Hvdraulic Food-Chain Models An approach to the study of food-we6 dynamics in large rivers Mary E. Power, Adrian Sun, Gary Parker, William E. Dietrich, and J. Timothy Wootton o habitats on Earth are more dynamic than floodplain riv- N ers. As large rivers flood and ebb, their floodplains and off-river water bodies are alternately con- nected and disconnected from the main channels. The area of inun- dated landscape can increase by two to three orders of magnitude from low to high water, and the surface area available for aquatic biological activity increases even more, because when floodplains are inundated, so are their grasses, trees, and the masses of dead organic material stored in these habitats. The enor- mous fluctuations in resources and habitats that accompany these flood pulses have profound effects on river-floodplain ecosystems (Bay ley 1989, Junk et al. 1989, Sparks 1992, Welcomme 1985), whose constitu- ent biota have a variety of feeding and life-history adaptations for sur- Mary E. Power is an associate professor in the Department of Integrative Biol- ogy, University of California at Berke- ley, Berkeley, CA 94720. Adrian Sun is a graduate student in the Department of Zoology NJ-15, University of Washing- ton, Seattle, WA 98195. Gary Parker is a professor at the St. Anthony Falls Hydraulics Laboratory, University of Minnesota, Minneapolis, MN 55414. William E. Dietrich is a professor in the Department of Geology and Geophys- ics, University of California at Berke- ley, Berkeley, CA 94720. J. Timothy Wootton is an assistant professor in the Department of Ecology and Evolution, University of Chicago, Chicago, IL 60637. 0 1995 American Institute of Biological Sciences. Simulations linking physical and biological processes may guide research on ecological effects of massive changes to rivers and floodplains viving and exploiting the fluctua- tions (Adis and Mahnert 1986, Goulding 1980, Junk et al. 1989, Lowe-McConnell 1975). Floodplain rivers are not only dynamic but also spatially, hydro- logically, and biologically complex (Figure 1; Junk et al. 1989, Wel- comme 1985). Ecologists have long recognized that spatial heterogene- ity and temporal fluctuation can play strong roles in maintaining the rich- ness and complexity of ecological communities. In heterogeneous, fluc- tuating environments, consumers are less likely to overeat and extermi- nate their prey (Hastings 1977, Huffaker 1958). Competitors that dominate under particular condi- tions are likely to lose their perfor- mance advantage before they can exclude lesser competitors (Connell 1978, Hutchinson 1961, Tilman 1994). Therefore, the hydrologic fluctuations that impose huge mor- tality on river biota (Welcomme 1985) may, paradoxically, enhance the persistence of ecological com- munities by reducing the chances that their constituent populations will go extinct (Sparks 1992, Welcomme 1985). There is growing appreciation of the need to understand and predict the responses of river ecosystems to their massive rearrangement by hu- mans. Nearly all of the major flood- plain rivers in the northern hemi- sphere, and many in the southern hemisphere, have been leveed and1 or impounded for navigation, agri- cultural development, power gen- eration, or flood control. The 1993 flooding of the Mississippi rekindled the national debate over what is to be gained by increasing the height and extent of levees and what has been lost (Belt 1975, Leopold and Maddock 1954, Mairson 1994). Levees cut off large rivers from their floodplains; dams and diversions artificially stabilize flows, eliminat- ing the natural flood cycles. These changes can be thought of as large- scale experiments that remove spa- tial and temporal heterogeneity, but they have been underexploited by ecologists, who could use them to study the importance of flood re- gimes for river and riparian ecosys- tems (Sparks et al. 1990). Whether rivers are restored by being recon- nected to their historic floodplains or continue to be disconnected, tools are needed for predicting the re- sponses of fisheries, nuisance spe- cies, and endangered species to hy- drologic manipulation. Even when the focus is on target species rather than whole ecosystems, a food-web perspective is necessary, because the March 1995 159 Figure 1.The Fly River of Papua New Guinea has a turbid (light brown) main channel and numerous, clear (dark) off-river water bodies. The channel is 200 m wide. population dynamics of any species depends critically on how its re- sources, prey, and potential preda- tors also respond to environmental change. Hydraulic food-chain models We are developing a modeling ap- proach that links the relatively well- understood responses of river width, depth, and velocity to changes in discharge and the poorly understood responses of river biota to these hy- draulic parameters. The food webs are modeled as modified Lotka- Volterra equations for food chains with three trophic levels and two energy sources-detritus and veg- etation. We recognize that the strengths of interaction between or among the trophic levels are modu- lated by the hydrologic changes ac- companying the flood pulse. Our purpose in this article is to illustrate a conceptual framework that, when tailored to specific ecosystems, can guide field observations and mea- surements. As a first exploratory step, we have aggregated entities in a hypo- thetical river food web into four functional groups intended to rep- resent the dominant consumers and resources in a river food chain. De- tritzis comprises dead plant mate- rial, both litter introduced into the river from riparian tropical forests or grassy floodplains and dead aquatic vegetation. Vegetation is defined as living aquatic plants or algae. Some of this vegetation floats Cross section a. Floodplain river (e.g., floating macrophytes, which are abundant in some tropical chan- , nels; Junk et al. 1989); some grows