Ecosystem Experiments Stephen R. Carpenter," Sallie W. Chisholm, Charles J. Krebs, David W. Schindler, Richard F. Wright Experimental manipulations of entire ecosystems have been conducted in lakes, catch- ments, streams, and open terrestrial and marine environments. Experiments have ad- dressed applied problems of ecosystem management and complex responses of com- munities and ecosystems to perturbations. In the course of some experiments, environ- mental indicators and models have been developed and tested. Surprising results with implications for ecological understanding and management are common. Predicting responses of ecosystems to per- turbation is among the greatest challenges to ecology. Experiments are necessary, yet many important features of ecosystems, such as wide-ranging predators or large- scale geochemical processes, cannot be in- cluded in small artificial systems. Other fea- tures, such as microbial metabolism or plankton populations, quickly reach unreal- istic levels when isolated in containers. Such difficulties are overcome by direct experimental manipulations of entire eco- systems, including the organisms and the abiotic environment (1). These experi- ments involve large areas, such as catch- ments or the natural ranges of mobile pred- ators, for extensive periods of time (Fig. I). Ecosystem experimentation is field sci- ence. The outdoor laboratory may not be replicable, is not controlled, and is subject to many fluctuations. Experimental design usually involves trade-offs of spatial extent, replication, and duration, all constrained by financial resources. Challenges include the possibility that changes in experimental ecosystems are the result of historical trends or regional patterns rather than the manip- ulation. The value of unmanipulated refer- ence ecosystems as a check for such prob- lems has long been recognized (2). Ideally, long-term pre- and postmanipulation data are gathered for several reference ecosys- tems. Experimental systems should be mon- itored for long enough to assess trends be- fore treatment (3). Manipulations are usually simple, direct, and sustained for long enough to detect changes against background variability. Natural perturbations and management ac- tions can be multifactored, modest, or brief S. R. Carpenter is at the Center for L~mnology, University of Wlsconsln, Madson, W 53706, USA. S. W. Chshom IS in the Departments of Civil and Environmental Engi- neerng, and Biology, Massachusetts Institute of Tech- nology, Cambridge, MA 021 39, USA. C. J. Krebs is n the Department of Zoology, Universty of British Columbia, Vancouver, BC V6T 1W5, Canada. D. W. Schindler is In the Department of Zoology, Unversity of Alberta, Ed- monton, AB T6G 2E9, Canada. R. F. Wright is at the Norwegan nsttutefor Water Research, Box 173 Kjelsas, 041 1 Oslo, Noway. 'To whom correspondence should be addressed 324 and consequently their effects are hard to interpret (4). For example, changes in har- vest regulations that are too small to have detectable effects are a maior source of un- certainty in assessments of fishery manage- ment practices (5). Informative manipula- tions are simple and powerful. Replication is often impossible because of costs, public policy, or uniqueness of the ecosystem to be studied. Ecosystem experi- ments are more often intended to measure responses and their ecological significance than to test null hv~otheses and their sta- , L tistical significance (6). Generalization be- yond a single system depends on knowledge of mechanisms or comparisons of many eco- systems. Repetition of important experi- ments by diverse research groups is a valu- able check on the generality of responses. For example, similar ecosystem experiments have been performed in several nations on lake eutrophication, lake biomanipulation, ecosystem acidification, and effects of forest management on catchment hydrology and biogeochemistry. Comparison of indepen- dent experiments reveals which conclusions are general and which depend on local con- ditions (7). From origins in limnology (8), whole- ecosystem experimentation has been ap- plied to diverse habitats. Initial work fo- cused on biogeochemistry and chemical stressors of ecosystems, but the scope has expanded to include community dynamics and interactions of community and ecosys- tem processes. Terrestrial Ecosystems Catchments are basic units of the landscape (9, 10). Because boundaries can be well- defined, fluxes of water and chemical com- ponents can be measured into and out of the ecosystem. Atmospheric inputs include gases, particles, water, and dissolved chem- icals. Water passes through the terrestrial ecosystem, and is thereby altered in chern- ical composition in ways that can be attrib- uted to known biological and geochemical processes. Runoff from catchments thus in- tegrates the net effect of terrestrial process- SCIENCE * VOL. 269 * 21 JULY 1995 es. Changes in the amount and composition of runoff reflect changes in atmospheric inputs or in the terrestrial ecosystem. The paired-catchment concept is central to whole-catchment research. Typically, runoff is compared from two ecosystems. After a pretreatment study, one catchment is manipulated while a second serves as untreated reference. In a classic experiment at Hubbard Brook Exnerimental Forest, trees were cut and regrokth was inhibited by administra- tion of herbicide, while a refirence catch- ment was undisturbed (9). This manipula- tion drastically altered hydrologic flows, erosion, and biogeochemical' cycles for sev- eral years. Extensive nutrient loss resulted from decom~osition and nitrification. The experiment showed the tight linkage be- tween terrestrial ecosvstems and down- stream aquatic ecosystems. It also led to snecific recommendations of tree harvest practices that would maintain water quality, wildlife habitat, and productive potential of forests ( 11 ). Paired catchments (or paired forest stands) have been used to investigate other environmental processes. The effects of acid deposition on surface water have been re- vealed by large-scale experiments. Among these is the RAIN (Reversing Acidification in Norway) study of the effects of acid deposition on soil and runoff chemistry (12). Exclusion of acid rain from a whole catchment bv means of a roof showed that the effects df acid inputs were reversible (Fig. 1A). A parallel experiment with in- creased inputs of acids to a pristine catch- ment showed that surface waters could be acidified after only a few years of acid dep- osition (Fig. 2). As a scientific experiment, the RAIN project revealed mechanisms and rates of change of runoff and streamwater chemistry after changes in inputs of acidic pollutants (1 3). As an environmental dem- onstration. the RAIN nroiect showed that . acidification of surface waters was due to acid de~osition and th$t abidification could be reversed when thig dsposition was re- duced. These experiments expanded into manipulations of coupled catchment-lake ecosystems and studies of llming and other measures to mitigate acidification (14). Concern about the effects of global change on ecosystems has led to new exper- iments. Facilities of the RAIN project have recently been converted to examine effects of elevated C02 concentration and temper- ature on catchments (15). Increased atmo- spheric CO, concentrations and N deposi-