Functional Trait Analysis and the Multi-metric Integrity Model, Based on Stream Fish Indicators, and Their Relations to Chemical Water Quality

Functional trait and biological integrity approaches in stream ecology enable the determination and prediction of aquatic community responses to a variety of environmental stressors, such as chemical pollution, habitat disturbance, and biological invasion. Here, we used multi-trait based functional groups (FGs) to predict the functional responses of fish assemblages to the physicochemical and ecological health gradients in a temperate stream. The multi-metric biological integrity model (mIBI model) was used to evaluate stream ecological health. The FGs were derived from the distance matrix of trophic, tolerance, and physical habitat traits among fish species. The leading water quality indicators (conductivity [EC], total suspended solids [TSS], and chlorophyll-a [CHL-a]) varied conspicuously with the stream gradient and anthropogenic pollution. The multi-metric water-pollution index (mWPI) showed differences in chemical health from upstream to downstream. Monsoon precipitation may have affected the variations in fish species and associated changes linked to irregular chemical health. The fish FGs varied more by space (longitudinal) than by season (premonsoon and postmonsoon). Functional metrics, which reflected trophic and tolerance traits, as well as vertical position preference, were strongly correlated with water quality degradation downstream. Changes were evident in FG (II, III, and IV) combinations from the upstream to downstream reaches. Furthermore, the structure of the fish assemblages from FG-II and FG-III was significantly correlated with chemical (R2 = 0.43 and 0.35, p < 0.001) and ecological health (R2 = 0.69 and 0.66, p < 0.001), as well as the metrics of mWPI. In conclusion, the results indicate significant variations in both trait-based FGs and biological integrity among stream-fish communities, influenced by chemical water quality gradients. The causes included longitudinal zones and intensifying degradation of water quality downstream. Therefore, multi-trait based FGs can facilitate ecological health assessment and develop the mIBI model based on fish assemblages by reflecting the prevailing chemical health status of streams and rivers.


