Partial Weir Opening Is Associated with Shifts in Benthic Diatom Diversity and Assemblage Reorganization in a Monsoonal River

Abstract

Using a coordinated multi-year monitoring dataset collected during the 2020–2024 partial-opening management period, we examined benthic diatom assemblages across the Sejong, Gongju, and Baekje weirs in the Geum River, Republic of Korea. Seasonal surveys at eight stations were used to evaluate spatiotemporal variation in water quality and benthic diatom community structure under this hydrological management regime. Annual basin-wide averages showed gradual interannual changes in water quality, including declines in total phosphorus, total nitrogen, chlorophyll-a, turbidity, and biochemical oxygen demand after 2021, accompanied by increased dissolved oxygen. Diatom community indices based on relative-abundance data showed corresponding temporal variation, with decreased dominance and increased Shannon diversity, evenness, and taxon richness. Ordination analyses indicated gradual differentiation between the earlier (2020–2021) and later (2022–2024) monitoring groups within the study period, whereas random forest models showed limited explanatory power and were treated as exploratory. Overall, the results support benthic diatoms as sensitive descriptors of ecological change in flow-regulated monsoonal rivers while underscoring the value of long-term monitoring where true pre-intervention biological baselines are unavailable.

1. Introduction

Weirs and dams, while historically instrumental in supporting water supply, irrigation, and flood control, have been increasingly recognized for their disruptive impacts on riverine ecosystems [1,2,3]. These structures fragment longitudinal hydrological connectivity, disrupt sediment and nutrient fluxes, and simplify habitat complexity, leading to biodiversity loss and ecological degradation [4,5,6,7,8]. In response, dam removal and weir modification are being implemented globally as strategies to restore ecological integrity and hydromorphological function [9,10,11].

Ecological responses to hydrological interventions are highly context dependent and vary with climate, flow regime, and biological assemblage composition. In temperate regions, previous studies have documented sediment redistribution, biofilm succession, and biotic recolonization following dam removal or flow restoration [12,13,14,15]. In Europe, restoration projects have reported gradual recovery of benthic communities, including macroinvertebrates and periphyton, under re-established flow heterogeneity [16,17]. In contrast, in monsoon-dominated regions such as East Asia, where seasonal hydrological variability is pronounced, empirical evidence on benthic community responses to partial weir opening remains limited, particularly for diatoms [18,19,20,21].

Diatoms are widely recognized as effective sentinels of environmental change due to their rapid responses to variations in hydrology, substrate composition, and water chemistry [22]. Because epilithic diatom assemblages have long been used in freshwater biomonitoring to diagnose hydrological alteration, eutrophication, and habitat disturbance across rivers and streams, they provide an established historical framework for interpreting ecological change rather than a newly introduced indicator group [23,24,25,26,27,28]. Their high taxonomic diversity, ecological specificity, and short generation times make them particularly suitable for tracking community-level responses to river restoration and hydrological modification [23,24,25,29,30]. Accordingly, diatom-based indices are commonly used to detect ecological change and community reorganization, rather than to provide direct measures of ecosystem “status” or restoration success. These indices form a core component of several bioassessment frameworks, including the European Water Framework Directive, and are increasingly applied in Asia and North America [26,27,28,31].

The Geum River, a major monsoonal river in the Republic of Korea, underwent large-scale hydrological alteration during the Four Major Rivers Restoration Project (2009–2011), including the construction of 16 weirs [32]. Among these, the Sejong, Gongju, and Baekje weirs substantially altered flow conditions, habitat structure, and physicochemical regimes in the mid- to lower reaches. Earlier operational adjustments had already occurred at individual weirs, particularly at Sejong Weir, before the coordinated 2020–2024 monitoring window considered here [21,33]. Accordingly, the present study does not constitute a strict before–after comparison with pre-construction or pre-opening epilithic diatom data. Instead, it evaluates assemblage trajectories during a coordinated partial-opening management period and interprets them in relation to published regional studies of Korean river diatom assemblages and previously documented hydromorphological and water-quality changes in the Geum River system [21,34,35,36,37]. This framing has broader relevance for flow-regulated rivers in monsoonal regions where management transitions are staggered and true pre-disturbance biological baselines are rarely available.

