From Metapopulations to Metacommunities: Linking Theory with Empirical Observations of the Spatial Population Dynamics of Stream Fishes Jeffrey A

Home , Insular biogeography, Metapopulation, Population dynamics

From Metapopulations to Metacommunities: Linking Theory with Empirical Observations of the Spatial Population Dynamics of Stream Fishes Jeffrey A. Falke* and Kurt D. Fausch Department of Fish, Wildlife and Conservation Biology, Colorado State University Fort Collins, Colorado 80523, USA

Abstract.—Stream fishes carry out their life histories across broad spatial and temporal scales, leading to spatially structured populations. Therefore, incorporating metapopulation dynamics into models of stream fish populations may improve our ability to understand mechanisms regulating them. First, we reviewed empirical research on metapopulation dynamics in the stream fish ecology literature and found 31 papers that used the metapopulation framework. The majority of papers applied no specific metapopulation model, or included space only implicitly. Al- though parameterization of spatially realistic models is challenging, we suggest that stream fish ecologists should incorporate space into models and recognize that metapopulation types may change across scales. Second, we considered metacommunity theory, which addresses how trade-offs among dispersal, environmental heterogeneity, and biotic interactions structure communities across spatial scales. There are no explicit tests of metacommunity theory using stream fishes to date, so we used data from our research in a Great Plains stream to test the utility of these paradigms. We found that this plains fish metacommunity was structured mainly by spatial factors related to dispersal opportunity and, to a lesser extent, by environmental heterogeneity. Currently, metacommunity models are more heuristic than predictive. Therefore, we propose that future stream fish metacommunity research should focus on developing testable hypotheses that incorporate stream fish life history attributes, and seasonal environmental variability, across spatial scales. This emerging body of research is likely to be valuable not only for basic stream fish ecological research, but also multispecies conservation and management.

Introduction their life history (Schlosser 1991; Fausch et al. 2002). Additionally, streams are linear habi- Stream fishes require multiple habitat types tats, arranged in hierarchical dendritic patterns (e.g., spawning, rearing, and refuge) to com- (Fagan 2002; Campbell Grant et al. 2007). As plete their life cycles (Schlosser and Anger- a result, natural and anthropogenic barriers meier 1995). These habitats are often dispersed can easily block movements among habitats in in space, throughout the riverscape, so that fish such branching networks (Ward 1983; Win- must move among habitat patches to carry out ston et al. 1991; Morita and Yamamoto 2002). * Corresponding author: [email protected] Therefore, the spatial arrangement of habitats

207 208 falke and fausch and the ability to move among them are critical and dispersal both vary in strength in differ- for stream fish persistence and recolonization ent communities and that trade-offs between dynamics. Moreover, incorporating spatial in- these two forces can help explain fundamental formation into models of stream fish popula- differences in community structure and the tions should improve our understanding of the processes that shape it. One tenet of the meta- mechanisms that control their dynamics. community approach is that local commu- The metapopulation concept has been used nities often do not have discrete boundaries by stream fish ecologists to help explain variabil- (Holyoak et al. 2005). In fact, patches within ity in local-scale population dynamics (Rieman which local communities reside are often tem- and McIntyre 1995; Schlosser and Angermeier porary and vary in position over time. Habitat 1995). Metapopulations are defined as groups patches in stream ecosystems may fit this mod- of habitat patches that support local popula- el well, due to variation in their persistence and tions, within a matrix that is not suitable habitat. location seasonally owing to fluctuations in Metapopulation theory states that all local pop- stream flow, and over longer time scales owing ulations eventually go extinct, but that dispersal to succession after disturbance. However, the among them has a quantitative influence on relative importance of spatial versus temporal population dynamics, including rescuing them habitat heterogeneity remains to be addressed from extinction or allowing for recolonization for stream fish assemblages in the context of after extinction (Hanski and Simberloff 1997). metacommunity theory. Therefore, a metapopulation approach may be Our objectives in this chapter are to review well suited for the study of fish population dy- and synthesize what is known about stream fish namics across the spatially dispersed, heteroge- metapopulations and metacommunities and to neous habitats common to streams (Schlosser provide a simple test of metacommunity mod- and Angermeier 1995). This approach has most els using empirical data for Great Plains stream commonly been used in the conservation and fishes in the western United States. Specifically, management of salmonids (Rieman and Dun- we first conducted a literature review of stream ham 2000). However, empirical support for fish metapopulation research to evaluate (1) metapopulation dynamics in most stream fish what empirical evidence exists for stream fish populations remains to be quantified, especially metapopulation dynamics, (2) what types of for nonsalmonid species. models have been used to test for metapopula- Because there are strong effects of spatial tion dynamics in stream fishes and the extent scale, habitat heterogeneity, and dispersal on to which these models incorporate space, stream fish populations, it stands to reason that and (3) what metapopulation type (Harrison stream fish communities also may be spatially 1991; Schlosser and Angermeier 1995) best structured. Recent work has built upon three fits stream fish metapopulations based on a decades of metapopulation theory to bridge consensus among empirical studies. Second, the gap between spatial population structure we reviewed stream fish and metacommunity and spatial community dynamics (Leibold et literature to evaluate (1) what empirical evi- al. 2004). Metacommunity models attempt dence exists for stream fish metacommunity to “scale up” the concepts set forth by meta- dynamics, and (2) whether four metacommu- population models to the community level. nity paradigms might apply to stream fish com- Metacommunity theory holds that biotic in- munities. Finally, we used data from our own teractions (e.g., competition and predation) research to evaluate how dispersal opportunity stream fish metapopulations and metacommunities 209 and habitat heterogeneity affect community ing patches (e.g., patch size, patch quality). structure across scales in a Great Plains stream The most well known is the incidence func- fish metacommunity. We also evaluated what tion model (Hanksi 1994). metacommunity model best fits these data. In addition to the type of metapopulation model used in each study, we classified Metapopulation Dynamics in whether or not the authors observed or mea- Stream Fish Populations sured metapopulation dynamics in the population and what metapopulation structure was Methods found. Metapopulation structure was catego- We reviewed the literature on stream fish ecol- rized among five types defined by Harrison ogy to evaluate empirical evidence for metapo- (1991), as modified by Schlosser and Anger- pulation dynamics. We included only primary meier (1995; Figure 1). research papers that dealt with lotic fishes Classic metapopulation.—The concept of a and included the term “metapopulation” or population of populations linked by dispersal “source-sink” in the title, abstract, or keywords. (i.e., metapopulation) was first proposed by We used the Web of Science © database as the Richard Levins (1969, 1970). Levins showed primary method of identifying these papers. that a metapopulation could be maintained Our review covers articles published from by dispersal among several subpopulations 1900 to 2008 from all journals included in the in discrete habitat patches. He assumed that Web of Science. all patches were of equal size, patches were Once located, papers were categorized by the same distance from one another, rates of the degree to which space was incorporated, colonization and extinction were equal, and based on model types identified in Hanksi subpopulations had independent dynamics and Simberloff (1997). Spatially implicit (Hanski and Simberloff 1997; Gotelli 2001). models are the simplest type of metapopu- Some patches can remain vacant at equilibri- lation model and assume all subpopulations um in this model type. Although this “classic” are identical and equally connected. No habi- metapopulation structure is unrealistic and tat heterogeneity is included in these models, probably rare, it serves as a null model and a and spatial locations of patches are not incor- basis on which to build more spatially realis- porated. An example is Levins’ metapopula- tic models of metapopulations. tion model (Levins 1969, 1970). In contrast, Source-sink metapopulation.—Several mod- spatially explicit models incorporate space els based on empirical evidence of metapo- or habitat heterogeneity into the model but pulation dynamics were proposed by Harrison not necessarily both. Examples are cellular (1991) and modified for stream fish metapop- automata, lattice grid, and raster-based geo- ulations by Schlosser and Angermeier (1995). graphic information system (GIS) landscape The first is a modification of the mainland-is- models. In these models, migration depends land concept from island biogeography theory on distance but is usually restricted to adja- (Figure 1B; MacArthur and Wilson 1967). By cent patches. The final model type, spatially relaxing the assumption of the classic meta- realistic models, is the most complex. These population model that each patch is the same models incorporate both the spatial position size and has an equal potential for produc- of each patch on the landscape and individual ing migrants, a more realistic model in which attributes of both focal patches and all outly- patches differ in demography (i.e., survival 210 falke and fausch