Introduction
Functional trait analysis of aquatic organisms allows estimation of the prevalent linkages among anthropogenic disturbances such as water pollution, hydromorphological alterations, and loss of biological community structure (Frimpong & Angermeier, 2010;Menezes et al., 2010). Accordingly, grouping species based on functional characteristics can be a helpful indicator in predictive models of biological integrity in combination with the environmental conditions and taxonomic measurements (Hoeinghaus et al., 2007;Petchey & Gaston, 2002). The clustering approach to evaluating functional traits can identify suitable ecological attributes and community structures to warn of multiple disturbances shaping the biological community (Jackson et al., 2010). According to the environmental filtering hypothesis (Cornwell et al., 2006), cluster species on set of functional traits can respond similarly to abiotic conditions (Dumay et al., 2004;Wang et al., 2019). Thus, they can provide for simplification of ecological model and give a better understanding on functional modification of aquatic communities due to a variety of environmental problems.
Fish are an excellent bioindicator of water quality disturbance owing to their sensitivity to pollutants and disturbance factors (Ibáñez et al., 2010). Fish diversity and abundance usually increase from upstream to midstream to downstream because of habitat diversity, physical gradients, and functional variations (Araújo et al., 2009;Vannote et al., 1980). The river continuum concept (RCC; Vannote et al., 1980) provides a framework for predicting how aquatic communities change with longitudinal zones. Shifts in energy inputs along the physical gradient form the trophic composition of aquatic communities according to the RCC. Based on this concept, fish species that feed as insectivores are found upstream, whereas omnivores, herbivores, and carnivores become relatively abundant in the lower reaches of streams. The serial discontinuity concept (SDC; Ward & Stanford, 1995) reflects the effect of dams and other anthropogenic interventions (land-use, waste-water treatment plants (WWTPs), and natural and human-induced surface erosion, etc.) that may disrupt the river continuum. Accordingly, the barriers in lotic systems that cause habitat alteration and fragmentation, hampering fish survival and movement (Helms et al., 2011;Mims & Olden, 2013).
Physicochemical degradation caused by intense land-use activities, moreover, pose a serious threat to biological communities in streams, such as compositional changes in functional fish structure (tolerance and trophic groups) and declines in fish diversity and abundance (Choudri et al., 2017;Duque et al., 2020;Kim et al., 2019;Shetty et al., 2015). Altered environments also favor the proliferation of invasive alien species (IAS) that have greater tolerance to elevated levels of nutrients and organic matter. Unfortunately, IAS threaten the persistence of native fish because of predation and the resulting unequal utilization of feeding and spawning grounds (Hermoso et al., 2011;Kim et al., 2021;Strayer, 2010). Hence, a clearer understanding of fish community responses (based on their functional traits) to environmental disturbances is vital for species conservation and mitigating ecological degradation.
In healthy environments, organisms form assemblages similar to those produced by long-term evolutionary processes, which have an elevated level of biological integrity (Angermeier & Karr, 1986). As a result of human activities, environments can lose their biological integrity and become unlivable to appropriate organisms. The multi-metric biological integrity model (mIBI model) is used to designate the ecological health status of streams and rivers (Angermeier & Karr, 1986;Herman & Nejadhashemi, 2015). The original IBI (Karr, 1981) and its adapted versions for various countries were based on multiple metrics of aquatic communities-including taxonomic richness, functional guilds, and species composition-that respond to anthropogenic alternations in lotic systems (Ruaro & Gubiani, 2013;Simon & Evans, 2017). Globally, the mIBI model based on fish assemblages has been accepted for measuring aquatic system health, including in Europe (Hermoso et al., 2010), Asia Jia et al., 2013), and North and South America (Long & Walker, 2005;Van Oosterhout and van der Velde, 2015).
During recent decades, fish functional traits in ecological studies are used to predict and describe species and community responses to environmental stressors; and species decline, invasions, and community homogenization (Frimpong & Angermeier, 2010;Villéger et al., 2017). Accordingly, functional trait-based analyses using multivariate statistical methods, principally correlated with biological assessment and monitoring, are conducted to identify interactions between environmental variables and fish community structure (Brind'Amour et al., 2011;Mouillot et al., 2013). One such approach is the cluster analysis for generating functional groups, based on a multidimensional distance matrix among species traits. Functional group clustering can reflect identical responses to the environmental conditions in an aquatic ecosystem due to similar functional roles of cluster species (Dumay et al., 2004;Wang et al., 2019), generating comprehensive and straightforward environmental fish community models. Overall, the fish-based mIBI model and functional trait analysis may yield variations in fish community attributes (functional groups, clusters of fish groups, richness, composition) on physicochemical gradients in the stream caused by both anthropogenic disturbances and the stream gradient.
This study, which was based on a large data set of water quality and fish collected from a temperate stream, analyzed the functional responses of fish communities to variations in chemical water quality by comparing biological integrity (mIBI model) and fish guilds to multiple trait-based functional groups. The main suggestion of the study is to show that multiple trait-based functional grouping can provide more responsible fish community metrics for the development of ecological modeling (mIBI model) and stream health assessment.

Study Area and Monitoring Sites
Dal Stream is located in the southernmost part of the Han River in South Korea. It originates from Mount Songnisan, passes through Goesan County, and joins the Namhan River west of Chungju City. The catchment area is 1614.4 km 2 and the river is almost 116 km in length (Song et al., 2016). The geographical location of the stream and its monitoring sites are shown in Fig. 1. For the purpose of the current study, samples were collected from eight fish-monitoring sites (F1-F8) and five water quality sites (W1-W5 and W1 and F3 are in the upstream zone, which has the highest proportion of forest area. Compared to the upstream sites, the midstream sites (W2 & F5 and W3 & F6) are affected by intense agricultural activities, mainly increasing paddy crop area. Also, a smaller stream (Dongjin) that passes through urban (Goesan Town) and industrial regions joins Dal Stream at W2 and F5. In addition, the Goesan Dam is located between sites F4 and W2 and F5. It was the first hydroelectric dam in South Korea, commissioned in 1957. The dam is 28 m in height, 41 m in width, and 171 m in length, with a maximum reservoir depth of 21.9 m. The downstream sites (W4 & F7 and W5 & F8) are located near Chungju City, and paddy fields are dominant in the area. Urban and industrial effluents also impact W5 and F8 and a wastewater treatment plant (WWTP) is located upstream to the east of the last site on a small stream (Chungju). Furthermore, the stream-order classification for the study sites was followed by the method of Strahler (1957). The first two sites (F1 and F2) were classified as the second-order stream, W1 and F3 as the third order, F4 and W2 and F5 as fourth order, and the other sites as the fifth order.

