia system, where the response function was not quadratic and the distributions often deviated from normality. Still, the consumption estimates from the second-order moment approximation were in good agreement with the partitioning estimates in most years. In this context, it is worth considering in what situations the effect of spatial variation will be largest and the MFA most erroneous. In general, the magnitude of the variance and covariance effects, and thus the error associated with using the MFA, increases with the degree of nonlinearity of the function and the degree of spatial variation of the population (Rastetter et al. 1992). In some instances, however, the net effect of heterogeneity may be small, although the degree of nonlinearity and/or spatial variation would indicate the opposite. This was the case in the SaduriaMonoporeia system, where the variance and covariance effects, which were both relatively large, in most years acted
in opposite directions. Consequently, the difference between the nonspatial (MFA) and spatial estimates of the consumption rate was fairly small, although the spatial variation of the system was substantial. This example highlights the importance of performing calculations when trying to judge whether spatial heterogeneity has important effects on process rates in an ecological system. The Effect of Spatial Variation on Population Dynamics In this study, we show how a dynamic population model based on empirical data can be made spatial using moment approximation. The model includes effects of both prey variance and predator-prey covariance. A central finding is that it is possible to model the dynamics of predatorprey covariance as a function of past and present densities and that this covariance may have important effects on the stability of the system. An approach related to moment approximation has been applied to the study of host-parasitoid interactions, assuming that spatial heterogeneity in attacks can be described by the aggregation parameter k of the negative binomial distribution. This approach has produced elegant model formulations and a general criterion for stability, CV 2 1 1 (CV is coefficient of variation), that often can be measured directly in the field (e.g., Pacala and Hassell 1991). Although early treatments confounded parasitoid variance and host-parasitoid covariance (May 1978; Has-
Plugging Space into Predator-Prey Models
Figure 5: Dynamics of regional mean prey density and the prey density experienced by a predator at the station scale. Predicted experienced densities were calculated using Nperceived p jN,P/P ⫹ N (Schneider et al. 1987).
sell 1980), later studies clarified how they should be separated (Hassell et al. 1991; Pacala and Hassell 1991). Most studies have treated variances and covariances as scaleindependent constants, but at least one study has modeled the host-parasitoid covariance as a function of parasitoid density (Jones et al. 1993). A weak stabilizing effect of the covariance was found in this study. Moment approximation has been used by theoretical ecologists (e.g., Chesson 1998; Holt and Barfield 2003), but not until recently has the method been used to estimate effects of heterogeneity in empirical systems (Melbourne et al. 2005). In the study by Melbourne et al., heterogeneity measures were treated as constants, which can give snapshot pictures of the systems. This certainly may improve our understanding of the effects of heterogeneity, but as shown here and elsewhere (Jones et al. 1993), variances and covariances can vary dramatically over time, which means that snapshots uncover only part of the picture. If moment approximation is to be used as a tool in population modeling, we need to be able to express variances and covariances as functions of the variables followed in
255
the model (Chesson 1998; Melbourne 2000; Melbourne and Chesson 2005). Variances usually do not pose a problem, because they can often be described as functions of mean density, while finding functions for covariances may be challenging. For the Saduria-Monoporeia system, we found that by including a time lag represented by the prey density from the previous year in the function, the covariance could be accurately predicted. This is the first empirical confirmation of the finding by Keeling et al. (2000) that the covariance in a spatial enemy-victim model with unstable dynamics may be expressed as a function of past densities. A difference between our covariance function and the one derived by Keeling et al. (2000) for a Nicholson-Bailey model is that past predator densities did not improve our model. We used the population model to examine how spatial heterogeneity may affect the Saduria-Monoporeia system. The nonspatial model can generate unstable dynamics (see Yodzis 1989), but it yields stable dynamics for parameter values considered realistic for the Saduria-Monoporeia system. Introducing spatial heterogeneity was destabilizing, and for realistic parameter values, the system oscillates with a cycle length corresponding to that observed in field data (Andersin