Читать книгу Ecology of North American Freshwater Fishes - Stephen T. Ross Ph. D. - Страница 15
ОглавлениеFOUR
Responses of Populations and Assemblages to Biotic and Physical Factors
CONTENTS
Relating Assemblages to the Environment
Multivariate Statistics and Fish Assemblages
Local versus Regional Effects on Assemblages
FISHES ARE CONFRONTED by an environment that is complex and heterogeneous, with components of their habitat changing on multiple temporal and spatial scales. For instance, water temperature or turbidity might change on an hourly or daily frequency, altering the suitability of certain habitats; in contrast, substrata might change over a longer time period of days, months, years, or decades, and basic structures, such as shoreline characteristics in lakes or riffle–pool sequences in streams, might vary on a scale of many months to tens or hundreds of years.
Various approaches have been used to describe the dynamics of local habitats, albeit most have been based on terrestrial systems (Wu and Loucks 1995). While developed initially for estuarine systems, a useful approach to understanding the freshwater habitat mosaic is to view it holistically as an environment possessing both dynamic (short-term physicalchemical and biotic variability, prey fields, and predator occurrence) and static (long-term structural variability, sediment type, and shoreline context) components, each having its own influence on fishes (Peterson 2003; Peterson et al. 2007). The timing, positioning, and amount of overlap between the dynamic and the stationary components thus control the suitability of local habitats or patches.
The distinction between the terms “habitat” and “environment” has garnered considerable debate (e.g., Ryder and Kerr 1989). Considered in the static-dynamic dimensions of Peterson (2003), habitat comprises the “localized structured component that acts as a template” for organisms, whereas environment is “the sum of the biotic and abiotic surroundings, including habitat and other organisms.” Static and dynamic features are, of course, not independent, and in the long run, dynamic features can influence “static” features. For instance, the bottom type in a stream, such as a coarse gravel substratum, is a function of the interactions of stream discharge, gradient, and bed materials. In the short term, changes in water depth, velocity, temperature, and quality across a gravel bar can alter the suitability of the coarse gravel substratum for particular fish species. For instance, in the Colorado River below Glen Canyon Dam, the hypolimnial release from the dam has greatly reduced ambient water temperature as well as turbidity. Annual water temperatures now range from 9 to 14° C and turbidity is very low; prior to the dam, summer water temperatures reached to near 30° C and the river carried extensive suspended sediments (Blinn and Poff 2005). As a consequence, native fishes such as Bonytail and Humpback chubs (Gila elegans and G. cypha) are now unable to spawn in what otherwise had been favorable main channel habitats and, in fact, are largely eliminated from the main channel (Minckley et al. 2003).
LANDSCAPE ECOLOGY
Aldo Leopold, one of the founders of the conservation movement in America, developed the concept of a land ethic, espoused, for instance, in the posthumously published A Sand County Almanac (Leopold 1949). The term “landscape ecology” was first used by a German scientist, C. Troll, in 1939 (Turner 1989), and the initial development of the field took place largely in Europe where it emphasized terrestrial patterns and humans as part of the landscape (Wiens 2002; Turner 2005). The field of landscape ecology thus deals with spatial patterns, the effect of temporal and spatial scales on how organisms perceive and respond to patchiness within the environment, and linkages among the elements within the pattern.
During the period from 1940 through the 1970s, terrestrial and aquatic ecologists, by and large, continued to view their study systems as distinct units rather than as being part of an interconnected landscape, although they might infer landscape-level processes as being important in creating the ecological pattern they were studying (Turner 1989). For example, longitudinal changes in fish faunas (the patterns) were related to changes in stream sizes and gradients. The maturation of the field of landscape ecology occurred when ecologists began to study the effects of spatial patterns on ecological processes (Turner 1989). The development of the River Continuum Model, treated later in this chapter, incorporated the role of spatial patterns in watersheds on processes occurring at different points or patches in the watershed, such as changes in energy sources and functional groups of organisms, and thus represents the incorporation of a maturing view of landscape ecology.
On one level, all of us have an inherent feeling of what constitutes a landscape. A more formal, albeit terrestrially biased, definition of a landscape is “a kilometers-wide area where a cluster of interacting stands or ecosystems is repeated in similar form” (Forman and Godron 1981); it is important, however, to emphasize that landscapes can vary greatly in size. Although terrestrially focused landscape ecologists have tended to view streams and lakes as boundaries, water bodies have their own internal heterogeneity, and it is important to recognize aquatic landscapes as well as terrestrial ones (Wiens 2002). A definition more amenable to aquatic systems considers a landscape as “an area that is spatially heterogeneous in at least one factor of interest” (Turner 2005).
Terrestrial and aquatic landscapes are usually defined by the most obvious geomorphic, hydrologic, vegetational, or land-use features. Boundaries between adjacent landscapes are defined by distinct changes in spatial elements, such as a change from a riverine landscape, with patches defined by substratum and water mass characteristics, to a riparian landscape, with complex bank side vegetation, to a drier upland landscape (Figure 4.1). Landscapes are, however, interrelated so that impacts in upland and/or riparian landscapes can affect aquatic landscapes. For instance, fish species richness in the Aspen Parkland Ecoregion of Manitoba, Canada, was positively related to the quality of the terrestrial landscape within the catchment (Wilson and Xenopoulos 2008).
FIGURE 4.1. A section of the San Juan River in Utah showing the complex patterns and juxtaposition of riverine, riparian, and upland landscapes. Aquatic habitats expand during high flow, which can also restructure the streambed and adjoining wetlands. Land and river imagery provided by R. Bliesner, Keller-Bliesner Engineering.
Landscapes themselves can be grouped into regions so that a region contains several to many landscapes and is a broad geographical area that may be ecologically diverse but that has a “common macroclimate and sphere of human activity and interest” (Forman 1995). Elements within a landscape include a background matrix, patches, and corridor, all of potentially varying shapes and sizes (Forman 1995). As perceived by organisms, what is termed a patch or a corridor can vary, so that patches for one species might become part of the surrounding matrix for another. However they are defined, the recognition that habitats are themselves embedded in an expanding hierarchy of other patches, and that the recognition of patches by organisms is scale dependent, are perhaps two of the most important outcomes from the emerging field of landscape ecology (Kotliar and Wiens 1990; Grossman et al. 1995; Wiens 2002).
In small headwater streams, three important landscape attributes are (1) interactions at the interface between terrestrial and aquatic habitats, (2) how habitats are related spatially on a large scale, and (3) how refuges that provide relief from harsh environmental conditions are spread across the landscape (Schlosser 1995a). With relatively minor modifications, these attributes remain important for fishes in other freshwater habitats, including lakes and large rivers (see Part 5).
Patches
Landscapes are mosaics of patches, which are spatial units that differ from other such units in terms of function or appearance (Kotliar and Wiens 1990). How fishes perceive patches is controlled to a large extent by their size and life-history stage. The smallest scale at which an organism responds to patch structure is referred to as “grain,” and the largest scale, equivalent to its lifetime home range, is referred to as “extent” (Kotliar and Wiens 1990). Grain and extent are organism-specific traits so that a benthic-oriented darter such as the Naked Sand Darter (Ammocrypta beani), which occurs primarily over clean sand (Ross 2001), might have a grain on the order of a millimeter or less as defined by the size of sand grains. In contrast, a water column minnow such as a Blacktail Shiner (Cyprinella venusta) might have a grain defined by water mass characteristics (e.g., depth, flow, and temperature) with a size on the order of centimeters or meters.
