Weigel_Dimick_river_invert_IBI_2011_0001.pdf

J-NABS

Journal of the North American Benthological Society

V

Development, validation, and application of a macroinvertebrate-based Index of Biotic Integrity for nonwadeable rivers of Wisconsin

Brian M. Weigel1

Fisheries and Aquatic Sciences Research, Bureau of Science Services, Wisconsin Department of Natural Resources, 2801 Progress Road, Madison, Wisconsin 53716 USA

Jeffrey J. Dimick2

Aquatic Biomonitoring Laboratory, College of Natural Resources, University of Wisconsin, Stevens Point, Wisconsin, 54481 USA

Development, validation, and application of a macroinvertebrate-based Index of Biotic Integrity for nonwadeable rivers of Wisconsin

Brian M. Weigel1

Fisheries and Aquatic Sciences Research, Bureau of Science Services, Wisconsin Department of Natural Resources, 2801 Progress Road, Madison, Wisconsin 53716 USA

Jeffrey J. Dimick2

Aquatic Biomonitoring Laboratory, College of Natural Resources, University of Wisconsin, Stevens Point, Wisconsin, 54481 USA

Abstract. Quantitative biological assessment protocols are needed for monitoring river status and evaluating river rehabilitation efforts. We conducted a standardized macroinvertebrate survey at 100 sites on 38 nonwadeable rivers in Wisconsin to construct, test, and apply an index of biotic integrity (IBI) intended to be such a bioassessment tool. We assigned independent samples to IBI development (n = 75) and IBI validation (n = 25) data sets. We placed Hester–Dendy artificial substrates at the sites for 6 wk and processed the samples of colonizing macroinvertebrates in the laboratory with a 500-target subsampling procedure plus a large–rare taxon search. Independent of the biota, we assigned an environmental disturbance score to each site based upon water chemistry, land cover, flow modification, and point-source pollution. Ten metrics that represent macroinvertebrate assemblage structure, composition, and function constitute the IBI: the number of taxa in: 1) Insecta or 2) Ephemeroptera, Plecoptera, Trichoptera (EPT); % individuals that were: 3) Insecta, 4) intolerant EPT, 5) tolerant chironomids, 6) gatherers, 7) scrapers, or from 8) the dominant 3 taxa; 9) the mean pollution tolerance value; and 10) the number of unique ecological functional trait niches. Analyses on both the validation subset of sites and all sites inclusive confirmed that least-disturbed sites had the highest IBI scores, severely disturbed sites had the lowest scores, and moderately disturbed sites had intermediate scores. Chironominae and Hydropsychidae taxa known to tolerate nutrient enrichment and overall degraded conditions dominated samples with low IBI scores. In contrast, a diverse assemblage that thrives in relatively undisturbed conditions was present in samples with high IBI scores. Comparison of the new macroinvertebrate IBI with an existing fish IBI suggested that the indices respond to different environmental stressors and illustrated the limitations of using only one taxonomic group for bioassessment. We discuss new macroinvertebrate methods, an IBI development process, and the refinement of metrics that may be useful in tailoring assessment tools for large rivers or wadeable streams in other regions. We also present applications of the IBI, including its potential use in comprehensive large river monitoring programs and for evaluating management efforts.

Key words: IBI, multimetric index, bioassessment, benthic macroinvertebrates, aquatic insects, large river.

Large nonwadeable rivers of the upper-midwestern USA are valuable ecosystems that provide many ecological services and societal benefits. Humans have used these rivers for drinking water, flood control, hydroelectric power, navigation, recreation, and waste assimilation, which ultimately degraded ecosystem integrity (Karr et al. 1985, Ward and Stanford 1989, Johnson et al. 1995, Sheehan and Rasmussen 1999).

Rivers have experienced large reductions in pollution because of the US Clean Water Act of 1972 and mitigation of hydroelectric disturbances because of the Electric Consumers Protection Act of 1986. The potential for additional pollution abatement exists via establishment of nutrient and biocriteria and a rigid 303(d) impaired-waters listing process that would trigger Total Maximum Daily Load (TMDL) modeling and implementation (USEPA 2000, WDNR 2010).

However, additional quantitative assessment tools are needed to determine river condition, prioritize

1 E-mail addresses: brian.weigel@wisconsin.gov 2 jdimick@uwsp.edu

J. N. Am. Benthol. Soc., 2011, 30(3):665–679 ’ 2011 by The North American Benthological Society

DOI: 10.1899/10-161.1

Published online: 31 May 2011 systems for protection or rehabilitation, assess river recovery after management application, and determine the overall efficacy of management programs.

A properly constructed and tested biological assessment framework has an integral role in the protection and rehabilitation of rivers (Courtemanch 1995, Barbour et al. 2000, Davies and Jackson 2006).

Indices of biotic integrity (IBIs) can be an assessment tool for determining current conditions and evaluat-ing restoration efforts (Ohio EPA 1987, Karr 1991).

IBIs are derived empirically by modeling relation-ships between the biotic assemblages and environ-ments of known condition. A standardized process can then be applied to sites of unknown environmen-tal condition to compare the assemblage found with that which is expected in the absence of degradation (Bailey et al. 2004). Biological monitoring is valuable for determining anthropogenic influences on rivers because biota respond integratively to stress from multiple spatial or temporal scales and pathways including water chemistry (Rosenberg and Resh 1993). Ultimately, IBIs quantify biological effects of human activities and rate environmental health.

Macroinvertebrate-based IBIs have been tailored widely to wadeable streams. However, few have been developed for large nonwadeable rivers (USEPA 2002). The Invertebrate Community Index (ICI) with scoring adjustments for the size of the water body was the first multimetric index applicable to rivers (Ohio EPA 1987, DeShon 1995), and it led to customization of an IBI for the Ohio River (Applegate et al. 2007).

