Introduction

Elevated nutrient concentrations in streams can lead to a variety of problems for human health, aesthetic impairment, water use, aquatic biota and downstream transport (Dodds & Welch, 2000). Ecological concerns include increased quantities of algae and aquatic plants (Houser & Richardson, 2010), decreased oxygen levels (Correll, 1998; Waite et al., 2019), toxic algal blooms (Lopez et al., 2008; Loftin et al., 2016), loss in important habitat due to excessive plant growth and resulting sedimentation (Sand-Jensen, 1998; Wharton et al., 2006), and a shift in aquatic food webs (Bracewell et al., 2019). Changes in stream primary production, habitat and food webs all contribute to potential increases in exotic species (Carpenter et al., 1998).

The eutrophication of surface waters is partially due to the increased loading of both nitrogen and phosphorus (Carpenter et al., 1998; USEPA, 1998, 2016a; NRC, 2000). A national assessment of 499 streams across the USA found that total nitrogen (TN) and phosphorus (TP) were 2–10 times higher than EPA recommended regional nutrient criteria, with highest TN concentrations associated with agricultural streams and TP elevated in both agricultural and urban streams (Dubrovsky et al., 2010). Point sources play a dominant role in nutrient loads in urban lands, comprising greater than 50% of N and P (Preston et al., 2011). Coastal areas of the USA are particularly impacted due to high transport of nutrients from inland rivers (Alexander et al., 2008) and from disproportionate population growth and urbanization along coastlines (Oelsner & Stets, 2019). Despite limited improvements, many streams and rivers in the USA are near or over nutrient saturation levels (King et al., 2014).

Challenges in managing nutrients involve determining the nutrient concentrations that elicit negative impacts to streams and how nutrient–biota interactions are influenced by other factors. Early work on establishing target nutrient levels was based on the distribution of nutrient concentrations (USEPA, 1998; Dodds & Welch, 2000) and use of reference sites for establishing “acceptable” concentrations (Herlihy & Sifneos, 2008). A major challenge of this effort is the determination of nutrient thresholds, which are defined as transition points of altered ecological conditions due to small changes in nutrient concentration (Huggett, 2005; Groffman et al., 2006; Dodds et al., 2010). Thresholds have been used to assess an ecological change in algal biomass (Dodds et al., 2010), specific taxa (Sunderman et al., 2015) and assemblage metrics (Black et al., 2011; Waite et al., 2019). Because species and ecosystem responses tend to be nonlinear (Clements et al., 2010; Dodds et al., 2010), studies have moved toward the use of nonlinear models, as in CART (Stevenson et al., 2012), gradient forest (GF) (Wagenhoff et al., 2017) and boosted regression tree (Waite et al., 2021). BRT models are among a family of techniques used to advance single-classification or regression trees by combining the results of sequentially fit regression trees to reduce predictive error and improve overall performance (Elith et al., 2008). Boosted regression trees (BRTs) are ideal for addressing the relative importance of variables that may be highly skewed (Elith et al., 2008) and has been shown to be useful in recent stressor studies (Munn et al., 2018; Waite et al., 2019, 2021).

This study builds upon previous work done as part of the U.S. Geological Survey Stream Assessment Program. Waite et al. (2021) examined the response of the multimetric index (MMI) metric to numerous stressors (68) in various stressor categories including habitat, pesticides (herbicides, fungicides, insecticides, and toxicity), nutrients, water quality (conductivity and dissolved oxygen) and stream flow. While Waite et al. (2021) provided important information as to what major groups were having a strong influence on biota, it was a coarse-grain approach in that MMI are nonspecific stressor response indicators and the high number of stressors resulted in not being able to fully parse out nutrient species and habitat. Our study represents a more fine-grain approach in that nutrient species can be separated out as to their relative importance among biota and regions. This study relied on a nutrient gradient design across a large geographic area (434 sites from five regions), repeated nutrient sampling prior to biological sampling and the comparison of three taxonomic assemblages (algae, invertebrates and fish). The objectives of this study were to determine 1) what nutrient species are most associated with each biological assemblage and 2) what is the relative influence of nutrients and instream habitat on biological assemblages. For the first objective, we hypothesize that dissolved forms of dissolved inorganic nitrogen (DIN) and orthophosphate (OP) will be more important to biota due to dissolved nutrients being biologically available. For the second objective, we hypothesize that nutrients will be more important for the diatoms and become less important as one moves up the food chain where the effects from nutrients become indirect (note: an important score is determined for each explanatory variable by the BRT models; this is used as evidence for the stated hypotheses).