To address this knowledge gap, we conducted a multi-year, seasonally resolved biomonitoring program across multiple sites along the Geum River. Specifically, our objectives were to: (1) quantify spatiotemporal variation in benthic diatom community structure during the coordinated partial-opening management period, without assuming direct causal restoration effects; (2) examine statistical associations between diatom diversity patterns and measured environmental variables using multivariate and exploratory machine-learning approaches; and (3) identify period-specific indicator taxa as diagnostic descriptors of assemblage reorganization, while explicitly avoiding inference of full ecological recovery or deterministic restoration success.

2. Materials and Methods

2.1. Study Area and Sampling Design

The Geum River (Geumgang), located in Republic of Korea, extends approximately 400 km from its source in the Sobaek Mountains to its estuary at the Yellow Sea. The river experiences a monsoonal flow regime characterized by pronounced seasonal variability in discharge and turbidity. During the Four Major Rivers Restoration Project (2009–2011), 16 weirs were constructed along the mid- to lower reaches, including the Sejong, Gongju, and Baekje weirs, resulting in substantial hydromorphological and ecological alterations [35,38].

Earlier operational adjustments had already occurred at some individual weirs, particularly Sejong Weir; however, the present study evaluates benthic diatom assemblages under coordinated partial-opening and related hydrological-management conditions across the Sejong, Gongju, and Baekje weirs during the 2020–2024 monitoring period [21,33]. To assess spatiotemporal ecological responses under these conditions, four principal site positions (SJ1, GJ1, GJ2, and BJ1) were established in relation to the three weirs (Figure 1). At each site position, paired sampling locations on the left (L) and right (R) riverbanks yielded eight bank-specific biomonitoring stations in the final sampling design, with bank orientation defined consistently according to downstream flow direction. Surveys were conducted on 11 campaign dates retained in the final analytical matrix: 22 June 2020, 27 April 2021, 9 May 2021, 27 May 2021, 29 September 2022, 25 October 2022, 20 June 2023, 8 August 2023, 10 September 2023, 27 October 2023, and 14 June 2024 (Table S3). Because the final analytical matrix includes only samples retained for analysis, some campaign dates are represented by incomplete site coverage. Peak monsoon periods were mostly excluded to improve comparability and field safety, although selected summer observations were retained when conditions allowed. Current velocity and discharge were not measured at each station during each survey and are therefore acknowledged as important limitations of the present study. Field surveys followed national river bioassessment protocols [39].

To evaluate whether riverbank position influenced community composition, epilithic diatom samples were initially collected from both banks (L and R). Preliminary bank-position comparisons were conducted using Bray–Curtis dissimilarities with PERMANOVA (adonis2) and PERMDISP (betadisper). Because bank position was not treated as a focal explanatory factor in the present study, subsequent analyses focused on temporal and spatial patterns represented in the final unit-sum relative-abundance matrix (Table S2), rather than on explicit left–right bank contrasts. The bank-position test results are summarized together with β-diversity partitioning outputs in Table S1 (see Section 2.4).

2.2. Epilithic Diatom Sampling and Identification

Epilithic diatom samples were collected from cobbles in riffles and shallow runs, targeting representative periphytic communities. Biofilms were brushed from a 5 × 5 cm area using sterilized toothbrushes, suspended in distilled water, and preserved with Lugol’s iodine. Samples were stored at 4 °C until laboratory analysis. The sampled area was standardized by placing a 5 × 5 cm acrylic template directly onto cobble surfaces during each sampling to ensure consistency across sites and years.

Organic material was oxidized using potassium permanganate (KMnO₄) and hydrochloric acid (HCl), and cleaned frustules were mounted on slides with Naphrax^® (Brunel Microscoopes Ltd., Chippenham, UK). At least 400 valves per sample were identified to the lowest possible taxonomic level under 1000× magnification, referencing standard floras [40,41,42,43]. All diatom taxa were verified against AlgaeBase [44], and valid author citations are provided at the first mention of each species name.

For all comparative and statistical analyses, diatom composition is expressed as relative abundances (%) derived from standardized counts, following established bioassessment practices that emphasize reproducibility and comparability across sites and years. Community attributes were assessed using these relative abundance data to calculate dominance [45], Shannon diversity [46], species richness [47], and Pielou’s evenness [48]. All indices were analyzed in their original, non-transformed form. The transformations applied to species abundance matrices for multivariate analyses are described in Section 2.4.