River macroinvertebrate IBIs or multimetric indices have been calibrated for use in Idaho (Royer et al.

2001) and Michigan (Wessell et al. 2008), for 6 rivers in the Midwest (Blocksom and Johnson 2009), and for mid-continental US great rivers (the Ohio, Missis-sippi, and Missouri rivers; Angradi et al. 2009b). River IBIs are rare because of the logistical complexities of adequately characterizing macroinvertebrates in large river reaches (e.g., Flotemersch et al. 2006) and because of the scarcity of relatively least-disturbed reaches needed to set reference expectations (Herlihy et al. 2008, Stoddard et al. 2008).

Many agencies responsible for assessing river health and management efforts desire a macroinver-tebrate-based assessment tool as a companion to fish-based methods. Our paper presents an IBI designed to assess the quality of macroinvertebrate assemblages in large, nonwadeable rivers. We started with the development of an IBI from an extensive database of standardized macroinvertebrate assemblage samples collected from a diverse set of river reaches that also represented different types and levels of anthro-pogenic stress. We evaluated the environmental condition of each reach with a semiquantitative procedure that incorporated land-cover, hydrologic, and water-chemistry measures. The variety of river types and stressors of different kinds and intensities acting on each river type made it possible to develop a robust and widely applicable IBI. We selected and scored macroinvertebrate metrics that compose the index to represent the structure, composition, and functional niches of large river macroinvertebrate assemblages. We validated accuracy of the IBI with independent ecological-condition and macroinverte-brate data collected from other river reaches not used in the development process. Last, we applied the IBI to the entire database to characterize index response to various stressors along the gradient of intensities.

Our paper offers several contributions to aquatic biomonitoring. In it, we discuss macroinvertebrate methods and an IBI-development process applicable to other regions. We introduce refinements of several macroinvertebrate metrics that may be useful in both large river and wadeable stream assessments. Our comparisons between the macroinvertebrate IBI with an existing fish IBI illustrate the value of bioassess-ment with several kinds of taxa to offer a more complete depiction of river health than assessment with only one metric. Last, we present applications that we envision for the IBI.

Methods

The study area covered Wisconsin, including rivers bordering the states of Iowa, Michigan, and Minne-sota (Fig. 1). Our definition of a nonwadeable river for management and assessment purposes is that it has §3 km of continuous channel too deep (.1.5 m) to sample fish effectively by wading during summer base flow. Wisconsin has §40 nonwadeable rivers with a combined length of .2500 km as riverine and 1500 km as impounded reservoir habitat (Lyons et al.

2001, Weigel et al. 2006). Watershed area for these rivers ranged from 480 to 206,414 km2 (median =

5856 km2). During the summers of 2003 through 2006, we collected macroinvertebrate samples from 100 sites on 38 nonwadeable rivers that spanned 4 Level III ecoregions of the upper Midwest (Omernik 1987;

Fig. 1).

Our 6-step approach to IBI development, valida-tion, and application largely followed established methods (Barbour et al. 1995, Hughes et al. 1998, Karr and Chu 1999, Stoddard et al. 2008). First, we identified and evaluated field sampling and labora-tory methods. Second, we classified river reaches from existing environmental data and selected sites to span natural and stressor gradients. For natural

666 B. M. WEIGEL AND J. J. DIMICK [Volume 30

FIG. 1. Map of the upper midwestern region of the US showing nonwadeable rivers and sites used to develop and test the macroinvertebrate index of biotic integrity (IBI). Sites were numbered starting at the river mouth. Each dot on the Mississippi River represents 2 sites. N. = northern, S.E. = southeastern, IA = Iowa, IL = Illinois, IN = Indiana, MI = Michigan, MN =

Minnesota, WI = Wisconsin.

2011] MACROINVERTEBRATE IBI FOR WISCONSIN RIVERS 667

features, we focused upon geographic location, river size, and dominant substrate. To estimate environ-mental stress, we targeted reaches with various levels of disturbance caused by fragmentation from dams, modified flow, nonpoint-source pollution, or multiple stressors including historical point-source pollution.

For contrast, we selected least-disturbed reaches.

Once sampling sites were selected based upon initial classification, we collected new water-chemistry and watershed data to quantify environmental distur-bance. Third, we used standardized methods to collect and process new macroinvertebrate samples.

Fourth, we used our macroinvertebrate assemblage and environmental data to evaluate potential metrics and develop an IBI. Least-disturbed sites established the benchmark for high-quality macroinvertebrate assemblages and illustrated how natural factors influence assemblage metrics. We evaluated macro-invertebrate-metric data with an independent assess-ment of environmental condition, quantifying metric range and sensitivity to degradation. We then selected the most informative complement of metrics, devel-oped metric scoring criteria, and completed our IBI.

Fifth, we tested the IBI with an independent set of sites not used in the development process. Last, we related macroinvertebrate assemblage and IBI data to environmental condition to interpret biological re-sponses to various stressors.

Macroinvertebrate sampling method

We selected modified Hester–Dendy (H–D) artifi-cial substrate samplers as macroinvertebrate collec-tion devices because they are uniformly applicable in a wide variety of rivers, including habitats where other methods will not work (Ohio EPA 1987).

Sampler construction and deployment were based upon Ohio EPA (1987) protocols. Each sampler consisted of an eyebolt that held eight 7.6 3 7.6-cm plates made of 3.2-mm-thick Masonite hardboard.

