Published June 24, 2014
Journal of Environmental Quality
TECHNICAL REPORTS SURFACE WATER QUALITY
A Comparison of Fish-based Classification Schemes for Reference Streams and Rivers in Nebraska Thomas Heatherly II,* Ken Bazata, David Schumacher, and Elbert Traylor
T
he accuracy and precision of stream bioassess-
Proper assessments of lotic ecosystems depend on our ability to isolate natural differences from anthropogenic disturbances. We examined Nebraska river and stream classification strength, based on fish species, using multiple response permutation procedures for common classification strategies: ecoregions, watersheds, hydrologic-landscape regions, and assemblage structure from cluster analyses. Next, we tested the ecological interpretability of classification schemes using nonmetric multidimensional scaling ordinations and ANOVAs. Finally, we used nonparametric ANOVA to identify environmental predictors of overall fish assemblage structure. Hydrologic-landscape regions had the highest classification strength, but cluster groups had the most ecological interpretability based on the discreteness of the groups in ordination space and on the large number of common species that had different abundances across cluster groups. In addition, presence/absence data provided groups with more classification strength and interpretability than abundance data. Temperature, stream size, total phosphorus concentrations, and the percentage of fine substrates were significantly correlated to nonmetric multidimensional scaling ordinations and to overall fish structure in the multivariate ANOVA models. Cluster analyses using presence/absence were therefore the best classification scheme, and we identified the environmental variables that are likely to be useful for determining whether streams should have similar biotic assemblages. This information will be a valuable guide for separating natural variability in biotic assemblages from anthropogenic influences.
ments depend on proper classifications of streams into types with naturally distinct biotic assemblages (Milner and Oswood, 2000; Hawkins et al., 2000). The traditional method has been to use landscape-based classifications (Roset et al., 2007; Stoddard et al., 2008) in which streams are grouped according to predetermined boundaries in the landscape, such as watersheds, Ecoregions (Omernik, 1987), or hydrologiclandscape regions (Winter, 2001). Watershed classifications assume that ecological attributes are influenced by properties of their catchments, such as geology, sediment supply, hydrology, and other factors (Hynes, 1975; Dunne and Leopold, 1978; Hawkins et al., 2000). Ecoregions assume that important factors correspond to identifiable regions within larger landscapes with similar terrestrial features (Omernik and Griffith, 1991). In practice, Ecoregions and watersheds often have explained minimal differences between streams (Hawkins et al., 2000; Wang et al., 2003). For example, Hawkins and Vinson (2000) compared the utility of five types of landscape classifications for explaining compositional differences of invertebrates from >2000 streams in the United States. They suggested that no existing landscape classification scheme was sufficient for partitioning invertebrate assemblages into discrete groups. Hydrologic-landscape regions (HLRs) are landscape-based classification schemes based on land surface area, soil and bedrock permeability, and precipitation and evapotranspiration potential (Winter, 2001; Wolock et al., 2004). If HLRs provide accurate classifications of hydrologic variability, they may prove superior to other classification schemes for the purposes of bioassessment. Wolock et al. (2004) found that HLRs were equal to or better than level II Ecoregions for describing land-surface forms, geologic texture, climate, land cover, and water quality and had some ability to characterize fish species richness. However, we are unaware of an explicit test of the ability of HLRs to describe discrete biotic communities. An alternative to landscape-based classifications is to use statistical classifications that do not rely on physiographic constraints (Gerritsen et al., 2000). Multivariate classifications,
Copyright © American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America. 5585 Guilford Rd., Madison, WI 53711 USA. All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher.
T. Heatherly, K. Bazata, D. Schumacher, and E. Traylor, Water Quality Division, Nebraska Dep. of Environmental Quality, 1200 “N” St., Suite 400, Lincoln, NE 68509; T. Heatherly, School of Natural Resources, Univ. of Nebraska-Lincoln, 3100 Holdrege St., Lincoln, NE 68583; T. Heatherly, current address: Departamento de Ecologia, Universidade do Estado do Rio de Janeiro, Rua São Francisco Xavier, 524, Rio de Janeiro, RJ, 20550, Brazil. Assigned to Associate Editor Patrick Inglett.
J. Environ. Qual. 43:1004–1012 (2014) doi:10.2134/jeq2013.03. 0102 Supplemental data file is available online for this article. Received 25 Mar. 2013. *Corresponding author ([email protected]).
Abbreviations: HLR, hydrologic-landscape region; MRPP, multiple response permutation procedure; NMDS, nonmetric multidimensional scaling.
1004
such as cluster analyses, are appealing because organismal data are inherently multivariate (Norris, 1995; Milner and Oswood, 2000). Gerritsen et al. (2000) found that streams in Wyoming were better characterized using clustering based on biotic similarity of invertebrates, although there were apparent Ecoregional affinities. Similarly, Milner and Oswood (2000) found that classifications based on multivariate techniques of invertebrates were better at classifying streams around Anchorage than multimetric indices that used regional classifications. Finally, Van Sickle and Hughes (2000) found that Ecoregions and watersheds were useful for classifying stream vertebrate assemblages, mostly because of spatial autocorrelation, but landscape-based approaches were not as effective as assemblage similarity-based approaches. We need to explicitly test different classification schemes to account for important natural differences between streams so that we may properly identify stressors. In this study, we first directly compared the utility of classifications of reference streams and rivers in Nebraska based on Ecoregions, watersheds, HLRs, and assemblage similarity using presence/absence and abundance data. Classifications are frequently performed with reference streams (Reynoldson et al., 1997; Hawkins et al., 2000; Bailey et al., 2004) to classify stream types by natural differences. Next, we used a combination of nonmetric multidimensional scaling ordination (NMDS) values and nonparametric multivariate ANOVA to identify environmental variables likely responsible for classifications. This research fills a gap in our knowledge of the best approaches for classifying streams according to natural differences. This knowledge will facilitate our ability to identify and manage anthropogenic impacts.
Materials and Methods Study Sites The streams in this study (Supplemental Table S1) were designated as reference sites by the Nebraska Department of Environmental Quality (Bazata, 2005, 2011) in an Environmental Management and Assessment Program study performed from 1997 to 2009. This designation was based on an index of habitat quality composed of 10 metrics of morphology (incision/width ratio, percent undercut banks, percent overhanging vegetation, percent pools, percent riffles, percent barren banks), substrate (percent sand substrate, percent silt substrate), and riparian quality (percent rowcrop within 30 m of stream, middle canopy density above stream). The streams that scored in or above the 90th percentile of all habitat index scores were used as reference streams. This method used only nonbiological characteristics, as suggested by Bailey et al. (2004) and Stoddard et al. (2008). We have much to learn about what constitutes a least-impaired condition and what constitutes a regionally representative set of sites (Whittier et al., 2007), especially in streams that cannot realistically be returned
to a historical condition. The reference streams in this study were chosen independently of this research and thus provide the opportunity to test classification schemes based on an existing management program. Furthermore, ongoing research in Nebraska has found that fish are better predicted by in- and nearstream habitat features than by land usage (T. Heatherly et al., unpublished observations). Thus, there is evidence that a reference system based on the physical template will be more meaningful for the management of fish than one based on features of the landscape. Sixty-eight sites were used in this study. These streams were within six level III Ecoregions and 11 of the 13 major watersheds (Fig. 1). Additionally, the reference streams were within 7 of the 14 HLRs that occur within Nebraska (Fig. 2).
Data Collection Each stream was sampled during summer baseflow conditions between the years 1997 and 2009, excluding 2002 and 2003 because of quality control concerns. We established a sampling reach of approximately 40 times the average stream width for sampling, with lower and upper limits of 150 and 300 m. We measured water quality parameters (temperature, pH, conductivity, and dissolved oxygen) with Hydrolab Quanta (Hach Hydromet) or Eureka Manta (Eureka Environmental Engineering) multiprobe meters. Water for nutrient chemistry analysis was collected the day of sampling in amber jars, preserved with concentrated sulfuric acid, and shipped overnight within 2 d of sampling to the USEPA Science and Technology Laboratory in Kansas City, Kansas. Samples were analyzed for dissolved and particulate phosphorus, nitrate-nitrogen, nitrite-nitrogen, and ammonium-nitrogen using standard methods (APHA, 1995). We used total phosphorus and total nitrogen in statistical analyses. Physical habitat measurements were taken at 11 evenly spaced transects along the study reach (Kaufmann and Robison, 1998). At each of the 11 transects, we measured stream width, depth, incision and bank angle,
Fig. 1. Map of Nebraska reference streams used in this study with delineations for level III Ecoregions, major watersheds, and assemblage groups assigned with hierarchical cluster analysis of fish assemblage presence/absence (P/A).