Figure 1. A conceptual model of metapopulation types for stream fishes across four spatial scales in riverscapes. Ovals represent habitat patches that are occupied (filled) or vacant (open). Insets show the most applicable original theoretical metapopulation types defined by Harrison (1991; after Schlosser and Angermeier 1995): (A) patchy population, (B) source-sink, (C) hybrid, and (D) nonequilibrium metapopulation types. Flow is from left to right in all stream networks. Arrows show dispersal among patches, and dashed lines indicate boundaries of “populations.” Symbols (e.g., triangles) represent different habitat types (e.g., spawning, rearing, and refuge) within patches that are required for stream fish to carry out their life history (not shown at the region scale). Some habitat patches that contain only a subset of these types may support only sink populations. Upstream and downstream movement that allows rescue and recolonization may be blocked by barriers (perpendicular line) or dry reaches (dot-dashed line). and/or reproductive rates) can be defined Patchy metapopulation.—Similar to the (Pulliam 1988). Source habitats have positive classic model, the patchy population model population growth rates (i.e., l $ 1.0), owing requires that subpopulation dynamics are to higher survival of adults and juveniles, and mostly independent (Harrison 1991). How- reproduction by adults. In contrast, subpopu- ever, the assumptions of identical habitat lations in sink habitats are unable to maintain patches and low dispersal among subpopula- themselves indefinitely (i.e., l < 1.0) without tions are relaxed. In fact, high dispersal among significant immigration of individuals from habitat patches and vastly different types of source habitats (e.g., “rescue effect”; Brown habitats characterize the patchy population and Kodric-Brown 1977). model (Figure 1A) and may result from taxa stream fish metapopulations and metacommunities 211 that use complementary habitats during differ- them, and that never goes extinct, then this is ent life stages (Dunning et al. 1992; Schlosser a hybrid metapopulation. Otherwise, a source- and Angermeier 1995). A key characteristic of sink metapopulation model may fit best, assum- the patchy metapopulation model is that no ing it fits the characteristics defined above for patches are ever unoccupied, because of high this type. dispersal rates. Hybrid metapopulation.—By incorporating Results and Discussion both patchy and source-sink population dynamics, a hybrid model can be developed that We identified 31 papers in the stream fish ecol- may be more realistic than either model alone ogy primary literature that contained the terms (Figure 1C; Harrison 1991). Within a group of “metapopulation” or “source-sink” in the title, “source” subpopulations, high dispersal among abstract, or keywords (Table 1). All papers different habitats required for carrying out the were published since 1995 and the majority species life history leads to persistence over (20) were published after 2000. In contrast, a time. In addition, satellite populations may act search for the term “metapopulation” alone in as sinks, due to the absence of critical habitats the Web of Science © database returned more required for persistence. than 3,000 research papers that contained the Nonequilibrium metapopulation.—Natural- term in their title, abstract, or keywords. There- ly or anthropogenically fragmented populations fore, papers in the stream fish literature make may be represented by a nonequilibrium meta- up less than 1% of the total body of metapopu- population model. Low or no dispersal due to lation research to date. reduced connectivity among patches, coupled Metapopulation dynamics for approxi- with deteriorating habitat quality, increases the mately 80 fish species or subspecies were con- rate of extinction among subpopulations, few sidered among the 31 papers. Fifty-seven of of which are recolonized (Figure 1D; Harrison these species were analyzed concurrently in 1991; Schlosser and Angermeier 1995). A non- two papers, Fagan et al. (2002) and Gotelli and equilibrium model often may be most appropri- Taylor (1999), though little species-specific ate for a species that shows a regional decline in data were provided. However, most papers distribution and abundance due to habitat frag- (24) addressed only one species, and 18 of the mentation and loss of connectivity. 31 papers focused solely on salmonids. Other Distinguishing among metapopulation types.— families represented included Cottidae, Cy- Based on their different characteristics, a short prinidae, Cyprinodontidae, Ictaluridae, Odon- dichotomous key can be developed to differen- tobutidae, Percidae, and Rivulidae. tiate among the five different metapopulation Many of the papers (14 of 31) discussed types. First, if dispersal and colonization among dynamics of the study populations in terms of patches is low or nonexistent, then a nonequilib- metapopulation theory, but no specific model rium metapopulation model fits best. Second, if was applied. We considered these papers to be patches never go extinct, then a patchy metapo- strictly observational (Table 1). Five papers pulation is appropriate. Third, if habitat patches used spatially implicit (Levins’ type) analy- are identical and their dynamics independent, ses to model metapopulation dynamics. For then a classic metapopulation model fits best. example, Gotelli and Taylor (1999) modeled Fourth, if there is a core area that contains differ- turnover in 36 species native to the Cimarron ent critical habitats with high dispersal among River, Oklahoma and found that colonization 212 falke and fausch Table 1. Summary of peer-reviewed literature on stream fish metapopulation ecology from 1995 through 2008. Presence or absence of the terms “metapopulation” or “source-sink” in the title, keywords, or abstract is indicated, along with the species studied. Papers with N/A in the abstract or keywords fields indicate that the journal does not contain these elements. Model type refers to the type of statistical model used: spatially implicit model (SIM), spatially explicit model (SEM), spatially realistic model (SRM), or observational (no modeling conducted; OBS). The type of metapopulation structure reported in the paper is also listed using codes (after Harrison 1991; Schlosser and Angermeier 1995): classic (CL), source-sink (SS), patchy (PM), nonequilibrium (NE), or not defined (ND). Model Metapopulation Reference Title Keywords Abstract Species type structure Chaumot et al. (2006) N N Y Bullhead Cottus gobio SRM NE Cooper and Mangel Y N/A Y Coho salmon SEM SS (1998) Oncorhynchus kisutch Crozier and Zabel N N Y Chinook salmon SEM NE (2006) O. tshawytscha Dunham and Rieman Y Y Y Bull trout Salvelinus SEM Rangea (1999) confluentus Dunham et al. (2002) N N/A Y Bull trout, Lahontan SEM ND cutthroat trout O. clarkii henshawi Fagan (2002) Y N Y Gila trout O. gilae SIM ND Fagan et al. (2005) N Y N Sonoran Desert fishes SIM ND (21 species) Gagen et al. (1998) Y N N Ouachita madtom Noturus OBS SS lachneri Garant et al. (2000) N Y Y Atlantic salmon Salmo OBS NE salar Gilliam and Fraser N N Y Giant rivulus Rivulus hartii OBS SS (2001) Gotelli and Taylor Y Y Y Cimarron River fishes SIM CL (1999) (36 species) Hilderbrand (2003) N N Y Cutthroat trout O. clarkii SEM SS Isaak et al. (2007) N Y N Chinook salmon SRM SS Isaak et al. (2003) N N/A Y Chinook salmon SIM PM Koizumi and Maekawa Y N Y Dolly Varden Salvelinus SEM ND (2004) malma Labbe and Fausch N Y N Arkansas darter SIM SS (2000) Etheostoma cragini Labonne and Gaudin N N/A Y Apron Zingel asper SRM SS (2006) Luttrell et al. (1999) N N/A Y Peppered chub OBS SS Macrohybopsis tetranema, shoal chub M. hyostoma Matsubara et al. (2001) Y N Y Donko Odontobutis SEM SS obscura McMahon and Matter N Y N Desert pupfishCyprinodon OBS SS (2006) macularius Morita and Yamamoto N N/A Y Whitespotted char SEM SS (2002) Salvelinus leucomaenis Neville et al. (2006) N Y Y Lahontan cutthroat trout OBS NE Policansky and Y Y Y Oncorhynchus spp. OBS ND Magnuson (1998) Rieman and Dunham Y Y Y Bull trout, Lahontan OBS ND (2000) cutthroat trout, Westslope cutthroat trout O. c. lewisi stream fish metapopulations and metacommunities 213 Table 1. Continued. Model Metapopulation Reference Title Keywords Abstract Species type structure Rieman and McIntyre N N/A Y Bull trout SEM SS (1995) Schlosser and N N/A Y Chinook salmon, creek OBS SS, PM Angermeier (1995) chub Semotilus atromaculatus Schlosser (1998) N Y N Creek chub OBS SS Shepard et al. (2005) N N/A Y Westslope cutthroat trout OBS NE Slack et al. (2004) N Y Y Bayou darter Etheostoma OBS SS rubrum Winston et al. (1991) N N/A Y Speckled chub OBS SS Macrohybopsis aestivalis, plains minnow Hybognathus placitus, Red River shiner Notropis bairdi, chub shiner N. potteri Young (1999) Y N/A Y Oncorhynchus spp. OBS NE a Range: a range of metapopulation types was found. and extinction probabilities were not corre- mended for some time (Moilanen and Hanski lated with the percentage of sites occupied. 1998; Hanski 2001; Ricketts 2001). Howev- Instead, site occupancy was related more to its er, only three papers in the stream fish ecol- longitudinal position along the river. These re- ogy literature incorporated characteristics of sults indicate that for these stream fishes, and neighboring habitats and their actual spatial probably others, a spatially implicit (nonspa- relationships to predict occupancy or popula- tial) approach is inadequate for characterizing tion parameters of focal habitats in a spatially metapopulation dynamics. Of the 17 papers realistic manner. These spatially realistic mod- that modeled metapopulation dynamics, 9 eling techniques included multiple logistic re- incorporated space explicitly. Most models gression with information-theoretic model se- were patch-based (see Dunham et al. 2002 lection (Isaak et al. 2007) and complex matrix for discussion) and were used to predict patch population models that incorporate dispersal occupancy based on variables such as patch and spatial processes (Chaumot et al. 2006; size, longitudinal position along the stream, Labonne and Gaudin 2006). These models of- isolation, and habitat quality. Multiple logis- fer great promise for modeling metapopulation tic regression was the most common method dynamics in complex landscapes, but their use used to model occupancy (e.g., Rieman and to date, especially in the stream fish ecology lit- McIntyre 1995; Dunham and Rieman 1999; erature, is limited. Koizumi and Maekawa 2004). Overall, results Metapopulation structure was an impor- of these studies indicated that habitats that are tant point of discussion in all of the papers we larger, less fragmented and isolated, and less reviewed (Table 1). Only 6 of the 31 papers degraded are more likely to be occupied by the did not classify their study metapopulations stream fish species studied. into one of the types identified by Harrison The realistic incorporation of space into (1991) and Schlosser and Angermeier (1995). metapopulation models has been recom- Gotelli and Taylor (1999) initially classified 214 falke and fausch their study metapopulations as classic Levins’ realistic models of stream fish population dy- type, though their analysis did not support namics based on principles from landscape this conclusion. It is not surprising that only and metapopulation ecology, although the one paper classified their metapopulation as term “metapopulation” was not included in the classic metapopulation type; as in other their title, abstract, or keywords. Clearly, some taxa, classic metapopulations are probably metapopulation-based research on stream fish rare in nature (Hanski 1996; Harrison and has been published recently that was not de- Taylor 1996). The most common metapopula- tected by our criteria. tion type reported was source-sink (16 of 31 Several patterns are common across re- papers). This also is not surprising, because search in stream fish metapopulations. The source-sink metapopulation dynamics are the first is that most papers (18 of 31) focused on simplest type that incorporates space. Patchy species in the family Salmonidae, even though metapopulations were discussed in only two salmonids make up less than 4% of fish species papers, Schlosser and Angermeier (1995) and native to North America (Froese and Pauley Isaak et al. (2003). This is surprising, because 2008). Rieman and Dunham (2000) reported the definition of patchy metapopulation types that salmonid life histories are well suited to (high dispersal among different habitat types) a