# Analysis of Environmental Factors Associated with Cyanobacterial Dominance after River Weir Installation

^{1}

^{2}

^{*}

## Abstract

**:**

_{3}–N, NH

_{3}–N, total nitrogen (TN), total phosphorous (TP), PO

_{4}–P, chlorophyll–a, Fe, total organic carbon (TOC), and SiO

_{2}content, along with biological and chemical oxygen demands. The results indicate that site-specific environmental factors contributed to the cyanobacterial dominance for each weir. In general, the physical characteristics of EC, APRCP7, Q7day, Temp, and ΔT were the most important factors influencing cyanobacterial dominance. The EC was strongly associated with cyanobacterial dominance at the weirs because high EC indicated persistent low flow conditions. A minor correlation was obtained between nutrients and cyanobacterial dominance in all but one of the weirs. The results provide valuable information regarding the effective countermeasures against cyanobacterial overgrowth in rivers.

## 1. Introduction

_{2}, CO

_{3}

^{2}

^{−}, and HCO

_{3}

^{−}) of cyanobacteria and affect their growth [5,29,30]. The pH has a dominant influence on the various forms of dissolved inorganic carbons (CO

_{3}

^{2}

^{−}, HCO

_{3}

^{−}, and H

_{2}CO

_{3}) employed in photosynthesis, and also has a high correlation with the C sources of phytoplankton [31,32,33].

## 2. Materials and Methods

#### 2.1. Site Description

^{3}), 14,248 (59 million m

^{3}), 15,074 (70 million m

^{3}), and 20,697 km

^{2}(101 million m

^{3}), respectively [54]. The CHW is the most downstream weir of the Nakdong River. Since the weirs were completed in 2012, algal blooms have occurred frequently due to cyanobacteria overgrowth in summer. The highest level of the harmful cyanobacteria cell density guideline of the World Health Organization (WHO), i.e., 20,000 cells mL

^{−1}, has been exceeded 36 (peak: 261,219 cells mL

^{−1}), 69 (peak: 495,360 cells mL

^{−1}), 95 (peak: 1,264,052 cells mL

^{−1}), and 80 times (peak: 715,993 cells mL

^{−1}) at GGW, DSW, HCW, and CHW respectively [55]. The overgrowth of harmful cyanobacteria in the Nakdong River has emerged as a social issue because this river is an important water source for 13 million citizens living in the Daegu, Busan, and Gyeongnam regions.

#### 2.2. Sampling and Analysis

^{−1}), and electrical conductivity (EC, µS cm

^{−1}) were measured at different depths at each site using a multi-item water quality meter (YSI-EXO, YSI-6600, YSI Pro plus). The pH, DO, and EC sensors were calibrated every week. The collected samples were stored at ≤ 4 °C and transported to the laboratory, where the properties (excluding water temperature, DO, EC, and pH) were assessed in accordance with the Standard Methods for the Examination of Water Pollution [56].

#### 2.3. Statistical Analyses

_{3}–N, and NO

_{3}–N. The organic matter and trace materials included the biological and chemical oxygen demands (BOD and COD, respectively), total organic C (TOC), Fe, and SiO

_{2}.

_{3}–N, NH

_{3}–N, TN, PO

_{4}–P, and Fe. The C.dominance represents the ratio of the cyanobacteria cell density to the total algae, indirectly representing the risk of algal blooms due to cyanobacterial dominance. The SMLR model excludes independent variables with low statistical significance in a step-by-step manner to create the multiple linear regression model with the best prediction performance. In this study, the stepwise regression method with forward selection was applied to the SMLR. For the first variable used in the model, the independent variable with the largest positive or negative correlation with the dependent variable was selected, and the independent variables with strong correlations were applied sequentially. The procedure was terminated if no variable satisfied the entry criteria [20]. For the model evaluation, the adjusted coefficient of determination (Adj. R

^{2}), root mean square error (RMSE), Mallows’ C

_{P}statistic, and Akaike information criterion (AIC) were used. The SMLR results were used to determine the important variables to be applied to the selection of a parsimonious multiple regression model.