on and detaches from river sub- strates. Herbivore-detritivdres or grazers feed on both vegetation and detritus and are envisioned in this case as being weak swimming, poorly streamlined organisms like prawns or crayfish (e.g., Macrobrachiurn spp. and Orconectes spp.). (Where fishes are dominant grazers, as in many tropical rivers, assumptions about grazer hydrodynamics would clearly change.) Predators are con- sidered to be large fish, such as bass or baramundi (Micropterus spp., Lutes calcifer), which are powerful swimmers and voracious consumers of animal prey. While issue can be taken with the simplifications and assumptions underlying our physi- cal and particularly our biological approximations, we view them as examples that should be switched or expanded when the model is tai- lored to fit a specific river ecosys- tem. In this article, we use the model to explore how the fundamental tem- poral and spatial features of flood- Hydrograph 21 b. Leveed river Figure 2. Cross-sections through idealized floodplain (a) and leveed (b) river channels, with corresponding hydrographs for channels with unregulated, sinusoi- dal flow. Dashed lines indicate channel depth at maximal and minimal discharge in the floodplain river; maximal and minimal discharge in the leveed river with an unregulated hydrograph; and chronic depths with artificially regulated low (Q,,,,,, = 100 m3/s) and average (Q,,, = 450 m3/s) discharge. 160 Bioscience Vol. 45 NO. 3 plain rivers influence the dynamics of our food chain. Hydraulic relationships Although floodplain rivers have complex morphologies, with off- river water bodies that vary in area, depth, shape, elevation, and fre- quency of connection to the chan- nel, we assume in this article that the single main channel and its flood- plain are simple rectangles in cross- section (Figure 2). We consider three cases: a natural river with access to its floodplain; a leveed river cut off from its floodplain by levees that retain high flows; and a river with artificially stabilized flow that never exceeds the tops of its bank, as might occur downstream of a diversion routing water out of the channel or as a regulated release stream from an impoundment. A comparison of the floodplain river and the leveed river with the same hydrograph iso- lates the effect of habitat expansion and contraction on modeled food- chain dynamics. The influence of temporal hydrologic fluctuation is shown by comparing food chains in the leveed river with a flood cycle and in the same channel with artifi- cially stabilized discharge. In the model floodplain river, dry- season flow is entirely contained within the main channel. When the depth of rising water in the channel exceeds bank-full depth, the flow spills over and instantly inundates the entire floodplain. Flow is con- veyed downstream rapidly through the channel and, after spillover, much more slowly over the flood- plain, where it is impeded by rough- ness from vegetation and organic detritus. The difference in flow rates between channel and floodplain can be calculated from empirical equa- tions relating the roughness of the channel to the flow velocity. Chan- nel and floodplain velocities deter- mine how the total discharge is apportioned between these two hab- itats (Henderson 1966, Parker 1993). Total discharge varies seasonally. We use an oscillating sine wave to represent a 12-month cycle with one dry season and one rainy season. In the natural river with access to its floodplain (Figure 2a), most of the Figure 3. Baramundi (Later calcifer) could be a top predator in the food web modeled in this article. The fish is being displayed by the true top predator in the Fly River food web. increase in discharge during the rainy season is absorbed by the huge ex- pansion of the river's width. In the leveed channel (Figure 2b), width cannot change, and changes in dis- charge are apportioned between river depth and flow velocity. Leopold et al. (1964) discuss the hydraulic ge- ometry of rivers (empirical relation- ships of width, depth,and velocity with discharge). Trophic relationships Like their physical setting, trophic networks in rivers are complex and dynamic. Trophic linkages between species form, break, and change in strength as environmental condi- tions change (Power 1992a, b). Food chains and aggregated functional groups are extremely simplified ab- stractions that ignore much of the real complexity of food webs. Ex- perimental and comparative stud- ies, however, have suggested that, at least in smaller rivers, there are chains of strong interactions within food webs that dynamically link predators through consumers to plants and detritus (Bowlby and Rolf 1986, Northcote 1988, Perrin et al. 1987, Power 1990, Power et al. 1985, Wootton and Power 1993). Evidence for the importance of strong chains through complex webs comes from trophic cascades, in which removal or reductions of predators release consumers, which in turn suppress populations of their own resources, producing alternat- ing release and suppression of trophic levels that often reach down to primary producers. Trophic cas- cades have been documented in lakes (Carpenter et al. 1985), subtidal marine habitats (Estes and Pal- misano 1974), and terrestrial com- munities (Kajak et al. 1968) as well as in rivers. In developing our hydraulic food- chain model, we initially focus on a three-level food chain with both detrital and producer energy sources, in hopes that interactions of these elements with each other and with their physical environment will cap- ture much of the dynamics to which other components of the river eco- system are entrained. Equations for biomass dynamics of each of the four trophic elements are tied to channel hydraulics because key pa- rameters in these equations are made functions of width, depth, velocity, or a combination of these variables (see box page 163 andTables 1 and2). Detritus standing stocks