Methods

From 2013 to 2017, the USGS evaluated multiple stressors on the ecological condition of streams in five regions of the USA: Northwest (Sheibley et al., 2017), coastal California (May et al., 2020), Upper Midwest (Garrett et al., 2017), Northeast (Coles et al., 2019) and Southeast (Journey et al., 2015) (Fig. 1). A total of 434 wadeable streams were sampled across the five regions. A key design component of assessing nutrients and ecological effects is to select sites that span a gradient of nutrient concentrations, with a portion of sites falling below published threshold levels for TN (0.5 mg l−1) and TP (0.03 mg l−1) (Dodds et al., 2002; Stevenson et al., 2006). Sites were selected along a gradient of urban or agriculture land use to capture a range of nutrient conditions. Site-specific geospatial boundaries and geospatial characteristics are available from Qi & Nakagaki (2020a and 2020b) (Table 1).

Fig. 1
figure 1

Location of sites within the five regions assessed in this study (modified from Waite et al. 2021)

Full size image
Table 1 Summary of sites, drainage area (DA) and land use of study sites; data represent mean and range
Full size table

Habitat

Habitat variables followed protocols by USGS (Fitzpatrick et al., 1998) for the Northeast, Southeast and Northwest, whereas USEPA protocols were used for coastal California and Upper Midwest (USEPA, 2010). Habitat variables used in this study were comparable among the two protocols. Stream reaches of approximately 150 m were established at each site, with instream habitat collected at 11 equidistant transects. Quantitative measures used in this study included stream width (m), gradient, percent fines, substrate size (D84) and percent canopy (Table 2). Flow related metrics (peak stage, interval peak, stream depth and depth change) were determined using either established USGS gages or pressure transducers. Median and maximum water temperature prior to sampling were determined using temperature thermistors.

Table 2 Assemblage metrics and explanatory variables used in BRT models
Full size table

Diatoms, invertebrates and fish

Diatoms, invertebrate and fish were sampled using USGS methods (Moulton et al., 2002) in the Southeast, Northeast and Northwest regions, with USEPA methods (USEPA, 2013) in the Midwest and California. Algae and invertebrates were collected in the richest targeted habitat (e.g., riffles) in the USGS protocols, whereas they were collected along set transects in the USEPA protocol. Concerns about the differences between the two sampling protocols have been addressed by a variety of researchers and have been found to be comparable, especially across large regional studies (Ode et al., 2008; Waite et al., 2014; Cuffney & Kennen, 2018). Fish were collected using standard electroshock methods along the 150 m reach (Moulton et al., 2002; USEPA, 2013). Method details for each region can be seen in Journey et al. (2015, Garrett et al. (2017), Sheibley et al. (2017), Coles et al. (2019) and May et al. (2020).

Nutrients

Discrete water samples were collected weekly over four weeks prior to the collection of biological sampling, with the median of the four samples used for each site. Water samples were collected using either grab samples or the equal-width method based on flow (U.S. Geological Survey, 2006). Colorimetry was used to determine total phosphorus (TP, DL = 0.002 mg/l) concentrations (O’Dell, 1993) with orthophosphate (OP, DL = 0.002 mg/l) by colorimetric analysis (Fishman, 1993). The colorimetric analysis method was also used for dissolved ammonia (NH3 + NH4+, DL = 0.005 mg/l) and nitrite (DL = 0.005 mg/l) (Fishman, 1993), whereas dissolved nitrate plus nitrite was determined by low-level enzyme reduction (DL = 0.02 mg/l) (Patton & Kryskalla, 2011). DIN is the sum of dissolved nitrate, nitrite and ammonia. TON is total organic nitrogen, calculated as the difference between TN and DIN concentrations.

