Northeastern Naturalist 532 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 22001166 NORTHEASTERN NATURALIST 2V3(o4l). :2533,2 N–5o5. 44 Validation of a Macroinvertebrate-based Index of Nutrient Status in Streams Using Macroinvertebrate, Water-chemistry, and Diatom Data Richard J. Horwitz1,*, Andrew Tuccillo1, Donald F. Charles1, Shane Neiffer1, and Thomas Belton1 Abstract - Determination of the causes of water-impairment is a critical part of bioassessment, and it is useful to be able to infer causes from the same sampling data used to assess the impairment. Determination of excess nutrient inputs to a waterbody as a cause of impairment is especially important because of the severity and ubiquity of nutrient-related water-quality problems nationwide. To that end, we tested and validated in New Jersey waters a macroinvertebrate-based nutrient biotic index (NBI) for phosphorus and nitrogen developed by the New York State Department of Environmental Conservation (NYSDEC). We used macroinvertebrate, water-chemistry, and diatom data from New Jersey streams collected in the state biomonitoring program and a study of diatom–nutrient relationships. We calculated tolerance values for widespread taxa based on frequency of occurrence in samples from sites with a range of nutrient concentrations. The NBI of a sample was calculated as a sum of the tolerance values of taxa in a sample weighted by the relative abundances of taxa. We developed tolerance values from the New Jersey data because relatively few taxa present in the New Jersey samples were rated in the New York study. NBIs for the New Jersey data calculated using the New Jersey-based tolerance values were significantly related to nutrient concentrations with correlations similar to, or greater than, those observed in the New York study. For taxa in common, the New Jersey-based tolerance values were only weakly correlated with the analogous New York values. To validate the NBI approach, we calculated NBI scores via a “leave-one-out” procedure for a data set not used to estimate tolerance values. These comparisons yielded statistically significant but weak correlations between the NBIs and nutrient concentrations. Factors that weaken these relationships are related to: (1) the specific data used (e.g., the lack of tolerance values for many taxa in independent data sets and weak temporal matching of macroinvertebrate and nutrient samples), (2) estimation issues (e.g., variability in estimates of tolerance values and NBIs), and (3) problems inherent in the approach (e.g., the effects of other factors on macroinvertebrate relationships). However, for all data sets examined, nitrogen and phosphorus concentrations were positively correlated, as were nitrogen and phosphorus tolerance values for taxa, and nitrogen and phosphorus NBI scores for sites. These correlations need to be considered in the selection of sampling sites for the development of tolerance values, the weighting of taxa in calculation of NBIs, and the interpretation of NBI values for the 2 nutrients. Introduction Bioassessment involves determination of both the level of water-body impairment and the causes of impairment. The determination of causes is essential for 1Academy of Natural Sciences of Drexel University, 1900 Benjamin Franklin Parkway, Philadelphia, PA 19103. 2Corresponding author - rjh78@drexel.edu. Manuscript Editor: Hunter Carrick Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 533 the development of effective plans for attaining water-quality standards and designated uses (Yoder and DeShon 2003). Excessive nutrient inputs are a widespread cause of eutrophic impairment, leading to the growth of nuisance levels of algae and decreased dissolved oxygen (DO) levels in nearby receiving waters as well as downstream rivers, lakes, and estuaries. However, problematic algal growth may occur where eutrophication is not as evident, (e.g., in streams with high rates of oxygenation). As a result, both nutrient and dissolved-oxygen criteria may be required to provide a full assessment for the protection of water quality. It would be efficient if data from existing sampling programs could be used both to quantify impairment and to distinguish causes of impairments, including excess nutrient inputs. For example, Yoder and Rankin (1995) used multi-metric indices to distinguish 9 different groups of anthropogenic activities that may lead to impairment. Excess nutrients were associated with many of these sources, such as conventional wastewater- treatment plants, combined-sewer overflows, nonpoint agriculture sources, and livestock access. Bryce and Hughes (2003) distinguished mining sources from agricultural sources using multi-metric fish, benthic macroinvertebrate, and diatom multi-metric indices. They noted that agricultural effects were associated with several stressors, such as sedimentation and nutrient enrichment. Other studies (e.g., Bae et al. 2011, Bedoya et al. 2011, Hering et al. 2006, Piliere et al. 2014) have correlated responses of indices to different stressors, including nutrients, but did not design specific indices for responses to groups of stressors. The New York State Department of Environmental Conservation (NYSDEC) developed nutrient biotic indices (NBIs) for evaluating nutrient conditions using data on benthic macroinvertebrate assemblages (Smith et al. 2007). These NBIs are directly linked to nutrient concentrations, rather than the activities that produce a variety of ecological impacts. In this paper, we test the precision and generality of the NBIs, using a methodology similar to Smith et al. (2007) on an independent macroinvertebrate and water-chemistry data set from New Jersey (NJ). The basic objectives of the study were to: (1) determine the generality of the NBI approach by applying it to data from NJ, (2) determine the robustness of tolerance values for macroinvertebrate species developed in New York (NY) by using them with the NJ data, and (3) validate the robustness of the NBI method by correlating NBIs to chemistry and inferred-chemistry data that are independent of the chemistry data used to calculate the NBIs. We calculated the NBIs as the average of the nutrient-tolerance values for taxa in each sample, weighted by the proportion of each taxon in the sample. We determined the nutrient-tolerance values for common macroinvertebrate taxa from modal values of nutrient concentrations where these taxa were collected. These NBIs are similar to inference models (ter Braak and Juggins 1993) in structure because they use observed nutrient-concentration–species-occurrence relationships to infer nutrient status from the sum of species occurrences in samples. However, the method does not directly estimate nutrients but defines an ordinal scale (the NBIs). The 2 NBIs correlate with increasing mean total phosphorus (TP) and nitrate + nitrite (NO3) values, respectively, and define a 3-tiered scale of nutrient Northeastern Naturalist 534 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 status (i.e., oligotrophic, mesotrophic, and eutrophic) using cluster analysis of invertebrate assemblage data. The NBIs appear to accurately reflect differences in stream trophic state. Aquatic macroinvertebrates have been used for freshwater monitoring and assessment for several decades by