_{3}–N, NH

_{3}–N, TN, PO

_{4}–P, and Fe.

_{3}–N, NH

_{3}–N, TN, TP, PO

_{4}–P, Chl–a, Fe, BOD, COD, TOC, and SiO

_{2}. For determination of the number of principal components, only the principal component axes having eigenvalues of 1.0 or higher—which represent the variation of the data normalized to the principal component axis—were considered [62,63,64]. In PCA, the factor axis is rotated by simplifying the factor pattern structure to facilitate the factor analysis. The factor axis rotation methods include oblique and orthogonal rotation. The orthogonal rotation methods include Equimax, Varimax, and Quartimax [65]. The Varimax rotation method was applied in this study. Furthermore, to determine whether the raw data were appropriate for PCA, Bartlett’s test of sphericity and the Kaiser-Meyer-Olkin (KMO) test were performed. Bartlett’s test of sphericity is performed under the hypothesis that the correlation matrix of the data used for analysis is an identity matrix. When this hypothesis is validated (p > 0.05), the data are inadequate for use in the PCA analysis. The KMO test result is a measure of the covariance among the factors inherent in the variables and data used for the analysis. The closer the result is to 1, the higher the analysis validity. The analysis can be performed only when this value is 0.5 or higher [63]. The analysis was performed after removing the variables with KMO values lower than 0.5 for each weir. The KMO test results showed that all the variables for HCW satisfied the KMO reference value (0.5 or higher). However, the KMO reference value was not satisfied by the Chl–a, BOD, NH

_{3}–N, and DO variables (four variables) for GGW; by DO, Temp, △T, PO

_{4}–P, NH

_{3}–N, Fe, SiO

_{2}, NO

_{3}–N, APRCP7, Q7day, and EC (11 variables) for DSW; and by Temp and TOC (two variables) for CHW. Thus, these variables were excluded from the analysis. From the analyses performed with the selected variables, all variables for all weirs satisfied the reference value (GGW: KMO value = 0.73, p < 0.05; DSW: KMO value = 0.57, p < 0.05; HCW: KMO value = 0.64, p < 0.05; and CHW: KMO value = 0.64, p < 0.05).

## 3. Results and Discussion

#### 3.1. Descriptive Statistics of Collected Data

^{−1}, the EC was 237.5–318 µS cm

^{−1}, the SS was 8.66–10.1 mg L

^{−1}, and the TN was 2.18–2.83 mg L

^{−1}. The strongest thermal stratification during the survey period was recorded at GGW. The DSW exhibited the highest pollution with a BOD of 2.23 (±2.85) mg L

^{−1}indicating a high organic matter loading, and a TP of 0.087 (±0.1) mg L

^{−1}indicating high nutrient concentrations, due to the effect of the inflow from a polluted tributary, the Geumho River. The highest Chl–a concentration was measured at DSW, at 36.5 (±162.7) mg m

^{−3}. The highest cyanobacterial cell densities were found at HCW and DSW at 10,761 (maximum: 453,283) cells mL

^{−1}and 9,236 (maximum: 694,667) cells mL

^{−1}, respectively.

#### 3.2. Correlation Analysis of Environmental Factors

_{2}and diatom cell density, which seems to have been because the SiO

_{2}concentration was sufficiently high and did not act as a limiting factor of diatom growth. According to Wetzel [35], SiO

_{2}contributes to the seasonal transition of diatoms and other algae species, and the diatoms are weaker in terms of food competition than other algae species in a water body with a SiO

_{2}concentration of approximately 0.5 mg L

^{−1}or lower. In the present study, the average SiO

_{2}concentrations were very high, at 4.77 (0.04–11.38), 4.02 (0–10.18), 3.53 (0.05–9.87), and 2.77 (0–9.85) mg L

^{−1}at GGW, DSW, HCW, and CHW, respectively.