metapopulation approach, due to discrete, seems to fit many stream fish metapopulations complementary habitats, frequent dispersal closely. Finally, nonequilibrium metapopula- among habitats, and strong population struc- tions were reported by six papers. These stream turing influenced by natal homing. However, fish metapopulations were typically in decline the complexity and diversity of life history across their ranges, and the papers were most strategies in salmonids, and the high degree of commonly oriented toward conservation (e.g., intraspecific variability in life history expres- Young 1999; Chaumot et al. 2006; Crozier and sion (e.g., migration patterns; Hendry and Zabel 2006). Stearns 2004), makes simple metapopulation Several papers that failed the criteria for types and models unsuitable for characterizing inclusion in our review nonetheless analyzed salmonid populations (Rieman and Dunham populations using a metapopulation approach 2000). In fact, Rieman and Dunham (2000) or discussed spatial population dynamics of suggested that metapopulation types of salmo- stream fishes in detail. For example, Scheurer nids may vary across a gradient of patch quality, et al. (2003) modeled persistence of brassy dispersal distance, and interpatch distance (see minnow Hybognathus hankinsoni as a func- their Figure 1), as opposed to conforming to a tion of habitat size, among-habitat connectiv- single type. Indeed, we found that most papers ity, and flow permanence at the river segment in the stream fish metapopulation literature scale. The authors found that brassy minnow used simple statistical models and classified were more likely to persist in deeper pools that metapopulations as one of the simplest types, were connected to other pool habitats and that that of a source-sink metapopulation. Many pa- persistence was higher in a segment with pe- pers were strictly observational and did not ex- rennial flow. These results are similar to those plicitly model metapopulation dynamics at all. reported above for other stream fish studies Although observational studies are important that used spatially explicit models. Similarly, for elucidating general patterns across stream Pichon et al. (2006), Letcher et al. (2007), and fish taxa, important advances in our knowledge Schick and Lindley (2007) presented spatially of stream fish population dynamics will occur stream fish metapopulations and metacommunities 215 only when rigorous studies are specifically de- the efficacy of different metapopulation -mod signed to evaluate spatial population dynamics els in explaining stream fish spatial population and distinguish among metapopulation types. dynamics. This includes quantifying parameters likely to affect population dynamics in a spatial frame- Metacommunity Dynamics in work, such as those proposed by Rieman and Stream Fish Assemblages Dunham (2000). The future of stream fish metapopulation A metacommunity is a set of local communi- research clearly lies with measurement and ties linked by dispersal of one or more species modeling of more spatially realistic variables, (Hubbell 2001). Therefore, in the simplest such as distance among patches, patch quality configuration, a metacommunity consists of (but see Isaak et al. 2007 for an alternative), several local communities that comprise a neighboring patch quality, and connectivity combined regional pool of species. It is the in- among patches. Additionally, knowledge of the terplay of dispersal and community dynamics characteristics and costs of dispersal by study within and between these two scales (i.e., local species is critical to incorporate into such mod- and regional) that is the focus of the metacom- els (Kareiva 1990; Hendry et al. 2004). Unfor- munity approach. Four paradigms have been tunately, complex spatially realistic population suggested as theoretical models for metacom- models are difficult to parameterize (Chaumot munity dynamics across taxa (Liebold et al. et al. 2006; Labonne and Gaudin 2006). More- 2004). Each paradigm differs in the degree to over, the scales at which stream fish metapo- which environmental heterogeneity, biotic in- pulations operate are often unknown, but are teractions, and dispersal processes are incor- a key to our understanding of these processes, porated (Figure 2). especially if metapopulation types do indeed Patch dynamics.—The patch dynamics vary across spatial and temporal scales (Rie- model builds on the equilibrium theory of is- man and Dunham 2000). land biogeography (MacArthur and Wilson In conclusion, due to the unique character- 1967; Figure 2A). This paradigm predicts that istics of stream fish life histories and habitats, a regional coexistence is maintained by a trade- refined theory that helps explain variability in off between competitive and dispersal ability stream fish metapopulation structure and dy- among members of the community (Hutchin- namics is warranted. However, formulation of son 1951; Hastings 1980; Tilman 1994). For this theory may prove challenging because of coexistence to occur, dispersal of strong com- the lack of detailed empirical data quantifying petitors must be limited so that they do not spatial population dynamics in stream fishes. drive other species to regional extinction. In For example, variables such as habitat size, contrast, inferior competitors must be better at location, and connectivity can be easily mea- colonizing than dominants when patches open sured, but patch- specific vital rates (e.g., births, up following disturbance (Caswell 1978; Yo- deaths, emigration, and immigration), and the dzis 1986). The patch dynamics paradigm is a rates and costs of movement among patches, spatially implicit model; patches are distribut- are difficult to measure across the large spatial ed uniformly and habitats are homogeneous. scales over which stream fishes carry out their Species sorting.—The species-sorting model life histories (Fausch et al. 2002). Nevertheless, incorporates the ideas that strong niche rela- these data are what will be required to evaluate tionships can occur between species and their 216 falke and fausch

Figure 2. Schematic representation of four metacommunity paradigms for two species, A and B (after Leibold et al. 2004). Arrows represent dispersal among habitat patches; heavy, solid arrows represent rapid dispersal, light, solid arrows moderate dispersal, and dashed arrows represent slow dispersal. Dominant species are indicated by large letters, and subdominant by small letters. Adaptation for specific niches is represented by shapes that match for species and habitat patches. The patch dynamics model (panel A) predicts that species persist by a trade-off between competition and colonization. Species A is a superior competitor, but species B is a better colonizer and can rapidly colonize empty patches. Under the species sorting model (panel B), species are well adapted to particular habitats (niches), and dispersal is slow. In contrast, under the mass effects model (panel C), dispersal is rapid among patches. Species persist in source habitats (larger symbols), to which they are well adapted and disperse to sink habitats (smaller symbols) to which they are not well adapted. Finally, the neutral model (panel D) predicts that all species are in all patches, and some will be lost due to slow random ecological drift, but this will be countered by moderate dispersal and speciation. habitats and that community structure may on the terms developed for metapopulation change along environmental gradients (Whit- models. taker 1972; Leibold 1998; Figure 2B). This Mass effects.—The mass effects paradigm model allows for dispersal by member spe- assumes that local community dynamics are cies, but occurrence is determined mainly by strongly affected by dispersal (Shmida and local abiotic conditions (e.g., habitat compo- Wilson 1985; Figure 2C). Moderate dispersal sition) independent of purely spatial effects. among patches that support local communi- By nature, habitats are heterogeneous under ties results in a source-sink dynamic (Holt the species-sorting paradigm, so this model 1985; Pulliam 1988; Holt 1993) within the would be considered spatially explicit based metacommunity, where sink populations are stream fish metapopulations and metacommunities 217 maintained by a rescue effect (Brown and and the “regional” scale over all ponds. Similar Kodric-Brown 1977). Mass effects models to typical stream habitats, connectivity among predict that species are different in their com- the ponds was variable depending on the flow petitive abilities in local habitat patches but into and out of them. The ponds were also het- are similar across the region among all habitat erogeneous in habitat, ruling out the patch dy- patches, on average, so that none are driven ex- namics and neutral models. The authors used tinct. Habitats are heterogeneous in the mass an integrated observational and experimental effects model, so it is spatially explicit. How- approach to investigate whether the mass ef- ever, spatial processes are predicted to be more fects or species-sorting paradigms better char- important than environmental factors in con- acterized this metacommunity. Overall, the trolling community composition, so that some authors found that despite high rates of zoo- species are present in habitat patches in which plankton dispersal within and among ponds, they could not persist without dispersal from there was little evidence for pure spatial mass source populations. effects (Cottenie and De Meester 2005). Most Neutral model.—The final metacommunity variation in zooplankton community structure paradigm is based on neutral theory (Hubbell could be explained by environmental variables, 2001; Figure 2D). The neutral model assumes conforming closely to the species-sorting that all species are equal in niche relations and model; zooplankton species occurred in the dispersal ability. This results in metacommu- habitats to which they were best adapted. nity dynamics influenced by slow random pat- Mouillot (2007) developed a theoretical terns of compositional change due to random perspective for coastal brackish lagoon fish extinction, dispersal, and speciation, termed metacommunities based on the four para- ecological drift. The neutral model assumes no digms introduced by Leibold et al. (2004). habitat heterogeneity and moderate dispersal The author discounted the neutral paradigm as and so is spatially implicit. being appropriate for this metacommunity be- Initial tests of metacommunity theory at- cause coastal fish species are not ecologically tempted to apply the four paradigms across eco- equivalent. His subsequent discussion focused logical systems and taxa (Leibold et al. 2004; on how coastal lagoons might be managed dif- Cottenie 2005; Holyoak et al. 2005). A few in- ferently depending on which metacommunity vestigators have tested the theory in aquatic sys- model fit the best (e.g., patch dynamics, spe- tems, but none have specifically tested hypoth- cies-sorting, mass effects). He concluded that eses for stream fishes. Therefore, here we briefly although it is unlikely that these lagoon habi- review the few applications of metacommunity tats would fit only one model best, one para- theory in other aquatic systems. digm might predominate. He also pointed out The most extensive and empirical inves- that movement beyond “conjecture and specu- tigation into metacommunity dynamics in lation” to careful tests of the metacommunity aquatic systems is that of Cottenie and others paradigms in real