2.3. Water Quality Measurements and Nutrient Analysis

In situ water-quality variables, including temperature (°C), pH, dissolved oxygen (DO, mg L⁻¹), electrical conductivity (EC, µS cm⁻¹), and turbidity (NTU), were measured using a YSI ProDSS multiparameter probe (YSI Inc., Yellow Springs, OH, USA; accuracy: ±0.1 °C for temperature, ±0.2 mg L⁻¹ for DO, and ±1% for EC). Calibration was performed prior to each field campaign following the manufacturer’s protocols. Water samples for physicochemical and nutrient analyses were collected on the same dates and at the same bank-specific biomonitoring stations as the diatom samples during each field campaign. Water samples were collected in sterile polyethylene bottles, immediately refrigerated, and transported to the laboratory within 24 h for chemical analyses.

Nutrient concentrations were determined according to standardized methods [39] and corresponding ISO protocols to ensure comparability across sites and sampling periods: nitrate nitrogen (NO³⁻-N; ISO 7890-3:1988 [49]), ammonium nitrogen (NH⁴⁺-N; ISO 7150-1:1984 [50]), total nitrogen (TN; ISO 11905-1:1997 [51]), orthophosphate phosphorus (PO₄³⁻-P; ISO 6878:2004 [52]), and total phosphorus (TP; ISO 15681-2:2003 [53]). Biochemical oxygen demand (BOD5) was determined using the dilution method (ISO 5815-1:2003 [54]). Chlorophyll-a (Chl-a) was measured spectrophotometrically using a Hitachi U-2900 spectrophotometer (Hitachi, Tokyo, Japan) following acetone extraction, with an analytical precision of ±0.5 µg L⁻¹. All analyses were performed in triplicate for quality control.

To facilitate comparison with diatom assemblage metrics based on relative abundances, nutrient and water-quality variables were treated as standardized continuous descriptors in multivariate analyses (Section 2.4). Because these measurements were campaign-based rather than high-frequency antecedent time series, they are interpreted here as contemporaneous environmental descriptors rather than exhaustive representations of short-term hydrological or nutrient dynamics. For visualization in Figure 2, annual mean values across the eight bank-specific biomonitoring stations were averaged to provide a basin-scale descriptive overview. Each parameter was normalized within its annual range (0–1) to emphasize relative interannual variability rather than absolute magnitudes. Detailed site-level spatiotemporal variability is presented in Figure S1.

2.4. Statistical Analyses and Ecological Modeling

All statistical analyses were conducted in R (v4.3.1) [55] and PAST (v4.11) [56]. All community-level analyses were based exclusively on relative-abundance data; absolute density values were not used in any statistical tests.

Raw diatom valve counts were standardized on a per-sample basis to generate a unit-sum relative-abundance matrix, in which the relative abundances of taxa within each sample sum to 1.0. This matrix (Table S2; see the Supplementary Analytical Framework) served as the sole input for all community-level analyses, thereby ensuring comparability among samples and consistency with the assumptions of distance-based and diversity analyses.

Transformations were applied only when required by the analytical framework. For ordination methods based on Euclidean geometry, including principal component analysis (PCA) and distance-based redundancy analysis (dbRDA), the relative-abundance matrix was Hellinger-transformed [57]. In contrast, Bray–Curtis-based analyses, including non-metric multidimensional scaling (NMDS), β-diversity partitioning, PERMANOVA, PERMDISP, and indicator species analysis, were conducted on non-transformed relative-abundance data. Diversity indices, including Shannon diversity, richness, evenness, and dominance, were calculated exclusively from non-transformed relative abundances. Log(x + 1) transformation was applied only for visualization and was not used as input for statistical testing.

Differences in Shannon diversity (H′), richness, evenness, and dominance between the earlier monitoring period (2020–2021) and the later monitoring period (2022–2024) were evaluated after testing for normality [58]. These temporal groupings represent analytical monitoring phases within the study window rather than true pre-intervention and post-intervention baseline states. Paired t-tests were used when parametric assumptions were satisfied; otherwise, Wilcoxon signed-rank tests were applied [59]. Effect sizes (Cohen’s d or r) were reported alongside p-values, and false discovery rate (FDR) correction [60] was applied to account for multiple comparisons.