Spacing between the plates was 3.2 mm between each of the first 3 plates, 6.4 mm between each of the next 3 plates, and 9.6 mm between the last 2 plates. We fastened 3 samplers to an 18-kg cinder block and either set it directly on rocky substrate or suspended it from a snag to maintain 0.75 to 1.5 m of water above the sampler at low flow. Sampler placement was consistent with the recommended minimum velocity of 0.09 m/s (Ohio EPA 1987). Macroinvertebrates colonized the samplers for 6 wk starting in mid-June.

After 6 wk, we retrieved the samplers, scraped off the organisms, combined the sample contents, and preserved them in ethanol. Laboratory processing incorporated a randomized grid-pan subsampling procedure that targeted 500 individuals (Hilsenhoff 1987) and a large–rare search for up to 10 min to include uncommon taxa (Courtemanch 1996, Vinson and Hawkins 1996). We identified all subsampled individuals to the lowest practical taxonomic level, usually species.

Environmental condition assessments

We classified rivers based upon natural and stressor conditions and, ultimately, had strong geo-graphic representation by sampling nearly every river in the state. We did not select sites at random. Rather, we chose sites deliberately to characterize the types of rivers on the landscape and the kinds and intensities of stress upon each river type. We especially targeted the least-disturbed and most severely disturbed river reaches to incorporate the greatest contrast of condi-tions. All sites were in warmwater reaches. We included only riverine reaches, but dams modified the flow regime at some locations. Multiple sites on a river or within the same drainage were not complete-ly independent. However, we maximized indepen-dence to the extent possible by targeting sites that were separated by dams. Sites not separated by dams within 1 drainage were included only if they had very different habitats (e.g., gradient, substrate, size) or stressors.

We developed an environmental disturbance score for each site based on in-stream nutrient concentra-tions, flow regulation, land cover, and historical point-source pollution (Table 1). At the time of macroinvertebrate sampler retrieval, we collected a water sample, preserved it, and had it analyzed according to standard protocols for total P (TP) and total N (TN; Wisconsin State Laboratory of Hygiene, Madison, Wisconsin). Nutrient concentrations for the environmental disturbance scores were based upon earlier work that identified an average concentration breakpoint above which fish and macroinvertebrate assemblages were consistently impaired (Weigel and Robertson 2007). Disturbance scores for TP and TN reflect doubling and tripling the breakpoint concen-trations. Land-cover proportions were measured at the watershed scale using geographic information system (GIS; Brenden et al. 2006) and Landsat thematic mapper data (Vogelmann et al. 1998, WDNR 1998). Agriculture and urban disturbance scores were roughly based on biotic and land-cover relationships in wadeable streams (Wang et al. 2003, Weigel 2003).

As reflected in the urban land-cover scoring in Table 1, the studies showed that good stream condi-tions were consistent with urban land cover ,5%, whereas fair to very poor stream conditions could

668 B. M. WEIGEL AND J. J. DIMICK [Volume 30 occur when urbanization §5%. Disturbance scores for flow regulation and historical point-source pollu-tion problems follow environmental classifications from a fish-based IBI for rivers of Wisconsin (Lyons et al. 2001). A composite score of the 6 environmental disturbance measures was calculated for each site, and an approximate trisection of the data yielded 32 sites classified as least-disturbed (score ƒ 2), 37 sites as moderately disturbed (score 3–6), and 31 sites as severely disturbed (score § 6). Site scores ranged from 0 to 10, but the possible range was 0 (lowest disturbance) to 14 (highest disturbance). The sites spanned a gradient of environmental disturbance as indicated by TP (0.02–0.50 mg/L) and TN (0.27–

5.72 mg/L) concentrations and % watershed area in agricultural (0–74) and urban (0–15) land cover. These environmental disturbance assignments were a mea-sure of environmental condition that was indepen-dent of the biota for purposes of IBI development and validation.

Macroinvertebrate metrics

After generating macroinvertebrate samples with the standardized methods, we calculated assemblage measures characterizing richness, tolerance, and ecol-ogy (Table 2). We included 47 derivations of the 14 basic metrics, e.g., enumerations by individuals and taxa and percentages by individuals and taxa. Several levels of dominance and tolerance were evaluated, including 10 combinations of the intolerant Ephemer-optera, Plecoptera, Trichoptera (EPT) metric by % individuals and taxa with maximum pollution toler-ance values (TV) ranging from 0 to 4. Likewise, we investigated % tolerant Chironomidae individuals, with minimum pollution tolerance values ranging from 7 to 10. The Hilsenhoff Biotic Index (HBI;

Hilsenhoff 1987), HBI10 (Hilsenhoff 1998), and mean pollution tolerance value (MPTV; Lillie and Schlesser 1994) represented the stress response of the assemblage to organic pollution. Pollution tolerance values were on a 0 to 10 scale, with 0 being the most intolerant of organic pollution. For aquatic insects, we tallied the unique functional trait niches (FTN) per sample as a community-level endpoint that represented 20 species traits (in 59 trait states) in life-history, morphological, mobility, and ecological categories (Poff et al. 2006). We also simplified the FTN tally to include only the 4 ecological traits (rheophily, thermal preference, habi-tat, and trophic status) in 17 trait states. In theory, high FTN equates to diversity and ultimately corresponds with environmental condition.

IBI development

For data analysis, we stratified sites by disturbance class (least, moderate, severe). Sites within each class had an equal probability of being assigned to the development or validation data set. This process established 2 independent data sets consisting of 75 sites for IBI development and 25 sites for validation.

