20. Biotic Manipulations of Aquatic Ecosystems
that the planktivores are removed (Hansson et al.
1998).
Most biotic manipulations of aquatic ecosystems
have been performed to estimate the magnitude and
variability of responses (i.e., ecological significance) rather than to test null hypotheses (i.e., statistical significance) (Carpenter et al. 1995a). Although it would be valuable to repeat the same
manipulation in several similar ecosystems (e.g.,
Bystrfim et al. 1998) this type of replication is often
impossible. It is possible to measure the response
in one manipulated ecosystem (Stewart-Oaten et al.
1986; Carpenter et al. 1989; Carpenter and Kitchell
1993a). To check alternative explanations for the
results, reference systems or additional manipulations should be included in the experimental design
(Stewart-Oaten et al. 1986; Walters et al. 1989).
In some situations, the null hypothesis of no response must be carefully considered. These situations raise questions of statistical power, or the
probability of detecting an effect that genuinely exists (McAllister and Peterman 1992). Statistical
power depends on the variance of the dependent
variables, the minimum magnitude of response that
the experiment is intended to detect, and number of
replicates. Replication may occur in space (by manipulating several ecosystems) or in time (StewartOaten et al. 1986; Walters et al. 1989, McAllister
and Peterman 1992). Although consideration of statistical power is becoming more prevalent in the
planning of replicated, small-scale experiments,
statistical power has rarely been estimated for replicated ecosystem experiments. Where spatial replication is possible, two studies suggest that about
10 lakes (5 treatment, 5 reference) provide adequate
power for experiments on trophic cascades and effects of habitat manipulation on fish growth (Carpenter 1989; Carpenter et al. 1995b). In cases in
which temporal replication is employed, long time
series before and after press treatments will maximize the prospects for detecting any effects that
occur. Sequences of manipulation may increase the
power. We are not aware of any formal power calculations for time series analyses of ecosystem experiments. Experience suggests that effects can be
detected with time series of about 80 or more observations, provided that autocorrelation is small
(Carpenter et al. 1989; Carpenter and Kitchell
1993b). In some situations, similar ecosystem experiments have been performed by different re313
search teams on very different ecosystems. Joint
analyses of these similar experiments (metaanalyses) could be very instructive.
Future Prospects
Field Guide to Keystones
Keystone species are central to understanding how
ecosystem structure is linked to processes. Power
et al. (1996) emphasize that keystones cannot be
identified reliably on the basis of biomass, productivity, nutrient content, or other indices of dominance. The field mark of a keystone is high sensitivity of ecosystem processes to changes in the
keystone's abundance. They define a keystone species as one "whose impact on its community or ecosystem is large, and disproportionately large relative to its abundance" (Power et al. 1996). A
number of attributes of keystones are identifiable
from basic ecology and natural history (Jones and
Lawton 1995; Power et al. 1996). However, field
experiments are necessary to determine which species are keystones and to measure their effects
(Paine 1980). Keystone species will often be wideranging, long-lived species that control, or are controlled by, spatially extensive processes or processes with long turnover times. Impacts of
keystone species are often transmitted through
complex, indirect effects among interacting species
(Power et al. 1996). Because indirect effects are
often mediated by the behaviors of wide-ranging
species and by habitat heterogeneity (Wootton
1994), many keystone interactions are impossible
to study in small-scale or brief experiments. Ecosystem manipulation is the only real option for
studying such keystone species.
Humans and Ecosystems
Humans are usually ignored or treated only as independent variables (sources of pollutants or agents
of perturbation) in ecosystem experiments. In fact,
humans are a dominant component of most ecosystems and interact (act as both independent and
dependent variables) with other system components
(McDonnel and Pickett 1993).
Some of the best examples of people as interactive ecosystem components come from fisheries
that the planktivores are removed (Hansson et al.
1998).
Most biotic manipulations of aquatic ecosystems
have been performed to estimate the magnitude and
variability of responses (i.e., ecological significance) rather than to test null hypotheses (i.e., statistical significance) (Carpenter et al. 1995a). Although it would be valuable to repeat the same
manipulation in several similar ecosystems (e.g.,
Bystrfim et al. 1998) this type of replication is often
impossible. It is possible to measure the response
in one manipulated ecosystem (Stewart-Oaten et al.
1986; Carpenter et al. 1989; Carpenter and Kitchell
1993a). To check alternative explanations for the
results, reference systems or additional manipulations should be included in the experimental design
(Stewart-Oaten et al. 1986; Walters et al. 1989).
In some situations, the null hypothesis of no response must be carefully considered. These situations raise questions of statistical power, or the
probability of detecting an effect that genuinely exists (McAllister and Peterman 1992). Statistical
power depends on the variance of the dependent
variables, the minimum magnitude of response that
the experiment is intended to detect, and number of
replicates. Replication may occur in space (by manipulating several ecosystems) or in time (StewartOaten et al. 1986; Walters et al. 1989, McAllister
and Peterman 1992). Although consideration of statistical power is becoming more prevalent in the
planning of replicated, small-scale experiments,
statistical power has rarely been estimated for replicated ecosystem experiments. Where spatial replication is possible, two studies suggest that about
10 lakes (5 treatment, 5 reference) provide adequate
power for experiments on trophic cascades and effects of habitat manipulation on fish growth (Carpenter 1989; Carpenter et al. 1995b). In cases in
which temporal replication is employed, long time
series before and after press treatments will maximize the prospects for detecting any effects that
occur. Sequences of manipulation may increase the
power. We are not aware of any formal power calculations for time series analyses of ecosystem experiments. Experience suggests that effects can be
detected with time series of about 80 or more observations, provided that autocorrelation is small
(Carpenter et al. 1989; Carpenter and Kitchell
1993b). In some situations, similar ecosystem experiments have been performed by different re313
search teams on very different ecosystems. Joint
analyses of these similar experiments (metaanalyses) could be very instructive.
Future Prospects
Field Guide to Keystones
Keystone species are central to understanding how
ecosystem structure is linked to processes. Power
et al. (1996) emphasize that keystones cannot be
identified reliably on the basis of biomass, productivity, nutrient content, or other indices of dominance. The field mark of a keystone is high sensitivity of ecosystem processes to changes in the
keystone's abundance. They define a keystone species as one "whose impact on its community or ecosystem is large, and disproportionately large relative to its abundance" (Power et al. 1996). A
number of attributes of keystones are identifiable
from basic ecology and natural history (Jones and
Lawton 1995; Power et al. 1996). However, field
experiments are necessary to determine which species are keystones and to measure their effects
(Paine 1980). Keystone species will often be wideranging, long-lived species that control, or are controlled by, spatially extensive processes or processes with long turnover times. Impacts of
keystone species are often transmitted through
complex, indirect effects among interacting species
(Power et al. 1996). Because indirect effects are
often mediated by the behaviors of wide-ranging
species and by habitat heterogeneity (Wootton
1994), many keystone interactions are impossible
to study in small-scale or brief experiments. Ecosystem manipulation is the only real option for
studying such keystone species.
Humans and Ecosystems
Humans are usually ignored or treated only as independent variables (sources of pollutants or agents
of perturbation) in ecosystem experiments. In fact,
humans are a dominant component of most ecosystems and interact (act as both independent and
dependent variables) with other system components
(McDonnel and Pickett 1993).
Some of the best examples of people as interactive ecosystem components come from fisheries