FIGURE 4.2. The distribution of Bayou Darters (Nothonotus rubrum) in Bayou Pierre, Mississippi, illustrating a three-level hierarchy of habitat patches. The closed circles in the top map show locations for Bayou Darters and the total range of circles thus defines the maximum extent. The bottom detailed map shows the distribution of riffles in one section of Bayou Pierre where each riffle would correspond to a habitat patch; within a single riffle (a size of approximately ≤ 10 m), Bayou Darters are selecting particular substratum sizes. Data from Ross et al. (1990, 1992b, 2001) and Slack et al. (2004).
The distribution of patches in an organism’s environment can be hierarchically nested between its grain and extent (Figure 4.2). For example, the Bayou Darter (Nothonotus rubrum) is endemic to Bayou Pierre, a tributary of the Mississippi River, where it is restricted to riffle habitats having a coarse, firm substratum (Ross et al. 1990, 1992b). The known range defines the maximum extent, even though actual extent on an individual basis is probably less. On a large scale, the extent of Bayou Darters constitutes a patch. At a finer scale, each riffle constitutes a habitat patch, and within each riffle there are smaller patches defined by the interplay of water depth, current velocity, and substratum size. For instance, Bayou Darters show strong selection for substratum particle sizes of 32–64 mm (Ross et al. 1992b), but during spawning, females select 1–2 mm diameter, coarse sand (Ross and Wilkins 1993). Consequently, the grain for Bayou Darters would seem to be approximately 1 mm.
FIGURE 4.3. Various conceptual models of metapopulation structure. Shaded circles are occupied; open circles are unoccupied; arrows indicate directions of movement. Based on Harrison and Taylor (1997), Wiens (1997), Fagan (2002), and Farina (2006).
Metapopulations
Given that aquatic habitats can be viewed as temporal and spatial mosaics of varying suitability, it is not surprising that populations of a species are not distributed uniformly across the aquatic landscape but instead tend to be aggregated in areas offering the most suitable habitat components. This view of how populations are distributed in space has been formalized as the metapopulation concept—namely a metapopulation is “any assemblage of discrete local populations with migration among them” (Hanski and Gilpin 1997) (Figure 4.3). As such, a metapopulation is “a set of populations that are interdependent over ecological time” (Harrison et al. 1988). Each local population or deme is subject to forces of local selection, including extinction, emigration, and immigration of individuals from other local populations. The balance of these forces determines the fate of local populations, the extent of sites occupied, and the overall size and genetic diversity of the metapopulation (Hanski and Gilpin 1997; Policansky and Magnuson 1998). The landscape concepts of patches and corridors and the ability of organisms to move between them are thus central to the metapopulation concept. (Movement is treated in Chapter 5.)
The term metapopulation (literally a population of populations) was first used by Richard Levins in 1970, although the mathematical description of a population of a single species comprising interconnected local populations appeared in 1969 (Hanski and Gilpin 1991). Levins (1970) described a landscape of occupied and vacant patches as a result of the colonization and extinction of local populations, with the overall collection of local populations comprising a metapopulation (Hanski and Gilpin 1991; Hanski and Simberloff 1997). This is often referred to now as the “classical” or “Levins style” metapopulation (Figure 4.3A) (Hanski and Simberloff 1997; Gotelli and Taylor 1999a). The concept has been readily, and often uncritically, applied to fish populations because the occurrence of habitat patches can result in spatial structuring of populations, such as linear or dendritic metapopulations in streams (Figure 4.3B, C) (Fagan 2002; Campbell Grant et al. 2007). However, without actual assessment of local colonizations and extinctions, rates of movement among local populations, and genetic structure, just because a species may comprise discrete local populations does not automatically indicate that it fits one of the metapopulation concepts (Hanski and Simberloff 1997; Gotelli and Taylor 1999a). For instance, a species existing as a number of discrete populations would, in the absence of movement, constitute a nonequilibrial population and would not be considered a metapopulation (Figure 4.3D).
Gotelli and Taylor (1999a) provide one of the few studies on freshwater fishes to rigorously test classical/Levins-style metapopulation predictions. They used a large data set of 46 fish species, censused 2–3 times per year for 11 years at 10 sites in the Cimarron River, Oklahoma. In the Levins-style metapopulation (Figure 4.3A), the probability of colonization should increase as the proportion of occupied sites increases because there would be more occupied sites from which dispersal could originate. Similarly, the probability of extinction should decrease as the proportion of occupied sites increases because of the increased odds of rescue from another site (Gotelli and Taylor 1999a). Using the 11-year data set for each species, Gotelli and Taylor constructed a matrix showing the occurrence of each fish species at a census site in each year. The proportion of occupied sites was the number of sites occupied in a year, divided by the number of censused sites. The probability of extinction was the number of occupied sites in year (t) that were not occupied in year (t + 1), divided by the number of occupied sites in year (t). Finally, the probability of colonization was the number of vacant sites in year (t) that were occupied in year (t + 1), divided by the number of censused sites in year (t).
In contrast to the prediction of a Levinsstyle metapopulation, the probability of extinction overall was not related to the proportion of occupied sites, although 5 of the 36 species in the analysis did show a significant negative relationship and thus fit the prediction. Similarly, the probability of local colonization was not related overall to the proportion of occupied sites; at the level of individual species, only one species showed a significant positive relationship. Rather than site occupancy, the position in the river system was a more important predictor of colonization and extinction, with the probability of extinction increasing in upstream areas and the probability of colonization increasing in downstream areas. Although the Cimarron River fishes did not fit the predictions of the Levins-style metapopulation model, they did fit predictions of an island-mainland metapopulation model where local extinctions are independent of each other and colonizations occur from outside of the smaller patches (Figure 4.3E). An island-mainland model is appropriate when there is wide variance in the size of the local populations or high variation in patch quality (Harrison and Taylor 1997).
Stream fishes are often viewed as having linearly arranged metapopulations (Figure 4.3B), and in some instances this may be appropriate when species, such as the Bayou Darter in Mississippi, occur almost exclusively in main-channel habitats (Ross et al. 1992b). In many other instances, fishes occur in a broader range of stream sizes so that a dendritic model (Figure 4.3C) would better capture their population structure. As modeled by Fagan (2002), linear and dendritic models may differ in responses to perturbations, depending on how dispersal occurs, and also differ in their responses to fragmentation. Dendritic models tend to show more severe responses to fragmentation, and fragment sizes tend to be smaller and have greater variance. However, if dispersal is sufficient, and occurs both upstream and downstream, the increased connectivity afforded by dendritic versus linear systems increases opportunities for repopulation of extirpated patches, thus increasing the overall persistence of a metapopulation (Fagan 2002; Campbell Grant et al. 2007). Because the union of two streams (the nodes in a dendritic network) may provide increased habitat diversity and a concentration of other resources, such areas can be characterized by increased species diversity (Campbell Grant et al. 2007). Finally, the spatial geometries of disturbance and dispersal can be quite different in dendritic systems. For instance, recovery of lost headwater populations 1 and 2 (Figure 4.3C) would require recolonization from level-2 population 5, rather than from the physically closer (but not connected by water) level-3 populations 3 and 4.
Changes in patch quality can result in areas that vary in favorability to growth, survival, and reproduction of fishes. In the most extreme cases, favorable patches, where successful reproduction exceeds mortality and where emigration exceeds immigration, can supply individuals to patches that do not allow long-term survival and reproduction (i.e., mortality exceeds successful reproduction and recruitment to the population). Such pairs of sites are referred to as sources and sinks (Pulliam 1988; Farina 2006). For example, in a small Minnesota stream, Schlosser (1995a, b) showed that Beaver dams functioned as source areas where most of the production of new individuals occurred. Stream sections between Beaver ponds tended to have lower retention and survival of fishes and thus operated as sinks. However, movement out of favorable source areas, which generally occurred during high-flow events, is also the mechanism whereby newly created Beaver ponds can be colonized by fishes, so there are potential advantages for movement out of areas with high population densities.