Physicochemical Variables and Precipitation Data
We evaluated the overall chemical water quality of the stream using monthly data of the physicochemical parameters between 2011 and 2019. The data on water quality were obtained from the Korean Ministry of Environment's Water Information Network. The physicochemical variables were water temperature (WT), electrical conductivity (EC), total suspended solids (TSS), dissolved oxygen (DO), biological oxygen demand (BOD), chemical oxygen demand (COD), total organic carbon (TOC), total nitrogen (TN), total phosphorus (TP), TN:TP ratio, and chlorophyll-a (CHL-a). A multi-parameter sensor (YSI Sonde 6600, Environmental Monitoring System, Yellow Springs, OH) was used to measure WT, DO, and EC at onsite. A standard procedure of the Korean Ministry of Environment (MOE, 2000) was followed for the sampling, preservation, and laboratory analysis of organic matter (BOD, COD, and TOC), TSS, nutrients (TN and TP), and CHL-a. TN was defined using the second derived procedure followed by persulfate digestion. A method using ascorbic acid after persulfate oxidation was used to determine the TP. Organic matter (BOD, COD, TOC) and TSS were analyzed using the standard methods of APHA (2005). The TSS was filtered through a glass microfiber filter with 1.2 μm pore size (grade GF/C, Whatman) and determined after drying at 105 °C for 1 h. CHL-a concentrations were measured using a spectrophotometer (DU-530; Beckman Coulter Inc., Brea, CA) after passing the water sample through a GF/C filter and extraction in ethanol (APHA, 2005). Also, monthly precipitation data during the study years was compiled from the Korean Meteorological Administration.

Multi-metric Water Pollution Index (mWPI) as an Integrated Chemical Indicator
The chemical health of the stream was assessed using the multi-metric WPI developed by Kim and An (2015) and modified by Atique and An (2018). The mWPI consists of seven metrics associated with four types of chemical health indicators, and each metric is assigned a score of 5, 3, or 1, based on its concentration level ( Table 1). The overall chemical health of the stream was assessed by total scores of all metrics. Accordingly, chemical health was categorized as excellent (31-35), good (25-29), fair (19-23), poor (13-17), or very poor (7-11).

Fish Sampling
The fish survey at study sites was conducted in two seasonal periods (premonsoon (April-May) and postmonsoon months (September-October)) from 2011 to 2016, and 2019. Fish sampling followed the modified wading method of Korean Ministry of Environment (MOE/NIER, 2008), which derived from the Ohio EPA method (EPA, 1993). The fish samplings at each site lasted for approximately 30 min based on the catch per unit effort (CPUE), but the time differed by stream orders. By using a cast net and kick net, we collected fish data from all microhabitats around study sites, including pools, riffles, and runs. The casting net is one of the oldest and most efficient ways to catch large numbers of small-to medium-sized fish in aquatic systems. In this study, a casting net (mesh size, 5 × 5 mm) was typically used in stream habitats of depth 2-3 m or < 0.5 m. A kicking net (mesh size, 4 × 4 mm) was used in habitats with fast currents and in-stream-vegetation zones, especially near banks. Identifying fish at species level were based on the regional identification of Korean freshwater fish (Kim & Park, 2002).

Ecological Stream Health Based on Fish-Based
Multi-metric Biological Integrity Model (mIBI Model) The mIBI model based on fish communities was applied to assess the stream's ecological health. It is composed of the following eight metrics: M 1 , total number of native fish species; M 2 , number of benthic riffle species; M 3 , number of sensitive species; M 4 , proportion of individuals of tolerant species; M 5 , proportion of individuals of omnivorous species; M 6 , proportion of individuals of native insectivore species; M 7 , total number of native individuals ; and M 8 , proportion of individuals of exotic species . Each metric was assigned a value of 5, 3, or 1, reflecting low, moderate, or high anthropogenic impact on the fish communities in the system, respectively (Simon & Evans, 2017). The expectation of M 1 -M 3 varies with the stream order and differs relative to the size of the stream according to the stream order classification. Overall, the ecological health was determined using the total score of the metrics and ranked as excellent (36-40), good (28-34), fair (20-26), poor (14-18), or very poor (8-13) .