ecosystems is warranted, a for zooplankton in a stream-like system of in- conclusion we support. terconnected ponds in Belgium (Cottenie et al. The only paper we found that addressed 2003; Cottenie and De Meester 2004, 2005). stream fish metacommunities was a meta- The authors investigated the relative effects of analysis by Cottenie (2005). The author evalu- habitat heterogeneity and dispersal on com- ated 158 published data sets in the ecological munity structure at the scale of local ponds, literature and used a multivariate variation 218 falke and fausch partitioning approach to categorize each meta- Metacommunity Dynamics in a Great community by the relative influence of local Plains Stream Fish Assemblage habitat heterogeneity and regional dispersal processes. Three of the data sets analyzed (www. Study System epa.gov; Marsh-Matthews and Matthews 2000; Our study system was the Arikaree River, a Townsend et al. 2003) included 11 stream fish principle tributary of the Republican River metacommunities. Five classifications resulted: that flows northeast from Colorado into Kan- (1) neutral model or patch dynamics, (2) spe- sas and has its confluence in southwest Ne- cies sorting, (3) species sorting and mass ef- braska with the North Fork Republican River fects, (4) undetermined metacommunity, and near Haigler, Nebraska (see Scheurer et al. (5) no metacommunity dynamics found. Of 2003 for basin map and study segments). Most the 11 stream fish metacommunities, one was of the Arikaree River basin (>96%) is located classified as neutral or patch dynamics, three as in Colorado. Eastern Colorado is a semi-arid species-sorting, four as species-sorting and mass region, averaging less than 44 cm of rainfall effects, and three as undetermined metacom- annually. The primary source of flow for the munities. These results indicate that at least for Arikaree is groundwater that originates in the small sample of stream fish metacommuni- the underlying High Plains Aquifer. The flow ties considered, all showed some form of meta- regime in the Arikaree is predictable across community dynamics and most were controlled months but variable among years. High flows by strong niche relationships, regional dispersal, generally occur in May and June from a com- or a combination of these. bination of groundwater recharge and spring What metacommunity processes, if any, precipitation. Flows decline during summer dominate in stream fish assemblages? The re- and are further reduced by pumping shallow view of metapopulation dynamics showed that alluvial groundwater for irrigation and riparian source-sink dynamics are widespread across evapotranspiration (ET). Once pumping and stream fish populations. This is not surpris- ET cease in the fall, flow resumes and increases ing because high environmental heterogene- gradually until the following spring. This is a ity and dispersal among habitats and streams harsh environment for fishes because water are common in stream fishes (Gowan et al. quality can be poor and flows low in the winter 1994; Poff 1997). Given this, we might expect and summer. During summer, water tempera- patch dynamics and neutral dynamics to be tures in shallow habitats can exceed 348C and rare in stream fish metacommunities, because dissolved oxygen levels often are less than 0.1 in these models patches are assumed to have mg/L (Scheurer et al. 2003). Conversely, dur- the same habitats. This leaves the species-sorting winter, shallow habitats can freeze entirely ing and mass effects models, which deal with (e.g., Labbe and Fausch 2000). As a result, fish- strong niche relationships and strong dispersal es native to the Arikaree include species toler- mediating assemblage structure, respectively. ant to extremes in environmental conditions. However, what is the relative influence of species sorting and mass effects on stream fish as- Fish and Habitat Characteristics semblage structure, and how might one determine this? In the following section, we attempt The Arikaree fish assemblage is mainly com- to answer these questions for a Great Plains posed of 11 native species collected since 2000. stream fish assemblage using empirical data. Most are from the family Cyprinidae, but oth- stream fish metapopulations and metacommunities 219 ers represent Catostomidae, Centrarchidae, factors (i.e., local habitat characteristics), Fundulidae, and Ictaluridae. Mesohabitat whereas spatial variables would not explain types consist of pools connected by shallow significant variation. Conversely, if spatial fac- runs. In the spring, flooded terrestrial vegeta- tors (e.g., regional dispersal opportunity or tion along the stream margin and in connected habitat connectivity) explain variation in the backwaters provide spawning habitats. During fish assemblage better than local habitat, we summer drying, pools become disconnected, would conclude that the mass effects model and some desiccate entirely. The remaining may be more appropriate. If neither environ- pools provide important refugia for fishes. As mental nor spatial factors explain significant a result, opportunity for fishes to disperse to variation, we would conclude that (1) patch both spawning and refuge habitats is critical dynamics or neutral models may be more ap- for survival and population persistence. Also, propriate, (2) metacommunity dynamics for flow permanence varies longitudinally along this system cannot be determined, or (3) no the river basin depending on inputs of ground- metacommunity dynamics are occurring in water. During summer drying, flow is highest this system. Table 1 in Cottenie (2005) pres- and most perennial in upstream reaches of the ents the decision tree used to evaluate the basin and decreases downstream as the allu- significance structure and metacommunity vium becomes disconnected from the regional types. aquifer (Falke 2009). This results in a gradient Fishes native to the Arikaree riverscape are of flow and environmental conditions along relatively small-bodied, short-lived, and able the Arikaree riverscape. to disperse rapidly to recolonize habitats that were previously dry (Scheurer et al. 2003). Predictions These fishes also have adaptations that allow Significant variation in habitat quality and sea- them to maximize their reproductive poten- sonal dispersal opportunity among habitats tial in this harsh, dynamic system such as (1) makes the Arikaree riverscape an ideal place egg placement strategies to avoid smothering to test metacommunity theory. Our analyses or abrasion in shifting substrates, (2) short focused on determining the relative influence egg incubation time, and (3) fast larval growth of environmental (habitat composition) and and maturation. Finally, most fishes native to spatial (dispersal opportunity) factors in ex- this system feed across trophic levels, so om- plaining variation in the fish assemblage of nivory is prevalent. Because many of these spe- the Arikaree River. We used a decision tree cies have similar resource requirements (and developed by Cottenie (2005) to determine presumably habitat requirements), and these metacommunity type based on the results habitats are extremely dynamic seasonally, we of statistical tests. Metacommunity type was predicted that local environmental conditions determined by the significance structure of would be less important in explaining fish as- the components of variation (based on a = semblage structure than the opportunity to 0.05; see Statistical Methods for details). For move among habitats when environmental example, from metacommunity theory, we conditions deteriorate or resources become expect that if the species-sorting model was limiting. Based on these factors, we predicted most applicable to the Arikaree metacom- that the Arikaree River fish metacommunity munity, variation in fish assemblage structure would conform most closely to the mass ef- would be best explained by environmental fects model. 220 falke and fausch Data Collection was estimated by making depth measurements (nearest 0.01 m) at three stations located at Our data collection focused on evaluating both one-sixth, one-half, and five-sixths of each re- local scale habitat characteristics and regional spective width transect. Pool area, volume, and dispersal opportunity for Arikaree River fish- average depth were calculated after Platts et al. es. We collected these data during the driest (1983). Additionally, at each depth station, period (August) of 2006 and 2007, along two substrate (sand, silt, gravel, and bedrock) and 6.4-km river segments (upstream and middle) aquatic vegetation (emergent, submergent, or identified by Scheurer et al. (2003). These floating) categories were recorded. Maximum segments vary in environmental conditions depth of each pool was measured with a stadia and are characterized by differences in stream rod (cm), and conductivity (mS) was mea- flows. During these 2 years, the upstream seg- sured using a multimeter (YSI Inc., model 85). ment was mostly perennial, whereas the mid- Ambient surface water temperature (nearest dle segment was mostly intermittent. 0.18C) was recorded with a digital thermome- The minimum unit for the habitat- mea ter (Cooper-Atkins, Versatuff Plus 396). Qual- surements was a pool. Pools along the entire itative estimates of the proportion of pool area upstream and middle segments were censused covered by each vegetation type (see above for in late July of both years. We identified and categories) and tumbleweeds were recorded, georeferenced each pool, and measured length as well as a count of the number of pieces of (m), width at the midpoint (m), and maximum small (#4 cm diameter) and large (>4 cm di- depth (cm). These measurements allowed esti- ameter) woody debris in each pool. mating maximum surface area (m2) and volume In addition to detailed local habitat mea- (m3) of each pool. Subsequently, in August of surements, we evaluated spatial factors that each year (about 2 weeks after the census), we may explain variation in the Arikaree River fish conducted a detailed survey to measure local assemblage. At the same time as habitat sam- scale habitat characteristics. In the upstream pling each year, we measured among-habitat segment, a subset of pools (N = 29 of 180 in connectivity as the length of flowing, inter- 2006, N = 19 of 218 in 2007) were randomly mittent, and dry reaches along each 6.4-km selected from two pool size categories (small segment. We georeferenced the start and end and large) for detailed habitat measurements. of each connectivity class (flowing, intermit- In the middle segment, all pools were sampled tent, and dry) and created a GIS line layer with in both years (N = 27 in 2006, N = 29 in 2007). reaches classified by connectivity. We also di- Within each pool, we measured pool area, vided each of the two segments into relatively volume, maximum depth, composition and homogeneous reaches based on geomorphol- distribution of substrate particles and aquatic ogy and patterns of drying. Geomorphology vegetation, presence and proportion coverage and drying were evaluated using U.S. Geo- of woody debris and tumbleweeds, up- and logic Survey topographic maps and maps of downstream flow connectivity, turbidity, con- monthly within-segment habitat connectivity ductivity, and surface temperature. Surface collected from 2000 to 2007 (see Scheurer et area was quantified by (1) measuring length al. 2003). Longitudinal position of each pool along the longest axis of the pool, (2) divid- along the riverscape was determined by mea- ing the length