Ecological metrics and stressor variables

For each taxonomic group, we selected up to six metrics for running the initial analyses. The final metrics were selected based upon their known response to nutrient and habitat stressors and their successful use in prior studies (Munn et al., 2018; Waite et al., 2019, 2020; Table 2). There was also an effort to include metrics that represented different assemblage characteristics, like overall assemblage diversity versus trophic status. Diatom metrics included the biological condition index group 4 and 5 (BC 4–5) which include tolerant diatoms; low phosphorus taxa (Potapova & Charles, 2007) which include diatom taxa that respond to low levels of total phosphorus; and relative abundance of moderate and highly motile diatom taxa which indicate increasing nutrient concentrations (Porter et al., 2007).

Invertebrate metrics used in this study were calculated using methods presented in Cuffney et al. (2003, 2007). Metrics selected included scraper richness as a trophic metric, the average tolerance of invertebrate taxa to represent overall invertebrate tolerance to nutrients and a multimetric (invertebrate MMI) which discriminates across land use disturbance gradients (Herlihy et al., 2020).

Fish metrics captured various features of assemblages including richness and tolerances outlined in Meador & Brown (2015) and Meador & Carlisle (2007). Metrics included the fish MMI (USEPA, 2016b), the average of species tolerance to nitrate (Meador, 2012) and the percent benthic invertivore taxa to reflect trophic status of fish assemblages.

Six nutrient species were included as environmental stressors (Table 2), including four measures of nitrogen (TN, DIN, ammonia and TON) and two measures of phosphorus (TP and OP). While many of these measures can be correlated in specific regions, each is used to represent various measures of nutrient condition. Ten physical habitat measures were included in the BRT modeling, including two geomorphic (wetted width and gradient), two substrate (percent fines and substrate size D84), percent canopy cover, median and maximum water temperature, two measures of stream depth (stream depth and change in stream depth) and two measures of stream flow (peak flow and interval peak flow).

Data analysis and model development

For all instream environmental stressors, we focused on summary metrics related to the median of the water quality sampling (Waite et al., 2019). Algal, invertebrate and fish metrics were used to create Boosted regression tree models (Gaussian) using the gbm library in R (Version 4.2.1; R Project for Statistical Computing, Vienna, Austria) (2018). Explanatory variables included 11 habitat and six nutrient variables (Table 2). A minimum of 1000 trees per model was used to optimize the BRT model, with evaluation of interactions and partial dependency responses to reduce potential explanatory variables and overfitting (Elith et al., 2008). Variables with Variable Importance ((VI) scores < 7 were removed to improve interpretation. VI scores are a measure of how often a variable is selected across all the trees and it’s deviance improvement vs. when it is removed compared to all other variables evaluated in the model. Final VI scores are re-scales so all variables add up to 100 (Waite et al., 2021). BRT models were run for each of the regions in this study. Studies have demonstrated the strong influence of regional differences in biological assemblages and therefore the need for separate assessments (Potapova & Charles, 2007; Munn et al., 2009; Herlihy et al., 2020). The VI scores were used to assess the two objectives of the study as to what explanatory variables are most important to the three taxonomic groups.

Partial dependency plots show the stressor plotted against the biotic metric and modeled response form (solid line curve), with Lowess smooth curves (dashed lines) added to aid in the overall interpretation (Waite et al., 2019). CV (cross-validation) R2 values, which are less likely to be inflated by overfitting, were calculated by holding 25% of the sites out of each regression tree split and then using those sites to test the amount of deviance explained by the split (Elith et al., 2008).

Results

Nutrients

All five regions in our study contained a wide range in TP and TN concentrations (Table 3) with sites falling below and above published threshold levels for TP (0.03 mg l−1) and TN (0.5 mg l−1) (Dodds et al., 2002; Stevenson et al., 2006) (Fig. 2). However, the percentage of sites below the threshold levels varied depending on the region, with a low of 9 and 7% for TN and TP, respectively, in the Midwest to a high of 35 and 45% for TN and TP, respectively, in the Northeast. Overall, TN concentrations ranged from < 0.05 to 36.4 mg l−1 and TP ranged from 0.01 to 2.82 mg l−1 for TP (Table 3). Highest average concentrations of nitrogen (TN, DIN, TON and ammonia) occurred in the Midwest, which is the more agriculturally dominated region, with highest averages of phosphorus (TP and OP) from California. Lowest average concentrations of TN and DIN occurred in the Northeast, with the Northwest having the lowest average concentrations of TON, ammonia, TP and OP.