many agencies including the New Jersey Department of Environmental Protection (NJDEP) and its Ambient Macroinvertebrate Network (AMNET) program. AMNET has over 760 non-tidal stations statewide for evaluating the attainment of aquatic-life use criteria under the US Clean Water Act, based on metrics of overall impairment calculated from assemblage data (http:// www.nj.gov/dep/wms/bfbm/amnet.html). The NJDEP has also collected water samples at a number of sites where macroinvertebrate sampling was conducted. NJ shares 5 ecoregions with NY, 3 of which are contiguous. Therefore, NJ represents a favorable situation for testing the applicability of the NYSDEC macroinvertebrate NBIs. For this study, we followed Smith et al. (2007) and used the relationships between NBIs and water-chemistry measurements to test the ability of the NBIs to detect nutrient impairment. However, it is important to note that parameters other than nutrient concentration may be equally important for bioassessment, such as excessive algal growth, changes in algal (diatom) community structure, and/or oxygen depletion. Subsequent development of the NBIs could be strengthened by matching NBI scores to these significant endpoints. For example, Smith et al. (2013) used break-point analyses to determine the nutrient concentrations that define thresholds for changes in macroinvertebrate and diatom assemblages. In addition, the biological condition gradient (BCG) was used for New Jersey streams to link macroinvertebrates (NJDEP 2007) to changes in ecosystem structure and function; algal–diatom metrics (trophic diatom index) were developed and augmented by BCG to quantify nutrient effects on water quality (Charles et al. 2010, Hausmann et al. 2016) Methods Tasks and work flow The basic approach of this study was to validate the NY method, both in terms of statistical approach and the specific macroinvertebrate tolerance values. Specific tasks were to: 1. Compile existing water-chemistry data for NJ sites, and develop a database containing linked water-chemistry and AMNET macroinvertebrate data. 2. Obtain NY nutrient-tolerance values for macroinvertebrate taxa and calculate NO3 and TP NBIs for NJ sites based on the NY tolerance value. 3. Relate the NO3 and TP NBIs to water-chemistry data, to validate use of the NY method and tolerance values. 4. Develop new tolerance-values based on NJDEP data and use these to calculate new NBIs. 5. Relate the new NBIs to water-chemistry data. 6. Validate the new NBIs using a leave-one-out approach. Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 535 7. Compare the macroinvertebrate NBIs with the diatom-based nutrient indices developed for NJDEP by PCER to validate the NY method. 8. Compare tolerance values developed for NY and NJ as a measure of the robustness of the estimation procedure. The data flow for these tasks is outlined in Figure 1. Data compilation The NJDEP AMNET data were collected at each site from multiple, traveling Dframe kick samples within riffle and run habitats. The "jab and sweep" method was employed (Barbour et al. 1999); a minimum of 20 jabs/sweeps were conducted, proportioned approximately to the numbers of each habitat type present. In all cases, stream-distance sampled approached, but did not exceed, 100 m. Level of effort was consistent for all sites (NJDEP 2012). These data consisted of relative abundances Figure 1. Schematic of data sources and calculations used for the 5 methods of NBI validation. Rectangles represent data sources, hexagons denote calculated indices, and circles indicate statistical comparisons. Single-headed, solid arrows indicate the use of data sources in or the direction of index estimation. Double-headed, solid lines show entities compared in the statistical analyses. A dashed line shows analyses performed using leave-one-out validation. A single line around a circle indicates that the regression variables are not independent because the chemistry data were used to estimate the nutrient index. A double line around a circle indicates independence of the regression variables. The suffixes d, c, and t refer to assemblage data, water chemistry data, and tolerance values, respectively. Northeastern Naturalist 536 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 of macroinvertebrate taxa. The individuals from the sample were identified to the lowest practicable taxonomic level, usually genus or species: 60% of records were identified to species level, 10% to morphospecies, 37% to genus, and 2% (Corixidae, Sphaeriidae, Enchytraeidae, and Chironomidae) to a higher level. Two data sets were available for our study. The main data set contained raw data for macroinvertebrates from 97 sites (Fig. 2). These data have also been used to relate macroinvertebrate assemblages to land-use and fish data (Flinders et al. 2008). The NJDEP provided water-chemistry data from 1996 to 2007 from these sites. We excluded data from southern NJ from most analyses because of possible differences in taxa composition between these sites and those in northern NJ. Our second data set was provided by D.F. Charles and consisted of water-chemistry and diatom samples taken at the same time at 29 AMNET sites. We employed nutrient concentrations and diatom-inferred nutrient concentrations from the second data set for comparison with NBIs. Macroinvertebrate data from the second data set were not included in the analyses discussed in this paper, although we included them in additional validation procedures not reported here (R.J. Horwitz et al., unpubl. data). Methods used to measure individual water-chemistry parameters differed among datasets. For consistency, we included only TP, NO3, and total nitrogen (TN) data in our analyses. These parameters were measured using methods listed in Table 1. For Figure 2. Map of NJ and NY study sites and the ecoregions in which sites are located. Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 537 Table 1. Parameters and method names for nutrient analyses in NJDEP and NJAI data sets. Agency Source Agency characteristic method Parameter Occurrence agency name code Agency method name Nitrate + nitrite NJDEP measurements USGS Nitrite + nitrate, water, I-2545-90 Nitrogen, nitrite + nitrate, colorimetry, (main and validation data sets) filtered, mg/L as N cadmium reduction-diazotization, automated-segmented flow Nitrate + nitrite NJDEP measurements EPA Nitrogen, nitrite (NO2) + 353.2 Nitrate (as N) Automated diazotization (main and validation data sets) nitrate (NO3) as N w/o Cd reduction column Total phosphorus NJDEP measurements EPA Phosphorus as P 365.4 Total phosphorus after block digestion (main and validation data sets) Total phosphorus NJDEP measurements USGS Phosphorus, water, I-4610-91 Determination of total phosphorus by a (main and validation data sets) unfiltered, mg/L Kjeldahl digestion method and an automated colorimetric finish that includes dialysis Nitrate + nitrite NJDEP measurements USGS Nitrogen, nitrite + nitrate, I-2546-91 Nitrogen, nitrite + nitrate, low ionic-strength (validation data set only) dissolved water, colorimetry, cadmium reductiondiazotization, automated-segmented flow Total phosphorus PCER measurements EPA Phosphorus as P 365.2 Phosphorus by single-reagent colorimetry (main data set and NJAI study) Nitrate + nitrite PCER measurements