Community composition was further examined using multivariate ordination. NMDS based on Bray–Curtis dissimilarities was used to summarize compositional patterns, with stress values around 0.3 interpreted as indicating moderate fit. PCA was applied to Hellinger-transformed species data and standardized environmental variables to identify major gradients, and environmental vectors were fitted using envfit() to aid interpretation. dbRDA was used to constrain assemblage composition by measured environmental variables, and the resulting ordination is presented in Figure S1.

β-diversity was partitioned into turnover and nestedness components using the betapart package [61], thereby quantifying the relative contributions of species replacement and nested species loss or gain. Group differences in community structure were evaluated using PERMANOVA and PERMDISP [62,63]. SIMPER analysis was used to identify taxa contributing most strongly to between-group dissimilarities [64]. In addition, site-level mean Bray–Curtis dissimilarities were correlated with local Shannon diversity (H′) to assess descriptive associations between compositional uniqueness and diversity.

Indicator species analysis (ISA) was conducted using the indicspecies package [65] with 999 permutations to identify taxa significantly associated with each temporal group.

Random forest (RF) regression and generalized additive models (GAMs) were used as complementary exploratory tools. RF regression (500 trees) was applied to summarize associations between campaign-based water-quality variables and Shannon diversity, rather than to infer primary environmental drivers. Variable importance was assessed using permutation-based changes in R², and partial dependence plots were used only to visualize non-linear patterns descriptively. GAMs with cubic regression splines (mgcv package) [66] were used to characterize non-linear temporal trends in diversity indices.

3. Results

3.1. Spatiotemporal Trends in Water Quality and Diatom Community Indices

Annual basin-wide means of water-quality parameters exhibited clear interannual variation across the eight monitoring sites during the study period (2020–2024) (Figure 2; Supplementary Figures S1 and S2). Total phosphorus (TP), total nitrogen (TN), and chlorophyll-a concentrations declined after 2021, while biochemical oxygen demand (BOD₅) and turbidity (NTU) also decreased. Dissolved oxygen (DO) increased over the same period, whereas electrical conductivity (EC) showed moderate interannual fluctuations. Because the water-quality dataset was collected at campaign intervals rather than at high frequency, these patterns are interpreted as basin-scale descriptive context for the biological results rather than as direct evidence of short-term drivers of diatom change.

Benthic diatom community indices calculated from relative-abundance data also varied among years (Figure 3). Mean dominance values decreased after 2021, whereas Shannon diversity (H′), Pielou’s evenness (J′), and Margalef richness (e′) showed higher mean values in 2022–2024 compared with 2020–2021. Interannual variability of all indices is summarized as mean ± SD, and no ecological interpretation is included at this stage (Figure 3).

3.2. Temporal Dynamics of Diatom Diversity and Community Dissimilarit

Mean Community Dissimilarity Index (CDI) values increased from 0.61 (2020) to 0.77 (2024), with the largest increase occurring between 2021 and 2022 (Figure 4). CDI values during 2022–2024 were higher than those during 2020–2021 (Wilcoxon signed-rank test, p < 0.01). Mean Shannon diversity (H′) also varied among years, with higher values observed in 2022 and 2024 (Figure 4).

β-diversity partitioning based on the Baselga framework [47] indicated that species turnover accounted for 72% of total dissimilarity, whereas nestedness accounted for 28% (Table S1). These values summarize the relative contributions of turnover and nestedness to compositional variation across the study period.

3.3. Ordination Analysis and Environmental Gradients

Non-metric multidimensional scaling (NMDS) based on Bray–Curtis dissimilarities showed partial separation between assemblages sampled during 2020–2021 and those sampled during 2022–2024 in ordination space (Figure 5a). The NMDS solution had a stress value of 0.32. Assemblages from 2022–2024 occupied a broader ordination space, consistent with greater dispersion in later years.

Principal component analysis (PCA) of environmental variables indicated that the first two axes explained 26.2% (PC1) and 23.3% (PC2) of total variance, respectively (Figure 5b). Nutrient- and turbidity-related variables (TP, TN, chlorophyll-a, NTU) exhibited high loadings along the major gradients. Distance-based redundancy analysis (dbRDA) yielded comparable gradient structures (Figure S3).