Fish Functional Traits and Group Clustering
Five functional traits based on 15 metrics representing various fish assemblages were used for functional group clustering. The functional traits included trophic traits (T 1 ), tolerance traits (T 2 ), stream-size preference (T 3 ), and vertical position preference (T 4 ), together with disturbance by exotic species (T 5 ) ( Table 2). The fish characteristics have represented the critical ecological functions of fish assemblages, including trophic interaction, species tolerance, and physical habitat preferences (Frimpong & Angermeier, 2010). Trait assignment for each species was based on regional fish identifications (Han et al., 2015;Kim & Park, 2002) and the most prominent global fish database (http:// www. fishb ase. org). The functional group clustering was determined using the dissimilarity distances of functional traits among fish, based on the Gower dissimilarity (Gower, 1971) and  Table 2 The metrics (M 1 -M 15 ) of fish functional traits (T 1 -T 5 ) selected for the functional group clustering Ward's linkage methods (Ward, 1963). The cophenetic coefficient for the quality of the dendrogram was 0.67, indicating a moderate correlation between the original dissimilarity matrix (Gower's distance) and the dissimilarity inferred from the clustering.

Statistical Analysis
Empirical regression models were performed on logtransformed TN, TP, and TN/TP values and CHL-a using SigmaPlot (ver. 14, Systat Software, Inc.). A canonical correspondence analysis (CCA) was also performed to identify relationships between fish functional traits and functional groups (FGs). The CCA is a direct-gradient-ordination method developed to explore unimodal responses of species abundance to environmental variables (Ter Braak, 1986). This analysis was performed using PAST software (ver. 4.03). Clustering to determine fish FGs was performed using the "hclust" function in the "stats" package of R. Moreover, a pearson correlation analysis was performed in R to assess the relationships among water quality parameters, the mIBI model, and the mWPI. Finally, we developed empirical regression models of the FGs, mIBI model, and mWPI.

Overall Trends in Water Quality
The summary statistics of physicochemical variables at the five sites are shown in Table 3. The spatial (site-based) analysis from 2011 to 2019 revealed longitudinal increases in WT, EC, TSS, COD, and CHL-a. Notably, the average EC (233.3 μS/cm), TSS (8.3 mg/L), COD (4.1 mg/L), CHL-a (10.7 μg/L), and BOD (1.4 mg/L) values increased at the last site (W5). The other parameters showed distinct patterns. The average TN in all areas was > 2 mg/L, alluding to an increasing trend from upstream to downstream. However, the average TP was significantly higher (54.27 μg/L) at the last site than at the other sites. In contrast, the average TN/TP ratio was lower (77.89 ± 48.88) at W5 than at the upper sites (W2-W4).

Spatio-Seasonal Variabilities of Physicochemical Parameters
High rainfall during the Asian monsoon (July-August) typically modulated stream water quality, as is the case for most rivers and streams in South Korea. The seasonality of TSS, EC, and organic-matter indicators (BOD and COD) is presented in Figure S1. The average EC decreased at all sites during the high-rainfall months during the monsoon season (July-August), but was stable in premonsoon and postmonsoon. In contrast, average TSS values were the highest at the study sites (16.3-21.9 mg/L) during July and at W5 (19 mg/L) in August. The seasonal patterns of the organic-matter indicators (BOD and COD) differed spatially. Similar seasonal BOD patterns observed among the sites (W1, W2, and W3), with its higher values during monsoon season. However, there was only slight seasonal variation in BOD values at W4. BOD values increased from January to June at the last site. BOD values decreased disproportionately during the monsoon, followed by a gradual decline. COD values of the sites were highest in July, except W5 (Fig. S1). At the last site (W5), COD values increased from January (2.5 mg/L) to June (5.9 mg/L), decreased slightly from August to September (5.3-4.8 mg/L), and declined by almost 50% in December (2.7 mg/L). The seasonality of nutrient and algal biomass indicators is presented in Fig. 2. Average TN values at W1, W2, and W3 tended to decrease from February to June and to increase from September to January, and also increased during the highest-rainfall month (July). At W5, TN levels decreased from March (3.7 mg/L) to September (2.4 mg/L) and then increased until February. By contrast, TP levels at all sites were strongly linked to significant rainfall events during the monsoon season. In contrast, the TN/TP ratio decreased noticeably during these months at all sites. Seasonal CHL-a values showed monomodal patterns at W1 and W3, where values increased from January (1.04 and 1.4 μg/L, respectively) to May (4.9 and 7.19 μg/L, respectively), and then decreased until winter. At W2, CHL-a values increased from January (1.2 μg/L) to July (5.1 μg/L), decreased in August (3.06 μg/L), increased in September (4.46 μg/L), and then declined until December (1.62 μg/L). At W4, CHL-a values increased from January (2.1 μg/L) to April (6.5 μg/L), were stable from May to September (4.4-5.2 μg/L) and decreased from October (2.3 μg/L) to winter.

