LRH: Kirsch and Peterson
RRH: Factors affecting fish assemblage structure
A multi-scaled approach to evaluating the fish assemblage structure within southern Appalachian streams USA
Joseph E. Kirsch1
Warnell School of Forestry and Natural Resources, University of Georgia, Athens, Georgia, 30602 USA
James T. Peterson2
Georgia Cooperative Fish and Wildlife Research Unit, US Geological Survey, Warnell School of Forestry and Natural Resources, University of Georgia, Athens, Georgia 30602 USA
1Current address: US Fish and Wildlife Service
850 S. Guild Ave., Suite 105Lodi, CA 95240 USA
E-mail address:
2To whom correspondence should be addressed.
Current address: Oregon Cooperative Fish and Wildlife Research Unit
US Geological Survey
104 Nash Hall, Corvallis, Oregon 97331 USA
E-mail address:
This draft manuscript is distributed solely for purposes of scientific peer review. Its content is deliberative andpredecisional, so it must not be disclosed or released by reviewers. Because the manuscript has not yet beenapproved for publication by the U.S. Geological Survey (USGS), it does not represent any official USGS finding orpolicy.
<A>Abstract
There is considerable uncertainty regarding the relative roles of stream habitat, landscape characteristics, and interspecific interactions in structuring stream fish assemblages. We evaluated the relative importance of environmental characteristics and species interactionson fish occupancy, at local and landscape scales, within the upper Little Tennessee River Basin, Georgia and North Carolina. Using a quadrat sampledesign, fishes were collected at 525 channel units within 48 study reaches during two consecutive years. Weevaluated the relative support for the influence of local- and landscape-scale factorson fish occupancy usingmulti-scaled,hierarchical, multi-species occupancy models. Modeling results suggestedthat the fish assemblage within the Little Tennessee River Basin was hierarchically structured and primarily influenced by stream size and spatial context, urban land coverage, and channel unit types.Landscape scale factors (i.e., urban land coverage, stream size and spatial context) largely constrainedthe fish assemblage structure at a stream reach and local scale factors (i.e., channel unit types)influenced fish distribution within stream reaches. Our study demonstrates the utility of a multi-scaled approach and the need to account for the inter-scale interactions of factors influencing assemblage structureprior to monitoring fish assemblages, developing biological management plans, or allocating management resources throughout a stream system.
North America possesses the richest diversity of temperate freshwater fishes within the world (Jelks et al. 2008). Unfortunately, high fish endemism coupled with environmental degradation has resulted insubstantial imperilment among native fishes with approximately 46% of the 1,187describedfreshwater and diadromous fish taxa classified as vulnerable, threatened, or endangered, a 92% increase from 1989(Jelks et al. 2008). Assuming no future catastrophic events, the extinction rate of North American freshwater fishes is projected to increase from 0.4% per decade (Ricciardi and Rasmussen 1999) to 3.2 % per decade (Burkhead 2012) due to habitat deterioration associated with anthropogenic land and water development. Sedimentation, nutrient loading, pollution, exotic species introduction, and altered hydrologic regimes are the anthropogenic stressors that have beenattributed to the precipitous increase inthe freshwater fish imperilmentin North America (Richter et al. 1997; Ricciardi and Rasmussen 1999;Contreras-Balderas et al. 2003; Dextrase and Mandrak 2006). To reduce these threats, effective management strategies need to be developed at the most effective local and/or landscape levels. Before these strategies are developed, it is essential to identify and quantify the primary factors influencing lotic fish assemblage structure to appropriately allocate management resources (Peterson and Dunham 2010).