increase as terrestrial plant litter falls into the channel or inundated floodplain and as vegetation growing within the river dies. Detritus is lost to March 199s 161 Trophic dynamics Hydraulic geometry I Vegetation Climate Geology nd use Figure 4. Overview of causal linkages assumed in the hydraulic food-chain model (diagram from Stella simulation software). In natural channels, local geology constrains channel and floodplain width by determining factors like bank cohe- siveness or location of mountain ranges or terraces bordering the active flood- plain. Alternatively, channel width can be determined by human land use, if levees have been constructed. In natural channels, climate (precipitation) controls discharge. Alternatively, discharge can be artificially regulated by humans by means of upstream diversion or retention structures. Discharge is divided between depth and velocity while flow is confined to the channel and among width, depth, and velocity after spillover in channels with access to their floodplains. Parameters in the trophic equations are functions of hydraulic variables (Table 1). A variety of alternative functional dependencies of trophic parameters on width, depth, velocity, or interactions of these variables could be important in particular food webs. grazers at a rate determined by their density and that of the detritus, as well as by the grazer's per capita ingestion rate of detritus. For sim- plicity, we have initially assumed constant per capita ingestion rates by all consumers and predators. In- gestion rates may, however, slow as consumers satiate, particularly if resource availabilities increase abruptly relative to consumer densi- ties. In this case, rates are better modeled with a saturating function such as the Type I1 functional re- sponse of Holling (1959). Detritus is also diminished as carbon is re- spired to the atmosphere as carbon dioxide. (Losses from the channel reach by downstream flushing of organic matter or living components of the food chain are assumed to be balanced by material washing into the reach from upstream, so outwash does not change local standing stocks.) Vegetation renews by logistic growth until it becomes self-limit- ing, for example, due to self-shad- ing, at a density equal to the environment's carrying capacity. Vegetation that dies without being grazed increases the detritus. (Mas- sive, abrupt die-off of aquatic veg- etation may accompany reconfine- ment of the channel. The amount of vegetation that becomes available as detritus to the river food web depends on factors, such as tem- perature, that control terrestrial decomposition and export of this material during the low-water pe- riod; Bayley page 153 this issue.) Grazers convert vegetation or detritus to offspring with an effi- ciency we assume here to be equal. although it could often be lower for detritus, whose food quality depends on the activity and abundance of associated microbial flora (Cummins 1973, Petersen et al. 1989). Grazers are killed by predators or die of other causes. Predators create off- spring from their prey and, being at the top of the food chain, suffer only nonpredatory mortality in the model presented here. (Clearly, human fish- ing adds another functionally sig- nificant trophic level in many rivers; Figure 3.) We assume that plants, but not grazers or predators, have growth rates that are sufficiently rapid, rela- tive to time scales of seasonal envi- ronmental change, to attain densi- ties at which competition has dynamic significance. This assump- tion can be modified to portray eco- systems in which competition oc- curs within other trophic groups. Details of our equations and param- eter values will be presented else- where.' Linkage of hydraulic and trophic dynamics In mathematical models, trophic interactions have often been investi- gated as somewhat disembodied entities, detached from realistic physical or temporal settings (but see Crowley 1978, DeAngelis 1992, Hastings 1977, Holt 1985, Oksanen 1990, Wootton and Power 1993). Most real food webs, however, oc- cur in patchy, gradually changing, or periodically disturbed environ- ments. Our model incorporates both abrupt and gradual changes affect- ing trophic interactions and biomass dynamics in one (leveed or regu- lated channel) or two (floodplain and river) habitats. Because we be- gin here with a one-dimensional model portraying large-river hydrau- lic and trophic dynamics at a single cross-section, our current model em- phasizes temporal dynamics rather than spatial heterogeneity. In the future, it is likely to be important to incorporate more spatial heteroge- neity by representing off-river water bodies important in the function of 'M. E. Power, G. Parker, W. E. Dietrich, and A. Sun, 1995, manuscript in preparation. 162 Bioscience Vol. 4.5 No. 3 river-floodplain ecosystems (Bay ley 1989, Junk et al. 1989, Sparks 1992). An overview of causal linkages assumed in our model is depicted in Figure 4. Local geomorphology (e.g., where terraces or mountains con- fine the floodplain) and land use (e.g., whether or not levees have been constructed) determine flood- plain width. Climate (e.g., precipi- tation) and land use (e.g., water storage or diversion) govern dis- charge. Width, depth, and velocity vary with discharge. How these vari- ables adjust to discharge depends on the cross-sectional dimensions of the channel and floodplain (Figure 2). Width, depth, and velocity influ- ence trophic dynamics by affecting key parameters in the biomass bal- ance equations (see box this page and Table 1) and by affecting the seasonal access of biota to the flood- plain in unleveed channels. Floodplain and channel habitats in unleveed rivers are connected and disconnected abruptly when flow depth spills overbank or ebbs below bank-full depth and is reconfined in the channel. At spillover, the large standing stock of dead plant mate- rial on the