Empirical Regression Model of Nutrients, TN:TP Ratio, and CHL-a
Spatial empirical models indicated that variations in CHL-a were significantly explained by TP and the TN/TP ratio (Fig. 3), but the p-and R 2 values of the regression models varied longitudinally. TP and the ambient ratio explained 30% and 32% of CHL-a variance (p < 0.05, n = 12), respectively, at the upstream zone (Us). However, the explanatory powers of these two significant variables increased to 52% and 59% at the midstream zone, respectively (p < 0.01, n = 12). At the downstream zone, TP and the ambient ratio explained 88% and 60% of the variance in CHL-a, respectively (p < 0.001, n = 12). Overall, CHL-a variance in the stream was significantly explained by TP (R 2 = 0.60, p < 0.001, n = 60) and the ambient TN/ TP ratio (R 2 = 0.50, p < 0.001). CHL-a values did not exhibit a significant relationship with TN overall or with the three longitudinal zones (p > 0.05). The variations in TN/TP ratios at all spatial scales were more strongly linked to TP values than to TN values.

Spatio-Seasonal Chemical Health Assessment Using Multi-metric Water Pollution Index (mWPI)
We investigated the influence of seasonal monsoon and other water quality stressors on the stream's chemical health based on mWPI (Table 4). The chemical criteria of the ambient TN metric were classified as oligotrophic (3 mg/L), mesotrophic (1.5-3 mg/L), and eutrophic (> 3 mg/L). The stream showed spatially and seasonally mesotrophic conditions based on TN concentrations. Only W5 displayed a eutrophic state in premonsoon. Similarly, based on the TP criteria (< 30, 30-100, and > 100 µg/L), four sites (W1 to W4) exhibited a mesotrophic state during the monsoon, with a shift to oligotrophic conditions in pre-and post-monsoon. By contrast, the status of W5 was mesotrophic in pre-and post-monsoon, and eutrophic in the monsoon season (101.72 µg/L, TP). Similarly, based on the TN:TP criteria (> 50, oligotrophic; 50-20, mesotrophic; < 20, eutrophic), the data showed seasonal and spatial changes between oligotrophic and mesotrophic, with prevailing mesotrophic conditions during the monsoon season at W1, W4, and W5. However, TSS values implied chemical health degradation at all sites during the monsoon season, with the highest average value (19.2 mg/L) at W1. Based on the EC and CHL-a scoring criteria, stream chemical health degraded more spatially than seasonally. CHL-a average values in mid-and up-stream locations revealed a mesotrophic state in pre-monsoon and during the monsoon, which shifted to oligotrophic condition in post-monsoon. The total mWPI scores indicated that the chemical health at upstream (W1) to midstream sites (W2 and W3) was good-excellent in pre-and post-monsoon, but "fair"   during the monsoon season. The chemical health status of W4 was fair both pre-and post-monsoon, whereas that of W5 was also fair pre-and post-monsoon, but poor in monsoon months.

Functional Group and Taxonomical Composition of Fish Communities
We collected 48 species representing 10 families ( were relatively abundant at the last study site (F8). The cluster analysis yielded a dendrogram with five distinct branches, which were termed functional groups (FGs, Fig. 4). Among the functional characteristics that explained the groups (FG-I to FG-V), trophic and tolerance metrics, as well as vertical position preferences, were the most influential. FG-I consisted of omnivore, intermediate, and water-column species (n = 7), whereas FG-II was the cluster of omnivore, tolerant, water column species (n = 15). Also, the FG-II contained two invasive alien species (L. macrochirus, and M. salmoides). FG-III composed of sensitive and riffle-benthic species with insectivores (n = 10) and carnivores (n = 2). FG-IV was the cluster of omnivore, sensitive, and intermediate species (n = 6). The last group (FG-V) consisted of insectivore and intermediate species (n = 8).
Based on relative abundance (RA%), the proportions of functional groups varied more longitudinally than pre-and post-monsoon (Fig. 5). Importantly, FG-III and FG-IV were the most abundant FGs in the stream and the abundance of which varied along downstream gradient. FG-III showed a higher RA% at upstream sites (F1, F2, and F3), but its abundance declined significantly at F5 and was absent in further downstream (F7  (PreMon, pre-monsoon, PostMon, post-monsoon) and F8). In contrast, the RA% of FG-IV increased from upstream to downstream. FG-II also showed a clear longitudinal increase along the downstream gradient.
The other two functional groups (FG-I and FG-V) had no observable variation for their RA% from upstream to downstream, and differed from site to site.