Least-disturbed sites helped determine if natural environmental conditions confounded interpretations of stressor and macroinvertebrate relations. We used the 32 least-disturbed sites to evaluate the variation in metric scores in relation to river size, geographic location, and substrate. Metric ranges can change according to size of the water body (e.g., DeShon 1995), and at least in wadeable Wisconsin streams, by geographic location (e.g., Weigel 2003). Metrics appropriate for inclusion in an IBI should be robust to differences in natural conditions among sites, or statistical relationships should exist to adjust metric scoring for these conditions (Karr and Chu 1999). We used basin area upstream of the sampling site as our measure of river size, latitude to test the presence of a natural north–south gradient, and % fines (sand or smaller) from visual estimates at the sampler location as the measure of substrate. We started with biplots between each candidate metric and the 3 natural

TABLE 1. Environmental disturbance measures used to classify sites as least-disturbed, moderately disturbed, and severely disturbed. The sum of scores (0–3) across the 6 measures became the site environmental disturbance score. TP = total P, TN = total N.

Measure of environmental disturbance Good (0) Fair (1) Poor (2) Very poor (3)

TP (mg/L) ƒ0.100 0.101–0.219 0.220–0.329 §0.330 TN (mg/L) ƒ0.957 0.958–1.913 1.914–2.874 §2.875 Flow regulation No dam ,30 km upstream Dam but no hydro-peaking Irregular hydro-peaking Hydro-peaking 23/d

Agricultural land cover ,10% 10–32% 33–65% .65% Urban land cover ,5% §5% Historical point-source pollution ,30 km of site No Yes

2011] MACROINVERTEBRATE IBI FOR WISCONSIN RIVERS 669

environmental factors to visualize any tendencies for metrics to correspond with natural conditions. We followed this inspection with formal statistical tests based on multiple linear regression with stepwise selection and Bonferroni correction for multiple comparisons (SAS, version 9.1; SAS Institute, Cary, North Carolina) with basin area, latitude, and % fines as factors. Variables were tested for normality before analyses and, if appropriate, were transformed to approximate normality (![x]-transformation for metric counts, arcsine![x]-transformation for proportional metrics, and loge[x] for basin area). Only % tolerant Chironomidae individuals with tolerance values = 9 (TolChir9) and 10 (TolChir10) were significantly (p ƒ

0.001) affected by a natural factor (latitude). For all other metrics, interpretation of the biplots did not reveal any consistent trends between metrics and natural factors, meaning that adjustments to metric scores for natural factors would not improve the response of the metric to disturbance.

We used the entire development data set to investigate metric correspondence along the gradient of environmental conditions and to select metrics for IBI inclusion. Appropriate metrics for an IBI had a linear relation with the environmental-disturbance score, in which optimal scores corresponded with least-disturbed conditions, worst scores corresponded with severely disturbed conditions, and intermediate scores corresponded with moderately disturbed con-ditions. The most informative metrics tended to have few sites with a metric value = 0 and a large range in metric values across a condition gradient. We evalu-ated each candidate metric with linear correlation and biplots against environmental disturbance scores. We further assessed ambiguity between metrics and environmental condition by creating box plots of metric values by the 3 environmental disturbance classes derived from 6 disturbance measures (Table 1).

Metrics with little overlap of interquartile ranges between disturbance classes showed a strong and clear relationship between the metric and environmental condition. We selected metrics showing the strongest relationship with environmental disturbance that represented assemblage richness, tolerance, trophic structure, and other ecological attributes (Karr and Chu 1999).

Metrics must to be standardized or scored to a common scale to create a multimetric index, and we evaluated the performance of 2 common approaches.

We calculated continuous scores with the methods of Blocksom and Johnson (2009) using 95th and 5th percentiles for metric ceiling and floor values, and then converted the range to 0 to 10. We also derived discontinuous scoring criteria for final IBI metrics

TABLE 2. Metrics considered for inclusion in the Wisconsin large-river macroinvertebrate index of biotic integrity (IBI).

Metric Description

Taxon richness and composition

Taxon richness (TR) Number of taxa Insect Number of individuals (I) and taxa (T), and proportion of individuals (%I) and taxa (%T) that are class Insecta Ephemeroptera, Plecoptera, Trichoptera (EPT) I, T, %I, and %T that are mayflies, stoneflies, and caddisflies Chironomidae (Chir) %I and %T that are midges Dominance (Domx) I and %I of the most dominant taxa where x denotes the number of dominant taxa included (x = 1…3) Shannon’s Diversity Index (Div) Assemblage richness and evenness: 2Sriln(ri) where ri = proportion of individuals represented by each taxon

Tolerance and composition

Hilsenhoff Biotic Index (HBI) Organic pollution tolerance: Snxvx/Snx where nx = number of individuals in taxon x and vx = tolerance value of taxon x

HBI10 Modified HBI where nx is limited to 10 Mean Pollution Tolerance Value (MPTV) Modified HBI: Svx/R where R = number taxa with a tolerance value Intolerant EPT (IntolEPTx) %I and %T that are intolerant EPT taxa where x = maximum tolerance value 0–4 Tolerant Chironomidae (TolChirx) %I that are tolerant Chironomidae taxa where x = minimum tolerance value 7–10

Ecology

Functional trait niches (FTN) Number of unique combinations of 20 insect species traits (in 59 trait states) Ecology FTN (EcoFTN) Number of unique combinations of the 4 ecology traits (rheophily, thermal preference, habitat, and trophic status) Trophic status %I and %T that are filterers (Fil), gatherers (Gath), shredders (Shr), scrapers (Scr), and predators (Pred)

670 B. M. WEIGEL AND J. J. DIMICK [Volume 30 based upon their frequency distributions for the environmental disturbance classes. Approximately the median metric value of the least-disturbed class set the scoring-criteria maximum (10 points), whereas the median metric value of the severely disturbed class set the scoring-criteria minimum (0 points).