floodplain becomes avail- able to river grazers. After spillover, these grazers, as well as their preda- tors, distribute themselves evenly over all inundated habitat (so their densities in channels temporarily drop). Here, we assume that mobile grazers and predators occupy the floodplain only when water is deeper than 0.2 m because this depth has been found to be a critical threshold below which larger prey are vulner- able to fishing birds. Armored cat- fish (family loricariidae) in Panama and grazing minnows (Campostoma) in Oklahoma avoid water shallower than 0.2 m, even when their food is abundant there and scant in deeper habitats (Power 1984, 1987). Simi- larly, crayfish (Orconectes propin- pus) in Michigan may avoid shal- low water because of risk from wading and diving birds, as well as terrestrial predators like raccoons (Creed 1990, 1994). Some vegetation in the channel is attached to the substrate and some floats freely. At spillover, the free- floating fraction is distributed over the floodplain, where it grows rap- idly because of the enormous in- Biomass balance equations for trophic dynamics subject to hydraulic constraints Detritus: a= I + m,V - ch HD - m,D dt Vegetation: fi = rV - c,, HV - m,V dt K Herbivore-detritivores (grazers): = b, ch HV + b, c,, HD - cp HP - m,,H dt Predators: - dP = bp cp HP - mpP dt Table 1. Symbols used in biomass balance equations. Symbol Meaning Units D V H P bh Detritus standing stock Vegetation biomass Grazer biomass Predator biomass Conversion efficiency for grazers eating Conversion efficiency for predators eating grazers Per capita grazing rate on detritus or vegetation Per capita predation rate on grazers Maximal intrinsic rate of increase for vegetation Input of allochthonous detritus Carrying capacity (asymptotic biomass) for vegetation Loss rate of detritus Nongrazing mortality of channel vegetation Mortality of grazers not due to predation Mortality of predators vegetation or detritus crease in habitat surface area. On the falling limb of the hydrograph, when flow drops to bank-full depth, the water is reconfined in the chan- nel, drying up the floodplain. Stand- ing stocks of detritus and vegetation available to the river consumers drop to zero on the floodplain, while standing stocks within the channel do not change. In contrast, mobile grazers and predators return to the channel, except for that fraction left stranded on the floodplain. Mortal- ity from stranding can be high. Bonnetto et al. (1969, cited in Welcomme 1985, p. 170) estimated that in the Parani River, Argentina, the biomass of fish that die annually by stranding is four times that caught by the fishery. Between the threshold transitions of spillover and reconfinement, gradual changes in width, depth, and velocity also affect the perfor- mance, gains, and losses of food- chain constituents. There are few data quantifying how hydraulic pa- rameters affect the performance of organisms likely to be dominant interactors in river food webs, so our discussion here is largely specu- lative. The elucidation of mechanis- tic linkages between physical envi- ronmental variables and species' performance and impacts is one of the most crucial areas of research for the eventual application of mod- els to actual problems. Trophic parameters in each of the biomass balance equations (in box) can be linked to hydraulic vari- ables. For example, we might expect the loss rate of detritus as respired carbon dioxide to decrease with in- creasing depth, because water tem- perature and microbial concen- trations would both be likely to decrease as depth increases. Vegeta- tion carrying capacity (K) should decrease with depth if vegetation is light-limited. If plants are nutrient- limited, their growth rate (r) might March 1995 163 Table 2. Hydraulic influences. Symbol Response Used in simulations `h 1 K r md Decreases linearly with velocity Yes after a certain threshold, and ramps down to zero at the slip speed Decreases with width, due to higher Yes proportion of refuge area on floodplain. In channel, increases above a threshold velocity at which flow dislodges and washes away refuges Increases with width, due to higher litter Yes input over floodplain Decreases with depth due to light limitation Increases with velocity due to increases in No No nutrient flux Decreases with depth due to temperature or No oxygen stratification increase up to a point with velocity, which would increase the flux of nutrients available to attached veg- etation (Whitford and Schumaker 1964). Above a certain velocity, however, local growth might be re- duced by sloughing, or by light limi- tation if high flows become more turbid. Arthropod grazers like cray- fish or prawns that are not particu- larly streamlined might use fewer of their appendages for food gathering and more for holding on to the sub- strate as flow velocity increased. In our model, we assume such grazers feed at a maximal rate until currents reach a threshold velocity, above which their per capita inges- tion rates decline linearly, until they lose their grip on the substrate and stop grazing altogether. We call that velocity the slip speed. For eight crayfish species, slip speeds in plexiglass flumes ranged between 26 and 50 cm/s (Maude and Will- iams 1983). Hart (1992) found that at near-bed velocities of more than 50 cm/s in a cobble-bedded Michi- gan stream, a dominant attached macroalga, Cladophora, escaped grazing from crayfish, which were able to suppress the alga at lower current velocities. (Hart attributed the higher apparent slip speed for crayfish in the field to the rough streambed.) Predator attack rates on grazers might also decline with current velocity because of con- straints on prey encounter or han- dling (Hansen et al. 1991). How- ever, we visualize the top predators as powerful, streamlined swimmers like predatory fish (Figure 4). Swim- ming power may compromise ma- neuverability, however, so