The distribution of Bayou Darters in Bayou Pierre of western Mississippi, with juvenile and adult fish occurring in discrete riffle patches (Figure 4.2) and not in intervening pools, suggested that Bayou Darters might comprise a linear metapopulation. Because a primary tenet of a metapopulation is movement among patches, Slack et al. (2004) first demonstrated that larval drift was occurring, such that larvae produced in an upstream patch could supply a downstream patch. Next, they tested the hypothesis that patches of riffle habitat lower in the stream system, where the substratum was softer and particle size generally smaller, functioned as sinks and that they were supplied by more upstream source areas (Ross et al. 2001; Slack et al. 2004). Although densities of Bayou Darters were greater in more upstream riffles, downstream riffles also supported Bayou Darters. The source-sink hypothesis was tested indirectly by comparing the age structure and somatic condition of the fish. In a sink, populations would be expected to have an altered age structure with fewer old individuals because of increased mortality. In fact, there were no differences in age structure between upstream and downstream riffles, with both supporting age-0 to age-3 fish in approximately the same proportions. Body condition also did not differ between upstream and downstream riffles. The data are consistent with a linear metapopulation model, but although apparent riffle quality is lower in downstream riffles, predictions of a source-sink hypothesis were not supported.
RELATING ASSEMBLAGES TO THE ENVIRONMENT
Conceptual and Statistical Models
Various conceptual and/or statistical models have been proposed relating the primary structure of assemblages (i.e., species presence and/or relative abundance), emergent assemblage structure (i.e., species richness, diversity, assemblage complexity, and trophic relationships), or ecological/life-history traits of species to environmental factors (Marsh-Matthews and Matthews 2000). Such models provide insight into how fish assemblages are formed and maintained. At the risk of oversimplification, these models can be placed into two groups: (1) conceptually based a priori approaches where general features of the habitat are used to make predictions of ecological traits of species or assemblages, or (2) databased a posteriori approaches where species occurrences and/or abundances are related to habitat features, using some type of univariate or multivariate analysis. In contrast to the first group, which is based on a mechanistic understanding of how communities operate, this second approach is largely nonmechanistic.
A Priori Models
There are three principal conceptual models (e.g., a set of predictions arising from basic ecological principles) that are prevalent in the literature on lotic systems relating species traits or the emergent structure of assemblages to general environmental features (Goldstein and Meador 2004)—habitat templates (Southwood 1977), landscape filters (Poff 1997), and the river continuum concept (Vannote et al. 1980).
Habitat Template
In 1977 and 1988, Southwood suggested that the habitat is a template providing a predictive pattern for the evolutionary assembly of communities and life-history traits thereof, much like the periodic table of elements in chemistry. A key presumption is that the present-day ecological traits of the organisms will match the current ecological conditions, which, as previous chapters have suggested, is not always the case. Although very much aware of the importance of historical factors, Townsend and Hildrew (1994) set out to develop testable predictions of the habitat template model for species as well as assemblage traits. They used two axes, temporal habitat heterogeneity and spatial heterogeneity, in developing predictions of how species traits (e.g., reproductive type, age and size, parental care, movement, etc.) or assemblage traits (e.g., importance of biotic interactions) would respond to the habitat template (Figure 4.4). They viewed temporal heterogeneity as primarily a measure of the frequency of disturbance and habitat heterogeneity as primarily a measure of the availability of refugia. Thus as disturbance increases, the availability of refugia would become more important. Their model was developed with the Rhône River drainage, France, in mind as a testing arena, but the general constructs should apply to other systems, both lentic and lotic. For instance, considering the species trait of life span, the model would predict that life span should be short on the unstable side of the template (Figure 4.4) and long or short on the stable side (see also Chapter 9). Similarly, body size should be small on the unstable side and large or small on the stable side. In terms of assemblage characteristics, the more stable and complex habitats on the left side should lead to greater specialization in resource use, greater importance of biotic interactions, and increased potential for coevolution. The unstable and low complexity habitats on the right side should favor more generalists in resource use, as well as little potential for biotic interactions or coevolution.
Tests of the species-trait predictions of Townsend and Hildrew’s habitat template model have been equivocal. For instance, an important prediction is that species traits such as body size and parental care should decrease in environments with low spatial heterogeneity and high temporal heterogeneity—the lower right side of the figure. Tests of these and other predictions based on the responses of 13 taxonomic groups of plants and animals occurring in the Rhône River drainage resulted in only mixed support, and support for fishes in this system was totally lacking (Resh et al. 1994). However, Resh et al. (1994) pointed out that the large preexisting data set used to test the predictions might have had methodological limitations that precluded a fair test of the model. Also, the occurrence of species in a habitat might reflect more of chance movement rather than actual habitat selection—resulting in a blurring of the match between species traits and the nature of the habitat.
FIGURE 4.4. The habitat template model showing relationships of species or assemblage traits to the frequency of disturbance and complexity of the physical habitat. The large shaded triangles show areas of different predicted traits. The dashed line shows the transition points between traits on the upper left and lower right. The transition can shift to the left for short-lived species and to the right for long-lived species. Adapted from Resh et al. (1994) and Townsend and Hildrew (1994).
Studies of northern U.S. midwestern streams (Poff and Allan 1995), and comparisons of functional convergence between European and eastern North American fish assemblages (Lamouroux et al. 2002), offer somewhat stronger support for the habitat template model—at least in terms of assemblage predictions. For instance, Poff and Allan (1995) found that two predictions of the habitat template model—variable habitats should contain more resource generalists and nonvarying habitats should contain more specialists (cf. Figure 4.4)—were supported for stream fish assemblages. Hydrologic variables used by Poff and Allan (1995) included flow predictability and variation, base flow stability, and frequency of spates.
Landscape Filters
Recognizing that assemblages are the end products of “interacting multiple causes” at “multiple spatial and temporal scales,” Poff (1997) proposed that functional attributes of species in assemblages are shaped by a hierarchical series of landscape filters that include both physicochemical and biotic factors. A particular community thus comprises species possessing the appropriate ecological “shapes” to have passed through the filters. The heuristic model presented earlier (see figure in Part 2) is similar to this approach. Key elements in the operation of Poff’s landscape filters are categorical niches, defined as discrete levels of species requirements along a given resource axis, and categorical filters, defined as the strength or resistance of a filter in allowing species to pass through it. The combination of categorical filter strength with the categorical niche determines the probabilities for a species to pass through a given filter (Figure 4.5). The example is based on four landscape scales, with one filter per each scale, and a source pool of three species. Two values are required to parameterize the model: a species-specific trait (the resistance of each species to being removed by the filter) and a landscape-level trait (the strength of the filter relative to its ability to remove species). Poff (1997) arbitrarily chose three resistance categories: 1-strong resistance (not removed by any filter), 0.5-intermediate resistance (affected only by the strongest filter, which has a 50% chance of removing the species), and 0-weak resistance (100% chance of removal by strong filters and 50% chance of removal by intermediate filters). Similarly, Poff chose three filter strengths: 1-strong, 0.5-intermediate, and 0-weak.
FIGURE 4.5. An example of the landscape filter model, based on four landscape scales and a source pool of three species, and illustrating how species traits interact with a hierarchical series of filters to determine species occurrences in assemblages. R = the resistance of each species to being removed by the filter; S = the strength of the filter relative to its ability to remove species. See text for additional explanation. Based on Poff (1997).