Metric values between the maximum and minimum criteria received an intermediate score of 5 points. The sum of metric scores constituted the overall IBI score.

We calculated IBI scores based upon each scoring approach and found that the IBIs were highly correlated (Spearman r = 0.97) for both development and validation data sets. Compared to the discontin-uous approach, the continuous-scoring approach compressed the range by 10 points among the development sites and 21 points among the validation sites, so we only used the discontinuous-scoring approach thereafter.

IBI validation

Correlation analyses on the 25-site validation data set revealed how strongly the IBI was related to environmental disturbance on a new set of sites. To assess differences among environmental disturbance classes, we categorized sites according to Table 1 and used an analysis of variance (ANOVA) with a Duncan multiple-comparisons test (DMC). We used distur-bance class as the main effect and IBI score as the response variable. The IBI was considered valid if IBI and environmental disturbance scores corresponded significantly and if differences among the classes were such that the least-disturbed systems had the highest scores and the severely disturbed systems had the lowest.

Macroinvertebrate and fish IBI responses to stressors

We evaluated the IBI as an assessment tool by pooling the development and validation data sets and characterizing the relative influence of various envi-ronmental stressors on IBI scores. Summary statistics of the environmental condition variables by class indicated stressor and IBI relationships. Correlation analyses showed how strongly the IBI was related to environmental-disturbance scores. We also correlated and plotted macroinvertebrate IBI vs fish IBI (fish data were from Lyons et al. 2001). The fish data were collected during 1996–1998 from sites within ,5 km of the macroinvertebrate sites. We anticipated that strong correspondence between the IBIs would confirm assessment accuracy. Alternatively, if large differences existed between the IBIs, site-specific investigations could elucidate biotic responses to various conditions or stressors. Last, we used patterns of taxa in relation to sites of known disturbance levels to aid in understanding biological responses to various stressors.

Results

The study rivers had a wide variety of macroinver-tebrate taxa. From all samples combined, we identi-fied 51,105 individuals representing 261 taxa. Chir-onomidae (Diptera) composed 42% of the individuals and 37% of the taxa overall. Single samples yielded 84 to 919 individuals that represented 10 to 57 taxa.

Summarizing the most abundant 3 taxa per sample illustrated the prevalence of certain groups. Taxa of the subfamily Chironominae were among the top 3 most abundant taxa in 66 (of 100) samples, with Polypedilum sp. (tribe Chironomini; TV = 5–7) among the top 3 in 23 samples, Glyptotendipes sp. (tribe Chironomini; TV = 10) in 22 samples, and Rheotany-tarsus sp. (tribe Tanytarsini; TV = 6) in 20 samples.

Hydropsychidae (Trichoptera) were among the top 3 most abundant taxa in 53 samples, with Cheumato-psyche sp. (TV = 5) among the top 3 in 43 samples, and Hydropsyche sp. (TV = 2–7) in 17 samples. The next most prevalent insect taxon was Maccaffertium sp.

(Heptageniidae:Ephemeroptera; TV = 1–5), which was among the top 3 in 18 samples. Most taxa were members of 1 of 9 orders in Class Insecta, but we also found 34 taxa from 16 invertebrate orders that were not insects. Tricladida (no TV assigned) were 1 of the top 3 in 20 samples.

Macroinvertebrate metrics and IBI development

Based upon analyses of all 75 sites in the develop-ment data set, 39 of the 47 metric derivations were significantly related to the environmental-disturbance score, making them candidates for inclusion in the IBI.

We wanted to keep 3 to 4 metrics from each of the 3 general metric categories in Table 2 and to create an IBI composed of 10 metrics that collectively repre-sented the number of individuals and taxa. We retained 4 metrics from the taxon richness and composition category (Table 2) that had the strongest correlation with disturbance: number of insect taxa (Insect-T; r = 20.55), % insect individuals (Insect-%I;

r = 20.44), number of EPT taxa (EPT-T; r = 20.44), and % individuals in the top 3 taxa (Dom3-%I; r =

0.58). Shannon’s Diversity Index was strongly corre-lated with disturbance (r = 20.53), but we excluded it because it was redundant with Dom3-%I (r = 20.95).

The 2 Insect metrics retained were not strongly redundant (r = 20.37). We retained 3 dissimilar metrics from the tolerance and composition category that were correlated with disturbance: mean pollution

2011] MACROINVERTEBRATE IBI FOR WISCONSIN RIVERS 671

tolerance value (MPTV; r = 0.54), % intolerant EPT individuals with maximum tolerance value = 2 (IntolEPT2-%I; r = 20.46), and % tolerant chironomid individuals with minimum tolerance value = 8 (TolChir8-%I; r = 0.23). From the ecology category, we retained number of unique combinations of the 4 functional trait niches (rheophily, thermal preference, habitat, and trophic status; EcoFTN; r = 20.49) instead of the all-inclusive FTN (r = 20.45). We also kept trophic measures that were correlated with disturbance: % gathering insects (Gath-%I; r = 0.28) and % scraper insects (Scr-%I; r = 20.38).

Scoring criteria for the final 10 metrics are present-ed in Table 3. The sum of the 10 metric scores composed the overall IBI score which ranged from 0 (worst) to 100 (best). Most state environmental agencies use 4 or 5 qualitative categories for general rating purpose, so we assigned qualitative ratings at 20-point increments from very poor (ƒ19), poor (20– 39), fair (40–59), good (60–79), to excellent (§80), and scored each site.

IBI validation

The 25-site test data set represented each environ-mental disturbance class and had IBI scores that ranged from 5 (very poor) to 90 (excellent; Fig. 2A).