we have made predator attack rates reflect limited ability to search in struc- tural refuges for prey. Prey refuges in the main channel are envisioned as log jams that, like grazers, can be dislodged by high flows. In model simulations, refuges cover a maximal proportion of the channel bed (5%) at low flow, but they begin to be dislodged at flow velocities of 1 m/s and are washed away when flows exceed 2 m/s. Ref- uges for prey from predators are assumed to cover 20% of the flood- plain, where current velocities never get high enough to dislodge them. As a consequence, predator attack rates are even lower after spillover than would be expected from prey dilution over the floodplain. Results from preliminary simulations Examples of simulation output for a floodplain river with one high and one low water period per year simu- lated by sinusoidal discharge, a lev- eed river with the same sinusoidal discharge, and a regulated river with artificially stabilized low (100 m3/s) and artificially stabilized average (450 m3/s) flow are shown in Figure 5. The floodplain river with both temporal and spatial seasonal dy- namics (Figure Sa) maintains, over the long term, the most stable popu- lations at higher trophic levels (graz- ers and predators). In the leveed channel with sinusoidal discharge, predators initially increase as they benefit from prey concentrated in channels with little refuge. This ini- tial advantage, however, allows predators to harvest their prey at an unsustainable rate, so prey are even- tually driven to low levels at which predators starve. In channels in which constant low discharge is maintained, grazers show damped oscillations but per- sist. Grazer oscillations are coupled with damping oscillations in their detrital and plant foods, as in many other trophic models that, like ours, are derived from classical Lotka- Volterra models of consumer-re- source dynamics. The grazer densi- ties that are eventually sustained in the low, constant discharge channel are not sufficient, however, to main- tain a viable predator population. Predators decline asymptotically to zero in this simulation (Figure 5c). In the channel with constant aver- age discharge, flows are chronically too high for the nonhydrodynamic grazers to feed effectively, and they starve, followed by the crash of their predator's population. Although these simulation results are consistent with some predictions about the influence of spatial het- erogeneity and temporal fluctuation on population and food-web persis- tence (Huffaker 1958, Hutchinson 1961), they are at odds with others. In mathematical models of food webs that lack explicit temporal or spa- tial contexts, longer food chains were found to be more dynamically frag- ile with longer return times to equi- libria following perturbations (Pimm 1982). Therefore, food webs with short chains were predicted to pre- dominate in nature, particularly where environments are perturbed (Pimm 1982, Pimm et al. 1991). Our simulations, in contrast, sug- gest that the longest (three-level) food chains are maintained only when the environment fluctuates. Predator populations persist longer in channels subject to discharge fluc- tuation than in channels with stabi- lized flow (Figure Sb versus Figures 5c and Sd), and they persist over the 164 Bioscience Vol. 45 No. 3 long term only when biota have pe- riodic access to inundated flood- plains (Figure Sa). These results are congruent with current understanding of the impor- tance of the flood pulse in flood- plain rivers (Bayley 1989, Junk et al. 1989, Sparks et al. 1990). A somewhat more unexpected out- come of these simulations is that biota do not simply track hydro- logic changes. Longer biotically driven cycles can be superimposed on the hydrologic cycles (e.g., Fig- ure 5b). In other simulations, preda- tors, grazers, and vegetation showed damped oscillations with wave- lengths much longer than the an- nual hydrologic cycles, which were closely tracked only by detritus. These cycles damped as the modeled food web adjusted to a perturbation caused by setting initial densities at levels away from the system's equi- librium. While aspects of this be- havior depend on arbitrary initial conditions, transient dynamics fol- lowing perturbations may be quite important in real rivers, which can experience striking year-to-year, as well as within-year, variation in dis- charge. In the simulations reported here, the hydrograph had seasonal, but no year-to-year, variation. Few river hydrographs are this regular, but some, for example, that of the Orinoco River at Ciudad Bolivar (Figure 2 1.5 in Vasquez and Wil bert 1993), are close. In future studies, we plan to use hydraulic food-chain models to explore the effects of runs of unusually dry or wet years on food webs. Future needs and directions To advance modeling efforts to the point where they might eventually address real ecological and manage- ment issues, we need better infor- mation on two aspects of river food webs. First, which are the key taxa or functional groups that are linked in the chains of strong interactions? Second, how do their performances and impacts in food webs vary un- der changing environmental condi- tions, such as the dramatic fluctua- tions in area, depth, and velocities of river habitats? These questions are interrelated. March 1995 a b Time Time Time I me C d L Time I Time L- Time - Time Figure 5. Simulation results for four cases: (a) a river with access to its floodplain and sinusoidal discharge (only channel shown); (b) a leveed river with sinusoidal discharge retained entirely within the channel; and regulated rivers with low (c) and average (d) flow that does not exceed bank-full depth. Properties of organisms, for example their size, are likely to affect their performance under certain environ- mental conditions, and environmen- tal conditions are likely in large part to determine which