To run the hypothetical model and determine the most likely species assemblage at the microhabitat level, I have provided likely species resistance values and filter strengths based on work done in the Pascagoula River drainage of southeastern Mississippi (Baker and Ross 1981; Ross and Baker 1983; Ross et al. 1987; Ross 2001). For instance, at the level of basin with the filter of nutrient enrichment, Brown Bullhead (Ameiurus nebulosus) would likely have a resistance of 1, allowing it to pass through the nutrient enrichment filter, which, because many species are strongly impacted by eutrophication, is also set at 1. At the basin level then, Brown Bullhead would have a score of 1 and be classed as abundant. At the reach level, the filter of flood interval has a strength of 0 (given the generally nonerosive floods of this region) and Brown Bullhead are assigned a resistance strength of 0.5. Therefore, at the reach level, Brown Bullhead would also have a score of 1 and be classed as abundant. At the channel-unit level, the filter of water velocity has a strength of 0.5 and Brown Bullhead have a resistance value of 0, given that they tend to occur in slowvelocity habitats, resulting in a score of 0.5 and being ranked as common. Finally, at the microhabitat level the filter of coarse substrate size has a strength of 1 (since bottom-inhabiting species generally respond strongly to particle size) and Brown Bullhead, which usually occur over fine substrata, have a resistance of 0, resulting in a score of 0 and a local abundance classed as rare. Poff (1997) cautioned that the model should not be used to predict absence because it does not incorporate all factors that might influence species presence or absence. Based on likely resistance levels and filter strengths as cited previously, two more species, Weed Shiner (Notropis texanus) and Blackbanded Darter (Percina nigrofasciata) can also be run through the four filter levels with the end result being a local assemblage most likely comprising Weed Shiner. Brown Bullhead are limited by the microhabitat filter of particle size, and Blackbanded Darter are limited by the basin filter of nutrient enrichment.
To expand the landscape model to real-life situations would require fairly extensive ecological information on the fish species potentially available at the basin-level species pool, along with knowledge of what filters are likely operating at each level of the hierarchy as well as their respective strengths. Although more research is needed to understand species susceptibility to various filters, for many regions of North America there is likely sufficient information to assess the relative resistance of species to various environmental filters as well as filters that are likely important at the different hierarchical levels. The information on composition of local assemblages that is provided by the mechanistic approach of landscape models is qualitative rather than quantitative. In other words, the models suggest relative abundance levels (i.e., abundant, common, or rare), but do not, and likely will not, provide quantitative measures of abundance (i.e., density).
Goldstein and Meador (2004) tested various predictions of the landscape model using their analyses of how fish species traits varied among different stream sizes. For the most part, their work supported the theoretical predictions from Poff’s (1997) landscape model. For instance, their work supported the landscape model prediction that fish morphology would be best predicted by local factors such as hydraulic stress. Some differences between the predictions of the landscape model and the findings of Goldstein and Meador (2004) occurred with how reproductive and substratum-use traits varied within and among streams. Landscape predictions, as interpreted by Goldstein and Meador (2004), were that reproductive traits “will vary with flow and substrate variability” and that “substrate preferences are driven largely by microhabitat scale factors.” In contrast, Goldstein and Meador (2004) found that both reproductive strategies and substratum use varied relative to stream size (i.e., among rather than within streams).
River Continuum Concept
The river continuum model (RCC) (Vannote et al. 1980) emphasizes continuity with gradual changes in species occurrences and functional groups from headwaters to downstream reaches (Figure 4.6). The model was conceived as an extension of the physical, geomorphic changes that occur longitudinally in rivers, with the idea that “over extended river reaches, biological communities should become established which approach equilibrium with the dynamic physical conditions of the channel.” In this sense the RCC is quite similar to the habitat template model.
FIGURE 4.6. The relationship between stream size and the physical and biotic components, as proposed in the river continuum model. CPOM = coarse particulate organic material; FPOM = fine particulate organic material; P = primary production; R = respiration. Based on Vannote et al. (1980) and Paller (1994). See text for further explanation.
Headwater streams, stream orders 1–3 (Box 4.1), are strongly influenced by the presence or absence of riparian vegetation. If riparian vegetation is well developed, or if the stream is otherwise strongly shaded by being in a deep canyon, energy input into the stream is largely derived from the surrounding terrestrial area (i.e., allochthonous input) rather than from within the stream (autochthonous input) so that instream measurements show greater respiration than production (P/R < 1). Such situations are common in many eastern and southeastern streams and montane western streams. As the stream increases in size, there is gradually more light penetration into the water and autochthonous production increases so that P/R > 1. In contrast, in headwater streams without well-developed riparian vegetation that could shade the stream, autochthonous production by submerged vascular plants, periphyton, or algae is well developed so that P/R > 1. Such streams are typical of high elevations and latitudes and arid regions in general.
BOX 4.1 • Stream Order
Streams can be categorized in various ways such as discharge, water depth, gradient, water quality, and branching pattern, to name but a few. The branching pattern of streams, or network analysis, has proven useful as a way to describe streams (Leopold et al. 1964; Leopold 1994). Pioneering work on network analysis of streams involved the concept of stream order, first proposed by Robert Horton (1945) and later modified by Strahler (1952, 1957). Using Strahler’s modification of Horton’s system, the smallest unbranched tributary is classified as first order. The union of two first-order streams results in a second-order stream, and the union of two second-order streams results in a third-order stream. To generalize, when two streams of equal rank join they form a segment of the next highest order. However, ordinal rank is not increased by the entrance of lower-order streams.
The decision as to what constitutes a first-order stream can be somewhat arbitrary and to a certain extent depends on the purpose of the study. For instance, a geomorphologist might be more interested in including small channels even though they are not perennial. Strahler (1952), in fact, suggested that first-order streams were wet-weather streams that were normally dry. Leopold et al. (1964) further refined this by suggesting that first-order streams are the smallest unbranched tributaries shown on a 1:24,000 scale topographic map. In contrast, a biologist might favor an approach where first-order streams are classified as the smallest perennial unbranched tributaries that have persisted long enough to contain plants and animals (Hynes 1970). Obviously the second approach requires field verification, so the methodology suggested by Leopold et al. (1964) tends to be the approach of choice for all but the smallest of watersheds. In addition, GIS technology now provides an automated means of determining stream order (Lu et al. 1996).
A useful property of stream order based on the Horton-Strahler system is that a number of physical properties of streams are correlated with stream order. For instance, discharge and drainage area tend to be positively correlated with stream order whereas gradient is negatively correlated (Knighton 1984). The number of stream lengths of each order are negatively related to order so that there tends to be 3–4 times more streams of order n-1 compared to order n. Comparing order n-1 and n, the former has sections that are about half as long and drain somewhat more than one fifth of the area (Hynes 1970). Of course, stream order is a geographical-level variable and does not give information on microhabitat features such as pools, runs, or riffles. In addition, while it is a useful descriptor of the drainage network, streams of equivalent orders but from wet versus arid climates tend to be greatly different in size and biological attributes.
Medium-sized streams (orders 4–6) tend to be more open, have increased autochthonous production, and have greater diversity of insect functional groups (Figure 4.6). This general trend increases in large streams (order 6 and greater), except that increased turbidity can reduce light penetration and thus reduce photosynthesis (P/R ≤ 1). Medium and large streams have increased amounts of fine particulate organic materials (FPOM) produced by the upstream processing of coarse particulate organic materials (CPOM), such as leaves being processed by aquatic insects and bacteria.