et al. 1978; Leonardsson et al. 2002). What, then, is the mechanism behind the destabilizing effect of heterogeneity in our model? In the simulations, the effects of prey variance mediated by either the functional response or the logistic growth function did not affect stability. Instead, it was the covariance and its effect on the rate of encounters between predators and prey that destabilized the dynamics (fig. 3). The effect of covariances on stability is usually explained as an effect on the average values of parameters that control stability (Murdoch et al. 2003). This explanation was not valid here because the mean field model is stable for any value of attack rate when using realistic values of r, K, and h. Instead, we suggest that the destabilization can be explained by the temporal correlation between covariance and mean densities; figure 4 shows that the covariance is positive when prey densities decrease and, conversely, that the negative covariances coincide with increasing prey densities. This means that the covariance increases prey mortality when prey density decreases and decreases mortality when prey density goes up. Thus, the covariance dynamics tend to enhance prey density oscillations. This also means that the functional response is different at increasing and decreasing phases of the cycle. Recently, Pascual et al. (2001, 2002) showed that the complex dynamics generated by an individual-based spatial model could be reproduced by the corresponding mean field model after adjustment of parameter values, while retaining the form of the functions (see also Cuddington and Yodzis 2000). Clearly, this approach is not
256
The American Naturalist
useful for the Saduria-Monoporeia system, as highly unrealistic parameter values are required for the mean field model to produce oscillations. The increase in model complexity required to reproduce the observed dynamics is, however, modest; one additional state variable, the lagged prey density, and five parameters to characterize variances and covariances were required. Compared to individualbased models, this is a major simplification that improves our ability to analyze and understand the role of space on population dynamics. Our results concur with the conclusion reached in a number of articles that the spatial correlation of predator and prey distributions affects stability (Murdoch and StewartOaten 1989; Donalson and Nisbet 1999; Keeling et al. 2000; Hosseini 2003). However, there is no consensus in this literature concerning the effects of positive and negative covariances on stability. For example, negative covariances can be either stabilizing (Keeling et al. 2000; Hosseini 2003) or destabilizing (Donalson and Nisbet 1999). We conjecture that one reason why divergent conclusions have been reached is that the covariance can affect stability through at least two different routes: by a direct effect on the average attack rate and by temporal correlations between the covariance and mean densities. The temporal correlation can be destabilizing, as in this study, or stabilizing, as appears to be the case in the host-parasite model analyzed by Keeling et al. (2000). Moreover, it is well known that the effect of altered attack rate in mean field models can depend on the type of functional response and the value of other parameters in the model (e.g., Yodzis 1989). Thus, we suspect that generalizations concerning the relationship between covariance and stability will be difficult to find unless the relative importance of the two routes is considered. The covariance dynamics may be considered an explanation when comparing the behavior of models that differ in their representation of spatial processes. However, in a natural system, it may be equally valid to explain covariance patterns as the result of density dynamics. Therefore, mechanistic explanations should be sought among processes operating on lower levels of organization, such as predator and prey movement behaviors and patterns of local reproduction and mortality. Thus, an important question posed by our findings is what kind of behavior leads to destabilizing covariance dynamics. The fact that periods with high predation pressure, positive covariance, and decreasing prey densities are followed by periods with low prey density and negative covariance seems to suggest that the dominating negative covariance at the station scale is caused by local consumption; that is, Saduria act as a clearing predator (sensu Schneider 1992). However, smallscale aquarium experiments have demonstrated that Monoporeia respond to the presence of large Saduria by