The structure of a lotic fish assemblage is the result of a combination of biotic interactions and environmental influences (Jackson et al. 2001).Biotic interactions (i.e., primarily predation and competition) have been shown to influence assemblage composition throughfish species deletions (Gilliam and Fraser 2001) andby alteringfish behavior, such as changing habitat use (Power et al. 1985; Taylor 1996). Environmental characteristics, such as in-stream hydrogeomorphology (i.e., water depth, current velocity, and substrate composition), influence fish assemblagecompositionby providinghabitatsthatallowcertain species to persist over others given species-specific morphological limitations and life history requirements (Schlosser 1982; Moyle and Vondracek 1985;Jackson et al. 2001;Peterson and Rabeni 2001a).Landscape charactersitics, such as topography and terrestrial landuse,can affect the structureof local in-stream habitat characteristics, which in turn influence fish assemblage structure (Schlosser 1991; Peterson and Kwak 1999; Paul and Meyer 2001). Despite the multitude of studies demonstrating the effects of both biotic interactions and environmental characteristics on fish assemblage structure, the percieved importance of mechanisms is not well understood and is typically governed by their association with the scale of study (Jackson et al. 2001; Peterson and Dunham 2010).
Scale is defined by the spatial and temporal resolution of the observational units (i.e., grain) and the dimensions of a study (i.e., extent; Peterson and Dunham 2010). All mechanisms influencing fish assemblage structure, including biotic interactions and environmental characteristics, potentially operate across multiple spatial and temporal scales (Figure 1; Frissell et al. 1986; Tonn 1990; Poff 1997). Theability to observe or detect the effects of thesefactors, however,depends on the within- and among-grain heterogeneity (Peterson and Dunham 2010). Consequently, the perceived importance of factors influencing fish assemblage structure is influenced by the spatial and temporal scale over which an investigation is conducted (Fausch et al. 1994; Crook et al. 2001; Fausch et al. 2002; Peterson and Dunham 2010). For example, studies conducted using large sample units in large spatial extents (e.g., stream reaches within watersheds) often conclude that mechanisms operating over broad scales (e.g., topography) are the primary processes affecting lotic fish assemblage structure (e.g., Fausch et al. 1994),whereas studies conducted with smaller sample units and spatial extents (e.g., microhabitat usewithin a stream reach) likely identify mechanisms operating at local levels (e.g., interspecific competitionor hydrogeomorphic characteristics) as primary processes affecting lotic fish assemblage structure (Jackson et al. 2001; Peterson and Dunham 2010). Though the association between spatial scale of a study and the perceived importance of factors influencing fish assemblage structure have been acknowledged (e.g., Jackson et al. 2001; Quist et al. 2005), few studies have elucidated the importance of biotic interactions and environmental characteristics in structuring lotic fish assemblages across spatial scales.
Theneed to understand the importance of factors influencing fish assemblagesis exemplified by thesoutheastern United States.The region possesses the richest diversity andgreatest number of endemic fishes within North America, north of Mexico (Mayden 1988; Warren et. al. 2000, Jelks et al. 2008). The Southeast also has one of the highest proportions of imperiled freshwater fish taxa, where approximately28% of the 560 described species are classified as imperiled or extirpated primarily due toa variety of anthropogenic stressors (Warren and Burr 1994; Warren et al. 1997; Warren et al. 2000;Jelks et al. 2008).Although lotic fish assemblage structurein the Southeasthas been well studied,there remains considerable uncertaintyas tothe importance ofbiotic and abioticfactors across spatial scales. Therefore, we evaluated the factors influencing thefish assemblage structure within a Southeastriver basin with the following objectives: (1) to develop multi-scale framework for evaluatingthe effects of biotic and abiotic factors operating at geomorphic channel unit and stream reach spatial scales; (2) todetermine the relative importance of scale-specific effects; and (3) to quantify the scale-specific relations between the most influential factors and the occupancy of multiple fish species.
<A>Methods
Study area.- We investigated lotic fish assemblage structureandhabitat use in the Little Tennessee River Basin located in western North Carolina and northeast Georgia, USA. The Little Tennessee River is a tributary of the upper Tennessee River and is located in theBlue-Ridge province of the southern Appalachian Mountain range. The basin reportedly contains more than 150 native freshwater fish species (Warren et al. 2000;Etnier and Starnes 2001). The climate is classified as marine humid temperate based on its relatively mild air temperatures and high moisture content (Swift et al. 1988). The Little Tennessee River Basin drains an area of approximately 4,117km² and is dominated by second growth forestsdue to extensive deforestation in 19th century (Scott et al. 2002). Anthropogenic land use within the basin includes both agriculture and urban development, with agricultural land use declining and urban development steadily increasing to support human immigration (Gragson and Bolstad 2006).