mIBI Model and Its Metrics Correlate with Water Chemistry Parameters
Ecological stream health was assessed for the period of 2011-2016 and 2019 using the mIBI model (Fig. 6). The health status fluctuated between good and excellent at upstream sites (F1, F2, and F3) with no noticeable seasonal variation. The midstream sites (F4 and F6) showed similar ecological health conditions that alternated from fair to good. Another midstream site (F5) had a poor to fair health status. Ecological health fluctuated from very poor to fair at F7, and very poor to poor at the last site during study years. Overall, the mIBI model indicated that the upstream and midstream sites had excellent to fair ecological health status, whereas the downstream sites had a very poor to poor status (Table 6). We performed a Pearson correlation analysis of the relationships among ecological health, water chemical parameters, and chemical health. The mIBI model and its constituent metrics were significantly correlated with EC, organic matter (BOD and COD), TP, CHL-a, and the mWPI ( Figure S2).
The total number of native fish species (M 1 ) showed a significant negative correlation with EC and COD. Together with the number of sensitive species (M 3 ), the number of riffle-benthic species (M 2 ) was highly responsive to variations in chemical water quality and the mWPI. They showed significant negative correlations with the chemical parameters EC, BOD, COD, TP, and CHL-a (p < 0.05). They also exhibited positive associations with the TN/TP ratio and the mWPI. In contrast, the proportion of tolerant individuals (M 4 ) had significant positive correlations with EC, BOD, COD, TP, and CHL-a (p < 0.05). Also, M 4 was negatively related to the TN/TP ratio and the mWPI. The proportion of individuals of omnivorous species (M 5 ) was significantly correlated with TSS values only. The proportion of individuals of native insectivore species (M 6 ) was sensitive to water quality variables and showed a significant negative correlation with WT, EC, organic matter (BOD and COD), and CHL-a and a positive correlation with the mWPI. The proportion of native-species individuals (M 7 ) was significantly and negatively correlated with WT and EC, and positively with the mWPI. The proportion of individuals of exotic species (M 8 ) exhibited a positive correlation with the parameters (EC, COD, TP, and CHL-a) and a significant negative correlation with the mWPI. Accordingly, the ecological stream health based mIBI model was negatively correlated with increased levels of ionic contents, organic matter, nutrients, and primary productivity indicator.

Responses of Functional Characteristics to Chemical Water Quality and Ecological Health
CCA and a regression model were used to evaluate the responses of functional traits and groups to the gradients of chemical water quality and ecological health along the stream (Fig. 7). The first axis of CCA on the responses of most functional traits to water quality parameters was only significant (p < 0.05). For trophic metrics (M 1 -M 3 ), insectivore species (M2) negatively correlated with TSS and TP and positively associated with TN:TP ratio. Among tolerance metrics, sensitive fish (M 4 ) species were negatively associated with EC, BOD, COD, TP, and CHL-a. In contrast, tolerant fish species (M 6 ) showed positive associations with these water chemical parameters. On the first axis of stream size preferences, only upstream species showed negative responses to EC, organic matter, TP, and CHL-a. For vertical position preferences, water column species were positively correlated with EC, organic matter, TP, and CHL-a. But riffle-benthic species tend to be a negative correlation with them.
In the CCA of fish functional groups (FGs) as clustered by the analysis (Fig. 8), the first axis was significant (p < 0.05) and explained 76.5% of the variation. The functional group (FG-II) consisting of omnivore, tolerant, and water column species had positive correlations with EC, BOD, COD, TP, and CHL-a, and from upstream to downstream. However, FG-III with insectivore, sensitive, and riffle-benthic species was negatively linked to the positive gradients of the above parameters.
Moreover, the regression models revealed that FG-II and III responded significantly to ecological health variations in the stream. As shown in Fig. 9, the RA% of FG-II was weakly negatively correlated with the mWPI (R 2 = 0.42, p < 0.001, F = 24.04) and strongly negatively correlated with the mIBI model (R 2 = 0.69, p < 0.001, F = 73.53). The RA% of FG-III showed a weak positive relationship with the mWPI (R 2 = 0.31, p < 0.001, F = 15.14) and a strong positive relationship with the mIBI model (R 2 = 0.66, p < 0.001, F = 63.63).