entering the water column, a behavior that may be used for long-distance dispersal (Sparrevik and Leonardsson 1995). Thus, it cannot be ruled out that this movement response also influences covariance dynamics at the station scale. Conclusions and Future Perspective It has repeatedly been concluded that a major challenge for ecology today is to develop techniques for incorporating effects of scale-dependent variation in extrapolating results from experiments to nature (Wiens 1989; Levin 1992; Kemp et al. 2001; Englund and Cooper 2003). Moment approximation provides such a framework for scaling up experimentally obtained functions to heterogeneous populations. The moments consist of familiar descriptors of heterogeneity that are easily accessible from field data. These are added as terms that correct the experimentally derived small-scale model, which makes it possible to distinguish the effects of different measures of heterogeneity. In a second-order approximation of the functional response, we can separate the effects of variance and covariance in prey and predator distributions, both of which can be coupled to well-understood biological mechanisms. Moment approximation can also be used to add space as a component to dynamic population models while still keeping model structure simple. This, however, requires that the variance and the covariance of the system can be expressed using the variables that are tracked by the model. We show that this may be done for the Saduria-Monoporeia system, where we have a tight coupling between predator and prey. It remains to be seen whether this method is successful for other systems as well. Moment approximation also opens up the exciting possibility of adding spatial variation to existing mean field models, provided that appropriate field data are available. This exercise could, besides giving valuable insights into the importance of heterogeneity for population dynamics, also serve to evaluate the benefits and limitations of moment approximation. While the density-variance relationship is a well-studied subject, there seems to be a large gap in knowledge concerning density-covariance relations. Indeed, our empirical results and the theoretical findings of Keeling et al. (2000) suggest that the covariance is a key to understanding why predator-prey systems exhibit bounded fluctuations. Thus, we urge both theoreticians and empiricists to investigate covariance dynamics and how the covariances relate to densities in oscillating systems. Do these patterns, for example, differ between regular predator-prey cycles and cycles driven by cohort competition or cannibalistic interactions? It can be hypothesized that the covariance dynamics are influenced by predator and prey movements
Plugging Space into Predator-Prey Models and therefore are scale dependent (Englund and Hamba¨ck 2004a, 2004b). It may thus be of interest to study whether covariance dynamics at different scales can be related to movement behaviors, for example, overall mobility, predator aggregation in prey patches, or predator avoidance by prey. Another question to be studied is to what extent moment approximations to the second order will suffice to capture relevant aspects of heterogeneity. Bringing in higher-order moments is possible, but model complexity will increase sharply, and the mechanistic understanding of the spatial effects will suffer. It should also be kept in mind that including higher-order moments does not necessarily generate better estimates (Rastetter et al. 1992). The possibilities and limitations of moment approximation for ecological applications are so far largely unexplored. Nonetheless, the technique certainly brings promise for advancing our empirical knowledge of the well-recognized, but little treated, problem of how spatial variation shapes population dynamics. Acknowledgments We want to thank Umea˚ Marine Sciences Centre for assistance in field sampling during 1990–2002. S. Diehl, P. Hamba¨ck, P. R. Hosseini, L. Persson, and three anonymous reviewers gave valuable comments on the manuscript. The data used in this article mainly come from an environmental monitoring program funded by the Swedish Environmental Protection Agency (to K.L.). Sampling and analysis of supplementary data were funded by a grant from the Swedish Council for Forestry and Agricultural Research to K.L. The study was made possible by a grant from the Swedish Research Council to G.E. Literature Cited Aljetlawi, A. A., E. Sparrevik, and K. Leonardsson. 2004. Preypredator size-dependent functional response: derivation and rescaling to the real world. Journal of Animal Ecology 73:239–252. Andersin, A.-B., J. Lassig, L. Parkkonen, and H. Sandler. 1978. Longterm fluctuations of the soft bottom macrofauna in the deep areas of the Gulf of Bothnia 1954–1974; with special reference to Pontoporeia affinis Lindstro¨m (Amphipoda). Finnish Marine Research 244:137–144. Bailey, V. A., A. J. Nicholson, and E. J. Williams. 1962. Interactions between hosts and parasites when some hosts are more difficult to find than others. Journal of Theoretical Biology 3:1–18. Barlow, N. D. 2000. Non-linear transmission and simple models for bovine tuberculosis. Journal of Animal Ecology 69:703–713. Bartell, S. M., W. G. Cale, R. V. O’Neill, and R. H. Gardner. 1988. Aggregation error: research objectives and relevant model structure. Ecological Modelling 41:157–168. Bergstro¨m, U., and G. Englund. 2002. Estimating predation rates in experimental systems: scale-dependent effects of aggregative behaviour. Oikos 97:251–259.