Based on ANOVA and DMC, mean scores differed significantly between environmental-disturbance classes (R2

= 0.48, F = 9.94, p , 0.001). The least-disturbed class was different from the moderately and severely disturbed classes, whereas the moderately and severely disturbed classes were not statistically different. Inclusive of all test sites, Pearson’s correla-tion showed a statistically significant relationship between IBI and environmental disturbance scores (r = 20.60, p = 0.002).

Macroinvertebrate and fish IBI stress responses

Box plots of all sites summarized the distribution of IBI scores within each environmental-disturbance class (Fig. 2B). Average nutrient concentrations and % agricultural land cover approximately doubled between the environmental-disturbance classes. A biplot of all sites showed correspondence between

TABLE 3. Final metrics and scoring criteria. Abbreviations as in Table 2.

Metric

Scoring criteria and rating (points)

Poor (0) Fair (5) Good (10)

Insect-T 0–21 22–31 .31 Insect-%I 0–89% 90–95% .95%

EPT-T 0–6 7–15 .15

Dom3-%I .66% 41–66% 0–40%

MPTV .6.440 5.876–6.440 0.000–5.875

IntolEPT2-%I 0% 0.1–3% .3% TolChir8-%I .16.0% 2.5–16.0% 0.0–2.4% EcoFTN 0–8 9–12 .12 Gath-%I .54% 16–54% 0–15% Scr-%I 0.0% 0.1–7.4% .7.4%

FIG. 2. Box plots of index of biotic integrity (IBI) scores by environmental disturbance classes for the validation data set with sites representing poor (n = 8), fair (n = 9), and good (n = 8) conditions (A) and for the combined development and validation data sets with sites represent-ing poor (n = 31), fair (n = 37), and good (n = 32) conditions (B). Thick lines in boxes are medians, ends of boxes show quartiles, whiskers show ranges, and open circles indicate outliers.

672 B. M. WEIGEL AND J. J. DIMICK [Volume 30

IBI and environmental-disturbance scores (Pearson’s r = 20.59, p , 0.001; Fig. 3A). Two situations appeared to cause large discrepancies between IBI and envi-ronmental-disturbance scores. Relative to their envi-ronmental-disturbance scores, rivers isolated in the Driftless Area ecoregion had high IBI scores, whereas sites downstream of eutrophic reservoirs had low IBI scores.

A biplot between macroinvertebrate IBI and fish IBI scores showed general correspondence of the 2 indices (Pearson’s r = 0.48, p , 0.001; Fig. 3B). The line of best fit indicated that macroinvertebrate IBI tended to be lower than fish IBI scores but that 75% of the sites had macroinvertebrate IBI and fish IBI scores within 1 or 2 assessment classes (20–40 points). Sites in the Driftless Area or in reaches regulated for hydroelectric power had high macroinvertebrate IBI relative to fish IBI, whereas sites with diverse fish habitat but high nutrient enrichment had low macro-invertebrate IBI relative to fish IBI.

Discussion

Our study was a first step towards an efficient and effective macroinvertebrate-based tool for assessing the ecological condition of a wide array of nonwade-able rivers. Developed from the full range of rivers and conditions statewide, the IBI is applicable to nonwade-able rivers of Wisconsin, its border rivers with Iowa, Michigan, and Minnesota, and possibly additional rivers of the upper Midwest within similar ecoregions.

We envision this IBI as part of a monitoring and assessment program to fulfill Clean Water Act report-ing, implementation, and evaluation goals.

Macroinvertebrate sampling method

We determined that H–D artificial substrates provided an adequate macroinvertebrate sample for river bioassessment. A common criticism of coloniza-tion samplers is that they require 2 site visits, 1 for installation and 1 for retrieval (Flotemersch et al.

2006). Typically, we spent 9 d afield to install and retrieve ,42 samplers so, on average, a 2-person crew sampled 2.3 sites/d, which is comparable to a transect method used in Michigan to sample 2 sites/d (Merritt et al. 2005). Sampling time in the transect method increases with river size, and M of the samples in our study were taken from rivers larger than the largest river in the Michigan study (Wessell et al. 2008).

Efficiency in the field appears similar for the transect and H–D methods in small rivers, but H–D methods become increasingly efficient when larger rivers are sampled. The greatest disadvantage of colonization samplers occurs when samplers are lost, vandalized, FIG. 3. Scatterplots for index of biotic integrity (IBI) vs environmental disturbance scores (Pearson’s r = 20.59) (A) and macroinvertebrate IBI vs fish IBI scores (Pearson’s r =

0.48) (B) at all 100 sites. Small dots represent a single site at the plotted value, medium dots represent 2 sites, and large dots represent 3 sites. Both macroinvertebrate and fish IBI categories were at 20-point increments (range for excellent = 80–100). The line of best fit is displayed on each panel. For panel A, environmental disturbance scores predicted least-disturbed (0–2), moderately disturbed (3–6), and severely disturbed conditions (7–10). Ellipses identify sites within Driftless Area ecoregion (a), and sites downstream of eutrophic reservoirs (b). For panel B, ellipses identify sites regulated for hydroelectric power or within Driftless Area ecoregion (a) and sites with diverse fish habitat but high nutrient enrichment (b).

2011] MACROINVERTEBRATE IBI FOR WISCONSIN RIVERS 673

inundated with fine substrate, or dried because of low water levels. Unusable samplers were uncommon in our study (8% of those installed), but their loss would have been disruptive in a paired experimental design, for example. Depending upon data needs, installation of backup samplers might be warranted to help ensure data are acquired. The greatest advantage of H–D samplers was that they provided a uniform technique equally applicable among a wide array of inherently different rivers. From bouldery rapids to u-shaped channels and braided channels of shifting fines, H–D samplers collected adequate numbers of individuals representing macroinvertebrate assem-blages consistent with independent measures of environmental condition.