Site selection.- This study was conducted in collaboration with the Coweeta Long Term Ecological Research Program'ssynoptic sampling and assessment project.A total of 48 sites (henceforth, study reaches) were chosen to represent the range of land uses, stream sizes, stream network positions, and elevations found within the Little Tennessee River Basin (Figure 2).Forty-six study reaches were sampled during the pilot phase of the project in the summer (July – September) of 2009. Based on the results of the pilot sampling, 8 of the 46 study reaches and two additional study reaches (10 total) were selected to represent common land use (i.e., forest, urban, and agriculture) and development types (i.e., valley development and hill-slope development) within the Little Tennessee River Basin. The ten reaches then were sampled during the spring (May) of 2010, summer (August) of 2010, and spring (May) of 2011.
We obtained riparian buffer condition datafor each study reach from Long and Jackson (2013). We noted if a study reach had an intact riparian buffer which was defined as fully forested cover extending 10m ormore on both sides of the channel.All other landscape data for each study reach were calculated using readily available Coweeta Long Term Ecological Research geographic information system (GIS) data (CLTER 2011) and ArcGIS software version 9.3.Land use data for each study reach were derived from 2006 land cover data classified from Landsat Thematic Mapper satellite imagery with 30m resolution. Land cover classeswere grouped into four land use classes: urban, agriculture, forest, and other. Study reachelevation, gradient, and hydrologic data were derived from a United States Geological Survey seamless digital elevation model with 9.353m resolution. For each study reach, the number of contributing tributaries and watersheds boundaries were delineated using protocol described by Jenson and Domingue (1988). Tributaries were identified using a minimum drainage area threshold of 0.0875km2 to reflect observations during the summer of 2009. Link magnitude (Link) and downstream link (D-Link) were calculated to represent study reach size and study reach network positions, respectively. Link magnitude was defined as the number of contributing tributaries headwater streams upstream of a study reach (Shreve 1966). Downstream link was defined as the link magnitude of the next downstream confluence from a study reach (Osborne and Wiley 1992).
Sampling design.- We used a multi-scale sample design in which samples were collected from individual geomorphic channel units (henceforth, channel units) nested within study reaches. At each study reach, the stream was stratified into distinct channel unit types and sampled using a quadrat sample design (terminology follows Williams et al. 2002). All channel units within a study reach were classified as either riffle, run, or pool following standardized definitions (Hawkins et al. 1993; Peterson and Rabeni 2001b). Riffle channel units were, on average,shallow areas with swiftcurrentvelocities, surface turbulence, and tended to have coarser substrate than pools or runs. Conversely, pool channel units were, on average,deep areas withslow current velocities, little to no surface turbulence, and tended to have finer substratethan runs or riffles. Run channel units were, on average, areas with moderate depth, moderate to swift current velocities, little to no surface turbulence, and mixed substrate.
Two to four replicates of each of the channel unit type present at each study reach were randomly selected for fish sampling and habitat measurement. During the summer of 2009, the randomly selected channel units were sampled during a single visit. During the spring and summer of 2010 and the spring of 2011, each randomly selected sample unit was sampled twice within each season to facilitate species detectionprobability estimation.When sampling occurred twice within a season, the time between the two sampling occasions was 5-7 days to allow fish to recolonize the channel units after sampling (Peterson and Bayley 1993).