types of organ- isms can be "strong interactors" (Paine 1980) at a given place and time. Slip speeds for grazers, for example, are likely to depend on whether large prawns or minuscule mayflies (order Ephemeroptera) are the dominant consumers of vegeta- tion and detritus in the food web. Which grazers dominate, in turn, is likely to be strongly influenced by which prove hydrodynamically com- petent under particular flow regimes (Hart 1992). Size structure and life-history stage structure are important fea- tures of biological populations (Ebenman and Persson 1988, Mittel- bach et al. 1988) not yet accounted for in our model. Welcomme and Hagborg (1977) modeled growth, mortality, and recruitment for a river-floodplain fish population with four discrete age classes, exploring the impacts on fish production of minimal dry season area and maxi- mal area inundated during floods. One promising area for future de- velopment would be to combine models of age- or size-structured populations, like theirs, with mod- els of multitrophic level interactions, like ours, in an exploration of how population and community dynam- ics interact in fluctuating environ- ments. A particularly crucial fea- ture to study would be seasonal life-history bottlenecks for key popu- lations. Different life-history stages of river organisms typically occupy different habitats. For example, off- river water bodies (e.g., oxbow lakes, billabongs, or varzea lakes) serve as rearing habitats for juveniles of spe- cies that occupy the main river chan- nels after reaching maturity (Wel- comme 1985). In addition, different fish species may segregate between off-river lakes and the main channel during low flow and then interact as both enter the floodplain after its inundation (Welcomme 1985). Our model does not represent off-river water bodies, but the consequences for food webs of the spatial hetero- geneity and potential refuges from predators and competitors that they contribute are important to explore in future modeling efforts. In addition to the general compli- 165 Channel with mnstant low discharge: Channel (leveed) with sinusoidal discharge: Channel with mnstant average discharge: m Channel and floodplain with sinusoidal discharge: DBbihJI Figure 6. In these simulation results, food chains are shortened to one or two levels in channels without access to their floodplains, and the predator level per- sists only when floodplains are periodi- cally inundated. cations introduced by size and habi- tat structure, site-specific natural history information is necessary to tailor models to specific ecosystems. We have attempted to model the habitat and trophic dynamics re- lated to spillover, reconfinement, and hydraulic changes driven by sea- sonal discharge fluctuations, which are macroscale attributes of all large rivers. It is harder to generalize the postulated trophic interactions among large rivers. We expect that some behaviors of our model are general, such as sea- sonal dynamics of the food web due to the tendency of grazers to profit from access to the inundated flood- plain and the tendency of predators to benefit when prey are concen- trated with them in channels. These dynamics underlie the tendency of temporal and spatial variation to promote the persistence of longer food chains (Figure 6), as Huffaker's (1958) experiments demonstrated. Other results are clearly sensitive to the natural history features postu- lated for specific strong interactors, such as the grazer slip speeds. Mod- eling efforts are useful if they focus attention on those natural history attributes of biota and ecosystems that warrant further study because of their potential importance to dy- namics. Conclusions Large rivers have been defined as "those large enough to intimidate research workers" (D. P. Dodge, cited by Hynes 1989). Modeling ef- 166 forts are particularly important to guide field studies in these large, dynamic systems, where sampling and experimental manipulations are difficult. We see the interplay of modeling and field investigation as the best approach to understanding the complex environmental prob- lems such as those that arise when levees eliminate the floodplain and the flood pulse, or regulation elimi- nates, alters, or dampens seasonal changes in discharge-the master variable that limits and resets river .populations throughout entire drain- age networks. We expect that future research is likely to support the ba- sic ecological paradox of rivers: that large, frequent hydrologic pertur- bations are crucial for long-term maintenance of their biodiversity, their enormous productivity, and the higher trophic levels, which include the biological populations most prized by humans. Acknowledgments A. Sun acknowledges support from the University of California Natural Reserve System and NSF BIR9256532 (a mathematical biology training grant to P. Kareiva, T. Daniels, and G. O'Dell at the University of Wash- ington). We thank David Hart, Dina Fonseca, Peter Bayley, Susan Gresens, and an anonymous reviewer for insightful comments; Ok Tedi Mining Limited, Papua New Guinea, for the photograph of the Fly River; and the National Science Founda- tion (DEB-9319924) for research support. References cited Adis, J. and V. Mahnert. 1986. On the natu- ral history and ecology of Pseudo- scorpiones (Arachnidae) from an Amazo- nian blackwater inundation forest. Amazoniana 9: 297-314. Bayley, P. B. 1989. Aquatic environments in the Amazon Basin, with an analysis of carbon sources, fish production, and yield. Can. Spec. l'ubl. Fish. Aquat. Sci. 106: . 1995. Understanding large river- floodplain ecosystems. Bioscience 45: Belt, C. B. Jr. 1975. The 1973 flood and man's constriction of the Mississippi River. Science 189: 681-684. 1969. Limnological investigations on bi- otic communities in the middle Parana river valley. Int. Ver. Theor. Angew. 39 9-408. 