EPA Nitrogen, nitrite (NO2) + 353.2 Nitrate-nitrite nitrogen by cadmium (main data set and NJAI study) nitrate (NO3) as N reduction Total Kjeldahl PCER measurements (NJAI only) EPA Total Kjeldahl nitrogen 351.2 Total Kjeldahl nitrogen by semi-automated nitrogen colorimetry Northeastern Naturalist 538 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 many sites, only N or P data were available. As a result, sample sizes for analyses involving each nutrient were smaller than the total number of sites in the database. We conducted 5 analyses to provide varied approaches to validation (Fig. 1): 1. We used NY tolerance values (from Smith et al. 2007) to calculate NBIs for the NJ data; regressions of these NBIs with the chemistry data from NJ provide a validation of the NY method. This was appropriate because data used to estimate the nutrient optima (i.e., the NY macroinvertebrate and chemistry data) were not used in the regression. A high correlation indicates the generality of both the NBI approach and the specific tolerance-values estimated by Smith et al. (2007); a low correlation could result from differences in tolerance of each taxon across the region, presence of many taxa in the NJ data for which NY did not estimate tolerance values, problems with taxonomic precision and accuracy, problems with relevance with the chemistry data, or lack of generality of the NBI approach. 2. We calculated NJ tolerance values using the NJ macroinvertebrate and chemistry data; the macroinvertebrate data and tolerance values were used to calculate NBIs; regressions of these NBIs and the NJ chemistry data are directly comparable to the evaluation technique used by Smith et al. (2007). However, the NBIs are not independent of the chemistry data, and thus do not provide an independent evaluation. The NBIs provide a test of the generality of the approach, but do not provide a true validation. 3. We used the NJ macroinvertebrate and chemistry data from the main data set to estimate tolerance values, macroinvertebrate data from the diatom sites and these tolerance values to calculate NBIs, and the regression of these NBIs along with the inferred-concentration indices derived from the diatom analyses to assess the consistency of the macroinvertebrate NBIs with the diatom-based nutrient indices. These tests provide a stronger test and validation of the NBI approach because they correlate NBIs with data independent of those used to estimate the NBIs, and they use inferred- chemistry data, which show fewer problems with relevance of the chemistry data to the macroinvertebrate assemblages. A low correlation suggests problems with the NBI approach, in terms of the basic approach or statistical estimation issues. 4. We used the NJ macroinvertebrate and chemistry data to estimate weighted- average nutrient concentrations for each taxon. The weighted-average nutrient concentrations were used to infer nutrient-concentration indices, which we then compared to the measured nutrient concentrations. We employed a leave-one-out procedure (see below) to conduct these calculations so that the inferred nutrient-indices were independent of the measured nutrient-concentrations. This analysis provides a validation of major elements of the NBI approach. The leave-one-out procedure efficiently uses the data, avoiding loss of precision in validating separate data sets. As with approach (3), a low correlation suggestions theoretical or statistical problems with the NBI approach. Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 539 5. We employed a procedure similar to approach (4), but we omitted the leave-one-out procedure. A comparison of approaches (4) and (5) indicates the importance of the non-independence of the inferred and measured chemistry values; if methods (4) and (5) produce similar values of correlation, the non-independence of the NBI and test data may not be a large problem. However, a much higher correlation for (4) than for (5) would suggest that the apparent success of the NBI approach may depend on the non-independence of the evaluation. Data matching and filtering Smith et al. (2007) used water-chemistry data collected within 90 days of the macroinvertebrate sampling. However, very few nutrient and macroinvertebrate samples from NJ sites were collected at the same or similar times. Therefore, we linked the water-chemistry data from the main data set to macroinvertebrate samples if the water samples were collected in the same season and ≤5 years of macroinvertebrate sample collection. We defined seasons as: spring = 1 March–31 May, summer = 1 June–31 August, fall = 1 September–31 November, and winter = 1 December–28/29February. At many sites, 1 macroinvertebrate sample corresponded to >1 nutrient measurement that met the temporal matching criteria. In these cases, we used the averages among all the eligible measurements of a nutrient. For the main database, averages of TP and NO3 were 2.65 and 4.15, respectively. We removed extremely high and low nutrient values prior to analyses using a qualitative technique whereby we viewed a distribution of the data points using JMP 7 software (SAS, Cary, NC), and identified points within the long tails at both ends for exclusion. After all the above criteria were applied, samples from 97 sites were available for analysis as part of the main database. We excluded taxa found in less than 2% of sites (Smith et al. 2007), resulting in a total of 254 invertebrate taxa. Calculation of nutrient optima for macroinvertebrate taxa and NBIs for samples We calculated the nutrient optima, NBIs, and NO3 and TP as follows (see Smith et al. 2007 for further details). We ordered the set of sample sites used to calculate optima by the appropriate nutrient concentration (i.e., NO3 or TP) and divided them into 15 bins. The nutrient ranges within bins were the same as those used by Smith et al. (2007). The distribution of NJ nutrient data differed somewhat from that of the NY dataset. As a result, the number (ni) of samples in each bin (bi.) differed somewhat, typically with more samples in the higher nutrient-concentration bins. We calculated the mean nutrient concentration of each bin (mi) and the frequency (fi) of each taxon (t) in each bin (i.e., the proportion of ni samples in which the taxon was present). The nutrient optima were calculated as: Ot = Σ(fi * mi) / Σ(fi). We ordered and divided into 11 bins nutrient optima with approximately equal Northeastern Naturalist 540 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 numbers of taxa in each bin. We defined the rank of each bin as the tolerance values (TVi) of each taxon in that bin. The TVis ranged from 0 to 10, with the lowest value corresponding to occurrence at low nutrient-concentrations. For each sample, we calculated a nutrient value (NBIT) from the sum of the tolerance values of taxa in the sample weighted by the proportion (pi) of each taxon in the sample: NBIT = Σ(pt * TVt). For samples in which tolerance values were not available for all taxa in the samples, we used the proportion of individuals among all rated individuals, instead of the proportion of individuals in the entire sample. Application of NY-based tolerance values to NJ data (method 1) The first application of the NBI strategy to NJ streams used the taxon-specific tolerance values