www.agronomy.org • www.crops.org • www.soils.org 1005
is the overall weighted mean of withingroup means of the pairwise dissimilarities among sampling units (Oksanen, 2011). Classification strength (Van Sickle, 1997) is the difference between within-group dissimilarities and average among-group dissimilarities calculated from a specified number of permutations of fish into random groupings and is chosen by setting the weight type of the mrpp function to 3 (Oksanen, 2011). We used 5000 permutations in each test. The classification strength is reported as within-group agreement, and significance is given by P value. Cluster analysis and MRPP provide information to assist in creating meaningful groups but do not provide details regarding environmental interpretations. We assessed the relationships between classifications and Fig. 2. Map of Nebraska reference streams used in this study with delineations for hydrologicenvironmental variables through vector fitting landscape regions and assemblage groups assigned with hierarchical cluster analysis of fish of environmental variables onto NMDS and by assemblage abundance. modeling the effect of environmental variables on overall composition structure using substratum size composition and embeddedness, bankfull width, nonparametric multivariate ANOVA. The ordination reduces the and depth and canopy cover. Additionally, we used visual surveys complexity of compositional data into a few dimensions (Oksanen, to estimate the percentage of riffle, run, and pool habitat as well as 2011) and thus may provide relatively simple interpretations of the the relative amounts of cover within the categories of filamentous overall multivariate dataset. We used the “metaMDS” function algae, macrophytes, brush, woody debris, overhanging vegetation, in the “Vegan” package to compare ordinations built from 20 undercut banks, boulders, and artificial subtrata (e.g., lowhead random starting configurations of fish presence/absence and used dams and rip-rap at bridges). Environmental variables with a large the Jaccard and Bray-Curtis indices as our dissimilarity measures number of zeroes were not included in subsequent analyses. for presence/absence and abundance data, respectively. We used We collected fish by multipass electro-fishing the entire reach; the function “envfit” to identify environmental variables with we used a backpack shocker for small, shallow streams (LR-24 significant correlations to the ordination. The direction of vector Electrofisher, Smith-Root Inc.) and a generator-powered shocker arrows on the ordination plot indicates the direction of most rapid mounted on a tote-barge (SDC-1 Stream Shocker Electrofishing change in the variable, and the length of the arrow is proportional Tow Barge System, ETS Electrofishing, LLC) for larger streams and to the strength of correlation (r2) between the ordination and rivers. We identified all large fish at the site before release. Smaller variable (Oksanen, 2011). Finally, we used the “ordihull” function fish were identified in the field or preserved in 10% formalin to draw polygons around each assemblage-based group of sites solution for later identification at the Sternberg Museum of Natural on the ordination plot as derived from cluster analyses. We used History, where they have been maintained as voucher specimens. Procrustes rotations to test each new iterative ordination derived Data Analysis from a different random starting configuration and standardized the abundance data before ordination with Wisconsin square-root We performed agglomerative, k-means cluster analyses using standardization (Oksanen, 2011). the hclust function in the base R package (R Development Core Nonparametric multivariate ANOVA imposes no constraints Team, 2012) with the Jaccard coefficient for presence/absence on compositional structure and does not assume linear data ( Jaccard, 1901) and Bray-Curtis dissimilarity (Bray and relationships between environmental variables and community Curtis, 1957) for abundance data. We grouped sites into clusters structure, whereas NMDS confines the community to a few axes using Ward’s linkage. We plotted the within-cluster sums of and vector fitting does assume linear relationships (Oksanen, squares against the number of potential cluster groups to choose 2011). We used the function “Adonis” in “Vegan” to model the the appropriate number of clusters. effect of 19 environmental variables (listed in Supplemental Next, we tested the classification strength of groups based on Table S2) on a dissimilarity matrix calculated with “vegdist” fish assemblage structure, Ecoregions (Level III; Omernik, 1987), using the Bray-Curtis dissimilarity index. Significance was watershed boundaries, and HLRs using a multiple response determined by comparing modeled results with 1000 random permutation procedure (MRPP) using the “mrpp” function in the iterations of the data. We began with a global model of all 19 “Vegan” package (Oksanen et al., 2012). Hydrologic-landscape variables and reduced the model by sequentially removing the region data were acquired as a GIS file through the USGS: http:// least significant parameter and rerunning the analysis until water.usgs.gov/GIS/dsdl/hlrus.tgz. We did not have enough the remaining parameters were significant. We used Pearson streams within Ecoregion 42 or 43 or HLR 6 or 13 for statistical correlation coefficients to ensure modeled variables were not analysis. All analyses were performed with presence/absence and correlated with each other. abundance data. The MRPP uses the test statistic delta, which 1006
Journal of Environmental Quality
We used one-way ANOVA to test for differences in the abundance of the 20 most abundant fishes among classification schemes. Tukey’s honestly significant differences test was used for pairwise comparisons among groups within classification schemes. Both of the tests were performed with the base R program. All abundances were log+1 transformed before analysis to meet normality and heteroscedasticity assumptions.
Results The plots of within-group sums of squares suggested that the presence/absence dendrogram be separated into four clusters and the abundance dendrogram be separated into two clusters (Fig. 3). The presence/absence groups were not limited to specific watersheds or HLRs, yet there did appear to be some regional similarities. For example, Group 1 occurred primarily in the northeast of the state, especially in the Elkhorn and eastern Niobrara River watersheds (Fig. 1). Group 2 largely represented the rest of the Niobrara watershed in the north-central portion of Nebraska. Group 3 mainly represented the southern watersheds, including the Republican, Little Blue, and Big Blue River watersheds, and Group 4 mainly occurred in the western Nebraska panhandle. The two abundance groups appeared more spatially overlapping, yet group 1 appeared to be more prevalent in the west and southwest (Fig. 2). The HLR classification scheme using presence/absence data had the highest within-group agreement (Table 1). The next highest within-group agreement values were for cluster analyses
of abundance data and then Ecoregions, watersheds, and clusters of presence/absence data. Ecoregion, watershed, and HLR classifications were not significant using abundance data. The NMDS ordination of presence/absence (Fig. 4) and abundance data (Fig. 5) had sufficiently low stress (0.169 and 0.187, respectively) to reduce fish community dissimilarity to two dimensions. For presence/absence data, only the cluster group (Fig. 4D) demonstrated both distinctness and compactness when plotted on NMDS ordinations. Nine of the 19 environmental variables tested were significantly correlated with the presence/absence ordination at p ≤ 0.100 (Supplemental Table S2). Streams within the White Hat River watershed, within Ecoregion 25, and within HLR 10 appeared to be associated with cobbles, overhanging vegetation, undercut banks, and fine substrates (Fig. 4). Streams within the Big Blue, Elkhorn, Middle Platte, South Platte, and Republican River watersheds were associated with higher nutrients and warmer temperature, and these were typically within Ecoregions 42 and 47 and HLRs 3, 5, and 8. For the abundance ordinations, all classification schemes were broadly overlapping, although the watershed groupings appeared to be the most distinct (Fig. 5C). The identity and direction of correlation of environmental variables were similar to the presence/absence ordination (Supplemental Table S3), except that the percentage of undercut bank habitat replaced the percentage of macrophyte cover in the abundance ordinations. Streams within the White Hat River watershed were also found
Fig. 3. Plots of the within-group sums of squares error against the numbers of cluster groups in hierarchical cluster analyses of fish presence/ absence (left) and abundance (right) collected from Nebraska reference streams. Table 1. Results of multiresponse permutation procedure to test the fit of classification of Nebraska streams by assemblage structure, Ecoregion, watershed, and Hydrologic Landscape Regiond (HLR) using fish species presence/absence and abundance. Grouping† Cluster-P/A Cluster-Abund Ecoregion-P/A Ecoregion-Abund Watershed-P/A Watershed-Abund HLR-P/A HLR-Abund
Within-group agreement
Observed delta
Expected delta
0.105 0.202 0.172 −0.012 0.117 0.151 0.460 0.012
5.412 434.7 5.005 551.1 5.337 462.5 3.265 538.2
6.044 544.8 6.044 544.8 6.044 544.8 6.044 544.8
Delta significance 0.001 0.001 0.001 0.545 0.001 0.070 0.001 0.383
† Abund, abundance; P/A, presence/absence. www.agronomy.org • www.crops.org • www.soils.org 1007
Fig. 4. Nonmetric multidimensional scaling ordination plots of Nebraska reference streams based on the similarity of fish species’ assemblages using presence/absence data. Polygons contain the sites of each assemblage group as determined by (A) level III Ecoregion, (B) hydrologiclandscape region, (C) watershed, and (D) hierarchical cluster analysis. Vectors show the direction and strength of environmental variable correlations to the ordination (P ≤ 0.100). BB, Big Blue; EL, Elkhorn; LO, Loup; MP, Middle Platte; MT, Missouri Tributary, NI, Niobrara; RE, Republican; SP, South Platte; TN, total nitrogen; TP, total phosphorus; WH, White Hat. Lodgepole Creek (SP2019) was not plotted in ordinations for clarity.
to be associated with cobbles and overhanging vegetation using abundance data. Streams within the North Platte River watershed were associated with undercut banks. As with presence/absence data, streams in the Big Blue, Elkhorn, Middle Platte, South Platte, and Republican River watersheds were associated with higher nutrient concentrations and warmer temperature. Nineteen of the 20 most abundant fish species had average abundances per site that were different among presence/ absence groups, and 12 of 20 of these fish were different among watersheds (Fig. 6). Six species were different among Ecoregions and HLRs, although HLRs did slightly better among the 10 most abundant fish. Additionally, only presence/absence groups had ANOVA F values that exceeded 5.0. We used Tukey’s honestly significant differences to examine how the most abundant fish differed among presence/ absence groups. Group 1 typically had high abundances of Notropis stramineus, Notropis dorsalis, Cyprinella lutrensis, and Hybognathus hankinsonui (Fig. 7). Group 2 had relatively high abundances of N. stramineus, N. dorsalis, Semotilus atromaculatus, Catostoma commersoni, and Rhinichthys cataractae. Group 3 had high abundances of N. stramineus, Pimephales promelas, S. atromaculatus, C. lutrensis, and Campostoma anomalum. Group 4 had relatively high abundances of R. cataractae. We performed nonparametric ANOVA analyses on fish presence/absence and abundance starting with the 10 environmental variables that were correlated to the ordinations. 1008
For the presence/absence of fish, 4 of the 19 environmental variables were significant predictors of overall assemblage dissimilarity (Table 2). Temperature explained the highest degree of variability, followed by average stream width, the percentage of fine substrata, and total phosphorus. Overall, these four variables explained about 15% of the total variation in dissimilarity (R2 = 0.15), and each individual variable explained ~3 to 4% of total variation. Only temperature and average width were significant predictors of the dissimilarity of fish abundances in the nonparametric ANOVA (Table 3), and each explained ~4% of total variation.
Discussion This study explicitly tested three common bioassessment classification schemes and another scheme that has not been applied for this purpose. Ecoregions, watershed boundaries, HLRs, and clusters provided potentially useful classifications of the fish assemblages of streams and rivers in Nebraska, and presence/absence data were more likely to produce discrete groups at a statewide scale than abundance data. Multiple response permutation procedure tests suggested that HLRs provided the most within-group agreement. However, presence/ absence clusters gave the most distinct groups when plotted on NMDS ordinations and were better than HLRs and Ecoregions at discriminating among the most abundant fish species. Boundaries of the major watersheds provided the next strongest Journal of Environmental Quality
Fig. 5. Nonmetric multidimensional scaling (NMDS) ordination plots of Nebraska reference streams based on the similarity of fish species’ assemblages using abundance data. Polygons contain the sites of each assemblage group as determined by (A) level III Ecoregion, (B) hydrologiclandscape region, (C) watershed, and (D) hierarchical cluster analysis. Vectors show the direction and strength of environmental variable correlations to the ordination (P ≤ 0.100). BB, Big Blue; EL, Elkhorn; LO, Loup; MP, Middle Platte; MT, Missouri Tributary, NI, Niobrara; RE, Republican; SP, South Platte; TN, total nitrogen; TP, total phosphorus; WH, White Hat. Lodgepole Creek (SP2019) was not plotted in ordinations for clarity.