Fishes in each channel unit at each study reach were sampled during daylight hours withSmith-Root LR 24 pulsed DC backpack electrofishers operating between 0.2 to 0.3 amperes. Fish sampling began at the farthest downstream channel unit at each study reach within a sampling occasion.In general, prior to electrofishing, each channel unit had a 5mm mesh seine deployed across the downstream end for riffles and runs or the upstream end forpools to minimize fish escape and maximize detection during sampling. The seine completely blocked off the end of the sample unit and was secured to the streambed. Fishes then were then sampled via electrofishing using a single pass starting from the opposite end of the channel unit from where the seine was deployed.In streams with average wetted widths ≤ 4 meters, onebackpack electrofisher was used by one crewmember with at least one additional crewmember collectingstunnedfish with a 5mm mesh dipnet. In streams with average wetted widths 4 meters, two backpackelectrofishers were used and at least two additional crewmembers collected stunned fish with 5mm mesh dipnets.To maximize detection while sampling, all crewmembers involved with electrofishing attempted to disturb the substrate with their feet to limit stunned fishes from becoming lodged on the substrate or woody debris to maximize detection. After the single pass was completed for each channel unit, fishes were collected from the seine and all dipnets.
Immediately after sampling each channel unit, all captured fishes were identified to species and released back into the stream. Fishes that could not be accurately identified in the field were initially preserved in a 10% formalin solution and brought back to the laboratory. Preserved fishes were later transferred to a 70% ethanol solution and identified to species.
Physical habitat measurements.- Water quality and physical in-stream habitat characteristics expected to influence fish occupancy orthe ability to detect fish were measured at each study reach and channel unit in conjunction with fish sampling. Prior to fish sampling, water quality characteristics were measured in flowing water using calibrated meters at each study reach for each sampling occasion. Turbidity was measured using an Oakton T-100 meter to the nearest 0.01 Nephelometric Turbidity Unit (NTU). Specific conductance was measured using an Oakton CON 400 Series meter to the nearest 0.01 microsiemens per centimeter (μs·cm-1).
Immediately after fish sampling, physical in-stream habitat characteristics were estimated within each channel unit at each study reach for each sampling occasion. Mean wetted width, mean water velocity, and mean water depth were estimated by averaging 3-5 randomly placed measurements within each channel unit (following Peterson and Rabeni 2001b). Water depths were measured to the nearest centimeter using a two meter top-set wading rod. Water velocities were measured at 0.6 depth using a Marsh McBirney model 2000 Flo-mate meter. The length and wetted widths of each unit were measured to the nearest centimeter using a standard measuring tape. The mean cross-sectional area of each channel unit was estimated by multiplying the channel unit mean width by the channel unit mean depth. The surface area of each channel unit was estimated by multiplying a channel unit mean width by length.
Substrate composition within a channel unit was visuallyestimated by two or more crewmembers and averaged (Peterson and Rabeni 2001b). Substrate composition was categorized based on particle diameter as fine sediment (<5mm), gravel (5-50mm), and coarse sediment (50mm; modified from Dunne and Leopold 1978). Wood density was estimated by counting the pieces of wood within the wetted channel unit that were at least 50cm in length and 10cm in diameter or aggregates of smaller pieces of wood with comparable volume and dividing by the channel unit area.
Statistical analysis.- The probability of detecting lotic fish species is rarely 100% and is often related to the factors that influence fish populations (Bayley and Peterson 2001; Peterson and Paukert 2010). To minimize the effect of incomplete species detection, weevaluatedthe relative importance of biotic interactions, in-stream habitat,andlandscape characteristics on fish assemblage structureusing multi-species, multi-scale occupancymodels (Mordecai et. al. 2011). Multi-scaleoccupancy models consist of three submodels thatestimate the probability that a species occupies a study reach (ψ),the probability that the species uses a channel unit within an occupied study reach (θ), and the probability of detecting a species given it occupies a channel unit (p) in a single modeling framework. All of the multi-scale occupancy submodels (i.e.,ψ, θ, and p) werefitusinglogit linear functions of predictor variables (e.g., stream habitat characteristics). The primary assumption of the occupancy estimator is that a species cannot colonize or abandon a study reach between sample occasions (MacKenzie et al. 2006).Webelieve that this study met this assumption because sample occasions were 5 -7 days apart within a season and did not include periods of seasonal fish migration or seasonal changes in habitat use (Peterson and Rabeni 2001a).