for NY (Smith et al. 2007) to calculate NBI scores for NJ sites. Taxon coding differences between NY and NJ identifications were resolved by eliminating differences below the species level and by including “nr. taxon x”designations (i.e., specimen was either species x or a very similar species). However, even after this level of resolution, there was relatively low overlap between the NJ and NY taxonomic lists, so that NY tolerance values could be applied to only 69 of 254 NJ taxa (about 27.5%). We calculated NBI scores of NJ sites using the process described in Smith et al. (2007); the abundances of unrated NJ taxa were ignored by the calculations. We regressed the TP and NO3 NBI scores against log-transformed TP and NO3 values. Calculation of tolerance values using NJ data and of new NBIs (method 2) We calculated a second set of tolerance values for all NJ taxa, except rare taxa, using macroinvertebrate-abundance and nutrient data collected in NJ. We determined tolerance values using the same methods as Smith et al. (2007) and described above, using their 15 original nutrient bins to calculate nutrient optima in the first phase of the calculations. We regressed the resulting NBI scores against corresponding log-transformed nutrient values as described above. Macroinvertebrate NBIs and diatom-based nutrient indices (method 3) We conducted a regression of the TP- and NO3-based NBI scores from sites used to develop the diatom nutrient index with the diatom-inferred TP and TN values. This analysis compares nutrient status estimated from macroinvertebrates with that from diatoms. Multiple diatom samples were taken at the majority of sites, resulting in multiple inferred nutrient values. In these cases, the inferred values were averaged before comparison with NBI scores. Nutrient inference from macroinvertebrate assemblages and leave-one-out validation We used the macroinvertebrate data in the main data set to infer NO3 and TP concentrations by calculating nutrient optima for each taxon (excluding rare Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 541 taxa) and then inferring nutrient-concentration indices for each sample from the nutrient optima: INFi = Σ(fit * Ot) / Σ(fi) where fit is the proportion of taxon t in sample i, and Ot is the nutrient optimum for taxon t. These estimates are not scaled to truly estimate nutrient concentrations, but are directly related to such estimates. The main use of these indices is for correlation with measured concentrations; thus, scaling these indices to form a concentration estimate is not necessary. For simplicity, we refer to these indices as inferred nutrients in the remainder of the paper. We performed this procedure in 2 ways. One (method 4) included all data, so that the nutrient inferences for each sample were based on nutrient optima which were partly based on data from that sample. The other (method 3), a leave-one-out procedure, calculated nutrient optima over all samples except one, and then used these optima to infer nutrient concentrations in the excluded sample. We employed this procedure over all samples (excluding each sample in turn), so that each inferred-nutrient concentration was independent of the measured-nutrient data for that sample. We undertook several other types of validation with different data sets to estimate tolerance values and evaluate NBIs derived from these tolerance values, which produced results similar to those reported here (R.J. Horwitz et al., unpubl. data). Statistical analyses We regressed NBI scores and inferred nutrient-concentrations against log-transformed nutrient values. In some cases, we regressed residuals against factors which might affect the accuracy of the regressions. We performed all data manipulations with Excel and Access (Microsoft), and statistical analyses using JMP 7 and SigmaPlot (Systat Software, Inc.). Results Application of NY-based tolerance values to NJ data The relationships between the NBI scores calculated with NY-based tolerance values and log-transformed nutrient concentrations were positive and significant but very weak (Table 2: rows 1a, 1b). The poor relationships were at least partially due to the relatively low proportion of taxa (2–67%) assigned tolerance values in any given sample, which in turn stemmed from the relatively low overlap between the 2 taxonomic lists and the occurrence of some taxa with high abundance in single samples. A few taxa that were widespread and abundant in the NJ samples, e.g., Dugesia tigrina (Girard) (up to 60% relative abundance) and Neocleon (up to 31% relative abundance), were not reported in the NY samples. Other taxa were found in only a few NJ samples, but in high relative abundance (10–50%) in single samples. These taxa include Amphinemura delosa (Ricker), Diamesa nivoriunda (Fitch), Dubiraphia, Eurylophella temporalis (McDunnough), Isonychia arida (Say), Leuctra truncate Claassen, Nemoura, Palucidella articulata (Ehrenberg), Northeastern Naturalist 542 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 Table 2. Relationships between tolerance values, NBI values, and nutrient concentrations. TP inflo and N inflo are the inferred concentrations using the leave-one-out procedure; TP infall and N infall are the inferred concentrations using all data.. Source of data Regression results Macroinvertebrate Chemistry Method Index Tolerance values data data n R2 P R2 R2 logTP & log(NO3 NBI-P & NBI-P vs log(TP) or TN) NBI-N 1a NBI NYSDEC Main Main 68 0.19 less than 0.0002 0.38 0.58 2a NBI Main dataset Main Main 68 0.45 less than 0.0001 0.62 3a NBI Main dataset Diatom Diatom inferred 29 0.62 less than 0.0001 0.61 0.73 TP inflo vs TP inf vs P inf vs log(TP) TP infall NO3 inf 4a Inferred nutrient Main dataset Main Main 68 0.43 less than 0.00001 0.63 0.73 5a Inferred nutrient Main dataset (leave-one-out) Main Main 68 0.12 less than 0.0039 0.72 Nitrogen ([Nitrate + nitrite] or [Total nitrogen]) 1b NBI NYSDEC Main Main 97 0.15 less than 0.0001 2b NBI Main dataset Main Main 97 0.51 less than 0.0001 3b NBI Main dataset Diatom Diatom inferred 29 0.66 less than 0.0001 N inflo vs N inf vs log(NO3 or TN) N infall 4b Inferred nutrient Main dataset Main Main 97 0.47 less than 0.00001 0.75 5b Inferred nutrient Main dataset (leave-one-out) Main Main 97 0.18 less than 0.000014 Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 543 Paralauterborniella, Paratendipes albimanus (Meigen), Perlesta, Plumatella, 2 Prosimulium spp., 2 Pesudocleon spp., and 5 species of Symphitopsyche. Other differences reflect different levels of identification between the NJ and NY programs, particularly of non-insect taxa. For example, species of the amphipod genera Caecidotea and Gammarus, the gastropod genera Physella and Pisidium, and several annelids were reported by NJ, while NY reported only generic identifications of amphipods and gastropods and suborder of annelids. Conversely, NY reported species of Drunella, while NJ reported only a generic identification. Some other genera, such as Ephemerella, Microtendipes, Nais, Orthcladius, Polypedilum, Simulium, and Stenelmis, were represented in both states, but each with at least 1 species not found in the other state. Residuals from the NBI–nutrient regressions were weakly negatively related to the number of taxa assessed (TP: n = 68, R2 = 0.11, P < 0.0051; NO3: n = 97, R2 = 0.08, P < 0.0039). The lack of a strong temporal association between the chemistry and macroinvertebrate samples may have also played a role in the weakness of the correlation, but there were no significant relationships between the residuals of TP or NO3 and the inter-sample interval. Calculation of tolerance values using NJ data The regressions of the NBI scores calculated with NJ-based tolerance values against log-transformed nutrient concentrations were positive, significant, and moderately strong (Table 2: rows 2a, 2b); R2 values (0.45 and 0.51 for TP and NO3, respectively) were similar or higher than those reported by Smith et al. (2007): R2 = 0.46 for P-NBI and TP and 0.32 for N-NBI and NO3. For shared taxa, the NJ-based tolerance values were only weakly correlated with the NY-based tolerance values (TP: n = 69, R2 = 0.13, P < 0.0026; NO3: n = 69, R2 = 0.06, P < 0.048). Relationships between NBIs and nutrient concentrations inferred from diatoms NBI scores calculated from NJ-based tolerance values were correlated with diatom-based inferred TN and TP indices (Ponader et al. 2007; Fig. 3). The relationships were fairly strong despite relatively low sample numbers and poor temporal association between samples (Table 2: rows 3a, 3b). Nutrient optima: Estimation and validation of NJ-based tolerance values The relationships between inferred and measured nutrients using all data were highly significant (Table 2: rows 4a, 4b). The r2 values were slightly lower than those between the NBIs and measured nutrients (rows 1 vs 3). However, the inferred nutrient-concentrations estimated using the leave-one-out procedure were much less correlated with the observed nutrient concentrations (Fig. 3; Table 2: rows 5a, 5b). For TP, the slope of both inferred TP–measured TP relationships were much less than 1 (0.086 for inference with all data, and 0.037 for leave-one-out inference). The difference between the logs of the 2 methods of inferring TP increased with nutrient concentration (Fig. 4). At low TP concentrations, the inferred concentrations Northeastern Naturalist 544 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 Figure 3. Relationship between NBI scores from macroinvertebrates and nutrient concentrations inferred from diatoms. (A) Total phosphorus and (B) NO3. Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 545 Figure 4. Relationships between nutrient concentrations (mg/L) inferred from macroinvertebrate data using the leave-one-out procedure versus measured nutrient concentrations. (A) Total phosphorus and (B) NO3. Northeastern Naturalist 546 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 were slightly greater than the measured concentrations. At high TP-concentrations, the inferred concentrations were much less than the measured concentrations. For nitrate, the slope of inferred NO3 versus measured NO3 were also much less than 1 (0.20 for inference with all data and 0.10 for leave-one-out inference; Fig. 3). As with TP, the difference between the inferred NO3 from the 2 methods increased with nutrient concentration (Fig. 5), with large differences for a few points at each end of the NO3 gradient. Discussion The biomass and/or species composition of benthic macroinvertebrates have been related to nutrient enrichment, either as sources of enrichment or as other chemical properties related to enrichment. Spieles and Mitsch (2000) found that average diel dissolved oxygen and specific conductivity were the best environmental predictors of invertebrate-community metrics in high- and low-nutrient constructed wetlands; these variables, along with chemical oxygen demand and nitrate-nitrogen, described nearly 90% of the invertebrate-community index (ICI) variation in a 4-predictor regression model. Biggs et al. (2000) found that biomass and species composition of invertebrates responded to nutrient-enrichment experiments, but the response was mediated by the type of fish predator present. Camargo et al. (2004) found that a multimetric assessment of nutrient enrichment in impounded rivers may be a useful technique for the biological assessment of nutrient enrichment in fluvial ecosystems. More recently, Smith et al. (2007) demonstrated a relationship between macroinvertebrate assemblages (described by NBI scores) and nitrogen and phosphorus concentrations for NY streams. These NBIs were subsequently used by Smith et al. (2013) to determine change points in biological assemblages. Using analogous methods for calculating NBIs, we found similar relationships between NBI scores and chemical concentrations for NJ streams. This similarity in strength of NBI-chemical data between the NY and NJ analyses suggests that differences between the NY and NJ protocols (e.g., wider temporal variation between macroinvertebrate and chemistry samples for NJ) did not greatly affect the results. However, in both of these analyses, relationships were tested using the same data that were used to estimate the nutrient optima for taxa. Use of the NY tolerance values on the NJ dataset provides 1 type of independent validation, i.e., tolerance values from NY data were tested using a second dataset. This analysis produced weak relationships between NBI and water chemistry. The relationship between estimates of nutrient optima of the same taxa from the NJ data and from the NY data also tests the robustness of the tolerance estimates. When we included taxa that were relatively common in samples from both states, the 2 sets of estimates were only weakly correlated. This finding suggests imprecision in estimation of optima, imprecision in measurement of water chemistry, interactions with other factors, or different responses of macroinvertebrate taxa in the 2 states. The leave-one-out procedure provided a second validation. There was a large decrease in model fit from the all-data inference to the leave-one-out inference. The difference in model fit shows the effect of extreme data points on estimation of Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 547 Figure 5. Relationship between the difference in inferred nutrient concentrations by 2 procedures (leave-one-out versus all data) and nutrient concentrations. (A) Total phosphorus and (B) NO3. Northeastern Naturalist 548 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 nutrient optima because extreme points have a large effect on estimates of nutrient optima. The low slopes of the inferred nutrient–measured concentration regressions are an example of the edge effect (Birks 1998, ter Braak and Juggins 1993), which commonly occurs in inference models. The edge effect may arise from the inherent asymmetry in the distribution of taxa near the extremes of environmental gradients, resulting in systematic bias of nutrient optima toward the mean. In many inference models, the true environmental gradients are statistically recovered by de-shrinking (Birks 1998). In addition, we performed split data-validation using several different pairs of training and validation data. Similar to our results for the leave-one-out tests, we detected significant, but weak, relationships between NBIs and water chemistry in these validations (R.J. Horwitz et al., unpubl. data). Although our analyses indicate a relationship between macroinvertebrate assemblages and nutrient