Discussion
Physicochemical degradation of aquatic environments causes changes in the biological community structure and loss of species diversity, including discharge of wastewaters (domestic, agricultural, and industrial effluents), and cultural eutrophication (Strayer & Dudgeon, 2010). Our study indicated that water quality parameters were locally affected by intensifying of human land-use, and they may have been drivers of the functional variation and biological integrity of stream fish community.
As expected, the site (W5 & F8) in downstream showed deterioration due to land-use impacts (industrial and agricultural activities and WWTP) to chemical water quality. These evidenced of the large increase in organic matter (BOD and COD), total phosphorus (TP), and total suspended solids (TSS). Also, a significant increase in primary productivity indicator (CHL-a) was evident at the last site (10.7 μg/L, eutrophic) that was related to a high availability of nutrients, particularly TP. The TN/TP ambient ratio indicated phosphorus limitation in the stream; a ratio > 32 indicates phosphorus limitation (Steinman & Duhamel, 2017). Seasonally, rainfall  Fig. 4 events during the Asian monsoon (July to August) were the primary influence on chemical water quality, as in prior studies of aquatic ecosystems in South Korea Kim et al., 2018;. COD, TSS, and TP values increased significantly (p < 0.05) during the monsoon season (July-August) at all sites. This indicates that the effects of agricultural, urban, and industrial effluents were intensified by monsoon runoff events. Agricultural runoff from highland fields and cattle farms was evident at W1 and F3, based on the increased TSS and TP values during the monsoon (Park & Shin, 2011). The mWPI showed that water quality status was degraded at sites affected by anthropogenic pollutants from the WWTP, agriculture, and industry, especially during the monsoon and in the downstream area. The chemical condition of the downstream study site was fair pre-and post-monsoon and poor during the monsoon season. Furthermore, spatial and seasonal empirical models approved that TP significantly regulated the variation in CHL-a in the stream, suggesting P-limitation system. Therefore, the potential aspect of water quality management is the nutrient regulation (particularly TP) and anthropogenic activities within the stream drainage.
In accordance with these physicochemical variations, we predicted the possible effect of chemical water quality on the stream fish community and ecological health based on trait-based functional groups (FGs) and multi-metric biological integrity model (mIBI model). In total, 48 fish species were sampled from the study sites. Species richness and abundance increase from upstream to downstream, as is typical for fish communities in river systems (Araujo et al., 2008;McManamay & Jager, 2019) with the increasing scope of the physical habitat availability (Vannote et al., 1980). However, despite relatively low richness at the site F1, richness and abundance did not increase from upstream to downstream. These findings were attributable to the dam. According to the previous study, fish communities in the upper and lower reaches of Goesan Dam differed markedly, based on species turnover rate and spatial autocorrelation (Song et al., 2016). They suggested that the loss of longitudinal continuum and disruption of flow regimes caused by the dam explain the fragmentation of fish communities in the stream. Moreover, the large decline in total fish individuals at W5 & F8 may have been affected by the effluents of wastewaters.
The result was particularly associated with sensitive and endemic species of Korean rivers. For instance, Zacco koreanus, which is sensitive and insectivorous endemic species, was absent at downstream sites. These findings were consistent with the results of previous studies (Choi et al., 2017;Kim et al., 2019).
The upstream and midstream sites showed excellent to fair ecological health status according to the multi-metric biological integrity model (mIBI model). By contrast, downstream sites were in very poor to poor condition. Ecological health was strongly connected with chemical health based on mWPI. Increased nutrients, organic matter, and CHL-a primarily affected ecological health . Moreover, the FGs based on multiple traits also revealed functional changes among the local fish communities and the compositional variation in the groups was more spatially than seasonally. Omnivore fish generally increase with high autochthonous primary production along downstream gradient in temperate streams, whereas insectivorous are tended to be higher abundance in upstream (Pease et al., 2012;Vannote et al., 1980). The groups (FG-III and V) with insectivore species showed, however, different responses to chemical water quality, ecological health and stream downstream gradient because of tolerance levels (intermediate and sensitive) and physical habitat preferences. The FG-III consisted of sensitive, insectivore, and riffle-benthic species longitudinally decreased by abundance and was absent from the last study site, which had poor to fair chemical health. The organic matter (BOD and COD), EC, TP, and CHL-a negatively correlated with the group. Alternatively, the FG-V with insectivore and intermediate species did not show any longitudinal variations and differed from site to site, as well as the FG-I with omnivore species an intermediate species. Also, the identical pattern revealed for the groups (FG-I, II, IV) on omnivore species that tied up with the differences among the cluster species by tolerance levels and habitat preferences. The FG-IV consisted of omnivore and intermediate species showed the increasing trend from upstream to downstream for its relative abundance. The FG-II with omnivore, tolerant, and water column species (included two IAS) showed a high proportion to the fish composition at the W5 and F8. And the group positively correlated with chemical water quality indicators (BOD, COD, EC, TP, and CHL-a), whereas it was negative correlations with ecological and chemical stream health. The present outcomes indicated that multiple trait-based functional grouping can provide more responsible community metrics for the development of ecological modelling and stream health analysis rather than the groups based one trait. Importantly, identifying cluster species that perform well on environmental gradients is crucially dependent on the choice of functional traits (Dumay et al., 2004;Villéger et al., 2017). It is of importance for aquatic conservation and management to predict how specific functional roles will respond to target abiotic factors (Frimpong and Angermeier, 2010;Mouillot et al., 2013). According to our study, the combination of trophic, tolerance metrics, and vertical position preferences can allow for appropriate responses to the heterogeneity of chemical water quality.
Physicochemical downstream degradation due to diverse wastewater effluents entering through lotic systems is one of the leading drivers of the loss of functional integrity of fish communities, stream health  and the proliferation of IAS (Kennard et al., 2005;Kim et al., 2021). Due to their widespread presence, urban and industrial runoffs can result in severe water quality concerns downstream, such as eutrophication (Wurtsbaugh et al., 2019). Therefore, the mitigation of humaninduced pollutants associated with intense rainfall runoffs, particularly suspended solids, nutrients, and organic matter, is required for further stream management to protect and restore the natural condition of the downstream environment. In recent years, increasing attention has been given to the application of nature-based systems (NbS) for alleviating the environmental consequences of major chemical pollutants and heavy metals in aquatic systems (Biswal et al., 2022). A number of studies have shown that the NbS such as constructed wetlands and bioretention systems efficiently eliminate the pollutants (suspended solids, organic matter, and nutrients) from anthropogenic-induced runoffs (Biswal et al., 2022;Choi et al., 2015;Mancuso et al., 2021). Developing NbS could facilitate the improvement of chemical stream health and have a positive effect on the ecological integrity of aquatic communities downstream and human well-being simultaneously.