Table 3 Median and range of concentrations for the six nutrient species assessed in this study
Full size table
Fig. 2
figure 2

Distribution of TP by region based on the median weekly value for four weeks prior to sampling biological assemblages. The vertical line represents an ecological threshold for TP (0.03 mg l.−1, Dodds et al. 2002)

Full size image

Pearson correlations of the six nutrient species indicated that all were significantly correlated (P < 0.05, n = 435, r = 0.28–0.95) with all regions combined; however, there were some differences at the regional scale. TN and TP were strongly correlated (P < 0.01) at r = 0.63–0.69 for Northwest, California and Northeast, whereas correlations were weaker in the Midwest and Southeast (r = 0.45 and 0.48, respectively). Studies normally focus on either total nutrients (TN and TP) or on the dissolved fractions (DIN and OP). Our study found that at the regional scale TN and DIN were highly correlated ranging from 0.92 to 0.99, whereas TP and OP ranged from 0.84 to 0.97. Therefore, using the total or dissolved forms of nitrogen or phosphorus provided similar results. In contrast, there were larger differences when comparing TN to either TON or ammonia. TN and TON correlations ranged from 0.43 to 0.87 with lowest correlations in the Midwest and Southeast and highest correlations in California and Northeast. The relationship between TN and ammonia was more varied with correlations ranging from r = − 0.1 (Midwest, P > 0.5) to r = 0.71 (California). However, the average r value was only 0.42 which indicates that ammonia can vary substantially with TN among regions. Correlations of TP with TON can be an indicator of TON source. In our study, TP was correlated with TON in all regions with lowest correlation in the Midwest (r = 0.48) and all other regions between r = 0.67 and 0.77.

BRT models

Forty-two BRT models were run using three taxonomic groups (algae, invertebrates and fish) in five regions; fish were not collected in California (BRT data in Supplemental Table). BRT was unable to construct four diatom models due to insufficient site separation, which reduced the number of BRT models to 38. Overall, invertebrate BRT models were the strongest with CV R2 values ranging from 0.25 to 0.63 (average = 0.47) with fish models the weakest, ranging from 0.04 to 0.55 (average = 0.26). A nutrient species was the dominant variable in most diatom models, with nutrients and habitat equal as the dominant variables for invertebrate models and habitat the dominant variables for fish models.

Diatoms

The low phosphorus diatom metric had the highest CV R values in the Midwest, Northeast and Southeast, whereas in the Northwest the diatom biological condition metric (BC 4–5) CV R2 was highest (Table 4). The motile diatom metric was consistently the poorest performer in all regions. None of the three diatoms metrics consistently outperformed others among the regions; however, diatom metrics were best explained by predictor variables in the Southeast. TP or OP was the dominant variable in 82% of diatom models, resulting in phosphorus having the highest average percent importance across all diatom models (Fig. 3); dissolved OP was found more important than TP in most models. In contrast, while various nitrogen species occurred in 9 of the 11 models (82%), it was total organic nitrogen (TON) that occurred in more BRT models with the highest average across all models (Fig. 3). While habitat variables were important in all diatom models, interval peak was the only habitat variable with the highest VI value for the BC-45 metric in the Northeast. Variables reflecting changes in streamflow were the most common habitat explanatory variable for diatoms, which included interval peak and change in depth; variables of lessor importance included stream width, substrate and peak flow.