evenly into three perpendicu- suring the distance from the downstream most lar transects, and (3) measuring width at the boundary of the middle segment upstream to midpoint of each transect. Pool volume (m3) stream fish metapopulations and metacommunities 221 each pool (m) using GIS. Finally, we measured best explained variation in the fish assemblage the distance from each focal pool sampled for data. We tested all variables for multicollinearity detailed habitat conditions to the nearest adja- using pairwise matrices based on Pearson’s cor- cent pool (m) and recorded whether or not the relation coefficient. If the correlation coefficient focal pool was isolated from the adjacent pools between a pair of variables was greater than 0.8, upstream and downstream. one of the variables was excluded from our anal- We sampled fishes from the pools that ysis. Variables to be included within each of the were previously surveyed for detailed habi- two final matrices were evaluated by forward tat measurements within 2 weeks of habitat selection (ter Braak and Verdonschot 1995; ter measurements. Fishes were collected with Braak and Smilauer 2002) with a Monte Carlo 4.8-mm mesh seines using three-pass deple- test (a = 0.05). Each CCA was computed using tion sampling. Pools were isolated at their CANOCO 4.5 (ter Braak and Smilauer 2002), up- and downstream ends with block nets of and analyses were based on interspecies distanc- the same mesh size to prevent emigration of es and biplot scaling. All Monte Carlo permuta- fishes during sampling. All seining passes were tion tests were done under the reduced model conducted from upstream to downstream. All with 999 permutations. fishes were identified to species and enumer- We tested 11 pool habitat variables using ated separately for each pass, and all fishes CCA for inclusion into our environmental were released unharmed following processing. matrix: maximum depth (MAXD), volume Abundance estimates and standard errors for (VOL), conductivity (COND), turbidity each species within individual pools were esti- (TURB), small woody debris (SMWOOD), mated using the generalized removal model in tumbleweeds (TUMBLE), submergent veg- Program CAPTURE (White et al. 1982). etation (SUBMERG), emergent vegetation (EMERG), and floating vegetation (FLOAT). Statistical Methods Only sand and silt substrate types were ob- We used canonical correspondence analysis to served in our field data, so we calculated the partition environmental and spatial variation in percent of the pool that was composed of sand the Arikaree River fish data (Palmer 1993; ter substrate (SAND). We adjusted pool tempera- Braak and Verdonschot 1995; Legendre and ture measurements taken at different times of Legendre 1998). Canonical correspondence day by calculating residuals from a regression 8 analysis (CCA) selects a linear combination of of pool ambient surface temperature ( C) ver- explanatory variables to maximize the disper- sus time of day (e.g., 1240 hours). This resulted sion of species scores (Jongman et al. 1995). in a variable (TEMP) that indicated whether Using this method, partitioning the variation a pool was relatively warm (positive residual) allowed us to test for the effects of one set of or relatively cool (negative residual) compared explanatory variables while holding the other to others and accounted for the potential influ- set constant. Consequently, we were able to ence of groundwater on pool temperature. All evaluate the relative influence of spatial versus environmental variables (excluding TEMP) environmental factors on variation in fish as- were transformed to ensure normal distribu- semblage structure. tion. Maximum depth, volume, and conductivity were log -transformed, and the other We began our statistical analyses by cal- 10 culating separate CCAs to determine which variables (all proportions) were transformed environmental and spatial variables in each set by calculating the arcsine of their square root. 222 falke and fausch Following transformation, all environmental We partitioned the variation in fish as- variables were standardized and centered by semblage structure among pools into the fol- calculating their z-scores (mean = 0; standard lowing components using partial canonical deviation = 1). After evaluation, eight variables correspondence analysis (pCCA; following explained significant environmental variation Borcard et al. 1992; Cottenie 2005; Langen- in the fish assemblage among pools: MAXD, heder and Ragnarsson 2007): environmental VOL, TEMP, TURB, TUMBLE, SUBMER, variation [E]; spatial variation [S]; total ex- and SAND. Subsequently, these variables com- plained variation [E+S]; pure environmental prised the environmental matrix [E] in subse- variation (proportion variation explained by quent partitioning analyses. environmental factors independent of space) Variables tested for inclusion into the spa- [E|S]; pure spatial variation (proportion ex- tial [S] matrix were those we considered might plained by space independent of environment) influence dispersal opportunity among habitats [S|E]; the spatial component of environmen- for fishes in the Arikaree metacommunity. This tal influence (variation shared by environment CCA was performed as described for [E]. To and space; [E ∩ S], calculated as [E] – [E|S]); evaluate species patchiness on the riverscape, and the total amount of unexplained variation we assigned each pool to one of the 11 geo- (1 – [E +S]). These calculations were based morphic reaches identified across the two study on eigenvalues resulting from the pCCA pro- segments. These variables were entered as 11 cedure. All results are presented as the propor- dummy variables (REACH0 to REACH10). tion of the total explained variation. To classify Additional discrete spatial variables included our Arikaree River fish metacommunity to one segment in which the pool was located (i.e., up- of the four MC paradigms, we used the deci- stream or middle; SEGMENT) and whether sion tree presented by Cottenie (2005). the pool was isolated from all adjacent habitats Results and Discussion (ISOLATED). We also included the longitudinal position of the pool along the riverscape Mean values of the 19 environmental and spa- (LONG) and the distance to the nearest pool tial variables within a segment were generally (DISTBW). Our final spatial variable indi- similar among years (Table 2), but values dif- cated the amount of flowing water habitat sur- fered between segments and reflected the het- rounding a focal pool at three different spatial erogeneity in habitats. Percent sand substrate scales. From our connectivity maps and geo- was higher in the middle segment in 2006, referenced pool locations, we calculated the and variables representing aquatic vegetation amount of flowing water habitat upstream and coverage were usually lower in the middle downstream from each pool at three scales: than upstream segment. Therefore on aver- 50, 250, and 500 m upstream and downstream age, the upstream segment contained siltier, (100M, 500M, 1000M, respectively). Finally, more vegetated pool habitats, whereas pools z-scores were calculated for all continuous in the middle segment were sandier and less variables (excluding LONG). After evalua- vegetated. Factors representing water quantity tion, 8 of the 18 variables explained significant in both environmental and spatial sets (e.g., variation in the fish assemblage and were in- VOL; 1000M) were usually higher in the up- cluded in the final spatial matrix [S]: 1000M, stream than middle segment. Also, more pools REACH1, REACH2, REACH3, REACH4, were isolated in the middle than in the up- LONG, DISTBW, and ISOLATED. stream segments and distance between pools stream fish metapopulations and metacommunities 223 Table 2. Mean (SE) values for environmental and spatial variables used to partition variation in the stream fish assemblage of the Arikaree River, Colorado across two 6.4-km stream segments and2 years. Variables in bold explained significant variation and were included in final modelsP ( < 0.05). Number of pools included from each year is shown. See text for detailed descriptions of variables. Upstream segment Middle segment Variable 2006 2007 2006 2007 Environment Maximum depth (m) 0.76 (0.07) 0.76 (0.06) 0.40 (0.03) 0.47 (0.03) Volume (m3) 12.63 (2.51) 14.96 (3.27) 3.61 (1.24) 7.58 (1.28) Residual temperature 0.18 (0.24) –0.05 (0.53) –0.19 (0.54) –0.17 (0.43) Conductivity (µS) 498.55 (16.95) 547.32 (27.11) 753.67 (30.86) 695.45 (28.18) Turbidity (%) 63.10 (6.28) 61.84 (7.28) 32.96 (5.87) 43.97 (4.75) Small woody debris (%) 4.66 (2.08) 1.58 (1.58) 3.52 (1.77) 1.03 (0.58) Tumbleweeds (%) 0.00 (0.00) 0.00 (0.00) 9.44 (2.30) 6.21 (1.73) Submergent vegetation (%) 27.59 (6.82) 29.74 (4.90) 31.11 (6.37) 23.28 (5.62) Emergent vegetation (%) 32.41 (4.90) 22.89 (2.82) 10.74 (2.38) 7.93 (1.52) Floating vegetation (%) 42.76 (7.44) 25.53 (7.95) 1.11 (1.11) 16.38 (4.88) Sand substrate (%) 25.29 (42.11) 42.11 (7.95) 74.49 (4.47) 49.43 (4.29) Space Segment – – – – – – – – 100 flow (m) 19.38 (6.73) 54.21 (10.55) 32.74 (8.82) 13.07 (5.45) 500 flow (m) 103.10 (27.83) 263.32 (51.18) 148.30 (38.21) 78.41 (17.80) 1,000 flow (m) 195.76 (43.02) 505.32 (87.19) 251.93 (55.14) 195.28 (27.47) Geomorphic reach – – – – – – – – Riverscape longitudinal position – – – – – – – – Distance to nearest pool (m) 24.76 (3.71) 17.47 (3.49) 73.63 (28.20) 36.28 (5.53) Number isolated 4 10 15 16 Number of pools 29 19 27 29 was higher in the middle segment than the When spatial variables were considered, or- upstream segment. Therefore in general, the angethroat darter Etheostoma spectabile, central upstream segment was wetter and contained stoneroller Campostoma anomalum, and creek more complex habitats than the middle seg- chub were associated with wet reaches (high ment, which was drier with simpler habitats 1000M), whereas white sucker was associated spaced farther apart. with pools located close together (low DIST- Ordination revealed patterns in fish as- BW) and pools that were not isolated (Figure semblage structure influenced by both envi- 3B). Fathead minnow Pimephales promelas was ronmental and spatial factors (Figure 3A and relatively ubiquitous so scores for this species 3B). For example, white sucker Catostomus were near the origin in both ordinations. commersonii and creek chub were associated All components of the variation partition- with large, deep pools with low conductivity, ing (E, S, E|S, S|E and E+S) explained signifi- whereas green sunfish Lepomis cyanellus and cant (P < 0.05) variation in the Arikaree fish black bullhead Ameiurus melas were associated assemblage, indicating that metacommunity with turbid, silty pools (Figure 3A; see Appen- dynamics do occur in this assemblage. The dix A for species codes). These habitat associa- total amount of variation [E + S] explained tions are similar to those reported for these spe- in the metacommunity by the environmental cies (Cross and Collins 1995; Pflieger 1997). and spatial factors was 50.8% (Appendix B). 224 falke and fausch

Figure 3. Two canonical correspondence analyses of fish assemblage data across 104 pool habitats located along the Arikaree River, Colorado. The top panel (A) shows the results of fish assemblage versus environmental factors, and the lower panel (B) shows fish assemblage versus spatial factors. Codes for environmental and spatial variables are explained in the text. Species codes are the first three letters of the genus and specific epithet, respectively. Species and codes are listed in Appendix A.