Predictions of RCC about fish assemblage composition is at best limited to trophic groups and is based on food resources available in low-order, medium-order, and high-order streams. Vannote et al. (1980) also reasoned that low-order streams would tend to be cooler than high-order streams, although this could certainly vary depending on canopy cover and geographic region. In low-order streams, food resources for fishes would primarily be terrestrial invertebrates, with less importance of aquatic invertebrates or fishes as prey. In medium streams, food resources would expand to include more aquatic prey, both insects and fishes, and in large streams, the increased amount of autochthonous production could lead, in the absence of high turbidity, to the presence of planktivorous and herbivorous fishes. Although Vannote et al. (1980) said nothing about how invertivores might vary in abundance in high-versus mediumorder streams, later work (e.g., Goldstein and Meador 2004) has inferred that RCC predicted a decline. Several early papers, not cited by Vannote et al. (1980), did provide information on changes in functional groups of fishes along a gradient of stream order, including a classic paper by Shelford (1911). Shelford’s work on several small Michigan streams showed that headwaters of the streams were consistently occupied by several species, especially Creek Chub (Semotilus atromaculatus), and that fish species composition changed longitudinally in streams as a function of the physical changes in habitat. The headwater occurrence of Creek Chub (and in general the other three species within the genus) turns out to be a general pattern (Starret 1950; Kuehne 1962; Lotrich 1973; Ross 2001; Boschung and Mayden 2004). As predicted by RCC, in very small headwater streams, Creek Chubs consume primarily allochthonous materials, including insects and even berries (Lotrich 1973; Moshenko and Gee 1973; Ross 2001).
As stream order increases, there is a general trend for both trophic groups and species richness to increase (e.g., Kuehne 1962; Sheldon 1968; Schlosser 1982, 1987; Paller 1994), although species richness may plateau or even decline in large streams (stream orders > 5) (Whiteside and McNatt 1972; Platts 1979; Fairchild et al. 1998). Changes in faunas are generally gradual and do not closely correspond with changes in stream order (Matthews 1986a). Also, the faunal characteristics of small streams that are tributary to lower reaches of large streams (termed adventitious streams) are different when compared to a similarly sized headwater stream (Gorman 1986). For instance, adventitious streams can have greater fish diversity than similarly sized headwater streams and, in fact, have faunas that are more similar to the main channel than to headwater streams (Thomas and Hayes 2006).
The RCC predictions of changes in functional groups, such as fewer insect predators and more planktivores and detritivores with increasing stream size, are generally supported (Goldstein and Meador 2004). For instance, in large eastern and southeastern rivers, including the Missouri and Mississippi rivers, fishes that primarily consume plankton include Paddlefish (Polyodon spathula) and Bigmouth Buffalo (Ictiobus cyprinellus) (Rosen and Hales 1981). Other species also consume phytoplankton or zooplankton, but from on or near the substratum, such as River Carpsucker (Carpiodes carpio) (Brezner 1958; Ross 2001). Detrital materials also contribute to the diet of Bigmouth Buffalo and its congeners, Smallmouth Buffalo (I. bubalus) and Black Buffalo (I. niger) (Walburg and Nelson 1966; McComish 1967; Ross 2001).
The RCC continues to be an important heuristic tool in understanding stream ecosystems. However, among its shortcomings, it did not treat a stream or river as a landscape such that the great variety and complexity of aquatic habitats was largely ignored (Fausch et al. 2002).
A Posteriori Models
Models that use large data sets on fish occurrence and environmental characteristics of lakes and streams in an attempt to find predictive suites of physical characters, or to identify fish assemblages characteristic of locations or environmental conditions, typify a posteriori approaches. Such studies depend on some form of multivariate statistical analysis (Box 4.2) so that their popular use has paralleled that of modern, high-speed computers, and especially the powerful personal computers that are now common. Although widespread use of such approaches is relatively recent, the underlying statistical techniques, such as factor analysis, were developed early in the twentieth century (Sokal and Rohlf 1995; Gotelli and Ellison 2004). The application of multivariate approaches in fish ecology also benefitted from the interest in numerical taxonomy during the same time period (Sneath and Sokal 1973). Many of the computer programs developed for numerical taxonomy, such as cluster analysis and ordination, were used as well in ecological applications.
BOX 4.2 • Multivariate Statistics
In contrast to univariate models, such as linear regression where the response of the dependent variable is related to change in the independent variable or multiple linear regression where the response of the dependent variable is related to change in two or more independent variables, multivariate statistics deal with the simultaneous variation in two or more dependent variables (Manly 1986; Sokal and Rohlf 1995). As such, multivariate statistical techniques are ideally suited to dealing with the complexity of fish assemblages, both in comparing assemblages across space and/or time and for examining relationships among species and the physical and biological components of their environment. Multivariate statistics have become easy to use given the wide choice and availability of statistical software programs. However, this ease of use belies the underlying statistical complexity and the need for the user to have at least a conceptual understanding of what the statistical program is doing. In addition, such tests often assume that each of the variables has a certain structure, such as showing a normal distribution, that all the variables combined have a multivariate normal distribution, that variances among variables are homogeneous, or that the samples were collected randomly. Given the complexity and specialized nature of multivariate approaches, I have given only a brief introduction to some of the approaches that are referred to in the text. Principal components analysis, factor analysis, discriminant function analysis, correspondence analysis, and nonmetric multidimensional scaling can be considered ordination techniques because all are ways of ordering objects based on an array of variables or of ordering variables based on objects. Classification analysis provides another approach for recognizing patterns in multivariate data. Grouping objects (i.e., sites, individual organisms, or sampling units) based on measured variables (i.e., current speed, water depth, turbidity, or species composition) is termed Q analysis. For both ordination and classification, grouping measured variables based on the objects is termed R analysis because the grouping of variables by objects is based on correlation coefficients, such as Pearson’s r (ter Braak 1995; Legendre and Legendre 1998).
ORDINATION TECHNIQUES
Principal Component Analysis
Principal component analysis (PCA) was first described by Karl Pearson in 1901, and in 1933 Harold Hotelling developed computational methods (Manly 1986; Gotelli and Ellison 2004). It is one of the simplest multivariate approaches and has been widely used in ecological studies, although more recently its use has been supplanted by other approaches (Ludwig and Reynolds 1988). The objective is to create linear combinations (components) of the original variables that are not correlated and which capture most of the variation. If successful, the original number of variables is replaced with fewer principal components, making interpretation of the data easier (Manly 1986). There is much to be gained in terms of data simplification if the original measured variables are highly correlated. Indeed, if the original variables are themselves not correlated (generally this would be rare in ecological studies), then the number of components would be the same as the number of original variables and nothing would be gained by the analysis (Manly 1986). Because PCA is sensitive to the magnitude of the original variables, they are usually standardized to means of zero and unit variances. In addition, because PCA is a linear model, its usefulness declines with data that are nonlinear.
Factor Analysis
Factor analysis was developed by Charles Spearman in 1904 for the purpose of measuring human intelligence (Gotelli and Ellison 2004). As with PCA, the goal is to reduce the original number of variables to fewer, noncorrelated, variables (factors). Unlike PCA where factors are linear combinations of variables, factor analysis assumes that the measured variables are a linear combination of underlying factors, with the number of factors usually being less than the original number of variables (Kim and Mueller 1978; Sokal and Rohlf 1995; Gotelli and Ellison 2004). Factor analysis is especially useful as an exploratory approach to identify possible causal factors behind the original correlations in the data set (Sokal and Rohlf 1995; Gotelli and Ellison 2004). Factor analysis can use principal components as initial factors and, as with PCA, variables are first standardized to means of zero and unit variances (Manly 1986).