Table 4 Diatom booted regression tree (BRT) model results across five regions; variables are presented in descending order of variable importance (VI) in each model. Variable definitions are found in Table 2. All nutrient variables are based on median values of four weekly samples; n = the number of stream sites; CV = cross-validation. The diatom metrics for the California region could not be completed
Full size table
Fig. 3
figure 3

Percent importance values of individual nutrient species from BRT models. Values are the average of the percent importance values across the number of models within a taxonomic group

Full size image

Invertebrates

Invertebrate models had higher CV R2 values than diatoms or fish, with only minimal differences among the three metrics (Table 5). The Southeast region generally had the highest BRT CV R2 values and the Midwest with the lowest CV R2 values. Average invertebrate tolerance (RichTol) had the highest CV R2 values in 3 of the 5 regions. Invertebrates were the only taxonomic group that had approximately equal frequency of nutrient species and habitat variables as importance variables. Of the nutrient species, TON had the highest average percent importance value; however, ammonia was second with an average of 11% (Fig. 3). Habitat was equally as important overall in the BRT models, with 53% of models having a habitat variable as the most important explanatory variable. Interval peak and percent fines were the most common habitat variables in the models, with 33% of the models having interval peak as the first explanatory variable.

Table 5 Invertebrate boosted regression tree (BRT) models across five regions; variables are presented in descending order of variable importance (VI) in each model. Variable definitions are found in Table 2. All nutrient variables are based on median values of four weekly samples; n = the number of stream sites; CV = cross-validation
Full size table

Fish

Fish metrics differed from diatoms and invertebrates in that habitat variables were generally more important than nutrient species (Table 6). TP occurred more often (50%) than TON (33%); both had similar average percent importance values at 9 and 8%, respectively (Fig. 3). While many habitat variables played a role in the BRT models, percent fines and maximum temperature (Tmax) were the dominant variables. While there were no strong patterns as to the strength of the three fish metrics, fish MMI tended to have a slightly higher score (as CV R2) in the Northwest, Midwest and Northeast, whereas fish invertivores were strongest in the Southeast.

Table 6 Fish results for boosted regression tree (BRT) models across five regions; variables are presented in descending order of variable importance (VI) in each model. Variable definitions are found in Table 2. All nutrient variables are based on median values of four weekly samples; n = the number of stream sites; CV = cross-validation
Full size table

Partial dependency plot comparisons

We used partial dependency plots to assess patterns of nutrient species found to be important in models. For comparison, we examined a single metric within each of the three taxonomic assemblages for three of the regions. The diatom low phosphorus model had the highest VI values among the diatom model with either OP or TP the highest nutrient importance value in the models. For all three regions, the diatom low phosphorus metric decreased with increased concentrations of OP or TP, with the breakpoint occurring around 0.12 mg l−1 (Fig. 4), with little change above 0.2 mg l−1.

Fig. 4
figure 4

Partial dependency plots from selected boosted regression models for the Northwest (left column), Midwest (middle column) and Southeast (right column). Metrics plotted include the diatom low phosphorus taxa metric (upper panel), average invertebrate tolerance metric (middle panel) and fish nitrate tolerance metric (lower panel). Solid line represents the BRT modeled line, and the dashed line represents a Lowess smooth curve

Full size image

Ammonia was found to be important in the average invertebrate tolerance BRT models. Partial dependency plots indicated that average invertebrate tolerance increased with initial increases in ammonia, but then leveled off rapidly (Fig. 4). Breakpoints for the three regions varied only slightly, with the Northwest at 0.02 mg l−1, Midwest at 0.015 mg l−1 and Southeast a bit higher at 0.1 mg l−1. The ammonia vs invertebrate tolerance metric was unique in that the Lowess curve indicated an initial increase followed by a decrease over increasing ammonia concentrations.

For fish, TP was used to assess nutrient increases versus the nitrate-tolerant metric; TP was found to be positively correlated with TN (r = 0.45, P < 0.001). The BRT modeled line showed no apparent relationship between TP and nitrate-tolerant fish; however, the Lowess curve increased near the higher concentrations of TP (Fig. 4). In contrast, the Midwest showed nitrate-tolerant fish increasing slightly above 0.15 mg l−1. The Southeast region indicated the opposite pattern to the Midwest in that nitrate-tolerant fish decreased at lower levels of TP, near 0.05 mg l−1.