257
———. 2004. Spatial scale, heterogeneity and functional responses. Journal of Animal Ecology 73:487–493. Bolker, B., and S. W. Pacala. 1997. Using moment equations to understand stochastically driven spatial pattern formation in ecological systems. Theoretical Population Biology 52:179–197. Bolker, B. M., S. W. Pacala, and S. A. Levin. 2000. Moment methods for ecological processes in continuous space. Pages 388–411 in U. Dieckmann, R. Law, and J. A. J. Metz, eds. The geometry of ecological interactions. Cambridge University Press, Cambridge. Boulinier, T., A. R. Ives, and E. Danchin. 1996. Measuring aggregation of parasites at different host population levels. Parasitology 112: 581–587. Cale, W. G. 1995. Model aggregation: ecological perspectives. Pages 230–241 in B. C. Patten and S. E. Jorgensen, eds. Complex ecology. Prentice-Hall, New York. Chesson, P. 1998. Making sense of spatial models in ecology. Pages 151–166 in J. Bascompte and R. V. Sole´, eds. Modeling spatiotemporal dynamics in ecology. Springer, Berlin. ———. 2000. General theory of competitive coexistence in spatiallyvarying environments. Theoretical Population Biology 58:211–237. Chesson, P. L., and W. W. Murdoch. 1986. Aggregation of risk: relationships among host-parasitoid models. American Naturalist 127:696–715. Cronin, J. T. 2003. Patch structure, oviposition behavior, and the distribution of parasitism risk. Ecological Monographs 73:283– 300. Cuddington, K. M., and P. Yodzis. 2000. Diffusion-limited predatorprey dynamics in Euclidean environments: an allometric individualbased model. Theoretical Population Biology 58:259–278. de Roos, A. M., E. McCauley, and W. G. Wilson. 1998. Pattern formation and the spatial scale of interaction between predators and their prey. Theoretical Population Biology 53:108–130. Dieckmann, U., R. Law, and J. A. J. Metz. 2000. The geometry of ecological interactions. Cambridge University Press, Cambridge. Dobson, A. P., and P. J. Hudson. 1992. Regulation and stability of a free-living host-parasite system: Trichostrongylus tenuis in red grouse. II. Population models. Journal of Animal Ecology 61:487– 498. Donalson, D. D., and R. M. Nisbet. 1999. Population dynamics and spatial scale: effects of system size on population persistence. Ecology 80:2492–2507. Downing, J. A. 1991. Biological heterogeneity in aquatic ecosystems. Pages 160–180 in J. Kolasa and S. T. A. Pickett, eds. Ecological heterogeneity. Springer, New York. Ellner, S. P., E. McCauley, B. E. Kendall, C. J. Briggs, P. R. Hosseini, S. N. Wood, A. Janssen, et al. 2001. Habitat structure and population persistence in an experimental community. Nature 412: 538–543. Englund, G., and S. D. Cooper. 2003. Scale effects and extrapolation in ecological experiments. Advances in Ecological Research 33: 161–213. Englund, G., and P. Hamba¨ck. 2004a. Scale dependence of emigration rates. Ecology 85:320–327. ———. 2004b. Scale dependence of movement rates in stream invertebrates. Oikos 105:31–40. Gause, G. F. 1934. The struggle for existence. Williams & Wilkins, Baltimore. Goldwasser, L., J. Cook, and E. D. Silverman. 1994. The effects of variability of metapopulation dynamics and rates of invasion. Ecology 75:40–47.
258
The American Naturalist
Green, J. 1957. The feeding mechanism of Mesidothea entomon (Linn.) (Crustacea: isopoda). Proceedings of the Zoological Society of London 129:245–254. Haahtela, I. 1990. What do Baltic studies tell us about the isopod Saduria entomon (L.)? Annales Zoologici Fennici 27:269–278. Hanski, I. 1987. Cross-correlation in population dynamics and the slope of spatial variance: mean regressions. Oikos 50:148–151. Hassell, M. P. 1980. Foraging strategies, population models, and biological control: a case study. Journal of Animal Ecology 49:603– 628. Hassell, M. P., R. M. May, S. W. Pacala, and P. L. Chesson. 1991. The persistence of host-parasitoid associations in patchy environments. I. A general criterion. American Naturalist 138:568–583. Holt, R. D., and M. Barfield. 2003. Impacts of temporal variation on apparent competition and coexistence in open ecosystems. Oikos 101:49–58. Holyoak, M., and S. P. Lawler. 1996. Persistence of an extinctionprone predator-prey interaction through metapopulation dynamics. Ecology 77:1867–1879. Horne, J. K., and D. C. Schneider. 