12 EXPERIMENTAL ECOLOGY Tappa. D. 1965. The dynamics of the association of six limnetic species of Daphriin in Aziscoos Lake. Maine. Ecological Monographs 35:395-423. Taylor, D. J.. and P. D. N. Hehert. 1993. Habitat-dependent hybrid parentage and differential introgression hetween neighboring sympatric Dnphrtin species. Proceedings of the Na- tional Academy of Science of the USA 90:7079-7083. Tessier. A. J., and M. A. Ixibold. In press. Habitat use and ecological specialization within lake Duplrrrin populations. Oecologia. Tessier. A. J.. and J. Welser. 1991. Cladoceran assemblages, seasonal succession and the importance of hypolimnetic refuge. Freshwater Biology 25:85-93. Threlkeld, S. T. 1979. The midsummer dynamics of two Dnphrlin species in Wintergreen Lake, Michigan. Ecology 60:165-179. . 1980. Habitat selection and population growth of two cladocerans in seasonal en- virnnnlents. Pages 346-357 in W. C. Kerfoot (ed.). Evolution and Ecology of Zoo- plankton Communities. University Press of New England, Hanover. New Hampshire. Tilman. D. 1986. A consumer-resource approach to community structure. American Zool- ogist 26:5-22. Underwood, A. J. 1997. Ecological experiment & their logical design and interpretation using analysis of variance. Cambridge University Press, Cambridge. Underwood. A. J.. and P. S. Petraitis. 1993. Structure of intertidal asseinhlages in different locations: how can local processes be compared? Pages 39-51 in R. E. Ricklefs and D. Scliluter (ed.), Species Diversity in Ecological Communities: Historical and Geograph- ical Perspectives. University of Chicago Press, Chicago, Illinois. Vaiice. R. R. 1978. Predation and resource pirtitioning in one predator-two prey model communities. American Naturalist 1 12:797-813. Werner, E. E.. and J. E. Gilliam. 1984. The ontogenetic niche and species interactions in size-structured populations. Annual Review of Ecology and Systematics 15393425. Wilson, D. S. 1988. Holism and reductionism in evolutionary ecology. Oikos 53:269-273. Woltereck, R. 1932. Races, associations and stratification of pelagic daphnids in some lakes of Wisconsin and other regions of the United States and Canada. Transactions of the Wisconsin Academy of Sciences 27:487-522. Wootton. J. T. 1994. Predicting direct and indirect effects: an integrated approach using experiments and path analysis. Ecology 75: 151-165. Ziv. Y., Z. Abramsky, B. P. Kotler, and A. Subach. 1993. Interktence competition and tcinporal aiid hahitat partitioning in two gerbil species. Oikos 66:237-246. 6 Experimentation, Observation, and Inference in River and Watershed I n vest i ga t i o n s MARY E. POWER, WILLIAM E. DIETRICH, & KATHLEEN 0. SULLIVAN Ecologists seek to understand the interactions between species and their heterogeneous, changing environments. This work is made more difficult by the fact that the underlying processes are often mediated through complex community- or ecosystem-level inter- actions. What tools do we have for unraveling this complexity? Paine (1994) and Walters (1986, 1992) have reviewed the three fundamental meth- ods available to field scientists: observation, modeling, and experimentation. All new information about nature is initially obtained through the first method: observation, which includes monitoring. mapping, and detecting correlations or other patterns. Ob- servation alone, particularly when done in a hypothesis-free manner, has in the past disappointed ecologists by leaving them with messy data sets open to alternative inter- pretations. For this reason, ecologists in recent decades considered inferences froni observations to be weak relative to inferences drawn from manipulative experiments (e.g.. Connell 1974). We believe that field observations have been undervalued in con- temporary community ecology, leaving ecologists poorly equipped to contrihute to problems at large spatial scales at which manipulative experiments are infeasible. We will elaborate on this point later. A second fundamental method for studying nature is modeling. We rekr here to mechanistic modeling, either mathematical or qualitative, which attempts to portray key process that underlie phenomena of interest. The limitations of modeling are well known (e.g., Starfield and Blelnch 1986. Walters 1992). There is usually an assumptioil that the system's context is constant. and this is never true of real ecosystems. Also, mod- elers must assume that many (most) details are unimportant, but inevitably wine omitted details will be important, probably more so than others chosen for representation in the model. Nevertheless, as Walters (1992) points out, modeling is unavoidahle. If we have an idea about how our system works, we have a model of it. Therefore. he advises that we model openly, making assumptions explicit to ourselves and others. 