Fig. 6. Plot of the 20 most abundant fishes in reference streams in Nebraska against the F values of ANOVA models testing for differences in the abundance of fish species among level III Ecoregions, hydrologic-landscape regions (HLR), presence/absence cluster groups (PA), and watersheds. From left to right, fishes are listed from the highest to the lowest average abundance per site. The gray vertical line denotes the level of F value associated with significant ANOVA models. www.agronomy.org • www.crops.org • www.soils.org 1009
Fig. 7. Boxplots of the abundance of the most abundant fishes from Nebraska reference streams against groups designated by cluster analysis of fish presence/absence (P/A) data. Letters indicate differences among groups (Tukey’s HSD, P ≤ 0.05). CPUE, catch per unit effort.
ability to create distinct groups based on community dissimilarity, and common fish species frequently had abundances that were distinctly different among watersheds. Therefore, the bulk of our results support using cluster analyses with presence/absence data for generating distinct and ecologically interpretable fish classifications. Hydrologic-landscape regions were built to better manage the quantity and quality of water resources through the integration of climate, soil, and hydrological sciences (Winter, 2001; Wolock et al., 2004; Santhi et al., 2008). They have rarely been
tested as important units for biota, so our results are difficult to place within context. Wolock et al. (2004) found that HLRs accurately represented ~31% of variation in fish species richness, which was slightly less than the 45% of level II Ecoregions. They also found that HLRs were as good as or better than Ecoregions at describing landscape surface form and texture, climate characteristics, land cover, and water quality. Santhi et al. (2008) found that the variables used to construct HLRs, such as relief, climate, and the percentage of sand in soils, were good predictors of stream baseflows, yet they were unsure whether these variables
Table 2. Results of nonparametric multivariate analysis of variance to test for relationships between environmental variables and the presence/ absence of fishes in Nebraska streams. Parameter
df
Sums of squares
Mean squares
F value
R2
P value
Temperature Average width Total P Fine substrate, % Residuals Total
1 1 1 1 64 68
0.663 0.593 0.453 0.580 13.294 15.584
0.663 0.593 0.453 0.580 0.208
3.189 2.855 2.183 2.793
0.043 0.038 0.029 0.037 0.853 1.00
0.007 0.008 0.044 0.004
Table 3. Results of nonparametric multivariate analysis of variance to test for relationships between environmental variables and the abundance of fishes in Nebraska streams. Parameter
df
Sums of squares
Mean squares
F value
R2
P value
Temperature Average width Residuals Total
1 1 66 68
0.952 0.825 23.542 25.318
0.952 0.821 0.357
2.668 2.313
0.038 0.033 0.930 1.00
0.002 0.007
1010
Journal of Environmental Quality
could represent the variability of flows. We found that they may provide significant classifications of fish species, but HLRs did not provide as much ecological interpretability as groups derived from cluster analyses or watershed boundaries. This is somewhat surprising given that temperature and discharge were the best predictors of overall assemblage structure in nonparametric ANOVA tests but may be partially explained by the large number of local physical habitat correlates to NMDS ordinations. Furthermore, McManamay et al. (2012) found that HLRs did a poor job of predicting flow class, which suggests that HLRs may not capture the importance of stream size to the structure of fish assemblages. Hydrologic-landscape regions may provide one level of representative information when placed into a hierarchical framework that incorporates local influences; however, it is possible that regional frameworks will not capture enough of the individual variability of ecosystems to be broadly useful at statewide levels (McManamay et al., 2012). Previous classification strength research has been conflicting, yet classifications based on cluster analyses usually provided better representations of natural groups. Van Sickle and Hughes (2000) found assemblage structure was superior to landscape classifications for vertebrates in Oregon streams, and they also found Ecoregions performed better than catchments. Van Sickle (1997) suggested that fish in Oregon streams were best classified by Ecoregions. Several investigations of aquatic invertebrates confirmed the superiority of assemblage-based classifications to Ecoregions and watersheds, and the latter two classifications did not consistently explain more variation than random groupings (Gerritsen et al., 2000; Hawkins et al., 2000; Milner and Oswood, 2000; Wang et al., 2003). Waite et al. (2000) found weak classification strengths of Ecoregions to describe invertebrates in Mid-Atlantic streams of the United States, which improved when Ecoregions were filtered first by stream order but did not have the strength of clusters. Landscape-based classifications may suffer if there are weak ecological filters at landscape levels and if there are widespread individualistic relationships between species and the environment (Resh, 1994; Fore et al., 1996; Heino et al., 2003; Bonada et al., 2006; Heino and Mykrä, 2006). Herlihy et al. (2008) suggested nine Ecoregions maximized withinregion similarity in macroinvertebrate assemblages across the conterminous United States. The majority of Nebraska belonged to the single “Southern Plains” region largely because a lack of reference sites according to their criteria prevented the use of smaller regions. Therefore, we may expect that homogeneity and stream degradation in Nebraska may make Ecoregions at any level disadvantageous. We did identify distinct stream types within Nebraska using dissimilarity measures, and classifications based on fish presence/absence appeared to incorporate site-specific differences in important environmental variables. Our explicit tests of the classification strengths of different grouping methods will help managers control for natural differences in reference streams, which has been identified as a major impediment to bioassessment using a reference-condition approach (Waite et al., 2000; Herlihy et al., 2008). Evidence that presence/absence data performed best for community analyses was found by Hirst and Jackson (2007), who recreated known configurations of community data for a variety of ordination methods. They suggested that although abundance data potentially contain more information, they
are more easily biased by natural and experimental error than presence/absence data (see also Jackson and Harvey, 1997). Additionally, the ability of abundance and presence/absence to represent assemblages may depend on the relative commonness and rarity of the species (Manel et al., 2001; Cushman and McGarigal, 2004) and on the spatial scale in question (Strayer, 1999; Cushman and McGarigal, 2004). For example, Cushman and McGarigal (2004) found that presence/absence data were superior for larger spatial scales and assemblages composed of numerous relatively uncommon species of birds. However, Bart and Klosiewski (1989) found that abundance and presence/ absence data performed similarly in describing bird assemblages, and Strayer (1999) found that presence/absence data generally were not effective at detecting such trends as population declines. Given our findings and the majority of those in the literature, we recommend the use of presence/absence data to classify streams of Nebraska by fish assemblages. Temperature, stream size (average width), total phosphorus, and the percentage of fine substrates appeared to influence the structure of fish assemblages in Nebraska based on nonparametric multivariate ANOVA tests. These variables were also significantly correlated to both ordinations (Supplemental Tables S2 and S3), which is further confirmation of their ecological importance. Temperature is a driver of the metabolism of ectotherms and influences the distributions, growth rates, and fecundity of aquatic organisms (Newell and Minshall, 1978; Sweeney and Vannote, 1978; Vannote and Sweeney, 1980). Stream size is also a fundamental component of stream ecology. For example, increases in habitat space are considered responsible for the logarithmic increase in fish richness with increases in discharge (Xenopoulos and Lodge, 2006; Oberdorff et al., 2011). Phosphorus is an essential macronutrient that is frequently limiting in freshwater ecosystems (Sterner and Elser, 2002), and the percentage of fine substrates may be an indicator of impairment to the watershed that results in sediment deposition to streams (Allan and Castillo, 2007). In this study, three independent tests suggested these variables were responsible for the observed structural differences of fish assemblages in streams and rivers of Nebraska.
Conclusions Cluster analyses using presence/absence data provided more ecologically meaningful classifications of Nebraska streams and rivers based on fish species than classifications based on Ecoregions, major watershed boundaries, hydrologiclandscape regions, or cluster analyses using fish abundances. This conclusion is based on the separation of presence/absence groups in ordination space and on the differences in abundance of the most common fish species among classification groups. Stream size and temperature, phosphorus concentrations, and the percentage of fine substrates are useful for isolating natural versus anthropogenic-driven variability in fish assemblages. These distinctions should be widely applicable to those tasked with managing biotic integrity in agricultural regions.
Acknowledgments This study was supported by U.S. Environmental Protection Agency CWA section 104(b)(3) and section 319 funds awarded to the Nebraska Department of Environmental Quality. The authors thank L. Johnson
www.agronomy.org • www.crops.org • www.soils.org 1011
for assistance in creating Figures 1 and 2 and Associate Editor M. Archer and two anonymous reviewers for constructive advice on earlier drafts.