^{−1}, TN = 0.3 mg L

^{−1}) in every weir. No correlation was found between Chl–a and TP, except at DSW (r = 0.702, p < 0.05). Similarly, no correlation was found between Chl–a and TN, except at DSW (r = 0.728, p < 0.05). In general, the algal biomass (Chl–a) exhibits a high correlation with TP in stagnant waters such as those of reservoirs and lakes; when the P supply is blocked by the photic zone in the reservoir surface layer, algal growth is P-limited [23,24]. However, unlike in reservoirs, nutrients are continuously supplied to the studied river from the watershed and therefore maintain a high concentration. Thus, algal growth and P concentration have a low correlation because the P-limitation on algal growth occurs only in periods when a low flow rate is sustained. It is difficult to define why algal growth was TP-limited only at DSW, however, the high TP variability (Table 1) could provide an explanation.

#### 3.3. Selection of Important Environmental Factors Associated with Cyanobacterial Dominance

#### 3.3.1. Step-Wise Multiple Linear Regression (SMLR)

^{2}, RMSE, Mallows’ C

_{P}statistics, and AIC values of the model according to the independent variables are also presented in Table 2.

_{3}–N, and TN as independent variables; the variability of C.dominance, (i.e., the dependent variable) was reproduced at 47.6%. Among the 12 models considered for DSW, the best performance was obtained for the model taking NO

_{3}–N, EC, TN, Temp, NH

_{3}–N, and Q7day as independent variables; the C.dominance variability was reproduced at 52.1%. The best performance for HCW as obtained for the model taking EC, NO

_{3}–N, Temp, TN, Q7day, Chl–a, APRCP7, and Fe as independent variables; the C.dominance variability was reproduced at 72.9%. Among the 12 models considered for CHW, the best performance was obtained for the model taking NO

_{3}–N, EC, APRCP7, Q7day, and TN as independent variables; the C.dominance variability was reproduced at 53.8%. The Adj.R

^{2}value of the multiple regression model selected at each weir ranged from 0.476 to 0.729, and the variability of the cyanobacterial dominance ratio was not reproduced sufficiently. These findings are interpreted as indicating that the cyanobacterial dominance environment is established through a combination of complex nonlinear relationships involving many variables, and cannot be sufficiently explained by a parametric regression model that assumes a linear combination of variables.

#### 3.3.2. Recursive Feature Elimination Based on Random Forest Model (RFE-RF)

_{4}–P, and APRCP7) were used (Table 3). For DSW, the lowest RMSE of 0.156% was also obtained using seven variables (EC, TOC, Temp, TP, TN, Q7day, and ΔT). For HCW, the lowest RMSE of 0.145% was obtained when eight variables (EC, ΔT, Temp, Q7day, APRCP7, TOC, Fe, and PO

_{4}–P) were used. For CHW, the lowest RMSE of 0.160% was obtained when five variables (EC, TN, APRCP7, ΔT, and Q7day) were used.

^{2}values compared with the SMLR model for all weirs (GGW: 0.910, DSW: 0.932, HCW: 0.949, and CHW: 0.902). This finding suggests that the RF model—which is a non-parametric method—reproduces the complex nonlinear correlations of many variables affecting the cyanobacterial dominance environment better than the SMLR model, which is a parametric method. Therefore, use of the RFE-RF method for the variable importance evaluation related to cyanobacterial dominance is considered a useful approach.

^{−1}and increased at higher concentrations. Furthermore, it increased in proportion to TP until the level of 0.2 mg L

^{−1}was reached; then, it remained constant. For HCW, the C.dominance increased with higher EC, ΔT, Temp, Fe, and APRCP7, and decreased with higher Q7day. In particular, the C.dominance increased sharply at a ΔT of 1 °C or higher. For CHW, the C.dominance increased with higher EC and when ΔT was 1 °C or higher, and decreased with higher APRCP7, Q7day, or TN concentrations.