Discriminant Function Analysis
This approach is really a special case of factor analysis, where the goal is to extract factors (now referred to as discriminant functions) that best separate identifiable groups that are recognized prior to the analysis (Cooley and Lohnes 1971). Groups could be individuals of the same species or sex, or fish assemblages in the same latitude. The discriminant functions are linear combinations of variables that best separate the groups, and each function is uncorrelated with other functions. A useful feature of discriminant analysis is that once functions have been determined, they can be used to classify new data to groups (Gotelli and Ellison 2004).
Correspondence Analysis
Correspondence analysis (CA), or reciprocal averaging, is another approach used to elucidate group characteristics, such as species or functional groups, to habitat characteristics (ter Braak 1995). As is true of the other ordination techniques discussed here, CA requires the assumption that groups show unimodal distributions across the environmental variables. In contrast to the other approaches, CA does a simultaneous ordination of rows and columns to maximize the separation of the groups along each axis (Gotelli and Ellison 2004). Mathematical properties of CA, and other ordination techniques, result in compressing the extremes of an environmental gradient and accentuating the middle, resulting in what is variously referred to as the “horseshoe” or “arch” effect (Wartenberg et al. 1987; ter Braak 1995). Modifications to CA, collectively referred to as detrended correspondence analysis (DCA), were designed to deal with the distortion (ter Braak 1995). However, methods used to remove the curvature of scaling all have limitations (Wartenberg et al. 1987; Gotelli and Ellison 2004).
Nonmetric Multidimensional Scaling (NMDS)
Unlike the previous ordination techniques, which generally retain the original spacing of observations in multivariate space, NMDS is based on ranked distances. It can be used with any distance measure, and the goal of NMDS is to maximize distances of dissimilar objects and minimize distances of similar objects. It is particularly useful in ecological studies because it performs well with data containing many zero values and is robust to deviations from multinormality (Gotelli and Ellison 2004; Paavola et al. 2006).
CLASSIFICATION ANALYSIS
In contrast to ordination, the goal of which is to separate objects or variables along meaningful axes, classification analysis seeks to form discrete groupings. Cluster analysis, based on hierarchical methods, is the most commonly used form of classification analysis in ecological studies (Gotelli and Ellison 2004).
Cluster Analysis
Cluster analysis is particularly useful as an exploratory data tool by creating hierarchical groupings of objects by variables (Q analysis) or variables by objects (R analysis). Approaches to hierarchical cluster analysis are based on similarity or dissimilarity matrices and most commonly form groups by nearest-neighbor joining (Legendre and Legendre 1998; Gotelli and Ellison 2004).
USEFUL REFERENCES
Gotelli, N. J., and A. M. Ellison. 2004. A primer of ecological statistics. Sinauer Associates. Sunderland, Massachusetts.
Jongman, R. H. G., C. J. F. ter Braak, and O. F. R. van Tongeren. 1995. Data analysis in community and landscape ecology. Cambridge University Press, New York.
Kim, J.-O., and C. W. Mueller. 1978. Introduction to factor analysis. Sage University Paper series on Quantitative Applications in the Social Sciences, series 07-013. Sage Publications, Beverly Hills, California.
Ludwig, J. A., and J. F. Reynolds. 1988. Statistical ecology. John Wiley and Sons, New York.
Manly, B. F. J. 1986. Multivariate statistical methods, a primer. Chapman and Hall, New York.
Multivariate Statistics and Fish Assemblages
Pioneering multivariate studies relating habitat characteristics and fishes include G. R. Smith and Fisher (1970), dealing with the distribution patterns of fishes in Kansas, and Stevenson et al. (1974), who studied 53 species of western and central Oklahoma fishes from 27 drainage units. Both studies were based on factor analysis, which treats the variation and covariation of the original variables as a linear combination of underlying factors, with the number of factors usually being less than the original number of variables (Sokal and Rohlf 1995; Gotelli and Ellison 2004). Factor analysis thus is a means of reducing the number of variables (i.e., the factors replace the original variables) and in identifying possible causal factors that are behind the original correlations in the data set (Sokal and Rohlf 1995).
The Oklahoma study (Stevenson et al. 1974) included fish and environmental data from tributaries of the Arkansas, South Canadian, and Red river drainages. Species diversity is generally low in this environmentally harsh region, but numbers of individuals can be quite high (Matthews 1988). The analysis identified six factors defined by their responses to variation in environmental variables and the occurrence of fish species. Environmental variables, 13 in all, included average stream flow, annual precipitation, average winter temperature, number of days above freezing, elevation, and measures of water chemistry. For example, Factor 1, named the Ghost Shiner group because Ghost Shiner (Notropis buchanani) showed the highest loading (correlation) with this factor, also included Longnose Gar (Lepisosteus osseus), average flow, Freshwater Drum (Aplodinotus grunniens), Slenderhead Darter (Percina phoxocephala), runoff, and depth of salt deposits (only variables with high loadings on this factor are given). In other words, much of the variation among the seven listed variables could be captured by a derived variable, Factor 1. Species with high loadings on Factor 2 included the centrarchids—Largemouth Bass (Micropterus salmoides), Orangespotted Sunfish (Lepomis humilis), Bluegill (L. macrochirus), Longear Sunfish (L. megalotis), and White Crappie (Pomoxis annularis). Other species with high loadings were Suckermouth Minnow (Phenacobius mirabilis) and Western Mosquitofish (Gambusia affinis); high evaporation and low elevation were the associated environmental variables. This factor thus captures a lower gradient and warmer fauna dominated by centrarchids. Overall, the study showed the strong impact of climatic and habitat factors on the distribution of Great Plains fishes.
Studies such as these provide a way to reduce the number of variables to a more easily handled number and also offer an objective assessment of the linkages among species and of associations among species and their environments. As with current applications of multivariate techniques, the outcomes depend on appropriate sampling designs, choosing appropriate spatial and temporal scales for analyses and meeting any assumptions (e.g., normality) of the statistical tests.
Another important feature of these studies is their predictive ability. For instance, the Oklahoma study of Stevenson et al. (1974) identified a group of fishes comprising Red River Pupfish (Cyprinodon rubrofluviatilis), Red River Shiner (Notropis bairdi), Speckled Chub (Macrhybopsis aestivalis), and Chub Shiner (N. potteri) that were positively related to indicators of natural brine. They surmised that, given increased salinity levels caused by oil and gas extraction, these species would show an expansion of their ranges, assuming that they had access to the new habitats. In support of this prediction, Red River Pupfish were introduced (perhaps by bait dealers) into a saline tributary (2.4 ppt) of the Cimarron River in northwestern Oklahoma. The pupfish are reproducing and appear to be established (McNeely et al. 2004).
Because different multivariate techniques have different strengths and weaknesses, more recent studies often combine several approaches. Rahel (1984) analyzed fish assemblages from 43 bog lakes in northern Wisconsin. Bog lakes are late successional-stage lakes in the transition from lakes to wetlands and are characterized by low pH, low oxygen levels, and generally low fish species diversity (in this case 20 species).
Fish species were grouped into assemblages using the multivariate technique of detrended correspondence analysis (DCA) on habitat distribution data (Box 4.2). Three assemblages were identified: the centrarchid assemblage consisting of bass and sunfish and associated species such as Northern Pike (Esox lucius), the cyprinid assemblage, and the Central Mudminnow (Umbra limi)-Yellow Perch (Perca flavescens) assemblage. The latter group, which occurred along with the other two assemblages, comprised a “core species group,” to which others could be added in lakes with less harsh environments.