Conclusions
We analyzed the relationships of biological integrity, fish guilds, and FGs with chemical water health in a large temperate stream. The FGs based on multiple fish traits provided insights into the responses of functional fish assemblages to variations in the chemical quality of the stream and its ecological health. Intensive monsoon rainfall events strongly affected stream water quality and promoted inflow of anthropogenic pollutants and nutrients from the stream watershed. These were evidenced of seasonal and spatial increases of EC, TSS, organic matter (BOD and COD), TP, and CHL-a. The mIBI model implied that ecological stream health status reflected variations in chemical water quality. The assemblage structures of FGs differed more longitudinally than seasonally. The FG-III (sensitive, insectivorous, and riffle benthic) showed a negative correlations with increases in EC, organic matter (BOD and COD), TP, and CHL-a values, while the FG-II (tolerant and lentic preference) showed a positive association with them. Also, the two groups were significantly associated with ecological health status based on mIBI model. It suggested that the combination of trophic, tolerance metrics, and vertical position preferences can allow for appropriate responses to the physicochemical heterogeneity along streams. Overall, trait-based functional grouping can provide insights into environmental integrity, functional responses of fish assemblages to deteriorating water quality, and other modifications of fish habitats, particularly those caused by the installation of dams and weirs. Also, it is vital to protect longitudinal stream connectivity, and the erection of barriers (dams and weirs) should not divide the stream habitat. The increasing number of invasive species must be addressed to conserve the functional diversity of stream fish assemblages.