153-158. Bonetto, A. A., W. Dioni, and C. Pignalberi. Limnol. Verh. 17: 1035-1050. Bowlby, J. N. and J. C. Roff. 1986. Trophic structure in southern Ontario streams. Carpenter, S. R., J. F. Kitchell and J. K. Hodgson. 1985. Cascading trophic inter- actions and lake productivity. Bioscience Connell, J. H. 1978. Diversity in tropical rain forests and coral reefs. Science 199: 13 02-1 3 10. Creed, R. P. Jr. 1990. The direct and indirect effects of omnivorous crayfish on stream benthic communities. Ph.D. dissertation, Michigan State University, East Lansing, MI. Creed, R. P. Jr. 1994. Direct and indirect effects of crayfish grazing in a stream community. Ecology 75: 209 1-2103. Crowley, P. H. 1978. Effective size and the persistence of ecosystems. Oecologia 35: Cummins, K. W. 1973. Trophic relations of aquatic insects. Annu. Rev. Entomol. 18: DeAngelis, D. L. 1992. Dynamics of Nutrr- ent Cycling and Food Webs. Chapman and Hall, London, UK. Ebenman, B. and L. Persson. 1988. Size- structured Populations. Springer-Verlag, Berlin, Germany. Estes, J. A. and J. F. Palmisano. 1974. Sea otters: their role in structuring nearshore communities. Science 185: 1058-1060. Goulding, M. 1980. The Fishes and the For- est: Explorations in Amazonia Natural History. California University Press, Ber- keley, CA. Hansen, R. A., D. D. Hart, and R. A. Merz. 199 1. Flow mediates predator-prey inter- actions between triclad flatworms and larval black flies. Oikos 60: 187-1 96. Hart, D. D. 1992. Community organization in streams: the importance of species in- teractions, physical factors, and chance. Oecologia 91: 220-228. Hastings, A. 1977. Spatial heterogeneity and the stability of predator prey systems. Theor. Popul. Biol. 12: 37-48. Henderson, F. M. 1966.Open Channel Flow. MacMillan Publishing Co., New York. Holling, C. S. 1959. The components of predation as revealed by a study of small- mammal predation of the European Pine Sawfly. Can. Entomol. 91: 293-320. Holt, R. D. 1985. Population dynamics in two-patch environments: Some anoma- lous consequences of an optimal habitat distribution. Theor. Popul. Biol. 28: Huffaker, C. B. 1958. Experimental studies on predation: dispersion factors and predator-prey oscillations. Hilgardia 27: Hutchinson, G. E. 1961. The paradox of the plankton. Am. Nut. 95: 137-146. Hynes, H. B. N. 1989. Keynote address. Can. Spec. Publ. Fish. Aquat. Sci. 106: Junk, W. J., P. B. Bayley, and R. E. Sparks. 1989. The flood pulse concept in river- floodplain systems. Can. Spec. Publ. Fish. Kajak, A., L. Andrzejewska, and Z. Wojcik. 1968. The role of spiders in the decrease of damages caused by Acridoidea on Ecology 67: 1670-1679. 35: 634-639. 185-195. 1 8 3-206. 18 1-208. 343-383. 5-10. Aq~at. Sci. 106: 110-127. BioScience Vol. 45 No. 3 meadows-experimental investigations. Leopold, L. B., and T. Maddock. 1954. The Flood Control Controversy. Ronald Press, New York. Leopold, L. B., M. G. Wolman, and J. P. Miller. 1964. Fluvial Processes in Geo- morphology. Freeman, San Francisco, CA. Lowe-McConnell, R. H. 1975. Fish Com- munities in Tropical Freshwaters. Longman, London, UK. Mairson, A. 1994. Great flood of '93. Nat. Geogr. Mag. 185: 42-81. Maude, S. H., and D. D. Williams. 1983. Behavior of crayfish in water currents: hydrodynamics of eight species with ref- erence to their distribution patterns in southern Ontario. Can. J. Fish. Aquat. Sci. 40: 68-77. Mittelbach, G. G., C. W. Osenberg, and M. A. Leibold. 1988. Trophic relations and ontogenetic niche shifts in aquatic eco- systems. Pages 219-235 in B. Ebenman and L. Persson, eds. Size-Structured Popu- lations. Springer-Verlag, Berlin, Ger- many. Northcote, T. G. 1988. Fish in the structure and function of freshwater ecosystems: a "top down" view. Can. J. Fish. Aquat. Oksanen, T. 1990. Exploitation ecosystems in heterogeneous habitat complexes. Paine, R. T. 1980. Food webs: linkage, inter- action strength, and community infra- structure. ]. Anim. Ecol. 49: 667-685. Parker, G. 1993. Howard Creek Diagnostic Study, White Sulphur Springs, West Vir- ginia. Report prepared for the Soil Con- servation Service, US Department of Ag- Ekol. Pol. 16: 755-764. Sci. 45: 361-379. Evol. Ecol. 4: 220-234. riculture, Morgantown, WV. Perrin, C. J., J. L. Bothwell, and P. A. Slaney. 1987. Experimental enrichment of a coastal stream in British Columbia: ef- fects of organic and inorganic additions on autotrophic periphyton production. Can. J. Fish. Aqrrat. Sci. 44: 1247-1256. Petersen, R. C. Jr, K. W. Cummins, and G. M. Ward. 1989. Microbial and animal processing of detritus in a woodland stream. Ecol. Monogr. 59: 21-39. Pimm, S. L. 1982. Food Webs. Chapman and Hall, London, UK. Pimm, S. L., J. H. Lawton, and J. E. Cohen. 1991. Food web patterns and their conse- quences. Nature 350: 669-674. Power, M. E. 1984. Depth distributions of armored catfish: predator-induced re- source avoidance? Ecology 65: 523-528. ~- . 1987. Predator avoidance by graz- ing fishes in temperate and tropical streams: importance of stream depth and prey size. Pages 333-352 in W. C. Kerfoot and A. Sih, eds. Predation: Direct and Indirect Impacts on Aquatic Communi- ties. University Press of New England, Hanover, NH. . 1990. Effects of fish in river food webs. Science 250: 411-415. . 1992a. Top-down and bottom-up forces in food webs: Do plants have pri- macy? Ecology 73: 733-746. . 1992b. Habitat heterogeneity and the functional significance of fish in river food webs. Ecology 73: 1675-1688. Power, M. E., W. J. Matthews, and A. J. Stewart. 1985. Grazing minnows, pis- civorous bass and stream algae: dynam- ics of a strong interaction. Ecology 66: 1448-1456. Sparks, R. E. 1992. Risks of altering the hy- drologic regime of large rivers. Pages 119-152 in J. Cairns Jr., B. R. Nieder- lehner, and D. R. Orvos, eds. Predicting Ecosystem Risk. Advances in Modern Environmental Toxicology-Vol. 20. Princeton Scientific Publishing Co., Prince- ton, NJ. Sparks, R. E., P. B. Bayley, S. L. Kohler, and L. L. Osborne. 