Discussion

The dominant role of nutrients in the diatom models reflects their position at the base of the food web where they are directly influenced by nitrogen and phosphorus (Hering et al., 2006). This supports our second hypothesis that nutrients can be more important than habitat for diatoms. The role of phosphorus in the diatom models indicates its importance as a limiting nutrient (Downing et al., 2001; Smith, 2003; Camargo et al., 2005). However, nitrogen can be the limiting nutrient in streams with low N:P loading ratios (Smith, 2003), which may explain the importance of DIN for diatom biological condition in the Northwest. N:P ratios were highly mixed depending on the site and region with California and the Southeast more nitrogen limited (N:P < 10) and the Midwest more phosphorus limited (N;P > 20). Overall, nutrient limitation varies substantially by taxonomic group and region (Borchardt, 1996; Mebane et al., 2021). Furthermore, anthropogenic factors greatly influence the loadings of nitrogen and phosphorus to streams thereby influencing the relative limitation of either nitrogen or phosphorus (Jarvie et al., 2018). The relative equal mix of TP and OP as dominant factors in models relates to the high correlation between the two measures.

While we hypothesized that DIN would be the dominant form of nitrogen for diatoms, we found TON had the greatest influence on diatom metrics followed by TN and DIN (highly correlated). TON (total organic nitrogen) is not commonly analyzed as part of algal-nutrient studies; however, organic nitrogen can comprise up to 80% of TN in surface waters and is common in both urban (Scott et al., 2007) and agricultural settings (Hussain et al., 2020). In our study, TON composed an average of 36% of TN but varied widely across sites (1–100%). The importance of TON was shown by it occurring in 37% of the diatom BRT models, with it the dominant variable in 50% of the invertebrate models. TON was associated with agricultural land use in California, Northwest and Midwest, whereas TON was associated with urban land use in the Northwest and Northeast. The source of organic nitrogen includes soil runoff, litter decomposition on land and biological production in streams (Hussain et al., 2020). Overall, TON provides a major pool of nitrogen in stream and rivers and is a function of sources, hydrological processes and instream biogeochemical processes (Scott et al., 2007), with the TON pool consisting of a mixture of organic nitrogen compounds. While algae can utilize nitrate and ammonia directly, organic nitrogen needs to be biologically transformed into NH4, urea or nitrate (Berman & Chava, 1999). Taken together, TON can provide an important pool of nitrogen in flowing water systems (Scott et al., 2007), which is reflected in the BRT models.

While nutrients enhance the accrual (colonization plus growth) of benthic algae, light and temperature are also key to algal development, with hydrologic stability and grazing important for biomass loss (Biggs, 1996). Of the habitat variables examined in the BRT models, we found metrics related to stream flow, as in median daily change in depth and the number of days between high flow events (Interval Peak Flow and Depth change), to be the most common, followed by measures of substrate (percent substrate fines and substrate D84). It is well documented that the combination of streamflow and sediment is important controlling factors of benthic algae (Lamberti, 1996; Peterson, 1996; Munn et al., 2018).

As indicated by the Lowess curve, the Subsidy–Stress relationship (Odum et al., 1979) best describes the interaction of nutrients with invertebrates; at low but increasing nutrient concentrations there is often an increase in algae which has a positive influence on invertebrates (Chetelat et al., 1999); however, when nutrients become excessive, streams become more eutrophic which leads to a negative influence on invertebrate assemblages (Wang et al., 2007; Ouyang et al., 2018; Waite et al., 2019). The negative influence of nutrients on invertebrates can be due to a combination of factors as in changes in substrate due to increased primary production (Sand-Jensen, 1998; Wharton et al., 2006), food quantity and quality (King & Richardson, 2007), and dissolved oxygen levels (Correll, 1998). The result is often a change in assemblage diversity and composition of loss of native species (Carpenter et al., 1998; Smith et al., 1999). BRT models indicated that nutrients explained approximately half of the invertebrate metrics, with ammonia the most frequent form of nitrogen. Ammonia includes ionized ammonia (NH4+) and unionized ammonia (NH3), and while NH4+ is normally the dominant form, the relative proportions of each are dependent on pH and temperature (USEPA, 2013b). NH3 is the toxic form of ammonia and has been shown to be particularly detrimental to mussels and some species of fish, but less toxic to aquatic invertebrates (USEPA, 2013b). While the frequency of ammonia in the BRT models (60%) could indicate potential toxicity, concentrations of ammonia only exceeded USEPA chronic criteria (1.3 mg/l at pH 7.7 and water temperature of 20.3 °C) at one site (USEPA, 2013b), and therefore, the importance of ammonia likely indicates overall eutrophication more than toxicity. However, it is important to note that ammonia varies rapidly in streams due to input and uptake, and therefore, concentrations can exceed levels of concern for short periods. Furthermore, the assessment of ammonia toxicity is based upon limited taxa and measures; therefore, low-level effects could be a factor. Ammonia is a key nitrogen species for biological processes given it is assimilated by bacteria and algae, but also can undergo nitrification which converts ammonia to nitrate for additional assimilation (Bernhardt et al., 2002). TN or nitrate, which often accounts for most of the TN, is often reported as an important nutrient associated with a negative (Yuan, 2010) or positive (Piggott et al., 2012) influence on invertebrates.