The lakes were grouped by their environmental characteristics using principal components analysis (PCA). Out of nine original variables, PCA was able to capture 71% of the environmental variation among the lakes with three derived variables (components) (Box 4.2). The first principal component reflected the influence of lake size and habitat diversity, with lakes having high correlations on PC-I being large and having well-developed, complex littoral zones (Figure 4.7). The second principal component was largely a measure of lake productivity and acidity, with lakes having high loadings on this axis characterized by higher pH, alkalinity, and conductivity values. The third axis (not shown in Figure 4.7) reflected lake depth and adjoining wetland development. Rahel then overlaid the distribution of fishes defined by the three DCA-identified assemblages on the ordination of lakes based on habitat characteristics (Figure 4.7). Centrarchid assemblages were characteristic of large, highly productive lakes, whereas the cyprinid assemblage tended to occur in smaller, less productive lakes, and the Central Mudminnow-Yellow Perch assemblage occurred in low productivity, highly acidic lakes. In terms of successional stages, the Central Mudminnow-Yellow Perch assemblage occupied late successional environments that were transitioning to wetlands (Figure 4.7), whereas the centrarchid assemblage was characteristic of lakes in an early successional stage. The successional pattern thus shows a change from high to low fish species diversity, as environmental conditions become more limiting.
The application of multivariate statistical approaches to understanding the distributions of fish species and to fish assemblage structure is now widespread and many of such studies are treated in other sections of this chapter and in other chapters. A more recent approach, again following the technological advances in computing power, applies point location data, such as from museum collections, and the information available in geographic information systems (GIS) to the prediction of potential species occurrences. Sometimes referred to as “niche modeling,” the approaches use information associated with actual species occurrences within the framework of a GIS to automatically generate a map of additional localities where the species is likely to occur. One of the first approaches used an iterative, artificial intelligence software package called GARP (Genetic Algorithm for Rule-Set Production) (Stockwell and Peters 1999; Peterson 2001). Genetic algorithms are useful in instances where the original data, generally museum records for species occurrences, and environmental data do not meet the assumptions of most multivariate statistics. Perhaps a chief distinction between these niche models and the multivariate approaches discussed earlier is that the former generally incorporate more environmental data (often 30 or more data layers), focusing especially on topographic and climate data that can be placed in a GIS mapping system. Another obvious difference is that such studies generally are done remotely, without actual fieldwork other than the initial fish collections. As such, niche modeling links the information of museum data on species occurrence with the power of modern GIS systems, but is generally limited to data that are available remotely in electronic databases. However, as long as environmental data can be provided that are suitable for GIS, there is really no limit on what could be included (McNyset 2005). Detailed information on local habitat use, such as focal-point water velocities, substratum selection, or vertical water column position are typically not included, nor are the influences on species local occurrences caused by interactions with other species. Thus niche modeling does not include local dimensions of the realized niche of species but instead is more analogous to the fundamental niche (Hutchinson 1957a). Wiley et al. (2003) properly refer to these as “partial niche models.”
FIGURE 4.7. Principal components analysis of Wisconsin lakes on the basis of environmental characteristics. Data show mean factor scores of three fish assemblages, defined by detrended correspondence analysis, on PC-I and PC-II. See text for further explanation. Based on data from Rahel (1984).
FIGURE 4.8. The prediction of the Kansas distribution of Bluntface Shiner (Cyprinella camura) using a niche model (GARP). The solid gray line surrounds the known distribution of Bluntface Shiner in Kansas; the dashed line shows the predicted distribution. Circles show the locality data used to build the model; triangles show data used to test the model predictions. Based on McNyset (2005). See text for additional explanation.
Predictive capabilities of GARP models, as with other multivariate models (e.g., Ross et al. 1987), can be tested by using only a random subset of the species occurrence data to build the models. The withheld set of occurrence data can then be plotted on the map along with the model predictions. Of course another way to test model output is to “ground truth” the information with follow-up field studies. By either approach, GARP models have generally proven quite successful in predicting species occurrences for North American birds (Peterson 2001), marine fishes (Wiley et al. 2003), Mexican freshwater fishes (Domínguez-Domínguez et al. 2006), and Kansas freshwater fishes (McNyset 2005), among others. For example, McNyset (2005) was able to closely predict the distribution of 12 species of fishes in Kansas using GARP modeling.
Errors in niche modeling include overprediction (including habitats where the species does not occur) and omission (not including known habitats). For Bluntface Shiner (Cyprinella camura), one of the species studied by McNyset (2005) in Kansas, the omission error was calculated to be 17%, although essentially all streams occupied by Bluntface Shiner were included. Overprediction errors are much more problematical because their actual verification would require knowing the true range of the species at the pixel level of the distribution map (in which case running a niche model would be somewhat superfluous). Although no training or test data for Bluntface Shiner were from Walnut River (Figure 4.8), they are known to occur there (McNyset 2005). However, GARP modeling also predicted that Bluntface Shiner would occur in Kansas tributaries of the Kansas and Marais des Cygnes rivers—drainages where they are not known to occur. In this situation, it could be that the species actually occurs in these areas but the areas have not been adequately sampled, that the species once occurred in these drainages but has been eliminated, or, as seems to be true in this instance (Miller and Robison 2004), Bluntface Shiner have never occurred in these drainages, perhaps because of geographic or ecological barriers.
Niche models are also useful in the context of predicting ranges, or locations for refugia, of rare, threatened, or endangered species, as well as determining the probability of invasions by alien species. For instance, Chen et al. (2007) successfully modeled the North American distribution of two large minnows native to eastern Asia, Silver Carp (Hypophthalmichthys molitrix) and Bighead Carp (H. nobilis) that were originally brought into North America for aquaculture purposes. First Chen et al. (2007) constructed niche models using data associated with the natural distribution of these fishes in China, and then projected this information onto North American streams showing where invasions were likely to occur. Such information is useful to resource managers to be alert to the occurrence of nonnative species. Presently, the field of niche modeling is extremely active and includes development and application of new algorithms, such as Maxent, that show different strengths and weaknesses when compared to GARP (A. T. Peterson et al. 2007).
One or Several Models?
No single model of how fish assemblages are formed or identified would likely ever be sufficient. Each approach illustrated here, as well as the many others available in the literature, offer different insights into assemblage formation and spatial or temporal occurrence. A priori models provide an opportunity to infer what functional groups or species traits are likely to occur in an area. These approaches can be particularly powerful in efforts to understand how faunas might change as a consequence of anthropogenic factors such as eutrophication, desertification, construction of impoundments, increasing aridity, or elevated temperatures. Although a priori models are useful because of their generality, in most cases they are limited in their ability to predict the occurrence and/or abundance of a particular species. A posteriori models tend to be more restricted in their application, such as to a particular stream or geographic region, with the notable exception of recent advances in niche modeling using GARP or similar approaches. What a posteriori models might lack in generality they gain in their ability to relate the occurrence of particular species, or suites of species, to particular physical factors and to provide an objective means of defining recurring groups of species (i.e., fish assemblages).
LOCAL VERSUS REGIONAL EFFECTS ON ASSEMBLAGES
Any fish assemblage is the outcome of a myriad of factors operating over a broad range of spatial and temporal scales. Depending on the strength of the factors, the structure of some assemblages might be determined more by local temporal or spatial effects, whereas others by historical or broadscale factors. Focusing on spatial dimensions, one recent approach compares the relative importance of local to regional effects. Local effects on fish assemblages include the nature of the local habitat—especially how it varies spatially and temporally. Regional effects could encompass the regional species pool, as well as landscape features such as climate, elevation, and geographic location.