concentrations, NBIs based on the NJ data did not precisely infer nutrient concentrations in independent samples. This outcome does not invalidate the general NBI approach. There are several aspects of the analyses which could contribute to the weakness of the observed NBInutrient relationships: statistical issues, such as the difficulty of producing independent estimates of tolerance values for all taxa in a set of samples, the quality of the water-chemistry data available for analysis, the particular mathematical definition of the NBI, and the availability of estimates of tolerance values for various macroinvertebrate taxa. More importantly, the weak relationships detected relate to conceptual issues associated with the NBIs, including the causal basis for and the inherent variability in relationships between macroinvertebrate assemblages and nutrient concentrations. Availability of tolerance values for taxa Typically, the NBI for a sample was based on only a portion of the taxa present in the sample. Tolerance values could not be accurately estimated for rare or infrequent taxa, so NBI scores were based only on the widespread taxa present in each sample. This issue is important in validations with separate data sets because those used for estimating tolerances and testing NBIs may share relatively few taxa. The leave-one-out validation avoided many of the problems associated with different taxonomic lists, but the results were also relatively weak. As noted above, the taxonomic resolution of macroinvertebrate data affects the availability of tolerance values. Larger data sets would provide more samples with greater commonality in taxa present, which would allow more-accurate tolerance estimates for a larger number of taxa. A larger data set could also allow creation of sub-sets comprised of the species common to all, which would minimize effects of other significant factors. Tolerance values could be calculated from compilations of data from diverse sources. This approach could provide robust information on a large number of taxa, and resultant tolerance values would be independent of the data used to calculate NBIs. However, such meta-analyses would encounter problems with taxonomic consistency and accuracy of identifications. Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 549 Quality and relevance of water-chemistry data The consistency and relevance of the nutrient data affect estimation of tolerance values and the relationship between NBI scores and nutrient concentrations. Most of the water-chemistry samples were taken independently of the macroinvertebrate samples, and the data for individual sites were several years apart in many cases. Also, chemistry measurements were made using a variety of different methods (see Table 1). We selected TP and NO3 as relevant nutrient forms that were measured consistently across different programs. Other forms of P and N might exhibit a different nutrient response by macroinvertebrates, but there were no consistent data for these forms. The response of macroinvertebrates to nutrients is temporally integrated, however, which is one of the main advantages for using biological indices. Therefore, we would not expect correlations between macroinvertebrate indices and a few point-samples of nutrients to be as high as correlations with averages of more measurements made over a longer period of time. For the diatom subset, the correlation between P-NBIs and diatom-inferred TP was higher than any of the relationships between NBIs and measured chemistry values. The inferred values represent temporally integrated diatom responses. Uptake of nutrients by periphyton reduces nutrient levels. In streams where P is limiting, P may be reduced to very low concentrations. As a result, observed Pconcentrations may be lower than the concentrations which stimulated periphyton growth and resultant macroinvertebrate assemblages, weakening relationships between biota and observed nutrient concentrations. In systems with correlated sources of N and P, N may more accurately indicate high nutrient-loading. The estimation of nutrient optima depends on the distribution of water chemistry among samples used to make the estimates. Gaps in the distribution of nutrient levels and low numbers of samples representing extreme nutrient levels weaken the precision and accuracy of the estimates of nutrient optima. The NJ data had a greater range in nutrient concentrations than the NYSDEC data used by Smith et al. (2007), which might allow better estimates of nutrient optima. However, we used the same nutrient bins as Smith et al., which may have obscured changes in macroinvertebrate assemblages at extreme nutrient concentrations. Causal basis of macroinvertebrate-nutrient relationships The approach used by Smith et al. (2007) and in this study is empirical and does not assume or incorporate causal bases for relationships between macroinvertebrate assemblages and nutrient concentrations. Unlike algal responses (e.g., Charles et al. 2010, Ponader et al. 2007), macroinvertebrate responses are likely to be indirect, reflecting either trophic effects or correlations between nutrients and other factors which directly affect macroinvertebrates, especially fluctuating oxygen concentrations and potentially other agricultural runoff-related inputs such as pesticides. Trophic responses could include differences in macroinvertebrate feeding groups (e.g., grazers) in response to the quantity and type of algal food. Macroinvertebrates could be affected by changes in predator (e.g., fish) densities Northeastern Naturalist 550 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 or diet in response to nutrient-driven changes in production. Increases in nutrients commonly occur in agricultural or urban watersheds, where changes in sedimentation, geomorphology, hydrology, and temperature could have direct effects on macroinvertebrates. Analyses of the relationships of estimated nutrient optima and characteristics of macroinvertebrate taxa (e.g., feeding type, tolerance ratings) may provide evidence of some of these causal relationships. Such analyses could be used to improve estimates of nutrient optima or calculation of NBI scores. The appropriate level of taxonomic resolution for these analyses is uncertain. Estimated tolerance values frequently differed among species within genera, so that species-level analyses should provide more-precise relationships than analyses using higher-level taxonomy. However, finer levels of resolution result in fewer points for the estimation of nutrient responses and fewer taxa with estimated tolerance values available for estimating NBI scores. Mathematical definition of NBIs For each macroinvertebrate taxon, we calculated the nutrient optimum based on frequency of occurrence across different nutrient levels. Use of relative abundance within samples could improve sensitivity of the optima. For each sample, we calculated the NBI from nutrient optima and relative abundance of each taxon (excluding rare taxa). Various ways of weighting different taxa, e.g., giving more weight to sensitive taxa, and more-precise estimates of optima could improve sensitivity of the NBI. NBIs are subject to 2 estimation issues. First, generalist macroinvertebrate taxa may be common but provide little information on