114 EXPERIMENTAL ECOLOGY The third method is experimentation. delined here in the narrow sense that com- i nltlnity ecologists typically use. Experinlellts (sensu strictu, Paine 1994) involve study or replicated sainple units which are suhject to at least two treatments. One treatment is ;I control intended to represent the unmanipulated or background condition. In the other treatment(s), one or more factors are altered by the experimentalist. and their ilifltlence is evaluated by comparing responses of manipulated to control treatments. Replicated experiments cannot be performed in many situations, either because of lo- pistical constraints (Matson and Carpenter 1990) or because adequate controls do not C` - - exist (see discussion of scale issues following). Combined or Neslcd Approaches While niuch has been written about the greater power or rigor of experilnental over observational approaches (e&. Paine 1977, Underwwd 1990), it is usually more powerful to combine these approaches in a nested fashion (Fig. 6-1; see also Frost et 11. I988:2S2. Carpenter 1996). How experiments, ohservations. and modeling are combined depends on the scale of the stcdy and the question addressed. A question that has served as an extremely productive opening gambit for community ecologists as they first explore a system has been: "What would happen if. . . ?" (Fig. 6-la): "What would happen if I alter the density of species A or change factor B?" moti- vating what Art Dunham has called "kick it and see" experiments. This is the 1 / a. `What would happen if....?' \ Exploratory manipulation 7 Inference, Prediction \- work the \ same way?' b. `Does this new \- System \ hypothesis testing Ex trapolation , (observation- c-1 C. `Is there quantitative agreement with predictions?` I \ Calibration. validation Figure 6-1 . Nested experiiiiental, observational. and modeling approaches to ask three types of questions. RtVFR AND WAFERSHED INVESltGhTtONS 115 approach that has revealed important surprises. such as keystone species (Paine 1966, 1969). When experimeiits like these are done in intertidal systems on exposed rocky shores, as were the experimental removals of the starfish Pisnsier that led to the original for- niulatioii of the keystone concept (Paine 1966, 1969). direct observations of underlying processes are difficult or inipossihle. Intertidal "action" (graz,ing, predation. growth, settlement, export, and reproduction) typically happens under conditions inconvenient for hiiinan observers (e.g., crashing surfs on moonless nights). Inferences must he drawn froni periodic ohservntions (typically at roughly monthly intervals) of changes in the states of assemblages. These interpretations are bolstered by knowledge of the local hiology and the physical environment. When investigators are not able to observe the underlying processes in action, however, uncertainty may arise as to which compinents of the excluded biota were responsible for treatment effects (e.g., Edwards et al. 1980, Menge 1980, Underwood and Fairweather 1986) or whether alteration of consumer densities or behavior (Menge and Sutherland 1976, Menge 1980) or unintended habitat modifications (Virnstein 1978. Dayton and Oliver 1980, Hulberg and Oliver 1980) have caused or contributed to changes. Consequently, questions and controversies nver the interpretations of experimental results persist. Direct observations, when possible, can illuminate experimental hlack hoxes, retlric- ing the danger of misinterpreting experimental results. They are no panacea. given the problem of witnessing, let alone sampling, rare events with high impacts. But even in what would seem unlikely arenas, such as bnttle experiments with microorganisms, direct ohservations have illiiminafed causality. Gause ( 1934) directly ohscrved the spil- rial sepnration of two competing Pnrnrnecirtnr species and their food resources (sus- pended bacteria vs. deposited yeast cells) and deduced his famous principle that this separation contributed to their coexistence. More recently, BalciOnas and Lawler (1995) used direct microscopic observation to detect an escape in size hy a prey protozoan, Colpidiicrrr, from its predator Euplafes. which occurred when nutrient levels were in- creased in bnttle experiments. Their observation uncovered the mechanisni by which nutrient addition blocked top-down food chain control and shortened the length of the functionally important food chain. in contrast to previous predictions from simple food chain considerations (e.g., Lindeman 1942, Fretwell 1977, Oksanen et al. I98 I). Clearly. direct observations can lead ecologists to both propose ecological geeneraliza- lions and question them. When we know more about a system, we can work within the framework of models, which are hypotheses ahout how the system works (Fig. 6-lh,c). For example. we can attempt to extrapolate. We might observe that a different system shares features with one that has heen partially understood and ask whether the new system works the same way. We may instead be interested in whether the previously studied system will con- tinue to work the same way under new conditions and whether our understanding is rohiist heyond the circumstances in which it was first attained. When we have a niodel of how the system works in mind, we can nest both observations ilnd experiments within