References APHA. 1995. Standard methods. 19th ed. American Public Health Association, Washington, DC. Allan, J.D., and M.M. Castillo. 2007. Stream ecology: Structure and function of running waters. 2nd ed. Chapman and Hall, New York. Bailey, R.C., R.H. Norris, and T.B. Reynoldson. 2004. Bioassessment of freshwater ecosystems: Using the reference condition approach. Kluwer Academic, Boston, MA. Bart, J., and S.P. Klosiewski. 1989. Use of presence/absence to measure changes in avian density. J. Wildl. Manage. 53:847–852. doi:10.2307/3809224 Bazata, K. 2005. Nebraska stream classification using fish, macroinvertebrates, habitat, and chemistry evaluations from R-EMAP data, 1997–2001. Project report for the Nebraska Department of Environmental Quality, Lincoln, NE. http:// www.deq.state.ne.us/Publica.nsf//NebrStreamBiolMonitorReport20042008. pdf (accessed Feb. 2013). Bazata, K. 2011. Nebraska stream biological monitoring program 2004–2008. Project report for the Nebraska Department of Environmental Quality, Lincoln, NE. http://deq.ne.gov/Publica.nsf//NebraskaWaterMonitoringProgramselectronic.pdf (accessed Feb. 2013). Bonada, N., N. Prat, V.H. Resh, and B. Statzner. 2006. Developments in aquatic insect biomonitoring: A comparative analysis of recent approaches. Annu. Rev. Entomol. 51:495–523. doi:10.1146/annurev.ento.51.110104.151124 Bray, J.R., and J.T. Curtis. 1957. An ordination of the upland forest communities of southern Wisconsin. Ecol. Monogr. 27:325–349. doi:10.2307/1942268 Cushman, S.A., and K. McGarigal. 2004. Patterns in the species-environment relationship depend on both scale and choice of response variables. Oikos 105:117–124. doi:10.1111/j.0030-1299.2004.12524.x Dunne, T., and L.B. Leopold. 1978. Water in environmental planning. Freeman, San Francisco, CA. Fore, L.S., J.R. Karr, and R.W. Wiseman. 1996. Assessing invertebrate responses to human activities: Evaluating different approaches. J. North Am. Benthol. Soc. 15:212–231. doi:10.2307/1467949 Gerritsen, J.M., M.T. Barbour, and K. King. 2000. Apples, oranges, and ecoregions: On determining pattern in aquatic assemblages. J. North Am. Benthol. Soc. 19:487–496. doi:10.2307/1468109 Hawkins, C.P., R.H. Norris, J. Gerritsen, R.M. Hughes, S.K. Jackson, R.K. Johnson, and R. Jan Stevenson. 2000. Evaluation of the use of landscape classifications for the prediction of freshwater biota: Synthesis and recommendations. J. North Am. Benthol. Soc. 19:501–517. doi:10.2307/1468111 Hawkins, C.P., and R.M. Vinson. 2000. Weak correspondence between landscape classifications and stream invertebrate assemblages: Implications for bioassessment. J. North Am. Benthol. Soc. 19:501–517. doi:10.2307/1468111 Heino, J., T. Muotka, H. Mykrä, R. Paavola, H. Hämäläinen, and E. Koskenniemi. 2003. Defining macroinvertebrate assemblage types of headwater streams: Implications for bioassessment and conservation. Ecol. Appl. 13:842–852. doi:10.1890/1051-0761(2003)013[0842:DMATOH]2.0.CO;2 Heino, J., and H. Mykrä. 2006. Assessing physical surrogates for biodiversity: Do tributary and stream type classifications reflect macroinvertebrate diversity in running waters? Biol. Conserv. 129:418–426. doi:10.1016/j. biocon.2005.11.009 Herlihy, A.T., S.G. Paulsen, J. Van Sickle, J.L. Stoddard, C.P. Hawkins, and L.L. Yuan. 2008. Striving for consistency in a national assessment: The challenges of applying a reference-condition approach at a continental scale. J. North Am. Benthol. Soc. 27:860–877. doi:10.1899/08-081.1 Hirst, C.N., and D.A. Jackson. 2007. Reconstructing community relationships: The impact of sampling error, ordination approach, and gradient length. Diversity Distrib 13:361–371. doi:10.1111/j.1472-4642.2007.00307.x Hynes, H.B.N. 1975. The stream and its valley. Verh. Int. Ver. Theor. Angew. Limnol. 19:1–15. Jaccard, P. 1901. Distribution de la flore alpine dans le Bassin des Drouces et dans quelques regions voisines. Bull. Soc. Vaud. Sci. Nat. 37(140):241–272. Jackson, D.A., and H.H. Harvey. 1997. Qualitative and quantitative sampling of lake fish communities. Can. J. Fish. Aquat. Sci. 54:2807–2813. doi:10.1139/ f97-182 Kaufmann, P.R., and E.G. Robison. 1998. Physical habitat characterizaton. In: J.M. Lazorchak, D.J. Klemm, and D.V. Peck, editors, Environmental monitoring
and assessment program–surface waters: Field operations and methods for measuring the ecological condition of wadeable streams. EPA/620/R94/004F. Office of Research and Development, USEPA, Washington, DC.
p. 77–118. Manel, S., H.C. Williams, and S.J. Ormerod. 2001. Evaluating presence/absence models in ecology: The need to account for prevalence. J. Appl. Ecol. 38:921– 931. doi:10.1046/j.1365-2664.2001.00647.x McManamay, R., D. Orth, C. Dolloff, and E. Frimpong. 2012. Regional frameworks applied to hydrology: Can landscape-based frameworks capture the hydrologic variability? River Res. Appl. 28:1019–1033. doi:10.1002/rra.1493
1012
Milner, A.M., and M.W. Oswood. 2000. Urbanization gradients in streams of Anchorage, Alaska: A comparison of multivariate and multimetric approaches to classification. Hydrobiologia 422/423:209–223. doi:10.1023/A:1017001812212 Newell, R.L., and G.W. Minshall. 1978. Life-history of a multivoltine mayfly, Tricorythodes minutus 1: An example of effect of temperature on life cycle. Ann. Entomol. Soc. Am. 71:876–881. Norris, R.H. 1995. Biological monitoring: The dilemma of data analysis. J. North Am. Benthol. Soc. 14:440–450. doi:10.2307/1467210 Oberdorff, T., P.A. Tededsco, B. Hugueny, F. Leprieur, O. Beuchard, S. Brosse, and H.H. Dürr. 2011. Global and regional patterns in riverine fish species richness: A review. Int. J. Ecol. 2011:1–12. doi:10.1155/2011/967631 Oksanen, J. 2011. Multivariate analysis of ecological communities in R: Vegan tutorial. cc.oulu.fi/~jarioksa/opetus/metodi/vegantutor.pdf (accessed Mar. 2013). Oksanen, J., F. Guillaume Blanchet, K. Roeland, P. Legendre, P.R. Minchin, R.B. O’Hara, G.L. Simpson, P. Solymos, M. Henry, H. Stevens, and H. Wagner. 2012. Vegan: Community ecology package. R package version 2.0–5. http:// CRAN.R-project.org/package=vegan( accessed Mar, 2013). Omernik, J.M. 1987. Ecoregions of the conterminous United States (map supplement). Ann. Assoc. Am. Geogr. 77:118–125. doi:10.1111/j.1467-8306.1987. tb00149.x Omernik, J.M., and G.E. Griffith. 1991. Ecological regions versus hydrologic units: Frameworks for managing water quality. J. Soil Water Conserv. 46:334–340. R Development Core Team. 2012. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna. Resh, V.H. 1994. Variability, accuracy, and taxonomic costs of rapid bioassessment approaches in benthic macroinvertebrate monitoring. Boll. Zool. 61:375–383. doi:10.1080/11250009409355910 Reynoldson, T.B., R.H. Norris, V.H. Resh, K.E. Day, and D.M. Rosenberg. 1997. The reference condition: A comparison of multimetric and multivariate approaches to assess water-quality impairment using benthic macroinvertebrates. J. North Am. Benthol. Soc. 16:833–852. doi:10.2307/1468175 Roset, N., G. Grenouillet, D. Goffaux, D. Pont, and P. Kestermont. 2007. A review of existing fish assemblage indicators and methodologies. Fish. Manag. Ecol. 14:393–405. doi:10.1111/j.1365-2400.2007.00589.x Santhi, C., P.M. Allen, R.S. Muttiah, J.G. Arnold, and P. Tuppad. 2008. Regional estimation of baseflow for the conterminous United States by hydrologic landscape regions. J. Hydrol. 351:139–153. doi:10.1016/j.jhydrol.2007.12.018 Sterner, R.W., and J.J. Elser. 2002. Ecological stoichiometry: The biology of elements from molecules to the biosphere. Princeton Univ. Press, Princeton, NJ. Stoddard, J.L., A.T. Herlihy, D.V. Peck, R.M. Hughes, T.R. Whittier, and E. Tarquino. 2008. A process for creating multimetric indices for large-scale aquatic surveys. J. N. Am. Benthol. Soc. 27:878–891. Strayer, D.L. 1999. Statistical power of presence/absence data to detect population declines. Conserv. Biol. 13:1034–1038. doi:10.1046/j.1523-1739.1999.98143.x Sweeney, B.W., and R.L. Vannote. 1978. Size variation and distribution of hemimetabolous aquatic insects: Two thermal equilibrium hypotheses. Science 200:444–446. doi:10.1126/science.200.4340.444 Van Sickle, J. 1997. Using mean similarity dendrograms to evaluate classifications. J. Agric. Biol. Environ. Stat. 2:370–388. doi:10.2307/1400509 Van Sickle, J., and R.M. Hughes. 2000. Classification strength of Ecoregions, catchments and geographic clusters for aquatic vertebrates in Oregon. J. North Am. Benthol. Soc. 19:370–384. doi:10.2307/1468101 Vannote, R.L., and B.W. Sweeney. 1980. Geographic analysis of thermal equilibria: A conceptual model for evaluating the effect of natural and modified thermal regimes on aquatic insect communities. Am. Nat. 115:667–695. doi:10.1086/283591 Waite, I.R., A.T. Herlihy, D.P. Larsen, and D.J. Klemm. 2000. Comparing strengths of geographic and nongeographic classifications of stream benthic macroinvertebrates in the Mid-Atlantic Highlands, USA. J. North Am. Benthol. Soc. 19:429–441. doi:10.2307/1468105 Wang, L., J. Lyons, P. Rasmussen, P. Seelbach, T. Simon, M. Wiley, P. Kanehl, E. Baker, S. Niemela, and P.M. Stewart. 2003. Watershed, reach, and riparian influences on stream fish assemblages in the Northern Lakes and Forests ecoregion, USA. Can. J. Fish. Aquat. Sci. 60:491–505. doi:10.1139/f03-043 Whittier, T.R., J.L. Stoddard, D.P. Larsen, and A.T. Herlihy. 2007. Selecting reference sites for stream biological assessments: Best professional judgment or objective criteria. J. North Am. Benthol. Soc. 26:349–360. doi:10.1899/0887-3593(2007)26[349:SRSFSB]2.0.CO;2 Winter, T.C. 2001. The concept of hydrologic landscapes. J. Am. Water Resour. Assoc. 37:335–349. Wolock, D.M., T.C. Winter, and G. McMahon. 2004. Delineation and evaluation of hydrologic-landscape regions in the United States using geographic information system tools and multivariate statistical analysis. Environ. Manage. 34(Suppl. 1):S71–S88. doi:10.1007/s00267-003-5077-9 Xenopoulos, M.A., and D.M. Lodge. 2006. Going with the flow: Using speciesdischarge relationships to forecast losses in fish biodiversity. Ecology 87:1907– 1914. doi:10.1890/0012-9658(2006)87[1907:GWTFUS]2.0.CO;2
Journal of Environmental Quality
Journal of Environmental Quality
TECHNICAL REPORTS SURFACE WATER QUALITY
A Comparison of Fish-based Classification Schemes for Reference Streams and Rivers in Nebraska Thomas Heatherly II,* Ken Bazata, David Schumacher, and Elbert Traylor
T
he accuracy and precision of stream bioassess-
Proper assessments of lotic ecosystems depend on our ability to isolate natural differences from anthropogenic disturbances. We examined Nebraska river and stream classification strength, based on fish species, using multiple response permutation procedures for common classification strategies: ecoregions, watersheds, hydrologic-landscape regions, and assemblage structure from cluster analyses. Next, we tested the ecological interpretability of classification schemes using nonmetric multidimensional scaling ordinations and ANOVAs. Finally, we used nonparametric ANOVA to identify environmental predictors of overall fish assemblage structure. Hydrologic-landscape regions had the highest classification strength, but cluster groups had the most ecological interpretability based on the discreteness of the groups in ordination space and on the large number of common species that had different abundances across cluster groups. In addition, presence/absence data provided groups with more classification strength and interpretability than abundance data. Temperature, stream size, total phosphorus concentrations, and the percentage of fine substrates were significantly correlated to nonmetric multidimensional scaling ordinations and to overall fish structure in the multivariate ANOVA models. Cluster analyses using presence/absence were therefore the best classification scheme, and we identified the environmental variables that are likely to be useful for determining whether streams should have similar biotic assemblages. This information will be a valuable guide for separating natural variability in biotic assemblages from anthropogenic influences.