Pure environmental effects accounted for only ([E∩S]) accounted for 12.9% of the variation 6.5% of the variation explained, whereas pure in fish assemblage structure. A total of 49.2% spatial effects accounted for 31.4%. The over- of the variation was undetermined. Overall, lap among environmental and spatial factors spatial factors and environmental factors that stream fish metapopulations and metacommunities 225 were influenced by space explained the major- tat quality. Clearly, there is a strong influence ity of variation that could be explained in the of space on structuring stream fish populations Arikaree fish assemblage (44.3% of 50.8%). and assemblages, but more research is needed Because both pure environmental and pure across different ecoregions and stream sizes spatial variation components were significant, before generalizations can be made about the we cannot rule out that both species-sorting relative influence of environmental and spatial and mass effects dynamics are occurring in this factors in determining assemblage structure. assemblage during this dry portion of the year along the Arikaree riverscape (Cottenie 2005). Future Directions However, because the majority of variation was Metapopulations.—Overall, we know that explained by spatial factors, we conclude that stream fishes require multiple habitat types mass effects most likely dominate metacom- that are often dispersed across riverscapes munity dynamics in this system. to carry out their life histories (Dunning et We found that regional-scale spatial fac- al. 1988; Schlosser and Angermeier 1995; tors (e.g., distances among habitats, longitu- Fausch et al. 2002). These habitats may vary in dinal position, and habitat isolation) were im- quality and quantity across space. Therefore, portant in structuring stream fish assemblages it makes sense that spatially structured stream in pools along the Arikaree River. Our results fish populations may be best described by a were not surprising given presumed strong ef- hybrid metapopulation model (Figure 1C). fects of drying within and between habitats on Within “source” areas, all habitats (spawning, colonization and recolonization dynamics in rearing, and refuge) required for a species to this system. These results for the entire com- carry out its life history are available. These munity agree with those reported by Scheurer habitats are well connected, allowing for et al. (2003) for this system. For example, they population persistence over time. In satellite found that the probability of brassy minnow “sink” areas, some habitat type (e.g., spawn- persistence in pool habitats was highest in deep ing, rearing, refuge) is often missing, leading pools connected on at least one end to other to extinction of the subpopulation over time habitats. Our results show that pool isolation without emigration of individuals from the was highly influential in structuring the entire source area. Alternatively, mortality in the fish assemblage along the Arikaree River and sink habitat could be high due to predation. indicate that spatial factors affecting dispersal The prevalence of reports of the source-sink opportunity (e.g., habitat connectivity and iso- model in the stream fish metapopulation lit- lation) were more important than habitat qual- erature may be because only certain habitat ity (i.e., local pool attributes) in shaping this as- types were considered (e.g., only refuge and semblage. Falke and Gido (2006) also showed nonrefuge) or because the metapopulation that a pure spatial factor (distance from a res- was considered at too fine a spatial scale. In ervoir) was important in explaining stream fish our opinion, most of the metapopulations in assemblage structure upstream of large reser- those studies might be more accurately classi- voirs in Kansas. Finally, Isaak et al. (2007) re- fied as the hybrid type. What is an appropri- ported that for Chinook salmon Oncorhynchus ate metapopulation model for stream fishes, tshawytscha, use of spawning habitats in Idaho, and how might one go about categorizing a factors such as habitat size and connection to population to a particular metapopulation other habitats were more important than habi- type? Below, we build the case for accepting 226 falke and fausch a hybrid metapopulation model for a Great extinctions are common due to harsh environ- Plains stream fish with which we are familiar. mental conditions (e.g., pool desiccation), but Metapopulation theory may be particu- these habitats, or newly created habitats, are larly applicable to plains fishes, given the het- quickly recolonized. Therefore, at the scale of erogeneous and dynamic nature of the habitats in the Arikaree River basin, the nonequilibrium they occupy. The brassy minnow in the Arika- model is probably not appropriate. Likewise, ree River is a good example of a species that although the patchy-population model is quite requires multiple habitat types. They spawn attractive for brassy minnow in the Arikaree, and their larvae rear in shallow vegetated back- given the heterogeneity of habitats and high waters (Copes 1975; Scheurer et al. 2003). An rate of dispersal among them, extinctions do ontogenic habitat shift occurs when juveniles take place, especially in harsher stream reach- are about 20 mm total length (Falke 2009). es (e.g., middle segment). This model would They move to the main channel where they probably describe well the spatial dynamics seek pools to continue growth during sum- of brassy minnow in a given stream reach that mer and presumably store energy for winter. contains all required habitats but would de- These pools vary in habitat quality depending scribe poorly the dynamics in reaches that do on their location along a river segment. Persis- not. tence of brassy minnow in a given pool over the What about the classic metapopulation summer is a function of how deep the pool is in model? The classic metapopulation model does June (which determines whether it dries com- not allow habitat patches to be heterogeneous pletely by August; Labbe and Fausch 2000) and requires that habitats be equally spaced. and how connected it is to other pools (Sch- In contrast, channel units (pools, riffles, runs, eurer et al. 2003). If they survive the summer, and backwaters) in Great Plains streams are brassy minnow must seek deep refuge pools the result of the complex interaction between in which to overwinter because shallow pools habitat forming events (e.g., intense thunder- in Great Plains streams can freeze to the bot- storms) and the habitat template represented tom during harsh winters (Labbe and Fausch by variability in geomorphology and ground- 2000). Therefore brassy minnow require at water connectivity. For example, spawning and least three specific habitat types (spawning, refuge habitats are aggregated in reaches with rearing, and refuge habitats) within a given high groundwater connectivity (Falke 2009). year, to reproduce and survive. Similarly, the source-sink model is probably This previous research showed that brassy too simple because it does not allow for com- minnow require complementary, heteroge- plementary habitats (Pulliam 1988). However, neous habitats, which indicates that a meta- it may be reasonable to assume a source-sink population model is needed. Which metapo- dynamic if the reach scale is being considered, pulation model best fits brassy minnow in the where reaches that contain all habitats required Arikaree River basin? Based on our key (see by brassy minnow are “sources” and those that Distinguishing among metapopulation types do not are “sinks.” above), we begin by ruling out the nonequi- Overall, metapopulation dynamics of librium model. The nonequilibrium model brassy minnow within the Arikaree River typifies a situation in which there are many lo- basin are best described by the hybrid meta- cal extinctions and infrequent recolonization population model, which incorporates both (Harrison 1991). Along the Arikaree, local patchy and source-sink population dynamics stream fish metapopulations and metacommunities 227 (Figure 1C). Patchy population dynamics oc- metacommunity dynamics (Mouillot 2007). cur in core segments (e.g., upstream segment), However, an important intermediate step where spawning, rearing, and refuge habitats would be to integrate the four metacommu- are available and well connected and allow nity paradigms into a model based on actual high among-habitat dispersal rates. Subpopu- life history attributes of a real metacommunity lations in core habitats do not go extinct. In and incorporate factors previously established contrast, downstream segments (e.g., middle to be important in regulating communities, segment) may not contain habitats needed for such as seasonal and environmental variabil- all life stages and act as outlying sinks. ity, across spatial scales. This approach could The metapopulation model that best de- generate hypotheses to test using empirical scribes the spatial population dynamics of research. brassy minnow also may change across spatial Streams are fluctuating environments with scales (Figure 1). Within a given reach, per- seasonal flow dynamics influenced mainly by sistence depends on availability of comple- climate. Although variable across stream eco- mentary habitats and high dispersal ability types, flow regimes in temperate streams typi- (connectivity) among those habitats. A patchy cally include periods of high and low flows, and metapopulation model would be most appro- fish assemblages in these streams have become priate within a reach. At the scale of multiple adapted to maximize reproductive capacity reaches, where complementary habitats are ag- and survival in these environments (Poff and gregated within a reach, a source-sink model Ward 1989; Poff 1996). Much of the variation may be appropriate. At the segment scale, a in stream habitat availability results from sea- hybrid model where within-segment dynam- sonal flow variability, and any model of stream ics are represented by patchy metapopulation fish metacommunity dynamics should explic- dynamics and among-segment dynamics are itly incorporate flow. represented by source-sink dynamics may be We predict that the relative importance most appropriate. And finally, at a regional of habitat heterogeneity and dispersal oppor- scale, given the decline in brassy minnow dis- tunity in structuring stream fish assemblages tribution across its range in eastern Colorado, changes along a gradient of seasonal variation a nonequilibrium model might be best. This in flows (Figure 4) and illustrate this con- model could account for reduced recoloniza- cept using our Great Plains fish assemblage. tion potential among basins resulting from In spring, species-sorting metacommunity permanent reductions in among-basin con- dynamics may dominate because individual nectivity due to diversions and reservoirs that species show strong preferences for specific create barriers and groundwater pumping that spawning habitats (Falke 2009). In the Arika- causes reaches to dry permanently. ree River basin, many species use specific Metacommunities.—Clearly, the four meta- habitats only available during spring, such as community paradigms proposed by Leibold et flooded terrestrial vegetation and off-channel al. (2004) are a heuristic tool, a starting point backwaters, and there may be limited move- for further exploration of how dispersal, habi- ment among them. The end of the spawning tat heterogeneity, and biotic interactions inter- season in the Arikaree coincides with drying act across scales to shape communities. Mov- of shallow habitats. During this period, fishes ing beyond these simple models to testable must move from spawning habitats into sum- hypotheses is the next step in understanding mer refugia. Drying can occur rapidly, so the 228 falke and fausch

Figure 4. Theoretical model of Great Plains stream fish metacommunity dynamics with emphasis on habitat heterogeneity, movement, and seasonal flow variation for three species, A–C (modified from Schlosser and Angermeier 1995; Leibold et al. 2004). Arrows connect seasonal macrohabitats (large open circles: spawning, summer refuge, and overwinter) and mesohabitats used within seasons (patches: smaller open circles, squares, and triangles). Solid arrows indicate rapid dispersal rates; dashed arrows indicate slower dispersal rates. During spring/early summer, species sort into appropriate spawning habitat patches based on life history requirements (denoted by small filled shape match- ing larger open shape). However, species A is a generalist that spawns in multiple habitat patch types. Metacommunity dynamics during this period are best described by the species sorting model. Dispersal among patches is restricted during this period. Individuals then quickly move from spring spawning to summer refuge habitat as flow declines. During summer, two habitat types are available, harsh (open circle) and benign (open triangle) refugia. The harsh refugia serve as sink habitats and may dry and re- wet frequently (dynamic habitat persistence represented by dashed border). Dispersal is rapid among harsh and benign refugia (mass effects) but is often impossible due to lost connectivity between habitats. As flows slowly resume into winter, individuals disperse to occupy all habitats (represented by an open hexagon). During winter, habitats are homogeneous, and dispersal among them is low due to cold water temperatures. Metacommunity dynamics during this period are best represented by the patch dynamics or neutral models. Following winter, individuals move slowly to spawning habitats, based on phenology. See text for a full description of metacommunity types. ability to disperse quickly from spawning to ref- if pathways among those habitats exist. Finally, uge habitats is critical. Therefore, we predict that during winter, flow in the Arikaree increases a transition occurs in metacommunity dynam- and habitats become reconnected. However, ics from species-sorting to mass effects, in which during this period, most habitats are fairly ho- dispersal dominates. As water quality condi- mogeneous, and dispersal may be limited due tions in refuge habitats degrade due to drying, to cold water temperatures. As a result, aspects movement to more suitable refugia may occur, of the neutral or patch dynamics models prob- stream fish metapopulations and metacommunities 229 ably better describe stream fish assemblage -dy funds and W. Burnidge of The Nature Conser- namics during winter flow conditions. vancy for logistical support.