Laboratory procedures that targeted 500 individu-als and included a large–rare subsample maintained a reasonable workload while retaining sufficient infor-mation for assemblage analyses. Our methods yielded substantial assemblage information (261 taxa from 100 samples on 38 rivers). In comparison, Angradi et al.

(2009a) took ,160 snag and benthic kick-net samples each from the Ohio, Mississippi, and Missouri rivers (,1000 total samples) and found 142–171 taxa in snag samples and 214–308 taxa in benthic samples per river. Despite many differences in scope and scale between the studies, the comparison suggests that our methods collected a substantial assemblage for bioassessment.

Our choice of sampling method was intended to gather a collection of macroinvertebrates that reflected disturbance independent of habitat rather than to collect a complete list of taxa. Our use of H–D samplers and the targeted subsampling methods probably represented drifting and colonizing macroinverte-brates better than the entire assemblage. The goal of collecting an exhaustive taxon list would drain agency resources and preclude the deployment of a rapid bioassessment program over a broad spatial scale.

Environmental-condition assessments

The independent assessment of environmental con-dition was useful for rough site classification, but in our judgment, large discrepancies between IBI and environmental disturbance usually reflected an inac-curate disturbance score rather than an inaccurate IBI score. Disturbance measures were mostly representa-tive of conditions over broad spatial scales because they incorporated several stressors within the water-shed. The assessment, meant to be rapid and inexpen-sive, probably overlooked some local-scale stressors that were important (e.g., riparian modifications or removal of woody debris). Furthermore, the single TP and TN sample per site was a gross estimate of nutrients, was unable to characterize seasonal changes or event pulses, and probably underestimated distur-bance severity at sites downstream from polluted reservoirs. Despite shortfalls of this assessment ap-proach, our analyses showed that a relationship existed between macroinvertebrates and environmental dis-turbance. Several of the environmental measures used in our study have been combined in various ways by others to yield a composite environmental disturbance score for the development of macroinvertebrate indices (e.g., Weigel et al. 2002, Weigel 2003, Wessell et al. 2008, Angradi et al. 2009b , Blocksom and Johnson 2009).

Macroinvertebrate metrics

We considered a wide range of metrics that represented richness, tolerance to pollution, and ecology of macroinvertebrates inhabiting nonwade-able rivers. The choice of final metrics reflected a balance between different types of metrics and different measures of assemblage characteristics. To represent multiple levels of biological hierarchy as recommended by Karr and Chu (1999), we incorporat-ed measures of richness (Insect-T, Insect-%I, EPT-T, Dom3-%I), tolerance (IntolEPT2-%I, MPTV, TolChir8- %I), trophic structure (Gath-%I, Scr-%I), and ecological attributes that reflect multiple dimensions of biological systems (EcoFTN).

Most metrics had been used previously for bio-monitoring or incorporated in stream and river IBIs elsewhere, but we modified some metrics to reflect what we knew of large-river macroinvertebrate assemblages in Wisconsin. Simple metrics of taxon richness, dominance, EPT, and Chironomidae re-spond ambiguously to degradation and needed refinement. The challenge with taxon richness was to maximize retention of useful information from the sample without compromising another metric. For example, if the exotic zebra mussel (Dreissena poly-morpha) dominated a sample, it was important to include that information in a dominance metric, but to exclude it from a richness metric. As a result, Insect-T and Insect-%I corresponded with environmental condition better than simple taxon richness. The EPT metrics have a long history of corresponding with least-disturbed, healthy streams and rivers (e.g., Lenat and Penrose 1996), but rather tolerant EPT taxa (TV =

5–7) dominated some samples from moderately to severely disturbed rivers. In other words, sites with abundant, moderately tolerant EPT taxa scored the same as sites with abundant intolerant EPT taxa in all-inclusive EPT metrics, but the EPT-%I metric limited to intolerant taxa (TV ƒ 2) used more information

674 B. M. WEIGEL AND J. J. DIMICK [Volume 30 about the assemblage and corresponded better with environmental condition. Likewise, chironomids span a large range of tolerance values, e.g., ranges of 4 points among the species of Polypedilum known to occur in Wisconsin and 9 points among species in the Tanytarsini tribe. Limiting the TolChir-%I metric to TV § 8 reduced scores of samples with abundant tolerant midges but not of samples with intolerant to moderately tolerant midges and retained valuable taxonomic information for Insect-T and Insect-%I metrics. These same extremely tolerant midge taxa are strong predictors of complex toxicity in streams and rivers of Ohio (Yoder and Rankin 1995). Other similar derivations of taxon richness, dominance, and chironomid metrics have been tested for utility in other large river indices with mixed results (e.g., Applegate et al. 2007, Angradi et al. 2009b, Blocksom and Johnson 2009).

Previous investigators laid the foundation for using invertebrate functional traits in biomonitoring, espe-cially trophic functions (e.g., Cummins 1973), but more recently, investigators have found combinations of other functional traits valuable in predicting environmental stress (e.g., Townsend et al. 1997, Dolédec et al. 1999, Merritt et al. 2002, Statzner et al.

2005). This work implies that functional diversity may be better than taxon diversity at predicting distur-bance. Our metrics that summed unique FTN combi-nations within each sample (FTN and EcoFTN) were measures of functional diversity, intended to comple-ment richness or insect richness metrics. Poff et al.