Stream habitat played an equally important role as nutrients influencing invertebrate metrics in California, Northwest and Northeast, with habitat having a larger influence on invertebrates in the Midwest and Southeast. Of the habitat variables, interval peak stage and substrate fines occurred in more of the invertebrate models than other habitat variables. In an experimental study, Juvigny-Khenafou et al. (2020) found that flow reduction had a larger influence on invertebrate assemblages than fine sediment, which was followed by nutrients. Flow alteration modifies the diffusion of materials and instream habitat (Calapez et al., 2018; Lange et al., 2018) and can dislodge invertebrates from their habitat. Fine sediment, which can change in relation to stream flow, can reduce habitat heterogeneity by filling interstitial spaces (Piggott et al., 2015) and scouring benthic habitat. Fine sediment was equally important for both invertebrates and fish metrics, occurring in 50% of models.

BRT models indicated that fish metrics were influenced by physical habitat more than by nutrients, which supports our second hypothesis. However, nutrients are well documented for influencing fish assemblages, including TP (Meador & Carlisle, 2007; Atique & An, 2018; Waite et al., 2019) and TN and its dissolved forms (Meador & Carlise, 2007; Meador & Frey, 2018; Waite et al., 2019). Except for ammonia as a potential toxicant, nitrogen and phosphorus influence fish assemblages indirectly in the same way they influence invertebrates via increases in primary production (Kratina et al., 2012) which influences the overall food resources (Bracewell et al., 2019; Blocher et al., 2020), while also influencing habitat (e.g., substrate, macrophyte cover, etc.) and dissolved oxygen concentrations (Meador & Carlise, 2007; Munn et al., 2020).

The key habitat variables of importance to fish assemblages included percent fines, water temperature and changes in stream flow. As with invertebrates, percent fines reduce important habitat and limit food resources for insectivorous fish (Sullivan & Watzin, 2010; Meador and Frey, 2018; Blocher et al., 2020). Water temperatures and streamflow are important variables for fish distribution (VanCompernolle et al., 2019); minimum daily temperatures are critical for permitting species to tolerate periods of higher temperature (Farless & Brewer, 2017).

Conclusions

This study demonstrated the varying influences of nutrients and habitat on three biological assemblages in streams. Our hypothesis that the dissolved forms of nutrients (DIN and OP) would dominant the relationships of nutrients to biological metrics was not fully supported. We found both OP and TP to be key variables, but with nitrogen we found TON to dominant as an explanatory variable. However, our second hypothesis was supported in that the BRT models indicated that diatoms were more influenced by nutrients than habitat; invertebrates were equally influenced by both nutrients and habitat, and habitat had a greater control than nutrients on fish. Furthermore, phosphorus (orthophosphate or total phosphorus) was the key nutrient for diatoms, with ammonia and total organic nitrogen of approximate equal importance for invertebrates, and a mix of total organic nitrogen and total phosphorus the dominant nutrients for fish metrics. Diatoms and invertebrates both had a distinct breakpoint in a metric in response to increasing nutrient concentrations; however, fish tended to have more muted changes. Strongest individual biotic metrics included low phosphorus diatom taxa, tolerant invertebrate abundance and fish multimetric index (Fish MMI). This study illustrates that various nutrient species can have a dominant influence on biota depending on the assemblage and conditions, and that nutrients play a more important role at the food base with habitat becoming more important in higher trophic levels.