Conclusions References Clearly, spatial factors have an important in- Borcard, D., P. Legendre, and P. Drapeau. 1992. Par- fluence on both population and community tialling out the spatial component of ecological variation. Ecology 73:1045–1055. dynamics of stream fishes. However, to date, Brown, J. H., and A. Kodric-Brown. 1977. Turnover research into how space influences population rates in insular biogeography: effect of immi- and community dynamics in stream fishes has gration on extinction. Ecology 58:445–449. been in a formative state. Future metapopula- Campbell Grant, E. H., W. H. Lowe, and W. F. Fa- tion level research should incorporate space gan. 2007. Living in the branches: population dynamics and ecological processes in dendritic realistically and also consider that metapo- networks. Ecology Letters 10:165–175. pulation types may vary across spatial scales. Caswell, H. 1978. Predator-mediated coexistence: Advances will likely be driven by better meth- non-equilibrium model. American Naturalist ods of estimating habitat-specific demography 112:127–154. Chaumot, A., N. Milioni, A. Abdoli, D. Pont, and S. (e.g., Hines 1994) and costs of movement in Charles. 2006. First step of a modeling approach terms of growth and mortality. We suggest that to evaluate spatial heterogeneity in a fish (Cottus metacommunity theory also could be useful gobio) population dynamics. Ecological Model- in developing hypotheses and experiments to ling 197:263–273. investigate the relative influence of spatial and Cooper, A. B., and M. Mangel. 1998. The dangers of ignoring metapopulation structure for the environmental factors in structuring stream fish conservation of salmonids. Fisheries Bulletin assemblages. However, less emphasis should be 97:213–226. placed on categorizing assemblages by meta- Copes, F. A. 1975. Ecology of the brassy minnow, community type, as this may vary seasonally, Hybognathus hankinsoni (Cyprinidae). Univer- and more on using appropriate statistical meth- sity of Wisconsin, Museum of Natural History, Stevens Point. ods to partition variation important in structur- Cottenie, K. 2005. Integrating environmental and ing assemblages during specific seasons. Finally, spatial processes in ecological community dy- we predict that this emerging body of research namics. Ecology Letters 8:1175–1182. could be valuable not only for basic research in Cottenie, K., and L. De Meester. 2004. Metacommu- stream fish ecology, but also multispecies con- nity structure: synergy of biotic interactions as selective agents and dispersal as fuel. Ecology servation and management. 85:114–119. Cottenie, K., and L. De Meester. 2005. Local interac- Acknowledgments tions and local dispersal in a zooplankton metacommunity. Pages 189–211 in M. Holyoak, M. A. Data collection and manuscript preparation Leibold, and R. D. Holt, editors. Metacommuni- were supported by a grant from the Colorado ties. University of Chicago Press, Chicago. Cottenie, K., E. Michels, N. Nuytten, and L. De Division of Wildlife (CDOW) to KDF. N. Meester. 2003. Zooplankton metacommunity Cathcart, J. Hammer, and A. Klug assisted with structure: regional vs. local processes in highly field data collection. K. Bestgen, D. Durnford, interconnected ponds. Ecology 84:991–1000. D. Winkelman, and two anonymous review- Cross F. B., and J. T. Collins. 1995. Handbook of ers made helpful comments that improved the fishes of Kansas (revised). University of Kansas Press, Lawrence. manuscript. We thank T. Nesler of CDOW for Crozier, L., and R. W. Zabel. 2006. Climate impacts supporting this research and administering at multiple scales: evidence for differential pop- 230 falke and fausch ulation responses in juvenile Chinook salmon. Gilliam, J. F., and D. F. Fraser. 2001. Movement in Journal of Animal Ecology 75:1100–1109. corridors: enhancement by predation threat, Dunham, J. B., and B. E. Rieman. 1999. Metapopula- disturbance, and habitat structure. Ecology tion structure of bull trout: influences of physi- 82:258–273. cal, biotic, and geometrical landscape charac- Gotelli, N. J. 2001. A primer of ecology. Sinauer As- teristics. Ecological Applications 9:642–655. sociates, Sunderland, Massachusetts. Dunham, J. B., B. E. Rieman, and J. T. Peterson. 2002. Gotelli, N. J., and C. M. Taylor. 1999. Testing meta- Patch-based models to predict species occur- population models with stream-fish assemblag- rence: lessons from salmonid fishes in streams. es. Evolutionary Ecology Research 1:835–845. Pages 327–334 in J. M. Scott, P. J. Heglund, M. Gowan, C., M. K. Young, K. D. Fausch, and S. C. L. Morrison, J. B. Haufler, M. G. Raphael, W. A. Riley. 1994. Restricted movement in resident Wall, and F. B. Samson, editors. Predicting spe- stream salmonids: a paradigm lost? Cana- cies occurrences: issues of accuracy and scale. dian Journal of Fisheries and Aquatic Sciences Island Press, Washington, D.C. 51:2626–2637. Dunning, J. B., B. J. Danielson, and H. R. Pulliam. Hanski, I. 1994. A practical model of metapopula- 1992. Ecological processes that affect population dynamics. Journal of Animal Ecology tions in complex landscapes. Oikos 65:169– 63:151–162. 175. Hanski, I. 1996. Metapopulation ecology. Oxford Fagan, W. F. 2002. Connectivity, fragmentation, and University Press, New York. extinction risk in dendritic metapopulations. Hanski, I. 2001. A spatially realistic theory of meta- Ecology 83:3423–3249. population ecology. Naturwissenschaften Fagan, W. F., P. J. Unmack, C. Burgess, and W. L. 88:372–381. Minckley. 2002. Rarity, fragmentation, and ex- Hanksi, I., and D. Simberloff. 1997. The metapopu- tinction risk in desert fishes. Ecology 83:3250– lation approach, its history, conceptual domain, 3256. and application to conservation. Pages 1–26 in Fagan, W. F., C. Aumann, C. M. Kennedy, and P. J. I. Hanski and M. E. Gilpin, editors. Metapopu- Unmack. 2005. Rarity, fragmentation, and the lation biology: ecology, genetics, and evolution. scale dependence of extinction risk in desert Academic Press, San Diego, California. fishes. Ecology 86:34–41. Harrison, S. 1991. Local extinction in a metapopula- Falke, J. A. 2009. Effects of groundwater withdrawal tion context: an empirical evaluation. Biological and drought on native fishes and their habitats in Journal of the Linnean Society 42:73–88. the Arikaree River, Colorado. Doctoral disserta- Harrison, S., and A. D. Taylor. 1996. Empirical evi- tion. Colorado State University, Fort Collins. dence for metapopulation dynamics. Pages 27–39 Falke, J. A., and K. B. Gido. 2006. Spatial effects of in I. Hanski, and M. E. Gilpin editors. Metapopu- reservoirs on fish assemblages in Great Plains lation biology: ecology, genetics, and evolution. streams in Kansas, USA. River Research and Academic Press, San Diego, California. Applications 22:55–68. Hastings, A. 1980. Disturbance, coexistence, history, Fausch, K. D., C. E. Torgersen, C. V. Baxter, and H. and competition for space. Theoretical Popula- W. Li. 2002. Landscapes to riverscapes: bridging tion Biology 18:363–373. the gap between research and conservation of Hendry, A. P., and S. C. Stearns. 2004. Evolution il- stream fishes. BioScience 52:483–498. luminated: salmon and their relatives. Oxford Froese, R., and D. Pauley. 2008. FishBase. Available: University Press, Oxford, UK. www.fishbase.org (July 2008). Hendry, A. P., V. Castric, M. T. Kinnison, and T. P. Gagen, C. J., R. W. Standage, and J. N. Stoeckel. 1998. Quinn. 2004. The evolution of philopatry and Ouachita madtom (Noturus lachneri) metapo- dispersal: homing versus straying in salmonids. pulation dynamics in intermittent Ouachita Pages 52–91 in A. P. Hendry, and S. C. Stearns, mountain streams. Copeia 1998:874–882. editors. Evolution illuminated: salmon and their Garant, D., J. J. Dodson, and L. Bernatchez. 2000. relatives. Oxford University Press, Oxford, UK. Ecological determinants and temporal stability Hilderbrand, R. H. 2003. The roles of carrying capac- of the within-river population structure in At- ity, immigration, and population synchrony on lantic salmon (Salmo salar L.). Molecular Ecol- persistence of stream-resident cutthroat trout. ogy 9:615–628. Biological Conservation 110:257–266. stream fish metapopulations and metacommunities 231 Hines, J. E. 1994. MSSURVIV users manual. National tribution to the conservation of an endangered Biological Service, Patuxent Wildlife Research percid (Zingel asper). Canadian Journal of Fish- Center, Laurel, Maryland. eries and Aquatic Sciences 63:650–659. Holt, R. D. 1993. Ecology at the mesoscale: the influ- Langenheder, S., and H. Ragnarsson. 2007. The role ence of regional processes on local communities. of environmental and spatial factors for the Pages 77–99 in R. E. Ricklefs, and D. Schluter composition of aquatic bacterial communities. editors. Species diversity in ecological commu- Ecology 88:2154–2161. nities: historical and geographical perspectives. Legendre, P., and L. Legendre. 1998. Numerical ecol- University of Chicago Press, Chicago. ogy. Elsevier, Amsterdam, Netherlands. Holt, R. D. 1985. Population-dynamics in 2-patch Leibold, M. A. 1998. Similarity and local co-existence environments: some anomalous consequences of species in regional biotas. Evolutionary Ecol- of an optimal habitat distribution. Theoretical ogy 12:95–110. Population Biology 28:181–208. Leibold, M. A., M. Holyoak, N. Mouquet, P. Ama- Holyoak, M., M. A. Leibold, and R. D. Holt. 2005. raskare, J. M. Chase, M. F. Hoopes, R. D. Holt, J. Metacommunities: spatial dynamics and ecolog- B. Schurin, R. Law, D. Tilman, M. Loreau, and ical communities. University of Chicago Press, A. Gonzalez. 2004. The metacommunity con- Chicago. cept: a framework for multi-scale community Hubbell, S. P. 2001. The unified neutral theory of bio- ecology. Ecology Letters 7:601–613. diversity and biogeography. Princeton Univer- Letcher, B. H., K. H. Nislow, J. A. Coombs, J. O’Donnell, sity Press, Princeton, New Jersey. and T. L. Dubreuil. 