1997. Spatial variance of mobile aquatic organisms: capelin and cod in Newfoundland coastal waters. Philosophical Transactions of the Royal Society of London B 352:633–642. Hosseini, P. R. 2003. How localized consumption stabilizes predatorprey systems with finite frequency of mixing. American Naturalist 161:567–585. Huffaker, C. B. 1958. Experimental studies on predation: dispersion factors and predator-prey oscillations. Hilgardia 27:343–383. Huffaker, C. B., K. P. Shea, and S. G. Herman. 1963. Experimental studies on predation: complex dispersion and levels of food in an acarine predator-prey interaction. Hilgardia 34:305–330. Jones, T. H., M. P. Hassell, and S. W. Pacala. 1993. Spatial heterogeneity and the population dynamics of a host-parasitoid system. Journal of Animal Ecology 62:251–262. Keeling, M. J., H. B. Wilson, and S. W. Pacala. 2000. Reinterpreting space, time lags, and functional responses in ecological models. Science 290:1758–1761. ———. 2002. Deterministic limits to stochastic spatial models of natural enemies. American Naturalist 159:57–80. Kemp, W. M., J. E. Petersen, and R. H. Gardner. 2001. Scale dependence and the problem of extrapolation: implications for experimental and natural coastal ecosystems. Pages 3–47 in R. H. Gardner, W. M. Kemp, and J. E. Petersen, eds. Scaling relations in experimental ecology. Columbia University Press, New York. King, A. W. 1992. Translating models across scales in the landscape. Pages 479–514 in C. G. Jones and J. H. Lawton, eds. Linking species and ecosystems. Chapman & Hall, New York. Leonardsson, K. 1986. Growth and reproduction of Mesidotea entomon (Isopoda) in the northern Bothnian Sea. Holarctic Ecology 9:240–244. ———. 1991a. Effects of cannibalism and alternative prey on population dynamics of Saduria entomon (Isopoda). Ecology 72:1273– 1285. ———. 1991b. Spatial size variation in adult females of Saduria entomon (Crustacea, Isopoda): a comparison between field observations and predictions of a life-history model. Ophelia 34:91– 104. Leonardsson, K., A. Laine, and A.-B. Andersin. 2002. Macrozoobenthos. Pages 69–73 in Baltic Marine Environment Protection Commission. Fourth periodic assessment of the state of the marine
environment of the Baltic Sea area, 1994–1998. Helsinki Commission, Helsinki. Levin, S. A. 1992. The problem of pattern and scale in ecology. Ecology 73:1943–1967. Liljesthro¨m, G., and J. Rabinovich. 2004. Modeling biological control: the population regulation of Nezara viridula by Trichopoda giacomellii. Ecological Applications 14:254–267. Lopez, G., and R. Elmgren. 1989. Feeding depths and organic absorption for the deposit-feeding benthic amphipods Pontoporeia affinis and Pontoporeia femorata. Limnology and Oceanography 34:982–991. Luckinbill, L. S. 1974. The effects of space and enrichment on a predator-prey system. Ecology 55:1142–1147. Matsuda, H., N. Ogita, A. Sasaki, and K. Sato. 1992. Statistical mechanics of population: the lattice Lotka-Volterra model. Progress of Theoretical Physics 88:1035–1049. May, R. M. 1978. Host-parasitoid models in a patchy environment: a phenomenological model. Journal of Animal Ecology 47:833– 844. Melbourne, B. A. 2000. Scaling up population dynamics: spatial variance and the scale transition for stream biofilm and grazers. PhD diss. Australian National University, Canberra. Melbourne, B. A., and P. Chesson. 2005. Scaling up population dynamics: integrating theory and data. Oecologia (Berlin) 145:178– 186. Melbourne, B. A., A. L. Sears, M. J. Donahue, and P. Chesson. 2005. Applying scale transition theory to metacommunities in the field. Pages 306–330 in M. Holyoak, M. A. Leibold, and R. D. Holt, eds. Metacommunities: spatial dynamics and ecological communities. University of Chicago Press, Chicago. Murdoch, W. W., and A. Stewart-Oaten. 1989. Aggregation by parasitoids and predators: effects on equilibrium and stability. American Naturalist 134:288–310. Murdoch, W. W., C. J. Briggs, and R. M. Nisbet. 2003. Consumerresource dynamics. Princeton University Press, Princeton, NJ. Nisbet, R. M., A. M. de Roos, W. G. Wilson, and R. E. Snyder. 1998. Discrete consumers, small scale resource heterogeneity, and population stability. Ecology Letters 1:34–37. O’Driscoll, R. L., D. C. Schneider, G. A. Rose, and G. R. Lilly. 2000. Potential contact statistics for measuring scale-dependent spatial pattern and association: an example of northern cod (Gadus morhua) and capelin (Mallotus villosus). Canadian Journal of Fisheries and Aquatic Sciences 57:1355–1368. O’Neill, R. V. 1979. Transmutations across hierarchical levels. Pages 58–78 in G. S. Innis and R. V. O’Neill, eds. Systems analysis of ecosystems. International Cooperative, Fairland, MD. Pacala, S. W., and M. P. Hassell. 