(2006) concluded that the most informative traits in analyses of community response to disturbance were relatively independent (not linked evolutionarily) and susceptible to environmental selective forces. Traits only weakly correlated with other traits were most notably trophic and habitat traits and ecological traits as a group (rheophily, thermal, habitat, and trophic).

Assemblage-level measures of specialized traits should represent condition because disturbance re-duces ecological heterogeneity. We found that EcoFTN was better than FTN at predicting distur-bance, possibly because EcoFTN consists of traits that are evolutionarily labile. Limiting the study to colonizing macroinvertebrates in large rivers also may have contributed to low correlation because several nonecological trait states were unrepresented, resulting in 0-values for those traits.

IBI validation and macroinvertebrate and fish IBI stress responses

An important result was that analyses on the test data set based on environmental condition measures independent of the macroinvertebrates showed that the highest IBI scores were at the least-disturbed sites and the lowest IBI scores were at the severely disturbed sites, whereas sites with intermediate IBI scores were moderately disturbed. Given that the test data were not used in the IBI development phase, these results provide strong evidence that the IBI accurately reflects colonizing macroinvertebrates of large rivers and, by implication, overall ecosystem quality (Karr and Chu 1999). These results also provide support for the use of an IBI based upon colonizing macroinvertebrates for rapid bioassess-ment of large rivers.

Correlation and ANOVA showed that the IBI and environmental disturbance were significantly related, but a substantial proportion of variation remained unexplained. The performance of our IBI was very similar to comparable multimetric indices tailored to large rivers (e.g., Angradi et al. 2009b, Blocksom and Johnson 2009). We believe that these studies are testament to the difficulty of large-river biomonitor-ing. The lack of replicates and least-disturbed reaches under all combinations of conditions contribute to unexplained variation. For example, some taxa might be associated with specific ecoregions, but this type of relationship is very difficult to quantify without multiple reference sites and replicates for each river type and disturbance within each ecoregion. More-over, large rivers often flow through multiple eco-regions, so discerning the relative influence of each ecoregion is difficult. This issue further complicates using an ecoregion approach to classify large rivers.

Our study included some of the same rivers studied by other investigators. Thus, our findings also functioned as an independent validation of the IBI.

Blocksom and Johnson (2009) noted a general trend of decreasing environmental condition in the down-stream direction among randomly selected sites on the Wisconsin River. Our results agreed with theirs that the upper reach of the Wisconsin was generally in excellent condition, but we found that the middle reach was in very poor condition, whereas the lower reach mostly was in fair condition. A fish IBI suggested that the upper reach of the Wisconsin was in excellent condition, the middle was in very poor condition, and the lower reach was in excellent condition (Lyons 2005). Historical point-source pollu-tion has plagued the middle reach of the Wisconsin, and now the impoundments on the Wisconsin River in the lower O of the state suffer from excessive nutrients and are listed as 303(d) impaired waters (WDNR 2010). The Lower Wisconsin River is a 148-km free-flowing reach, rich in fish species and heterogeneous fish habitat. However, it has some of

2011] MACROINVERTEBRATE IBI FOR WISCONSIN RIVERS 675

the highest suspended chlorophyll levels among rivers statewide (Robertson et al. 2008). Here, perhaps the fish IBI reflected the extraordinary fish habitat, whereas the macroinvertebrate IBI corresponded to mediocre water quality (Fig. 3B). Authors of other macroinvertebrate studies on the St Croix River generally agreed that it was a least-disturbed system and that the downstream-most reach was relatively degraded (Boyle and Strand 2001, Blocksom and Johnson 2009).

As would be expected, most least-disturbed sites had the highest IBI scores and most severely disturbed sites had the lowest IBI scores, but some sites were exceptions. For example, the relatively small rivers limited to the Driftless Area ecoregion were evaluated as moderately to severely disturbed because of their highly agricultural watersheds and excessive nutrient concentrations, yet the macroinver-tebrate assemblages at these sites were comparable to those at relatively least-disturbed sites around the state (Fig. 3A). It is possible that the IBI is not particularly sensitive to moderate levels of nonpoint-source pollution in this area, and that the macroin-vertebrate assemblage could maintain a high level of integrity unless other human activities stress the river ecosystem. It is also possible that these Driftless Area rivers historically had exceptional macroinvertebrate fauna, far more diverse and intolerant than what is currently present, but no historical records or less-disturbed conditions exist for comparison. The abun-dance of cold-water tributary streams to rivers of the Driftless Area may be sources for a diverse and intolerant macroinvertebrate assemblage. Further in-vestigations should determine whether wadeable stream assessment methods are more appropriate for the Driftless Area rivers.

The macroinvertebrate assemblage appeared to be quite sensitive to semilacustrine reaches, and this sensitivity resulted in poor IBI scores indicative of severe impairment, even though some of these sites have good water chemistry and little other distur-bance. Several reaches became semilacustrine as they approached larger downstream water bodies, e.g., the most downstream sites on the St Croix and Wolf rivers where environmental disturbance scores indi-cated good conditions but the macroinvertebrate IBI was relatively low (Fig. 3A). The macroinvertebrate assemblage and consequent IBI appeared to under-score river reaches in which flow was naturally semilacustrine. However, our philosophy was that if a dam restricted flow then it was modified and probably impaired, which should be reflected in a low IBI score (e.g., Black 2, Mississippi River). The IBI scoring was tailored to riverine macrohabitats.

Macroinvertebrate IBI scores usually were consistent with scores produced with the fish IBI, but the outliers highlighted differences in macroinvertebrate and fish responses to environmental conditions or stressors.

Hydroelectric-peaking flow regimes caused fish-habitat degradation and were associated with low fish IBI relative to water quality (Lyons et al. 2001; Fig. 3B).