2007. Population response to Hutchinson, G. E. 1951. Copepodology for the orni- habitat fragmentation in a stream-dwelling brook thologist. Ecology 32:571–577. trout population. PLoS One 11:1–11. Isaak, D. J., R. F. Thurow, B. E. Rieman, and J. B. Dun- Levins, R. 1969. Some demographic and genetic con- ham. 2003. Temporal variation in synchrony sequences of environmental heterogeneity for among chinook salmon (Oncorhynchus tshaw- biological control. Bulletin of the Entomologi- ytscha) redd counts from a wilderness area in cal Society of America 15:237–240. central Idaho. Canadian Journal of Fisheries Levins, R. 1970. Extinction. Pages 77–107 in M. and Aquatic Sciences 60:840–848. Gesternhaber, editor. Some mathematical prob- Isaak, D. J., R. F. Thurow, B. E. Rieman, and J. B. lems in biology. American Mathematical Soci- Dunham. 2007. Chinook salmon use of spawn- ety, Providence, Rhode Island. ing patches: relative roles of habitat quality, Luttrell, G. R., A. A. Echelle, W. L. Fisher, and D. J. size, and connectivity. Ecological Applications Eisenhour. 1999. Declining status of two species 17:352–364. of the Macrhybopsis aestivalis complex (Teleo- Jongman, R. H. G., C. J. F. ter Braak, and O. F. R. van stei: Cyprinidae) in the Arkansas River basin Tongeren. 1995. Data analysis in community and related effects of reservoirs as barriers to and landscape ecology. Cambridge University dispersal. Copeia 1999:981–989. Press, Cambridge, UK. MacArthur, R. H., and E. O. Wilson. 1967. The theo- Kareiva, P. 1990. Population dynamics in spatially ry of island biogeography. Princeton University complex environments: theory and data. Philo- Press, Princeton, New Jersey. sophical Transactions of the Royal Society of Marsh-Matthews, E., and W. J. Matthews. 2000. Geo- London B 330:175–190. graphic, terrestrial and aquatic factors: which Koizumi, I., and K. Maekawa. 2004. Metapopulation most influence the structure of stream fish as- structure of stream-dwelling Dolly Varden charr semblages in the Midwestern United States? inferred from patterns of occurrence in the So- Ecology of Freshwater Fish 9:9–21. rachi River basin, Hokkaido, Japan. Freshwater Matsubara, H., H. Sakai, and A. Iwata. 2001. A river Biology 49:973–981. metapopulation structure of a Japanese fresh- Labbe, T. R., and K. D. Fausch. 2000. Dynamics of water goby, Odontobutis obscura, deduced from intermittent stream habitat regulate persistence allozyme genetic indices. Environmental Biol- of a threatened fish at multiple scales. Ecological ogy of Fishes 61:285–294. Applications 10:1774–1791. McMahon, T. E., and W. J. Matter. 2006. Linking Labonne, J., and P. Gaudin. 2006. Modelling popula- habitat selection, emigration and population tion viability in fragmented environments: con- dynamics of freshwater fishes: a synthesis of 232 falke and fausch ideas and approaches. Ecology of Freshwater Pulliam, H. R. 1988. Sources, sinks, and population Fish 15:200–210. regulation. The American Naturalist 1988:652– Moilanen, A., and I. Hanski. 1998. Metapopulation 661. dynamics: effects of habitat quality and land- Ricketts, T. H. 2001. The matrix matters: effective scape structure. Ecology 79:2503–2515. isolation in fragmented landscapes. American Morita, K., and S. Yamamoto. 2002. Effects of habitat Naturalist 158:87–99. fragmentation by damming on the persistence Rieman, B. E., and J. B. Dunham. 2000. Metapopula- of stream-dwelling charr populations. Conser- tions and salmonids: a synthesis of life history vation Biology 16:1318–1323. patterns and empirical observations. Ecology of Mouillot, D. 2007. Niche-assembly vs. dispersal-as- Freshwater Fish 9:51–64. sembly rules in coastal fish metacommunities: Rieman, B. E., and J. D. McIntyre. 1995. Occurrence implications for management of biodiversity in of bull trout in naturally fragmented habi- brackish lagoons. Journal of Applied Ecology tat patches of varied size. Transactions of the 44:760–767. American Fisheries Society 124:285–296. Neville, H. M., J. B. Dunham, and M. M. Peacock. Scheurer, J. A., K. D. Fausch, and K. R. Bestgen. 2003. 2006. Landscape attributes and life history vari- Multiscale processes regulate brassy minnow ability shape genetic structure of trout popula- persistence in a Great Plains river. Transactions tions in a stream network. Landscape Ecology of the American Fisheries Society 132:840–855. 21:901–916. Schick, R. S., and S. T. Lindley. 2007. Directed con- Palmer, M. W. 1993. Putting things in even better nectivity among fish populations in a riverine order: the advantages of canonical correspon- network. Journal of Applied Ecology 44:1116– dence analysis. Ecology 74:2215–2230. 1126. Pflieger, W. L. 1997. The fishes of Missouri. Missouri Schlosser, I. J. 1990. Environmental variation, life Department of Conservation, Jefferson City. history attributes, and community structure in Pichon, C. L., G. Gorges, P. Boet, J. Baudry, F. Gore- stream fishes: implications for environmental aud, and T. Faure. 2006. A spatially explicit management and assessment. Environmental resource-based approach for managing stream Management 14:621–628. fishes in riverscapes. Environmental Manage- Schlosser, I. J. 1998. Fish recruitment, dispersal, and ment 37:322–335. trophic interactions in a heterogeneous lotic en- Platts, W. S., W. F. Megahan, and G. W. Minshall. vironment. Oecologia 113:260–268. 1983. Methods for evaluating stream, riparian, Schlosser, I. J., and P. L. Angermeier. 1995. Spatial and biotic conditions. U.S. Department of Ag- variation in demographic processes of lotic fish- riculture, Forest Service, Intermountain Forest es: conceptual models, empirical evidence, and and Range Experiment Station, General Techni- implications for conservation. Pages 392–401 in cal Report INT-138, Ogden, Utah. J. L. Nielsen, and D. A. Powers, editors. Evolution Poff, N. E. 1996. A hydrogeography of unregulated and the aquatic ecosystem. American Fisheries streams in the United States and an examina- Society, Symposium 17, Bethesda, Maryland. tion of scale-dependence in some hydrological Shepard, B. B., B. E. May, and W. Urie. 2005. Status descriptors. Freshwater Biology 36:71–91. and conservation of westslope cutthroat trout Poff, N. L. 1997. Landscape filters and species traits: within the western United States. North Ameri- towards mechanistic understanding and pre- can Journal of Fisheries Management 25:1426– diction in stream ecology. Journal of the North 1440. American Benthological Society 16:391–409. Shmida, A., and M. V. Wilson. 1985. Biological de- Poff, N. L., and J. V. Ward. 1989. Implications of terminants of species-diversity. Journal of Bio- streamflow variability and predictability for lot- geography 12:1–20. ic community structure: a regional analysis of Slack, W. T., S. T. Ross, and J. A. Ewing. 2004. Ecolo- streamflow patterns. Canadian Journal of Fish- gy and population structure of the bayou darter, eries and Aquatic Sciences 46:1805–1818. Etheostoma rubrum: disjunct riffle habitats and Policansky, D., and J. J. Magnuson. 1998. Genetics, downstream transport of larvae. Environmental metapopulations, and ecosystem management Biology of Fishes 71:151–164. of fisheries. Ecological Applications 8:S119– ter Braak, C. J. F., and P. F. M. Verdonschot. 1995. S123. Canonical correspondence analysis and related stream fish metapopulations and metacommunities 233 multivariate methods in aquatic ecology. Aquat- White, G. C., D. R. Anderson, K. P. Burnham, and ic Sciences 57:255–289. D. L. Otis. 1982. Capture-recapture and removal ter Braak, C. J. F., and P. Smilauer. 2002. CANOCO methods for sampling closed populations. Los reference manual and CanoDraw for Windows Alamos National Laboratory, Report LA-8787- user’s guide: software for canonical community NERP, Los Alamos, New Mexico. ordination (version 4.5). Microcomputer Power, Whittaker, R. H. 1972. Evolution and measurement Ithaca, New York. of species diversity. Taxon 21:213–251. Tilman, D. 1994. Competition and biodiversity in Winston, M. R., C. M. Taylor, and J. Pigg. 1991. Up- spatially structured habitats. Ecology 75:2–16. stream extirpation of four minnow species due Townsend, C. R., S. Doledec, R. Norris, K. Peacock, to damming of a prairie stream. Transactions of and C. Arbuckle. 2003. The influence of scale the American Fisheries Society 120:98–105. and geography on relationships between stream Yodzis, P. 1986. Competition, mortality, and com- community composition and landscape vari- munity structure. Pages 480–491 in J. M. Dia- ables: description and prediction. Freshwater mond and T. J. Case, editors. Community ecol- Biology 48:768–785. ogy. Harper and Row, New York. Ward, J. V. 1983. The serial discontinuity concept of Young, K. A. 1999. Managing the decline of Pacific lotic ecosystems. Pages 29–32 in T. D. Fontaine salmon: metapopulation theory and artificial and S. M. Bartell, editors. Dynamics of lotic recolonization as ecological mitigation. Cana- ecosystems. Ann Arbor Science, Ann Arbor, dian Journal of Fisheries and Aquatic Sciences Michigan. 56:1700–1706.

Appendix A. Stream fish species common names, scientific names, and codes used in a partial canonical correspondence analysis of fish assemblage data collected in the Arikaree River, Colorado. Common name Scientific name Code Black bullhead Ameiurus melas AMEMEL Central stoneroller Campostoma anomalum CAMANO White sucker Catostomus commersonii CATCOM Orangethroat darter Etheostoma spectabile ETHSPE Plains killifish Fundulus zebrinus FUNZEB Brassy minnow Hybognathus hankinsoni HYBHAN Green sunfish Lepomis cyanellus LEPCYA Fathead minnow Pimephales promelas PIMPRO Creek chub Semotilus atromaculatus SEMATR

Appendix B. Eigenvalues and P values for a partial canonical correspondence analysis of stream fishes in the Arikaree River, Colorado. Total inertia = 0.957. Variation Eigenvalues Variation explained (%) P-value [E + S] 0.504 50.8 <0.001 [E] 0.186 19.4 <0.001 [S] 0.425 44.4 <0.001 [E | S] 0.062 6.5 0.005 [S | E] 0.301 31.4 <0.001 [E ∩ S] 12.9 1 – [E + S] 49.2