1990. Disturbance and recovery of large floodplain rivers. Environ. Manage. 14: 699-709. Tilman, D. 1994. Competition and hiodiver- sity in spatially structured habitats. Ecol- ogy 75:- 2-16. Vasouez. E.. and W. Wilbert. 1993. The Orinoco: physical, biological and cul- tural diversity of a major tropical allu- vial river. Pages 448-471 in P. Calow and G. E. Petts, eds. The Rivers Hand- book. Blackwell Scientific Publications, London, UK. Welcomme, R. L. 1985. River fisheries. Fish- eries Technical Paper 252. UN Food and Agricultural Organization, Rome, Italy. Welcomme, R. L., and D. Hagborg. 1977. Towards a model of a floodplain fish population and its fishery. Environ. Biol. Fishes 2: 7-24. Whitford, L. A., and G. J. Schumacher. 1964. Effect of a current on respiration and mineral uptake in Spirogyra and Oeclo- gonium. Ecology 45: 168-170. Wootton, J. T., and M. E. Power. 1993. Productivity, consumers, and the struc- ture of a river food chain. Proc. Natl. Acad. Sci. USA 90: 1384-1387. Zaret, T. M., and R. T. Paine. 1973. Species introduction in a tropical lake. Science 182: 449-455. MflNfiitES flRE IN IOU BE... fFnd The Tmble Is Us. Help save the manatee from extinction. Contact: Save the Manateeo Club, 594 500 N. Maitland Avenue o Maitland, FL 32751 or call: 1-800-432-JOIN Many manatee mortalities are human-related. This manatee was fatally injured in a watercraft collision. Other causes of human- related manatee mortalities include being crushed and/or drowned in canal locks and flood control structures; ingestion of fish hooks, litter, and monofilament line; entanglement in nets or crab trap lines; and vandalism. Currently, there are only about 1,800 manatees left. March 199.5 167 Need for Ecosystem Management of Large Rivers and Their Floodplains These phenomenally productive ecosystems produce fish and wildlife and preserve species Richard E. Sparks M ost of the 79 large river- floodplain ecosystems in the world have been al- tered by human activities; the rest are likely to be altered soon (Welcomme 1985). These complex ecosystems are composed of the flowing channels that most people associate with rivers, together with the floodplain lakes, backwaters, forests, and wetlands that harbor much of Earth's terrestrial and fresh- water biodiversity (Figures 1-3). River-floodplain ecosystems, unlike most lakes, are characterized by sea- sonal floods that promote the ex- change of nutrients and organisms among a mosaic of habitats and thus enhance biological productivity (Bayley page 153 this issue, Junk et al. 1989). Annual flood pulses are so pre- dictable and long-lasting that plants, animals, and even human societies have adapted to take advantage of them. In ancient Egypt and Meso- potamia, the fertility of the soils was renewed each year by the an- nual overflow of the rivers, thereby sustaining large populations in one place for millennia and permitting the development of great civiliza- tions. Outside these floodplains, the fertility was exhausted by a few years of steady cultivation, so people had to move on. Richard E. Sparks is director of the River Research Laboratory, P. 0. Box 590, Havana, IL 62644, a unit of the Center for Aquatic Ecology, Illinois Natural History Survey. 0 1995 Ameri- can Institute of Biological Sciences. Ecosystem management works to guide, rather than thwart, natural processes Despite centuries of alteration in the developed world, remnant river- floodplain ecosystems still exist. Central Europe's largest river, the Danube, retains, 650 km2 of its former floodplain in Slovakia and Hungary despite changes caused by dredging, channelization, and dam- building (Bacalbasa-Dobrovici 1989, Pearce 1994). In the United States, most of the 98,000-square-kilometer floodplain along the Mississippi downstream from the mouth of the Ohio has been leveed and drained for agricul- ture, but sizable floodplains have been preserved along the upper Mis- sissippi north of St. Louis as part of the National Fish and Wildlife Ref- uge System (NRC 1992). Two large tributaries of the Mississippi-the Illinois River, and the lower portion of the Missouri River-retain flood pulses and floodplains, and a major distributary (branch) of the Missis- sippi-the Atchafalaya-is building new deltaic floodplain in the Gulf of Mexico, thereby increasing what is already North America's largest re- maining (5700 km2) river overflow swamp (Hesse et al. 1989, 1993, NRC 1992). In the developing world, exten- sive river-floodplain ecosystems re- main, but they are diminishing at increasing rates as land use intensi- fies and as many countries attempt to follow the western model of eco- nomic development through the use of massive water resource projects (Sparks 1992). There now are dams on virtually all the large rivers in Africa (Obeng 1981). In South America, the upper Parani is dammed, but the middle and lower reaches still retain natural flood- plains, and the Parani's largest tribu- tary, the Paraguay, remains free- flowing.' Earth's largest river (in terms of flow), the Amazon, remains undammed, but it has been affected by clearing of the upland and flood- plain forests. Also, 100 planned tributary dams may block or impede fish migrations (Fearnside 1989, Junk and de Mello 1987). In 1993 and 1994, international attention was focused on large riv- ers and their floodplains when di- sastrous floods occurred in Bang- ladesh, western Europe, and the United States. Now questions are being asked about the effectiveness and cost of current flood and flood- plain management policies and about the potential for reducing fu- ture flood damage by preserving and restoring large river-floodplain eco- systems and their tributary water- sheds and wetlands (Sklar 1993, Sparks and Sparks 1994). This focus on flood-damage re- `Edmundo Drago, 1994, personal communi- cation. Instituto Nacional de Limnologia, Santa Fe, Argentina. 168 BioScience Vol. 45 No. 3