3.4. Exploratory Associations Between Diatom Diversity and Campaign-Based Water-Quality Descriptors

Random forest regression explained very little variation in Shannon diversity (out-of-bag R² = −0.03; RMSE = 0.42; Figure 6a), and the resulting variable rankings should therefore be interpreted only as exploratory summaries of association. Permutation-based variable importance ranked chlorophyll-a, dissolved oxygen, biochemical oxygen demand, and pH as measured variables with relatively higher importance, whereas total nitrogen and turbidity showed lower importance. Given the low predictive performance of the model and the campaign-based frequency of the environmental dataset, these results are interpreted only as exploratory associations rather than as robust evidence for the primary environmental correlates of diatom diversity.

Partial dependence analysis indicated a weak, non-linear relationship between chlorophyll-a concentration and predicted Shannon diversity, with higher predicted values at intermediate concentrations (Figure 6b). This pattern is shown for descriptive completeness only and should not be overinterpreted mechanistically.

3.5. Indicator Taxa Associated with Earlier and Later Monitoring Periods

Indicator species analysis (ISA; 999 permutations) identified five taxa with indicator values > 0.6 (p < 0.01). Navicula gregaria, Stephanodiscus hantzschii, and Achnanthidium gracillimum were associated with the earlier monitoring period (2020–2021), whereas Pseudostaurosira elliptica and Cyclotella choctawhatcheeana were associated with the later monitoring period (2022–2024). The mean relative abundances (±SD) of these taxa differed between the two monitoring periods, and their corresponding indicator values are presented in Figure 7.

4. Discussion

4.1. Hydrological Change and Ecological Responses

During the coordinated partial-opening management period, benthic diatom assemblages exhibited gradual and measurable changes in community structure rather than abrupt shifts. Diversity-related indices, including Shannon diversity, richness, and evenness, tended to increase during the later monitoring period, whereas dominance declined. These temporal patterns coincided with partial re-establishment of flow connectivity; however, they do not imply direct causality. Instead, the observed responses are more appropriately interpreted as progressive community reorganization rather than immediate or complete ecological recovery, thereby avoiding overstatement of system resilience.

Similar associations between enhanced hydraulic variability and benthic community responses have been reported in other river restoration studies, where increased flow heterogeneity and substrate diversity were linked to gradual changes in periphytic and macroinvertebrate assemblages [10,16,67]. Comparable patterns have also been documented in Korean rivers influenced by weir regulation and subsequent flow modification [21,37]. Collectively, these studies indicate that hydrological interventions are often associated with slow, multi-year ecological adjustment rather than rapid system-wide recovery. The magnitude and timing of responses are known to depend strongly on local flow regimes, geomorphic context, and disturbance history [67,68].

4.2. Broader Implications, Historical Context, and Baseline Constraints

The temporal trajectories observed in this study align broadly with international river restoration literature while also highlighting an important interpretive constraint: a site-specific pre-construction or pre-opening epilithic diatom baseline is not available for the exact reaches studied here. For that reason, the present results are interpreted against two contextual benchmarks: (i) published regional work showing how benthic diatom assemblages respond to environmental gradients in Korean rivers [37], and (ii) prior studies documenting hydromorphological and physicochemical changes associated with weir operation in the Geum River system [21,34,35,36]. Viewed in that context, the observed assemblage shifts are informative not because they reconstruct a pristine historical state, but because they show how diatom communities reorganize under partially reconnected, still-managed monsoonal river conditions.

These contrasts suggest that while general restoration principles may be shared across regions, the pace and direction of community change are strongly modulated by climatic and hydrological regimes. This underscores the importance of regionally calibrated assessment frameworks that explicitly account for hydrological variability and disturbance intensity [69,70,71]. Diatom-based indices remain useful for tracking temporal ecological change, but direct cross-regional comparisons should be approached cautiously due to differences in species pools, disturbance regimes, environmental context, and baseline availability [24,27].

4.3. Exploratory Associations with Measured Environmental Variables

Random forest analyses identified chlorophyll-a, dissolved oxygen, biochemical oxygen demand, and pH as relatively influential variables; however, the very low explanatory power of the models indicates that these relationships are exploratory rather than predictive. In addition, because the water-quality data were collected only during each sampling campaign rather than as higher-frequency antecedent series, these variables should be interpreted as contemporaneous descriptors of site conditions rather than as definitive drivers of diatom diversity. Accordingly, the RF results are interpreted as descriptive associations rather than evidence for environmental control over diatom diversity.