ments depend on proper classifications of streams into types with naturally distinct biotic assemblages (Milner and Oswood, 2000; Hawkins et al., 2000). The traditional method has been to use landscape-based classifications (Roset et al., 2007; Stoddard et al., 2008) in which streams are grouped according to predetermined boundaries in the landscape, such as watersheds, Ecoregions (Omernik, 1987), or hydrologiclandscape regions (Winter, 2001). Watershed classifications assume that ecological attributes are influenced by properties of their catchments, such as geology, sediment supply, hydrology, and other factors (Hynes, 1975; Dunne and Leopold, 1978; Hawkins et al., 2000). Ecoregions assume that important factors correspond to identifiable regions within larger landscapes with similar terrestrial features (Omernik and Griffith, 1991). In practice, Ecoregions and watersheds often have explained minimal differences between streams (Hawkins et al., 2000; Wang et al., 2003). For example, Hawkins and Vinson (2000) compared the utility of five types of landscape classifications for explaining compositional differences of invertebrates from >2000 streams in the United States. They suggested that no existing landscape classification scheme was sufficient for partitioning invertebrate assemblages into discrete groups. Hydrologic-landscape regions (HLRs) are landscape-based classification schemes based on land surface area, soil and bedrock permeability, and precipitation and evapotranspiration potential (Winter, 2001; Wolock et al., 2004). If HLRs provide accurate classifications of hydrologic variability, they may prove superior to other classification schemes for the purposes of bioassessment. Wolock et al. (2004) found that HLRs were equal to or better than level II Ecoregions for describing land-surface forms, geologic texture, climate, land cover, and water quality and had some ability to characterize fish species richness. However, we are unaware of an explicit test of the ability of HLRs to describe discrete biotic communities. An alternative to landscape-based classifications is to use statistical classifications that do not rely on physiographic constraints (Gerritsen et al., 2000). Multivariate classifications,
Copyright © American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America. 5585 Guilford Rd., Madison, WI 53711 USA. All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher.
T. Heatherly, K. Bazata, D. Schumacher, and E. Traylor, Water Quality Division, Nebraska Dep. of Environmental Quality, 1200 “N” St., Suite 400, Lincoln, NE 68509; T. Heatherly, School of Natural Resources, Univ. of Nebraska-Lincoln, 3100 Holdrege St., Lincoln, NE 68583; T. Heatherly, current address: Departamento de Ecologia, Universidade do Estado do Rio de Janeiro, Rua São Francisco Xavier, 524, Rio de Janeiro, RJ, 20550, Brazil. Assigned to Associate Editor Patrick Inglett.
J. Environ. Qual. 43:1004–1012 (2014) doi:10.2134/jeq2013.03. 0102 Supplemental data file is available online for this article. Received 25 Mar. 2013. *Corresponding author ([email protected]).
Abbreviations: HLR, hydrologic-landscape region; MRPP, multiple response permutation procedure; NMDS, nonmetric multidimensional scaling.
1004
such as cluster analyses, are appealing because organismal data are inherently multivariate (Norris, 1995; Milner and Oswood, 2000). Gerritsen et al. (2000) found that streams in Wyoming were better characterized using clustering based on biotic similarity of invertebrates, although there were apparent Ecoregional affinities. Similarly, Milner and Oswood (2000) found that classifications based on multivariate techniques of invertebrates were better at classifying streams around Anchorage than multimetric indices that used regional classifications. Finally, Van Sickle and Hughes (2000) found that Ecoregions and watersheds were useful for classifying stream vertebrate assemblages, mostly because of spatial autocorrelation, but landscape-based approaches were not as effective as assemblage similarity-based approaches. We need to explicitly test different classification schemes to account for important natural differences between streams so that we may properly identify stressors. In this study, we first directly compared the utility of classifications of reference streams and rivers in Nebraska based on Ecoregions, watersheds, HLRs, and assemblage similarity using presence/absence and abundance data. Classifications are frequently performed with reference streams (Reynoldson et al., 1997; Hawkins et al., 2000; Bailey et al., 2004) to classify stream types by natural differences. Next, we used a combination of nonmetric multidimensional scaling ordination (NMDS) values and nonparametric multivariate ANOVA to identify environmental variables likely responsible for classifications. This research fills a gap in our knowledge of the best approaches for classifying streams according to natural differences. This knowledge will facilitate our ability to identify and manage anthropogenic impacts.
Materials and Methods Study Sites The streams in this study (Supplemental Table S1) were designated as reference sites by the Nebraska Department of Environmental Quality (Bazata, 2005, 2011) in an Environmental Management and Assessment Program study performed from 1997 to 2009. This designation was based on an index of habitat quality composed of 10 metrics of morphology (incision/width ratio, percent undercut banks, percent overhanging vegetation, percent pools, percent riffles, percent barren banks), substrate (percent sand substrate, percent silt substrate), and riparian quality (percent rowcrop within 30 m of stream, middle canopy density above stream). The streams that scored in or above the 90th percentile of all habitat index scores were used as reference streams. This method used only nonbiological characteristics, as suggested by Bailey et al. (2004) and Stoddard et al. (2008). We have much to learn about what constitutes a least-impaired condition and what constitutes a regionally representative set of sites (Whittier et al., 2007), especially in streams that cannot realistically be returned
to a historical condition. The reference streams in this study were chosen independently of this research and thus provide the opportunity to test classification schemes based on an existing management program. Furthermore, ongoing research in Nebraska has found that fish are better predicted by in- and nearstream habitat features than by land usage (T. Heatherly et al., unpublished observations). Thus, there is evidence that a reference system based on the physical template will be more meaningful for the management of fish than one based on features of the landscape. Sixty-eight sites were used in this study. These streams were within six level III Ecoregions and 11 of the 13 major watersheds (Fig. 1). Additionally, the reference streams were within 7 of the 14 HLRs that occur within Nebraska (Fig. 2).
Data Collection Each stream was sampled during summer baseflow conditions between the years 1997 and 2009, excluding 2002 and 2003 because of quality control concerns. We established a sampling reach of approximately 40 times the average stream width for sampling, with lower and upper limits of 150 and 300 m. We measured water quality parameters (temperature, pH, conductivity, and dissolved oxygen) with Hydrolab Quanta (Hach Hydromet) or Eureka Manta (Eureka Environmental Engineering) multiprobe meters. Water for nutrient chemistry analysis was collected the day of sampling in amber jars, preserved with concentrated sulfuric acid, and shipped overnight within 2 d of sampling to the USEPA Science and Technology Laboratory in Kansas City, Kansas. Samples were analyzed for dissolved and particulate phosphorus, nitrate-nitrogen, nitrite-nitrogen, and ammonium-nitrogen using standard methods (APHA, 1995). We used total phosphorus and total nitrogen in statistical analyses. Physical habitat measurements were taken at 11 evenly spaced transects along the study reach (Kaufmann and Robison, 1998). At each of the 11 transects, we measured stream width, depth, incision and bank angle,
Fig. 1. Map of Nebraska reference streams used in this study with delineations for level III Ecoregions, major watersheds, and assemblage groups assigned with hierarchical cluster analysis of fish assemblage presence/absence (P/A).