Much of the role of regional or landscape factors in shaping fish assemblages has been discussed earlier in this chapter. In this section I focus on what has been a primary issue in the study of local versus regional factors, the relationship of the species richness of the local fauna to that of the regional fauna. The principal question is whether local faunas are saturated with species and thus resistant to the addition of new elements, or whether local faunas will increase in species richness as a function of regional species richness (Ricklefs 1987; Cornell and Lawton 1992). In other words, the question is to what degree local fish assemblages are determined by processes acting within the local area versus processes operating at a regional level. Asymptotic relationships between local and regional species diversity (i.e., saturation) suggest the primacy of local control over assemblages; in contrast, if the relationship remains linear (i.e., unsaturated), then local processes would be subordinate to the effect of the regional species pool (Ricklefs 1987). Although it seems clear to contrast local versus regional effects, local assemblages obviously contribute to the regional species pool thus creating a “chicken and egg” problem (Cornell and Lawton 1992).
FIGURE 4.9. The relationship between native fish diversity of local assemblages to regional fish diversity in Virginia streams at the drainage and local scales (based on Angermeier and Winston 1998) and Wisconsin lakes (based on Tonn et al. 1990). Dashed lines indicate a hypothetical direct relationship between regional and local diversity; for Virginia streams, solid lines indicate actual relationships between regional and local diversity. The closed circle and vertical line indicate the mean and 95% confidence interval of local species richness for Wisconsin lakes. Used from Ross and Matthews (in press) with permission from Johns Hopkins University Press.
A study on stream fish assemblages in Virginia showed that regional diversity generally explained more of the variation in local native fish assemblages than did local variables, but local variables also showed some, albeit reduced, explanatory power, with the most important being habitat complexity (Angermeier and Winston 1998). Graphs of local versus regional species richness, although not reaching asymptotes, had low slopes when comparisons were on large spatial scales, such as the drainage level, and all intercepts with the y-axis were significantly greater than zero, both suggesting a tendency for an asymptotic relationship (Figure 4.9). However, when analyzed at a local scale such as site, diversity was strongly related to regional diversity, indicating that the local sites were not saturated with species. The number of introduced species in a local area was also positively related to the regional number of introduced species and, in contrast to native species, showed no evidence of saturation. In addition, the number of native fish species did not influence the number of nonnative species, suggesting that high nativefish diversity does not preclude invasion by nonnative fishes.
The strong influence of regional compared to local factors has also been shown for lakes. Jackson and Harvey (1989) demonstrated that fish faunas of watersheds within the Laurentian Great Lakes showed the effect of large-scale regional processes reflective of postglacial dispersal or climate but were much less related to measures of environmental similarity (e.g., lake depth, area, and pH), although such factors likely have some role in affecting species composition. In a study comparing small lakes in Wisconsin and Finland, Tonn et al. (1990) showed that species richness in individual lakes was related to regional species richness but that local richness reached an asymptote, suggesting that individual lake faunas became saturated with species (Figure 4.9). However, regional factors alone could not explain local species composition because biotic factors, particularly the presence of large predators, also influenced species composition. Tonn and Magnuson (1982) also showed the effects of predator composition, as well as lake morphometry and winter oxygen levels, on the structure of fish assemblages in small Wisconsin lakes.
These studies all suggest a general, but highly variable, link between regional and local species richness. In contrast, in a study of fishes of the Interior Highland region (Ozark and Ouachita mountains of Arkansas, Oklahoma, Missouri, and Kansas) Matthews and Robison (1998) found that regional (river basin) species richness accounted only marginally for species richness at local sites if all species were considered, and that the regional-local species richness relationship was nonexistent within the minnow or darter families. Overall, within all levels of basin (regional) richness, there was great variation in numbers of species in local assemblages. They attributed the lack of strong regional effects to local physical factors at sites within basins, or to within-basin zoogeographic chance in movement or distributions of species. In an analysis of midwestern stream fishes at 65 sites in 13 drainages from Nebraska and Iowa south to Texas, local factors had more effect on species richness than did the overall size of the regional species pool (Marsh-Matthews and Matthews 2000). However, in contrast to emergent assemblage properties (i.e., species richness), primary assemblage structure (i.e., the occurrence of particular species) was strongly influenced by broad geographic factors, primarily latitude, reflective of the fact that many species have restricted north-south distributions (Conner and Suttkus 1986; Cross et al. 1986).
Regional and historic filters, as emphasized by Tonn et al. (1990), clearly can have a major influence on local assemblages and in some cases, especially southeastern streams and northern lakes, the richness of local fish assemblages is strongly affected by regional diversity. Species composition also can be influenced by large-scale factors such as latitude or divisions between major river basins. However, not all assemblages show a relationship between regional and local diversity, as evidenced by harsh midwestern streams and speciose upland streams. In addition, the scale of the study influences the outcome—namely, how small is the local area and how broad is the regional area.
SUMMARY
Fish populations and assemblages are shaped broadly by landscape features and by the mosaic pattern of patches within a landscape. The scale on which fishes perceive patches varies among taxa and life-history stage. Many fish populations likely comprise linear or dendritic metapopulations, although rigorous tests of metapopulation structure of North American freshwater fishes are extremely uncommon and one of the most rigorous studies supports an island-mainland metapopulation model.
Statistical models relating fish species and assemblages to environmental factors are increasingly widespread and generally fall into two groups. A priori models attempt to predict assemblage characteristics and, less often, species occurrence from general environmental features. Three common a priori models are the habitat template, the river continuum, and the landscape filter, and all have some predictive successes. A posteriori models use multivariate statistical techniques to find suites of environmental variables (both physical and biotic) that best predict the occurrence or abundance of species, assemblages, or functional groups. Such models rely on the approximately simultaneous collection of fishes and potentially predictive variables. However, a new class of multivariate models, “niche models,” uses museum collections of fishes in concert with independently collected digitized environmental data sets suitable for GIS programs. Because the various models have different strengths and weaknesses, the use of several models is often appropriate, with the choice based on the research questions and the nature of the data.
Fish assemblages are a product of local conditions and regional factors, including the regional fish fauna. Studies of both lentic and lotic fish assemblages suggest that the relative influence of local versus regional factors varies, although there is support for the influence of regional factors in both, as well as the importance of local factors, including the presence or absence of predators and habitat complexity. The importance of local versus regional effects seems to be heightened in harsher systems. However, as with population models, the number of studies of local versus regional effects on fish assemblages is quite limited, especially in terms of representation of different geographical regions.
SUPPLEMENTAL READING
Gotelli, N. J., and A. M. Ellison. 2004. A primer of ecological statistics. Sinauer Associates. Sunderland, Massachusetts. A useful source for understanding ecological statistics.
Hanski, I., and M. E. Gilpin. 1991. Metapopulation dynamics: Brief history and conceptual domain. Biological Journal of the Linnean Society 42:3–16. Provides a background of the metapopulation concept.
Leopold, A. 1949. A sand county almanac. Oxford University Press, New York, New York. A “must read” for students interested in conservation.
Turner, M. G. 2005. Landscape ecology: What is the state of the science. Annual Reviews of Ecology and Systematics 36:319–44. A good source for current views of landscape ecology.
Wiley, E. O., K. M. McNyset, A. T. Peterson, C. R. Robins, and A. M. Stewart. 2003. Niche modeling and geographic range predictions in the marine environment using a machine-learning algorithm. Oceanography 16:120–27. Provides a good background and examples of niche modeling.