The weak, non-linear association between chlorophyll-a concentration and Shannon diversity—characterized by higher diversity at intermediate concentrations—resembles patterns often discussed in the context of the intermediate disturbance hypothesis. However, this similarity should not be interpreted as mechanistic support for that hypothesis in the present system [72,73]. In flow-regulated rivers, hydrological variability, substrate heterogeneity, and disturbance frequency are likely to exert stronger influence on benthic diatom assemblages than water-quality variables alone [74,75,76].

Future modeling efforts would benefit from integrating physical habitat descriptors, hydrological metrics, and biological trait information to better characterize drivers of community dynamics and recovery trajectories following hydrological interventions [77,78,79].

4.4. Indicator Taxa and Ecological Trajectories

Indicator species analysis identified taxa that were statistically associated with the earlier and later monitoring periods, serving as diagnostic descriptors of assemblage composition rather than quantitative measures of species turnover. Earlier-period assemblages (2020–2021) were characterized by Navicula gregaria, Stephanodiscus hantzschii, and Achnanthidium gracillimum, whereas later-period assemblages (2022–2024) were associated with Pseudostaurosira elliptica and Cyclotella choctawhatcheeana.

These temporal shifts in characteristic taxa coincided with changes in hydrological and nutrient conditions, but they do not constitute direct evidence of deterministic species replacement or improved ecological status. Similar transitions from eutrophic- or low-flow-associated taxa toward more flow-associated diatoms have been reported in Korean rivers [37] and in European restoration contexts [16,80].

4.5. Methodological Considerations and Limitations

Several methodological limitations should be acknowledged. First, no site-specific epilithic diatom dataset was available from before weir construction or from a fully comparable pre-opening phase, which limits direct comparison with a historical “normal” community state. Second, current velocity and discharge were not measured at each station and survey, and the water-quality dataset represents campaign-based observations rather than high-frequency antecedent conditions. These constraints likely reduced our ability to resolve short-term hydraulic drivers of community change and should be kept in mind when interpreting water-quality associations.

Third, exclusion of peak monsoon periods—implemented to minimize flood-related sampling constraints—may have reduced sensitivity to the short-term disturbance and recovery dynamics characteristic of monsoonal rivers [37,81]. Finally, taxonomic diversity indices alone cannot fully capture functional, productivity-related, or ecosystem-level responses. Future restoration assessments would benefit from multi-metric approaches integrating functional traits, molecular tools, and ecosystem process indicators [78,81,82,83].

5. Conclusions

This study shows that partial weir opening was temporally associated with measurable changes in benthic diatom diversity and community composition, expressed as gradual shifts in assemblage structure over multiple years within a coordinated monitoring period. Using multivariate ordination, random forest analysis, and indicator species analysis, we identified statistically supported patterns of community variation that coincided with partial hydrological reconnection, without implying direct causal restoration effects.

Shannon diversity increased by approximately 25%, dominance declined by ~32%, and taxon richness increased by ~33% during the later monitoring period based on unit-sum relative-abundance data. These changes are most appropriately interpreted as evidence of progressive community reorganization under changing hydrological-management conditions, rather than as proof of complete ecological recovery or enhanced resilience. The observed trajectories suggest that partial re-establishment of longitudinal connectivity may be associated with gradual ecological adjustment in flow-regulated rivers, particularly where full dam removal is not feasible and true historical biological baselines are unavailable.

From a management perspective, benthic diatom assemblages provide a sensitive and cost-effective means of tracking ecological change when interpreted as descriptors of community structure rather than as direct measures of ecological status or restoration success. However, limitations include exclusion of peak monsoon periods, reliance on taxonomic and relative-abundance metrics, and the limited explanatory power of water-quality-based models.

Future restoration assessments would benefit from integrating hydrological metrics, habitat-scale descriptors, functional or trait-based indicators, and emerging molecular approaches. Overall, this study indicates that partial weir opening can be associated with measurable ecological change in monsoonal rivers, while underscoring the need for cautious interpretation and regionally calibrated, long-term monitoring frameworks.