www.agronomy.org • www.crops.org • www.soils.org 1005
is the overall weighted mean of withingroup means of the pairwise dissimilarities among sampling units (Oksanen, 2011). Classification strength (Van Sickle, 1997) is the difference between within-group dissimilarities and average among-group dissimilarities calculated from a specified number of permutations of fish into random groupings and is chosen by setting the weight type of the mrpp function to 3 (Oksanen, 2011). We used 5000 permutations in each test. The classification strength is reported as within-group agreement, and significance is given by P value. Cluster analysis and MRPP provide information to assist in creating meaningful groups but do not provide details regarding environmental interpretations. We assessed the relationships between classifications and Fig. 2. Map of Nebraska reference streams used in this study with delineations for hydrologicenvironmental variables through vector fitting landscape regions and assemblage groups assigned with hierarchical cluster analysis of fish of environmental variables onto NMDS and by assemblage abundance. modeling the effect of environmental variables on overall composition structure using substratum size composition and embeddedness, bankfull width, nonparametric multivariate ANOVA. The ordination reduces the and depth and canopy cover. Additionally, we used visual surveys complexity of compositional data into a few dimensions (Oksanen, to estimate the percentage of riffle, run, and pool habitat as well as 2011) and thus may provide relatively simple interpretations of the the relative amounts of cover within the categories of filamentous overall multivariate dataset. We used the “metaMDS” function algae, macrophytes, brush, woody debris, overhanging vegetation, in the “Vegan” package to compare ordinations built from 20 undercut banks, boulders, and artificial subtrata (e.g., lowhead random starting configurations of fish presence/absence and used dams and rip-rap at bridges). Environmental variables with a large the Jaccard and Bray-Curtis indices as our dissimilarity measures number of zeroes were not included in subsequent analyses. for presence/absence and abundance data, respectively. We used We collected fish by multipass electro-fishing the entire reach; the function “envfit” to identify environmental variables with we used a backpack shocker for small, shallow streams (LR-24 significant correlations to the ordination. The direction of vector Electrofisher, Smith-Root Inc.) and a generator-powered shocker arrows on the ordination plot indicates the direction of most rapid mounted on a tote-barge (SDC-1 Stream Shocker Electrofishing change in the variable, and the length of the arrow is proportional Tow Barge System, ETS Electrofishing, LLC) for larger streams and to the strength of correlation (r2) between the ordination and rivers. We identified all large fish at the site before release. Smaller variable (Oksanen, 2011). Finally, we used the “ordihull” function fish were identified in the field or preserved in 10% formalin to draw polygons around each assemblage-based group of sites solution for later identification at the Sternberg Museum of Natural on the ordination plot as derived from cluster analyses. We used History, where they have been maintained as voucher specimens. Procrustes rotations to test each new iterative ordination derived Data Analysis from a different random starting configuration and standardized the abundance data before ordination with Wisconsin square-root We performed agglomerative, k-means cluster analyses using standardization (Oksanen, 2011). the hclust function in the base R package (R Development Core Nonparametric multivariate ANOVA imposes no constraints Team, 2012) with the Jaccard coefficient for presence/absence on compositional structure and does not assume linear data ( Jaccard, 1901) and Bray-Curtis dissimilarity (Bray and relationships between environmental variables and community Curtis, 1957) for abundance data. We grouped sites into clusters structure, whereas NMDS confines the community to a few axes using Ward’s linkage. We plotted the within-cluster sums of and vector fitting does assume linear relationships (Oksanen, squares against the number of potential cluster groups to choose 2011). We used the function “Adonis” in “Vegan” to model the the appropriate number of clusters. effect of 19 environmental variables (listed in Supplemental Next, we tested the classification strength of groups based on Table S2) on a dissimilarity matrix calculated with “vegdist” fish assemblage structure, Ecoregions (Level III; Omernik, 1987), using the Bray-Curtis dissimilarity index. Significance was watershed boundaries, and HLRs using a multiple response determined by comparing modeled results with 1000 random permutation procedure (MRPP) using the “mrpp” function in the iterations of the data. We began with a global model of all 19 “Vegan” package (Oksanen et al., 2012). Hydrologic-landscape variables and reduced the model by sequentially removing the region data were acquired as a GIS file through the USGS: http:// least significant parameter and rerunning the analysis until water.usgs.gov/GIS/dsdl/hlrus.tgz. We did not have enough the remaining parameters were significant. We used Pearson streams within Ecoregion 42 or 43 or HLR 6 or 13 for statistical correlation coefficients to ensure modeled variables were not analysis. All analyses were performed with presence/absence and correlated with each other. abundance data. The MRPP uses the test statistic delta, which 1006
Journal of Environmental Quality
We used one-way ANOVA to test for differences in the abundance of the 20 most abundant fishes among classification schemes. Tukey’s honestly significant differences test was used for pairwise comparisons among groups within classification schemes. Both of the tests were performed with the base R program. All abundances were log+1 transformed before analysis to meet normality and heteroscedasticity assumptions.
Results The plots of within-group sums of squares suggested that the presence/absence dendrogram be separated into four clusters and the abundance dendrogram be separated into two clusters (Fig. 3). The presence/absence groups were not limited to specific watersheds or HLRs, yet there did appear to be some regional similarities. For example, Group 1 occurred primarily in the northeast of the state, especially in the Elkhorn and eastern Niobrara River watersheds (Fig. 1). Group 2 largely represented the rest of the Niobrara watershed in the north-central portion of Nebraska. Group 3 mainly represented the southern watersheds, including the Republican, Little Blue, and Big Blue River watersheds, and Group 4 mainly occurred in the western Nebraska panhandle. The two abundance groups appeared more spatially overlapping, yet group 1 appeared to be more prevalent in the west and southwest (Fig. 2). The HLR classification scheme using presence/absence data had the highest within-group agreement (Table 1). The next highest within-group agreement values were for cluster analyses
of abundance data and then Ecoregions, watersheds, and clusters of presence/absence data. Ecoregion, watershed, and HLR classifications were not significant using abundance data. The NMDS ordination of presence/absence (Fig. 4) and abundance data (Fig. 5) had sufficiently low stress (0.169 and 0.187, respectively) to reduce fish community dissimilarity to two dimensions. For presence/absence data, only the cluster group (Fig. 4D) demonstrated both distinctness and compactness when plotted on NMDS ordinations. Nine of the 19 environmental variables tested were significantly correlated with the presence/absence ordination at p ≤ 0.100 (Supplemental Table S2). Streams within the White Hat River watershed, within Ecoregion 25, and within HLR 10 appeared to be associated with cobbles, overhanging vegetation, undercut banks, and fine substrates (Fig. 4). Streams within the Big Blue, Elkhorn, Middle Platte, South Platte, and Republican River watersheds were associated with higher nutrients and warmer temperature, and these were typically within Ecoregions 42 and 47 and HLRs 3, 5, and 8. For the abundance ordinations, all classification schemes were broadly overlapping, although the watershed groupings appeared to be the most distinct (Fig. 5C). The identity and direction of correlation of environmental variables were similar to the presence/absence ordination (Supplemental Table S3), except that the percentage of undercut bank habitat replaced the percentage of macrophyte cover in the abundance ordinations. Streams within the White Hat River watershed were also found
Fig. 3. Plots of the within-group sums of squares error against the numbers of cluster groups in hierarchical cluster analyses of fish presence/ absence (left) and abundance (right) collected from Nebraska reference streams. Table 1. Results of multiresponse permutation procedure to test the fit of classification of Nebraska streams by assemblage structure, Ecoregion, watershed, and Hydrologic Landscape Regiond (HLR) using fish species presence/absence and abundance. Grouping† Cluster-P/A Cluster-Abund Ecoregion-P/A Ecoregion-Abund Watershed-P/A Watershed-Abund HLR-P/A HLR-Abund
Within-group agreement
Observed delta
Expected delta
0.105 0.202 0.172 −0.012 0.117 0.151 0.460 0.012
5.412 434.7 5.005 551.1 5.337 462.5 3.265 538.2
6.044 544.8 6.044 544.8 6.044 544.8 6.044 544.8
Delta significance 0.001 0.001 0.001 0.545 0.001 0.070 0.001 0.383
† Abund, abundance; P/A, presence/absence. www.agronomy.org • www.crops.org • www.soils.org 1007
Fig. 4. Nonmetric multidimensional scaling ordination plots of Nebraska reference streams based on the similarity of fish species’ assemblages using presence/absence data. Polygons contain the sites of each assemblage group as determined by (A) level III Ecoregion, (B) hydrologiclandscape region, (C) watershed, and (D) hierarchical cluster analysis. Vectors show the direction and strength of environmental variable correlations to the ordination (P ≤ 0.100). BB, Big Blue; EL, Elkhorn; LO, Loup; MP, Middle Platte; MT, Missouri Tributary, NI, Niobrara; RE, Republican; SP, South Platte; TN, total nitrogen; TP, total phosphorus; WH, White Hat. Lodgepole Creek (SP2019) was not plotted in ordinations for clarity.
to be associated with cobbles and overhanging vegetation using abundance data. Streams within the North Platte River watershed were associated with undercut banks. As with presence/absence data, streams in the Big Blue, Elkhorn, Middle Platte, South Platte, and Republican River watersheds were associated with higher nutrient concentrations and warmer temperature. Nineteen of the 20 most abundant fish species had average abundances per site that were different among presence/ absence groups, and 12 of 20 of these fish were different among watersheds (Fig. 6). Six species were different among Ecoregions and HLRs, although HLRs did slightly better among the 10 most abundant fish. Additionally, only presence/absence groups had ANOVA F values that exceeded 5.0. We used Tukey’s honestly significant differences to examine how the most abundant fish differed among presence/ absence groups. Group 1 typically had high abundances of Notropis stramineus, Notropis dorsalis, Cyprinella lutrensis, and Hybognathus hankinsonui (Fig. 7). Group 2 had relatively high abundances of N. stramineus, N. dorsalis, Semotilus atromaculatus, Catostoma commersoni, and Rhinichthys cataractae. Group 3 had high abundances of N. stramineus, Pimephales promelas, S. atromaculatus, C. lutrensis, and Campostoma anomalum. Group 4 had relatively high abundances of R. cataractae. We performed nonparametric ANOVA analyses on fish presence/absence and abundance starting with the 10 environmental variables that were correlated to the ordinations. 1008
For the presence/absence of fish, 4 of the 19 environmental variables were significant predictors of overall assemblage dissimilarity (Table 2). Temperature explained the highest degree of variability, followed by average stream width, the percentage of fine substrata, and total phosphorus. Overall, these four variables explained about 15% of the total variation in dissimilarity (R2 = 0.15), and each individual variable explained ~3 to 4% of total variation. Only temperature and average width were significant predictors of the dissimilarity of fish abundances in the nonparametric ANOVA (Table 3), and each explained ~4% of total variation.
Discussion This study explicitly tested three common bioassessment classification schemes and another scheme that has not been applied for this purpose. Ecoregions, watershed boundaries, HLRs, and clusters provided potentially useful classifications of the fish assemblages of streams and rivers in Nebraska, and presence/absence data were more likely to produce discrete groups at a statewide scale than abundance data. Multiple response permutation procedure tests suggested that HLRs provided the most within-group agreement. However, presence/ absence clusters gave the most distinct groups when plotted on NMDS ordinations and were better than HLRs and Ecoregions at discriminating among the most abundant fish species. Boundaries of the major watersheds provided the next strongest Journal of Environmental Quality
Fig. 5. Nonmetric multidimensional scaling (NMDS) ordination plots of Nebraska reference streams based on the similarity of fish species’ assemblages using abundance data. Polygons contain the sites of each assemblage group as determined by (A) level III Ecoregion, (B) hydrologiclandscape region, (C) watershed, and (D) hierarchical cluster analysis. Vectors show the direction and strength of environmental variable correlations to the ordination (P ≤ 0.100). BB, Big Blue; EL, Elkhorn; LO, Loup; MP, Middle Platte; MT, Missouri Tributary, NI, Niobrara; RE, Republican; SP, South Platte; TN, total nitrogen; TP, total phosphorus; WH, White Hat. Lodgepole Creek (SP2019) was not plotted in ordinations for clarity.