1991. The persistence of hostparasitoid associations in patchy environments. II. Evaluation of field data. American Naturalist 138:584–605. Pacala, S. W., and D. Tilman. 1994. Limiting similarity in mechanistic and spatial models of plant competition in heterogeneous environments. American Naturalist 143:222–257. Pascual, M., P. Mazzega, and S. A. Levin. 2001. Oscillatory dynamics and spatial scale: the role of noise and unresolved pattern. Ecology 82:2357–2369. Pascual, M., M. Roy, and A. Franc. 2002. Simple temporal models for ecological systems with complex spatial patterns. Ecology Letters 5:412–419. Perry, J. N., M. S. Noh, Y. Lee, R. D. Alston, H. M. Norowi, W. Powell, and K. Rennolls. 2000. Fitting host-parasitoid models with
Plugging Space into Predator-Prey Models CV2 1 1 using hierarchical generalized linear models. Proceedings of the Royal Society of London B 267:2043–2048. Rastetter, E. B., A. W. King, B. J. Cosby, G. M. Hornberger, R. V. O’Neill, and J. E. Hobbie. 1992. Aggregating fine-scale ecological knowledge to model coarser-scale attributes of ecosystems. Ecological Applications 2:55–70. Roland, J., and P. D. Taylor. 1997. Insect parasitoid species respond to forest structure at different spatial scales. Nature 386:710–713. Ruel, J. J., and M. P. Ayres. 1999. Jensen’s inequality predicts effects of environmental variation. Trends in Ecology & Evolution 14: 361–366. Ruxton, G. D., and W. S. C. Gurney. 1994. Deriving the functional response without assuming homogeneity. American Naturalist 144: 537–541. Schneider, D. C. 1992. Thinning and clearing of prey by predators. American Naturalist 139:148–160. ———. 2001. The rise of the concept of scale in ecology. BioScience 51:545–553. Schneider, D. C., and D. C. Duffy. 1985. Scale-dependent variability in seabird abundance. Marine Ecology Progress Series 25:211–218. Schneider, D. C., and J. F. Piatt. 1986. Scale-dependent correlation of seabirds with schooling fish in a coastal ecosystem. Marine Ecology Progress Series 32:237–246. Schneider, D. C., J. M. Gagnon, and K. D. Gilkinson. 1987. Patchiness of epibenthic megafauna on the outer Grand Banks of Newfoundland. Marine Ecology Progress Series 39:1–13. Sokal, R. R., and F. J. Rohlf. 1995. Biometry. W. H. Freeman, New York. Sparrevik, E., and K. Leonardsson. 1995. Effects of large Saduria entomon (Isopoda) on spatial distribution of their small S. entomon and Monoporeia affinis (Amphipoda) prey. Oecologia (Berlin) 101: 177–184.
259
———. 1998. Recruitment in the predacious isopod Saduria entomon (L.): alternative prey reduces cannibalism. Journal of Experimental Marine Biology and Ecology 221:1117–1130. ———. 1999. Direct and indirect effects of predation by Saduria entomon (Isopoda) on the size-structure of Monoporeia affinis (Amphipoda). Oecologia (Berlin) 120:77–86. Taylor, L. R., I. P. Woiwod, and J. N. Perry. 1980. Variance and large scale spatial stability of aphids, moths and birds. Journal of Animal Ecology 49:831–854. Tilman, D., and P. Kareiva. 1997. Spatial ecology: the role of space in population dynamics and interspecific interactions. Princeton University Press, Princeton, NJ. Tobin, P. C., and O. N. Bjørnstad. 2003. Spatial dynamics and crosscorrelation in a transient predator-prey system. Journal of Animal Ecology 72:460–467. Underwood, A. J. 1997. Experiments in ecology: their logical design and interpretation using analysis of variance. Cambridge University Press, Cambridge. Van Veen, F. J. F., C. B. Muller, I. C. T. Adriaanse, and H. C. J. Godfray. 2002. Spatial heterogeneity in risk of secondary parasitism in a natural population of an aphid parasitoid. Journal of Animal Ecology 71:463–469. Wiens, J. A. 1989. Spatial scaling in ecology. Functional Ecology 3: 385–397. Wilson, W. G., and L. D. Harder. 2003. Reproductive uncertainty and the relative competitiveness of simultaneous hermaphroditism versus dioecy. American Naturalist 162:220–241. Yodzis, P. 1989. Introduction to theoretical ecology. Harper & Row, New York. Associate Editor: Benjamin M. Bolker Editor: Jonathan B. Losos
The predator, Saduria entomon, and the prey, Monoporeia affinis, are dominating the benthic community of the northern Baltic Sea and constitute a tightly coupled predator-prey system. Photograph by Ulf Bergstro¨m.