nutrient status. Second, estimates of nutrient optima for uncommon species are more likely to be imprecise. To reduce the effects of these issues, the weight of each taxon to the overall index could be inversely related to some measure of the specificity and reliability of that taxon’s nutrient tolerance. For example, the weight could be inversely related to the range or standard deviation of the distribution of the taxon across the nutrient gradient or weighted by the square root of the sample size. As noted above, use of information on feeding type, and sensitivity to other factors or other characteristics of macroinvertebrate taxa could also improve the NBI. For many bioassessment uses, accurate estimation of nutrients across the range of observed concentrations is less important than detection of situations with very high nutrients. Both nutrient indices and tests of nutrient indices may be designed to predict and evaluate such conditions. As noted above, a large dataset is necessary to provide a sufficient number of samples from sites with high nutrient-conditions to develop a specific high-nutrient index. Inherent variability between macroinvertebrate assemblages and nutrient concentrations Even with well-matched chemistry and macroinvertebrate data, and improved methods of calculating NBIs, the NBI–nutrient relationships could be weakened by a variety of factors including the influence of other variables on macroinvertebrate assemblages such as habitat, hydrology, contaminants, pH, alkalinity, Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 551 temperature, and oxygen concentrations, and by the tolerance of many taxa to a range of nutrient conditions. As noted above, development of a larger data set for estimating and testing NBIs would allow filtering or subsetting data to limit or incorporate effects of other factors. Habitat quality may also play a role. For example, Ashton et al. (2014) examined N associated with macroinvertebrate communities in wadeable streams in Maryland, and found that an index of biological integrity (IBI) was significantly associated with multiple nutrient measures; however, the response of intolerant taxa was predominantly influenced by a nutrient–forest-cover gradient, whereas habitat quality had a greater effect on tolerant taxa. Thus, multiple lines of evidence may be needed to highlight the effects of excessive nutrients in streams on macroinvertebrate communities and taxa in Maryland, whose loss may not be reflected in metrics that form the basis of biological criteria. Refinement of indicator taxa and development of a nutrient-sensitive index is therefore warranted before water-quality thresholds for aquatic life can be accurately quantified. In addition, a recent USEPA expert workshop on the development of nutrient- enrichment indicators in streams (USEPA 2014) noted that the indicators most sensitive to nutrient pollution in streams and most predictive of impacts to higher trophic levels include TN and TP, chlorophyll-a, percent visual coverage of algae and in-stream macrophytes, and measures of algal assemblages (e.g., diatoms and soft-bodied algae). However, the USEPA (2014) report also acknowledged that most states routinely monitor only fish and macroinvertebrates, and that the public recognizes the linkage between adverse effects on fish and invertebrates and the impairment of aquatic life. The experts participating in the workshop (USEPA 2014) concluded that commonly used fish and macroinvertebrate indices may be less sensitive nutrient-pollution indicators than other indicators (e.g., algae) and that refined and/or species-level metrics for macroinvertebrates (specifically calibrated to be responsive to nutrient effects) continue to show promise. However, there can be a significant temporal lag between high nutrient-concentrations and adverse effects to some higher trophic levels, making it difficult to proactively prevent nutrient impairment. The experts identified the following research needs: (1) development of a single, standardized, primary-producer indicator that integrates the productivity of various producers into a single indicator; (2) improved understanding of the linkages between nutrient measures, primary-producer measures, and higher trophic levels that often are used to quantify aquatic-life impairment; (3) identification of a minimum data-set size necessary to characterize stressor–response relationships; and (4) development of regional stressor–response relationships linking nutrient concentrations to algal-assemblage indicators, algal abundance, and nutrient-sensitive macroinvertebrate indicators. It should also be noted that correlations between N and P concentrations are common among sites (e.g., Bae et al. 2011, Bedoya 2011), and responses of macroinvertebrate taxa to the 2 nutrients may also be correlated. Both types of correlations lead to inherent correlations in N and P indices and prevent Northeastern Naturalist 552 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 complete separation of responses to each nutrient individually. For example, if all sites with high N also had high P and vice versa, there would be no way to determine which nutrient was causing a response among macroinvertebrate taxa, even if taxa were responding to single nutrients. Analogously, if some taxa are most frequent at lower concentrations of both nutrients, absence of those taxa (which would drive up the NBI) would not be useful to distinguish high-P from high-N conditions (although presence would be informative). Smith et al. (2013) related N endpoints to the NBI-P. There were positive correlations between TP and NO3 optima and between the corresponding tolerance values of taxa for both the NY- and NJ-based estimates. For the NY estimates, the coefficient of determination (R2) between TP and NO3 nutrient optima was 0.58 and the R2 between tolerance values was 0.50. For the NJ-based estimates (estimated from the main data set), the R2 between the 2 nutrient- optima was 0.27 and the R2 between tolerance values was 0.26. In all of the chemistry data sets, there were positive correlations between the N and P values. These values are underestimates of the true correlation between nutrient concentrations because the estimates are affected by measurement errors. For example, for the diatom subset, the correlation between measured N and P values in the NJ algal index (NJAI) was higher than the correlation between the nutrient concentrations in the NJDEP main data set. This difference may reflect the combination of data sources and methods for chemistry measurements for the latter data. The positive correlations between nutrient responses and the correlations between nutrient concentrations together resulted in positive correlations between NBI scores for nitrogen and phosphorus at a site. These correlations should be considered in the interpretation of NBI analyses. Forming a single index of nutrient response, such as the eutrophication index of Hering et al. (2006), avoids the problem of separating the responses. Conclusion The NBI approach approximated the same success in NJ waters as in the NY study (Smith et al. 2007), as measured by the relationship between NBI scores and nutrient concentrations. However, the robustness of these relationships is uncertain