Fig. 6. Plot of the 20 most abundant fishes in reference streams in Nebraska against the F values of ANOVA models testing for differences in the abundance of fish species among level III Ecoregions, hydrologic-landscape regions (HLR), presence/absence cluster groups (PA), and watersheds. From left to right, fishes are listed from the highest to the lowest average abundance per site. The gray vertical line denotes the level of F value associated with significant ANOVA models. www.agronomy.org • www.crops.org • www.soils.org 1009
Fig. 7. Boxplots of the abundance of the most abundant fishes from Nebraska reference streams against groups designated by cluster analysis of fish presence/absence (P/A) data. Letters indicate differences among groups (Tukey’s HSD, P ≤ 0.05). CPUE, catch per unit effort.
ability to create distinct groups based on community dissimilarity, and common fish species frequently had abundances that were distinctly different among watersheds. Therefore, the bulk of our results support using cluster analyses with presence/absence data for generating distinct and ecologically interpretable fish classifications. Hydrologic-landscape regions were built to better manage the quantity and quality of water resources through the integration of climate, soil, and hydrological sciences (Winter, 2001; Wolock et al., 2004; Santhi et al., 2008). They have rarely been
tested as important units for biota, so our results are difficult to place within context. Wolock et al. (2004) found that HLRs accurately represented ~31% of variation in fish species richness, which was slightly less than the 45% of level II Ecoregions. They also found that HLRs were as good as or better than Ecoregions at describing landscape surface form and texture, climate characteristics, land cover, and water quality. Santhi et al. (2008) found that the variables used to construct HLRs, such as relief, climate, and the percentage of sand in soils, were good predictors of stream baseflows, yet they were unsure whether these variables
Table 2. Results of nonparametric multivariate analysis of variance to test for relationships between environmental variables and the presence/ absence of fishes in Nebraska streams. Parameter
df
Sums of squares
Mean squares
F value
R2
P value
Temperature Average width Total P Fine substrate, % Residuals Total
1 1 1 1 64 68
0.663 0.593 0.453 0.580 13.294 15.584
0.663 0.593 0.453 0.580 0.208
3.189 2.855 2.183 2.793
0.043 0.038 0.029 0.037 0.853 1.00
0.007 0.008 0.044 0.004
Table 3. Results of nonparametric multivariate analysis of variance to test for relationships between environmental variables and the abundance of fishes in Nebraska streams. Parameter
df
Sums of squares
Mean squares
F value
R2
P value
Temperature Average width Residuals Total
1 1 66 68
0.952 0.825 23.542 25.318
0.952 0.821 0.357
2.668 2.313
0.038 0.033 0.930 1.00
0.002 0.007
1010
Journal of Environmental Quality
could represent the variability of flows. We found that they may provide significant classifications of fish species, but HLRs did not provide as much ecological interpretability as groups derived from cluster analyses or watershed boundaries. This is somewhat surprising given that temperature and discharge were the best predictors of overall assemblage structure in nonparametric ANOVA tests but may be partially explained by the large number of local physical habitat correlates to NMDS ordinations. Furthermore, McManamay et al. (2012) found that HLRs did a poor job of predicting flow class, which suggests that HLRs may not capture the importance of stream size to the structure of fish assemblages. Hydrologic-landscape regions may provide one level of representative information when placed into a hierarchical framework that incorporates local influences; however, it is possible that regional frameworks will not capture enough of the individual variability of ecosystems to be broadly useful at statewide levels (McManamay et al., 2012). Previous classification strength research has been conflicting, yet classifications based on cluster analyses usually provided better representations of natural groups. Van Sickle and Hughes (2000) found assemblage structure was superior to landscape classifications for vertebrates in Oregon streams, and they also found Ecoregions performed better than catchments. Van Sickle (1997) suggested that fish in Oregon streams were best classified by Ecoregions. Several investigations of aquatic invertebrates confirmed the superiority of assemblage-based classifications to Ecoregions and watersheds, and the latter two classifications did not consistently explain more variation than random groupings (Gerritsen et al., 2000; Hawkins et al., 2000; Milner and Oswood, 2000; Wang et al., 2003). Waite et al. (2000) found weak classification strengths of Ecoregions to describe invertebrates in Mid-Atlantic streams of the United States, which improved when Ecoregions were filtered first by stream order but did not have the strength of clusters. Landscape-based classifications may suffer if there are weak ecological filters at landscape levels and if there are widespread individualistic relationships between species and the environment (Resh, 1994; Fore et al., 1996; Heino et al., 2003; Bonada et al., 2006; Heino and Mykrä, 2006). Herlihy et al. (2008) suggested nine Ecoregions maximized withinregion similarity in macroinvertebrate assemblages across the conterminous United States. The majority of Nebraska belonged to the single “Southern Plains” region largely because a lack of reference sites according to their criteria prevented the use of smaller regions. Therefore, we may expect that homogeneity and stream degradation in Nebraska may make Ecoregions at any level disadvantageous. We did identify distinct stream types within Nebraska using dissimilarity measures, and classifications based on fish presence/absence appeared to incorporate site-specific differences in important environmental variables. Our explicit tests of the classification strengths of different grouping methods will help managers control for natural differences in reference streams, which has been identified as a major impediment to bioassessment using a reference-condition approach (Waite et al., 2000; Herlihy et al., 2008). Evidence that presence/absence data performed best for community analyses was found by Hirst and Jackson (2007), who recreated known configurations of community data for a variety of ordination methods. They suggested that although abundance data potentially contain more information, they
are more easily biased by natural and experimental error than presence/absence data (see also Jackson and Harvey, 1997). Additionally, the ability of abundance and presence/absence to represent assemblages may depend on the relative commonness and rarity of the species (Manel et al., 2001; Cushman and McGarigal, 2004) and on the spatial scale in question (Strayer, 1999; Cushman and McGarigal, 2004). For example, Cushman and McGarigal (2004) found that presence/absence data were superior for larger spatial scales and assemblages composed of numerous relatively uncommon species of birds. However, Bart and Klosiewski (1989) found that abundance and presence/ absence data performed similarly in describing bird assemblages, and Strayer (1999) found that presence/absence data generally were not effective at detecting such trends as population declines. Given our findings and the majority of those in the literature, we recommend the use of presence/absence data to classify streams of Nebraska by fish assemblages. Temperature, stream size (average width), total phosphorus, and the percentage of fine substrates appeared to influence the structure of fish assemblages in Nebraska based on nonparametric multivariate ANOVA tests. These variables were also significantly correlated to both ordinations (Supplemental Tables S2 and S3), which is further confirmation of their ecological importance. Temperature is a driver of the metabolism of ectotherms and influences the distributions, growth rates, and fecundity of aquatic organisms (Newell and Minshall, 1978; Sweeney and Vannote, 1978; Vannote and Sweeney, 1980). Stream size is also a fundamental component of stream ecology. For example, increases in habitat space are considered responsible for the logarithmic increase in fish richness with increases in discharge (Xenopoulos and Lodge, 2006; Oberdorff et al., 2011). Phosphorus is an essential macronutrient that is frequently limiting in freshwater ecosystems (Sterner and Elser, 2002), and the percentage of fine substrates may be an indicator of impairment to the watershed that results in sediment deposition to streams (Allan and Castillo, 2007). In this study, three independent tests suggested these variables were responsible for the observed structural differences of fish assemblages in streams and rivers of Nebraska.
Conclusions Cluster analyses using presence/absence data provided more ecologically meaningful classifications of Nebraska streams and rivers based on fish species than classifications based on Ecoregions, major watershed boundaries, hydrologiclandscape regions, or cluster analyses using fish abundances. This conclusion is based on the separation of presence/absence groups in ordination space and on the differences in abundance of the most common fish species among classification groups. Stream size and temperature, phosphorus concentrations, and the percentage of fine substrates are useful for isolating natural versus anthropogenic-driven variability in fish assemblages. These distinctions should be widely applicable to those tasked with managing biotic integrity in agricultural regions.
Acknowledgments This study was supported by U.S. Environmental Protection Agency CWA section 104(b)(3) and section 319 funds awarded to the Nebraska Department of Environmental Quality. The authors thank L. Johnson
www.agronomy.org • www.crops.org • www.soils.org 1011
for assistance in creating Figures 1 and 2 and Associate Editor M. Archer and two anonymous reviewers for constructive advice on earlier drafts.