this niodel to test it (Fig. 6-lb). If a model has been developed to the poi111 of making quantitative predictions, we can also use nested experiments and observations to calibrate it (Fig. 6-IC) and eventually to validate it (to test the match between pre- dictions of a fully calibrated model and observations from nature). Nested experimental 116 EXI'ERIMENTAL ECOLOGY niaiiipulations may be needed if parameters require calihration under a range of partially controlled conditions. We will illustriite these three types of nested approaches with case histories drawn from river food web investigations and then discuss constraints on the application of these methods as the spatial extent of the system under study increases. Nested Experimental and Observational Studies of River Food Webs "What happens if. . . I": The Eel River of Northern California In the summer of 1989, Power experimentally manipulated fish in the South Fork Eel River to ask: "What happens if.. . the two most common species are excluded?" Enclosures and exclosures were distributed over a I-km reach within the forested wa- tershed of the Angelo Coast Range Reserve (formerly the Northern California Coasf Range Preserve). Only two fishes-juvenile steelhead (Oncorlryrrclrrts triykiss) and CaI- ifornia roach (Hcspcrolectccrs syttirrierrirus)-are common after winters with normal scouring floods. Surprisingly dramatic differences arose between fish enclosures and exclosures 5 to 6 weeks after the onset of the experiments. In the presence of fish, the dominant alga. C/OdO/dlfJ~fl, which grew as 40 to 60cm high turfs attached to boulder and bedrock at the start of the experiment, had collapsed down to a prostrate webbed mat no more than I to 2 ctn high. The algae remained erect in the fish exclosures and became overgrown with nitrogen-fixing bluegreens and diatoins (Power 1990). Densi- ties of benthic insects differed niarkedly between treatments as well. The fish enclosures were heavily infested with midges, fsefidocl,irorrornus riclmrSoni, that lived within the algae and wove it into tufts around its body. Heavy infestations of tuft-weaving midges collapsed the algal mats and produced a webbed and knotted architecture. This occurred several weeks later in the open river. Tuft-weaving midges were rare in the fish exclosures, where large numbers of small predators (lestid nymphs, sialid larvae, and young-of-the-year roach and stickleback) had recruited. These small predators were rare in the open channel and in fish enclosures but recruited in large numbers where larger fish were excluded and apparently suppressed the midge. Power tested this last inference with a nested experiment and direct observations. She stocked 24 screened (I-mm mesh) buckets with ca. 7 g of cleaned Clndoplrorfl (picked free under IOX magnification of conspicuous macroinvertebrates). Six huckets received four roach fry, six received four stickleback fry, six received four lestid nymphs, and six were left as predator-free controls. After 20 days, the predator-free controls had been colonized by four times more midges than had any of the predator treatments (Power 1990). This nested experiment supported the interpretation that it was the guild of small predators that had, in fact, suppressed the recruitment of tuft- weaving midges to fish exclosures. Subsequent direct observations of feeding hy larger fish and invertchrate predators revealed that the comnion predatory invertebrates (lestid damselflies, aeshnid dragonflies, and naucorid bugs) a11 are able to detect iiiidges and extract them from their algal tufts. The odonates, after watching tufts for several minutes. shot their mouthparts in and extracted midges with a "surgical strike." The RIVER AND WAIERSHFII INVESTIGATIONS 1 17 naucorid hugs probed cocoons with their beaks until they encountered, the midge. In contrast. the larger fish in the Eel River did not seem able to detect chironomids within their algal tufts. although when midges were extracted and exposed these fish ate them readily (Power et al. 1992). These behavioral observations documented the predator- specific vulnerability of the prey, which was the mechanism that produced four func- tionally significant trophic levels in the El River. Herbivorous mayflies were the dominant prey in the guts of the larger fishes (Power et al. 1992). Observations of gut contents alone would have suggested that fish should exert control from the third trophic level and enhance, rather than suppress, plant biomass. Clearly, a combination of ex- perimental and ohservational results produced a better Understanding of food web in- teractions than either approach would alone, but at this early stage of investigating a poorly understood system experiments were particularly critical. "Can we extrapolate?" The Eel River during Drought and Brier Creek, Oklahoma Power and colleagues repeated these fish manipulations during the summers of 1990 and 1991, doubling the enclosure numbers from 12 to 24 and expanding tlie design to study the separate as well as the combined effects of roach and steelhead. In