because the relationships in both cases were measured using the same data sets used to estimate nutrient responses of taxa. Estimation of nutrient optima and tolerance values from different data than those used for calculation of NBI scores produced significant relationships between NBIs or inferred nutrients and nutrient concentrations, but with high variance (low correlation) around the relationships. Furthermore, for individual taxa, estimates of nutrient-tolerance values from the NY data were not highly correlated with the tolerance values estimated from NJ data. These results indicate that macroinvertebrate assemblages are related to nutrient values but the relationships are not yet sufficiently characterized to allow them unambiguously to infer nutrient values from assemblage data. Northeastern Naturalist Vol. 23, No. 4 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 553 Acknowledgments We thank Leigh Lager, Paul Morton, and Kevin Berry of NJDEP for providing macroinvertebrate and chemistry data. A.J. Smith of NYSDEC provided estimates of tolerance values and explained aspects of the NY study, Shane Moser assisted with statistical analyses and preparation of figures, David Velinsky provided advice on selection and interpretation of nutrient data, Stephanie Kroll reviewed the manuscript, and Scott Haig prepared the map. This work was funded by a grant from the NDEP and by the Ruth Patrick Chair in Environmental Science at ANSDU. Literature Cited Ashton, M.J., R.P. Morgan, and S. Stranko. 2014. Relations between macroinvertebrates, nutrients, and water-quality criteria in wadeable streams of Maryland, USA. Environmental Monitoring and Assessment 186(2):1167–82. Bae, M.-J., Y. Kwon, S.-J. Hwang, T.-S. Chon, H.-J Yand, I.-S. Kwak, J.-H. Park, S.-A. Ham, and Y.-S. Park. 2011. Relationships between three major stream assemblages and their environmental factors in multiple spatial scales. Annals of Limnology 47:S91– S105. Barbour, M.T., J. Gerritsen, B.D. Snyder, and J.B. Stribling. 1999. Rapid bioassessment protocols for use in streams and wadeable rivers: Periphyton, benthic macroinvertebrates, and fish. Second edition. EPA 841-B-99-002. Available online at https://archive. epa.gov/water/archive/web/html/index-14.html. Accessed 2 December 2016. Bedoya, D., E.S. Manolakos, and V. Novotny. 2011. Characterization of biological responses under different environmental conditions: A hierarchical modelling approach. Ecological Modeling 222:532–545. Biggs, B.J., S.N. Francoeur, A.D. Huryn, R. Young, C.J. Arbuckle, and C.R. Townsend. 2000. Trophic cascades in streams: Effects of nutrient enrichment on autotrophic and consumer benthic communities under two different fish predation regimes. Canadian Journal of Fisheries and Aquatic Sciences 57:1380–1394. Birks, H.J.B., D.G. Frey, and E.S. Deevey. 1998. Review #1: Numerical tools in palaeolimnology— progress, potentialities, and problems. Journal of Paleolimnology 20:307–322. Bryce, S.A., and R.M. Hughes. 2003. Variable assemblage-responses to multiple disturbance gradients: Case studies in Oregon and Appalachia, USA. Pp. 539–560, In T.P. Simon (Ed.). Biological Response Signatures: Indicator Patterns Using Aquatic Communities. CRC Press, Boca Raton, FL. 576 pp. Camargo, J.A., A. Alonso, and M. De La Puente. 2004. Multimetric assessment of nutrient enrichment in impounded rivers based on benthic macroinvertebrates. Environmental Monitoring and Assessment 96:233. Charles, D.F., A.P. Tuccillo, and T.J. Belton. 2010. Diatoms and the biological-condition gradient in New Jersey rivers and streams: A basis for developing nutrient guidance levels. Patrick Center for Environmental Research Report 10-03 for NJDEP. Available online at http://www.state.nj.us/dep/dsr/nutrient/bcg-talu-final-report.pdf. Accessed 2 December 2016. Flinders, C.A., R.J. Horwitz, and T.J. Belton. 2008. Relationship of fish and macroinvertebrate communities in the mid-Atlantic uplands: Implications for integrated assessments. Ecological Indicators 8:588–598. Hausmann, S., D. Charles, T. Belton, and J. Gerritsen. 2016. A diatom-based biologicalcondition gradient (BCG) approach for assessing impairment and developing nutrient criteria for streams. Science of the Total Environment 562:914–927. Northeastern Naturalist 554 R.J. Horwitz, A. Tuccillo, D.F. Charles, S. Neiffer, and T. Belton 2016 Vol. 23, No. 4 Hering, D., R.K. Johnson, S. Kramm, S. Schmutz, K. Szoszkeiwicz, and P.F.M. Verdonschot. 2006. Assessment of European streams with diatoms, macrophytes, macroinvertebrates, and fish: A comparative metric-based analysis of organism response to stress. Freshwater Biology 51:1757–1785. New Jersey Department of Environmental Protection. (NJDEP). 2007. Development of the New Jersey high gradient macroinvertebrate index (HGMI). Final Report. Available online at http://www.state.nj.us/dep/wms/bfbm/download/HGMI_report.pdf. Accessed 2 December 2016. NJDEP. 2012. Ambient biomonitoring network, Northwest Region, round 4: Benthic macroinvertebrate data, Vol. 1 of 2. Division of Water Monitoring and Standards Technical Report. Available online at http://www.state.nj.us/dep/wms/bfbm/amnetRnd4.html. Accessed 12 December 2016. Pilière, H., A.M. Schipper, A.M. Breure, L. Posthuma, D. de Zwart, S.D. Dyer, and M.A.J. Huijbregts. 2014. Comparing responses of freshwater fish and invertebrate community integrity along multiple environmental gradients. Ecological Indicators 43:215–226. Ponader, K.C., D.F. Charles, D.R. Winter, and T. J. Belton. 2007. Diatom-based TP and TN inference models and indices for monitoring nutrient enrichment of NJ streams. Ecological Indicators 7:79–93. Smith, A.J., R.W. Bode, and G.S. Kleppel. 2007. A nutrient biotic index (NBI) for use with benthic macroinvertebrate communities. Ecological Indicators 7:371–386. Smith, A.J., R.L. Thomas, J.K. Nolan, D.J. Velinsky, S. Klein, and B.T. Duffy. 2013. Regional nutrient-thresholds in wadeable streams of New York State protective of aquatic life. Ecological Indicators 29:455–67. Spieles, D.J., and W. J. Mitsch. 2000. Macroinvertebrate community structure in high- and low-nutrient constructed wetlands. Wetlands 20:716. ter Braak, C.J.F., and S. Juggins. 1993. Weighted-averaging partial least squares regression (WA-PLS): An improved method for reconstructing environmental variables from species assemblages. Hydrobiologia 269/270:485–502. US Environmental Protection Agency (USEPA). 2014. USEPA expert workshop: Nutrient enrichment indicators in streams proceedings 16–18 April 2013. Office of Water EPA- 822-R-14-004. Available online at https://www.epa.gov/sites/production/files/2013-09/ documents/indicatorsworkshop.pdf. Yoder, C.O., and J.E. DeShon. 2003. Using biological-response signatures within a framework of multiple indicators to assess and diagnose causes and sources of impairments to aquatic assemblages in selected Ohio Rivers and streams. Pp. 23–55, In T.P. Simon (Ed.). Biological Response Signatures. CRC Press, Boca Raton, FL. 600 pp. Yoder, C.O., and E.T. Rankin. 1995. Biological-response signatures and the area of degradation value: New tools for interpreting multi-metric data. Pp. 263–286, In W.S. Davis and T.P. Simon (Eds.). Biological Assessment and Criteria: Tools for Water-Resource Planning And Decision Making. Lewis Publishers, Boca Raton, FL.