References APHA. 1995. Standard methods. 19th ed. American Public Health Association, Washington, DC. Allan, J.D., and M.M. Castillo. 2007. Stream ecology: Structure and function of running waters. 2nd ed. Chapman and Hall, New York. Bailey, R.C., R.H. Norris, and T.B. Reynoldson. 2004. Bioassessment of freshwater ecosystems: Using the reference condition approach. Kluwer Academic, Boston, MA. Bart, J., and S.P. Klosiewski. 1989. Use of presence/absence to measure changes in avian density. J. Wildl. Manage. 53:847–852. doi:10.2307/3809224 Bazata, K. 2005. Nebraska stream classification using fish, macroinvertebrates, habitat, and chemistry evaluations from R-EMAP data, 1997–2001. Project report for the Nebraska Department of Environmental Quality, Lincoln, NE. http:// www.deq.state.ne.us/Publica.nsf//NebrStreamBiolMonitorReport20042008. pdf (accessed Feb. 2013). Bazata, K. 2011. Nebraska stream biological monitoring program 2004–2008. Project report for the Nebraska Department of Environmental Quality, Lincoln, NE. http://deq.ne.gov/Publica.nsf//NebraskaWaterMonitoringProgramselectronic.pdf (accessed Feb. 2013). Bonada, N., N. Prat, V.H. Resh, and B. Statzner. 2006. Developments in aquatic insect biomonitoring: A comparative analysis of recent approaches. Annu. Rev. Entomol. 51:495–523. doi:10.1146/annurev.ento.51.110104.151124 Bray, J.R., and J.T. Curtis. 1957. An ordination of the upland forest communities of southern Wisconsin. Ecol. Monogr. 27:325–349. doi:10.2307/1942268 Cushman, S.A., and K. McGarigal. 2004. Patterns in the species-environment relationship depend on both scale and choice of response variables. Oikos 105:117–124. doi:10.1111/j.0030-1299.2004.12524.x Dunne, T., and L.B. Leopold. 1978. Water in environmental planning. Freeman, San Francisco, CA. Fore, L.S., J.R. Karr, and R.W. Wiseman. 1996. Assessing invertebrate responses to human activities: Evaluating different approaches. J. North Am. Benthol. Soc. 15:212–231. doi:10.2307/1467949 Gerritsen, J.M., M.T. Barbour, and K. King. 2000. Apples, oranges, and ecoregions: On determining pattern in aquatic assemblages. J. North Am. Benthol. Soc. 19:487–496. doi:10.2307/1468109 Hawkins, C.P., R.H. Norris, J. Gerritsen, R.M. Hughes, S.K. Jackson, R.K. Johnson, and R. Jan Stevenson. 2000. Evaluation of the use of landscape classifications for the prediction of freshwater biota: Synthesis and recommendations. J. North Am. Benthol. Soc. 19:501–517. doi:10.2307/1468111 Hawkins, C.P., and R.M. Vinson. 2000. Weak correspondence between landscape classifications and stream invertebrate assemblages: Implications for bioassessment. J. North Am. Benthol. Soc. 19:501–517. doi:10.2307/1468111 Heino, J., T. Muotka, H. Mykrä, R. Paavola, H. Hämäläinen, and E. Koskenniemi. 2003. Defining macroinvertebrate assemblage types of headwater streams: Implications for bioassessment and conservation. Ecol. Appl. 13:842–852. doi:10.1890/1051-0761(2003)013[0842:DMATOH]2.0.CO;2 Heino, J., and H. Mykrä. 2006. Assessing physical surrogates for biodiversity: Do tributary and stream type classifications reflect macroinvertebrate diversity in running waters? Biol. Conserv. 129:418–426. doi:10.1016/j. biocon.2005.11.009 Herlihy, A.T., S.G. Paulsen, J. Van Sickle, J.L. Stoddard, C.P. Hawkins, and L.L. Yuan. 2008. Striving for consistency in a national assessment: The challenges of applying a reference-condition approach at a continental scale. J. North Am. Benthol. Soc. 27:860–877. doi:10.1899/08-081.1 Hirst, C.N., and D.A. Jackson. 2007. Reconstructing community relationships: The impact of sampling error, ordination approach, and gradient length. Diversity Distrib 13:361–371. doi:10.1111/j.1472-4642.2007.00307.x Hynes, H.B.N. 1975. The stream and its valley. Verh. Int. Ver. Theor. Angew. Limnol. 19:1–15. Jaccard, P. 1901. Distribution de la flore alpine dans le Bassin des Drouces et dans quelques regions voisines. Bull. Soc. Vaud. Sci. Nat. 37(140):241–272. Jackson, D.A., and H.H. Harvey. 1997. Qualitative and quantitative sampling of lake fish communities. Can. J. Fish. Aquat. Sci. 54:2807–2813. doi:10.1139/ f97-182 Kaufmann, P.R., and E.G. Robison. 1998. Physical habitat characterizaton. In: J.M. Lazorchak, D.J. Klemm, and D.V. Peck, editors, Environmental monitoring
and assessment program–surface waters: Field operations and methods for measuring the ecological condition of wadeable streams. EPA/620/R94/004F. Office of Research and Development, USEPA, Washington, DC.
p. 77–118. Manel, S., H.C. Williams, and S.J. Ormerod. 2001. Evaluating presence/absence models in ecology: The need to account for prevalence. J. Appl. Ecol. 38:921– 931. doi:10.1046/j.1365-2664.2001.00647.x McManamay, R., D. Orth, C. Dolloff, and E. Frimpong. 2012. Regional frameworks applied to hydrology: Can landscape-based frameworks capture the hydrologic variability? River Res. Appl. 28:1019–1033. doi:10.1002/rra.1493
1012
Milner, A.M., and M.W. Oswood. 2000. Urbanization gradients in streams of Anchorage, Alaska: A comparison of multivariate and multimetric approaches to classification. Hydrobiologia 422/423:209–223. doi:10.1023/A:1017001812212 Newell, R.L., and G.W. Minshall. 1978. Life-history of a multivoltine mayfly, Tricorythodes minutus 1: An example of effect of temperature on life cycle. Ann. Entomol. Soc. Am. 71:876–881. Norris, R.H. 1995. Biological monitoring: The dilemma of data analysis. J. North Am. Benthol. Soc. 14:440–450. doi:10.2307/1467210 Oberdorff, T., P.A. Tededsco, B. Hugueny, F. Leprieur, O. Beuchard, S. Brosse, and H.H. Dürr. 2011. Global and regional patterns in riverine fish species richness: A review. Int. J. Ecol. 2011:1–12. doi:10.1155/2011/967631 Oksanen, J. 2011. Multivariate analysis of ecological communities in R: Vegan tutorial. cc.oulu.fi/~jarioksa/opetus/metodi/vegantutor.pdf (accessed Mar. 2013). Oksanen, J., F. Guillaume Blanchet, K. Roeland, P. Legendre, P.R. Minchin, R.B. O’Hara, G.L. Simpson, P. Solymos, M. Henry, H. Stevens, and H. Wagner. 2012. Vegan: Community ecology package. R package version 2.0–5. http:// CRAN.R-project.org/package=vegan( accessed Mar, 2013). Omernik, J.M. 1987. Ecoregions of the conterminous United States (map supplement). Ann. Assoc. Am. Geogr. 77:118–125. doi:10.1111/j.1467-8306.1987. tb00149.x Omernik, J.M., and G.E. Griffith. 1991. Ecological regions versus hydrologic units: Frameworks for managing water quality. J. Soil Water Conserv. 46:334–340. R Development Core Team. 2012. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna. Resh, V.H. 1994. Variability, accuracy, and taxonomic costs of rapid bioassessment approaches in benthic macroinvertebrate monitoring. Boll. Zool. 61:375–383. doi:10.1080/11250009409355910 Reynoldson, T.B., R.H. Norris, V.H. Resh, K.E. Day, and D.M. Rosenberg. 1997. The reference condition: A comparison of multimetric and multivariate approaches to assess water-quality impairment using benthic macroinvertebrates. J. North Am. Benthol. Soc. 16:833–852. doi:10.2307/1468175 Roset, N., G. Grenouillet, D. Goffaux, D. Pont, and P. Kestermont. 2007. A review of existing fish assemblage indicators and methodologies. Fish. Manag. Ecol. 14:393–405. doi:10.1111/j.1365-2400.2007.00589.x Santhi, C., P.M. Allen, R.S. Muttiah, J.G. Arnold, and P. Tuppad. 2008. Regional estimation of baseflow for the conterminous United States by hydrologic landscape regions. J. Hydrol. 351:139–153. doi:10.1016/j.jhydrol.2007.12.018 Sterner, R.W., and J.J. Elser. 2002. Ecological stoichiometry: The biology of elements from molecules to the biosphere. Princeton Univ. Press, Princeton, NJ. Stoddard, J.L., A.T. Herlihy, D.V. Peck, R.M. Hughes, T.R. Whittier, and E. Tarquino. 2008. A process for creating multimetric indices for large-scale aquatic surveys. J. N. Am. Benthol. Soc. 27:878–891. Strayer, D.L. 1999. Statistical power of presence/absence data to detect population declines. Conserv. Biol. 13:1034–1038. doi:10.1046/j.1523-1739.1999.98143.x Sweeney, B.W., and R.L. Vannote. 1978. Size variation and distribution of hemimetabolous aquatic insects: Two thermal equilibrium hypotheses. Science 200:444–446. doi:10.1126/science.200.4340.444 Van Sickle, J. 1997. Using mean similarity dendrograms to evaluate classifications. J. Agric. Biol. Environ. Stat. 2:370–388. doi:10.2307/1400509 Van Sickle, J., and R.M. Hughes. 2000. Classification strength of Ecoregions, catchments and geographic clusters for aquatic vertebrates in Oregon. J. North Am. Benthol. Soc. 19:370–384. doi:10.2307/1468101 Vannote, R.L., and B.W. Sweeney. 1980. Geographic analysis of thermal equilibria: A conceptual model for evaluating the effect of natural and modified thermal regimes on aquatic insect communities. Am. Nat. 115:667–695. doi:10.1086/283591 Waite, I.R., A.T. Herlihy, D.P. Larsen, and D.J. Klemm. 2000. Comparing strengths of geographic and nongeographic classifications of stream benthic macroinvertebrates in the Mid-Atlantic Highlands, USA. J. North Am. Benthol. Soc. 19:429–441. doi:10.2307/1468105 Wang, L., J. Lyons, P. Rasmussen, P. Seelbach, T. Simon, M. Wiley, P. Kanehl, E. Baker, S. Niemela, and P.M. Stewart. 2003. Watershed, reach, and riparian influences on stream fish assemblages in the Northern Lakes and Forests ecoregion, USA. Can. J. Fish. Aquat. Sci. 60:491–505. doi:10.1139/f03-043 Whittier, T.R., J.L. Stoddard, D.P. Larsen, and A.T. Herlihy. 2007. Selecting reference sites for stream biological assessments: Best professional judgment or objective criteria. J. North Am. Benthol. Soc. 26:349–360. doi:10.1899/0887-3593(2007)26[349:SRSFSB]2.0.CO;2 Winter, T.C. 2001. The concept of hydrologic landscapes. J. Am. Water Resour. Assoc. 37:335–349. Wolock, D.M., T.C. Winter, and G. McMahon. 2004. Delineation and evaluation of hydrologic-landscape regions in the United States using geographic information system tools and multivariate statistical analysis. Environ. Manage. 34(Suppl. 1):S71–S88. doi:10.1007/s00267-003-5077-9 Xenopoulos, M.A., and D.M. Lodge. 2006. Going with the flow: Using speciesdischarge relationships to forecast losses in fish biodiversity. Ecology 87:1907– 1914. doi:10.1890/0012-9658(2006)87[1907:GWTFUS]2.0.CO;2
Journal of Environmental Quality
