Experimental Design in Aquatic Ecology - Freshwater Ecology: Concepts and Environmental Applications - page 480

to assume that the chemical caused the deaths. A similar release is not likely
to be done on a nearby stream to replicate the event. Likewise, when large-
scale changes are observed in large lakes, river systems, or aquifers there is
no feasible way to perform replicated experiments on these systems, so other
techniques are required. Correlation is one of these techniques.
The primary problem with correlative approaches lies in the inability
to separate cause from effect. A tongue-in-cheek example of this is a plot
of the number of churches per town against the number of saloons per
town for a variety of towns with different populations. The plot will show
a positive correlation. The correlation occurs not because saloons cause
more churches to be built but because larger towns have more churches
and more saloons. Correlation can support causation but does not un-
equivocally prove ecological hypotheses.
A classic example of the power of correlation from limnology is the con-
troversy regarding the causes of eutrophication. More phosphorus certainly
was correlated with greater algal biomass in lakes. However, not until whole
lake phosphorus addition experiments were conducted (see Fig. 17.1) did pol-
luters have a difficult time denying that increased phosphorus led to eutroph-
ication. After the causation was determined, the correlation was more likely
to be used to predict responses of lakes to alterations in phosphorus supply.
Natural experiments can be strengthened in several ways (Carpenter,
1989). Time series can be used to provide replication. For example, if a
lake exhibits one level of fish production for several years, then nitrogen is
added and fish production increases, we can be fairly certain that the ni-
trogen addition caused the increase in production. The certainty is greater
if the nitrogen addition is discontinued and the fish production decreases
to its original level. Another way to strengthen such observations is to
compare two similar, if not identical, ecosystems.
Even given the potential problems, natural experiments have undeni-
able benefits. Perhaps the strongest of these benefits is that the observations
and correlations occur in entire systems, not in a beaker or a bottle. Nat-
ural experiments may be most likely to have relevance to community- or
ecosystem-level processes that operate in the real world (Carpenter et al.,
1995), and the results may be contrary to those extrapolated from small-
scale replicated experiments (Schindler, 1998).
SIMULATION MODELING
An additional method that can be used to explore possible hypotheses
is simulation modeling. Perhaps these could be called virtual experiments.
In this case, computer models of a system with the desired level of detail
and representation of processes are built, and the system can be perturbed
as desired. This is the main approach used to investigate global climate
change, atmospheric dynamics, and large-scale physical oceanography.
Benefits to modeling are that it is cheap, easily replicated, and may indi-
cate critical factors in complex systems that control the observed behavior.
Once the critical factors are determined by modeling, more detailed stud-
ies targeting these critical factors can occur. Thus, modeling can improve
efficiency of environmental research.
The results of such models can be strengthened in several ways. Sensi-
tivity analysis is used to systematically test the sensitivity of the model to