contrast to the 1989 results, however, fish had no functionally significant impacts on algae in either 1990 or 1991. In both the presence and the absence of fish, algae collapsed down to detritus within the first weeks of the experiment. A multiyear drought had begun in 1990. and in the absence of scouring floods large numbers of armored nntl sessile invertebrate grazers, invulnerable to most predators in the river, survived over the win- ter. When Clndophora began to grow in the late spring, these grazers quickly nibbled it back. These natural history observations finally motivated the definitive experinlent, which was a 2 X 2 factorial manipulation of steelhead and the dominant armored caddisfly. Dicosnioecrts gilvipes. The results showed that Dicosriiorc.rrs, not fish. con- trolled algal biomass during drought. Steelhead still had a statistically significallt neg- ative effect on algae (suggesting they were still at the fourth trophic level), hut their effect was small in comparison to the two-level impact of the predator resistant grazers (Woolton and Power, unpublished data). Cross-watershed surveys of algae and inver- tebrates in two regulated channels with artificially stabilized flow and four unregulated rivers that all scoured in 1989 were also consistent with the inference that scouring floods reset primary consumers to earlier successional stages that are more vulnerahle to predation and set the stage for trophic cascades that affect algal hiomass (Power 1992). The Eel River food web obtained under "noriiial" Mediterranean winter-flood, sunimer-drought conditions did not extrapolate to the same system during drought (Power 1995) or to regulated channels in the region that had heen subject to anthro- pogenic "disturhance removal experiments." A food chain tliat had four functional trophic levels with respect to impacts of predators mediated through consumers on plants collapsed to a drought food chain with two functional trophic levels (Fig. 6-2). despite the fact that all the key species were still represented in tlie coniiiiiinity. Extrapolation is useful even when it fails, because expectations that one system may resemble another "prepare the mind" to make focused observations tliat eithcr support a RIVER AND WATERSHFD INVESTIGATIONS 119 or refute the expectations. In contrast to the attempt to extrapolole frolii flood to tlrouglit years for the Eel River web, properties of a fond web in an Oklahoma prairie stream were predictable by analogy with subtidal food webs in the northeastern Pacific. where sea otters suppress sea urchins and indirectly maintain kelp forests (e.g.. Estes and Pnlniisano 1974). In Brier Creek, Oklahoma, some pools were filled with filanientous green algae, while adjacent pols were nearly barren. Observations of the distributions of predators and grazers quickly confirmed expectations from the sea otter-urchin-kelp model: piscivorous hass occurred in the green pols and were absent from the barren pnols. where schools of grazing minnows (Carrrposiornn nrrorrmlrorr) occurred (Power and Matlhews 1983). Subsequent experimental transfers of bass and Corrrpo.tiortrn among stream pools demonstrated that a trophic cascade did nnderlie the complemen- tarity of bass, Cnrnposrortro, and algae. We electroshocked bass out of a green pool. split it down the middle, and added Cnmnposfornn to one side; within S weeks, algae on that side were grazed down to a barren state, while the control side without the grazing minnows remained green (Power et al. 1985). Clearly, herbivory accounted for the barren condition of pools with Cnrnposmraa. It was not obvious, however, whether the mechanism for the complementarity between bass and their minnow prey was pre- dation or predator avoidance because, unlike algae, minnows have potential escape behavior. We resolved ;his question about causality with direct observations. We added bass to a pool with a school of Camposrontu but, before doing so, fenced off the upstream and downstream ends of adjacent pools, which were linked to the Cortrpo.tr~rtto pool by riffles which minnows could cross but which were too shallow for bass passage. These fenced areas served as potential "escape ports" for Cnrirlmiorrtn. We also grid- ded the siibstrate of the Cumposrorrrn pool and, before bass addition, niade behavioral observations of space use by adults and, during a spring repetition of the experiment, by young-of-the-year minnows. The adult fish tended to graze the deepest parts of the pool, with the young in slightly shallower water. After bass addition. both size classes moved into shallower water. which accounted for transient dynaniics in the spatial distribution of the algae. which initially declined in these shallow areas (Power et al. 1985, Power 1987). Over the next 5 or 6 weeks, however, the entire pool became overgrown with green algae. Predator avoidance was an important contributing factor: in the spring experiment, we found 40 of the initial 74 adult minnows in the upstream escape port just I week after bass had been added. Whether bass convert C`nrrt/w.