_{4}–P in an environment with high C.dominance results from cyanobacterial overgrowth. Therefore, accurate interpretation of the data mining analysis results requires expert analysis and judgment.

#### 3.4. Characterizing Environmental Conditions of High Cyanobacterial Dominance

#### 3.4.1. Decision Tree Analysis

^{−1}, the average C. dominance was 81% in 9 out of 108 data points in total (8%). The EC was also the most important variable for DSW. In that case, when EC was ≥ 336 µS cm

^{−1}and TP was ≥ 0.082 mg L

^{−1}, the average C.dominance was 77% in 8 out of 108 data points in total (7%). When EC was ≥ 336 µS cm

^{−1}, TP and TN were less than 0.082 and 2.4 mg L

^{−1}, respectively, and the average C.dominance was 55% in 9 out of 108 data points in total (8%). For HCW, EC and ΔT were important variables regarding environmental conditions for cyanobacterial dominance. The environmental conditions for a high cyanobacterial share were as follows—when EC was ≥ than 443 µS cm

^{−1}, the average share was 87% in 10 out of 96 data points in total (10%). When EC was < 443 µS cm

^{−1}and ≥ 325 µS cm

^{−1}, and when ΔT was ≥ 0.7 °C, the average share was 54% in 11 out of 96 data points in total (11%). For CHW, EC, TN, and Q7day were found to be important environmental variables for with the growth of a high cyanobacterial population. When EC was ≥ 385 µS cm

^{−1}at CHW, the average share was 71% in 9 out of 93 data points in total (10%). When EC was < 385 µS cm

^{−1}, TN was < 2 mg L

^{−1}, and Q7day was < 167 m

^{3}s

^{−1}, the average share was 52% in 14 out of 93 data points in total (15%).

#### 3.4.2. Principal Component Analysis

^{−1}or higher. The principal axes with eigenvalues of 1.0 or higher in the PCA results were extracted as the main components. Hence, five, two, five, and five principal components with eigenvalues of 1.0 or higher were selected for GGW, DSW, HCW, and CHW, respectively. The analysis showed that for GGW, the first, second, third, fourth, and fifth principal components contributed 31.9%, 18.8%, 9.6%, 8.2%, and 6.7%, respectively, to the total water quality change. For DSW, the first and second principal components contributed 44.3% and 21.7%, respectively, and for HCW, the first, second, third, fourth, and fifth principal components contributed 21.9%, 20.1%, 15.8%, 9.3%, and 7.4%, respectively, to the total water quality change. For CHW, the first, second, third, fourth, and fifth principal components contributed 25.2%, 18.9%, 12.6%, 8.2%, and 6.1%, respectively, to the total water quality change.

^{−1}), warning (1,000–10,000 cells mL

^{−1}), and alarm (higher than 10,000 cells mL

^{−1}) levels.

_{2}, Q7day, TN, TP, NO

_{3}–N, COD, and TOC have the greatest effect on the positive direction of PC1. As apparent from the bi-plot for PC1 and PC2, APRCP7, Q7day, SiO

_{2}, and COD form the same cluster in the opposite direction of pH (Figure 8a). Furthermore, the APRC7 and Q7day values for the summer can be divided into large and small values. The cyanobacteria tended to dominate when the precipitation and flow were low. The C.dominance vector has the same direction as EC, ΔT, and Temp, thus explaining the high EC, strong thermal stratification, and high cyanobacterial share at high temperatures (Figure 8a). The data corresponding to a cyanobacterial cell density higher than 10,000 cells mL

^{−1}mainly appear under high-EC conditions, strong thermal stratification, and a high cyanobacterial share. The results are in a good agreement with the findings in the lowland rivers of south-eastern Australia [10,11,13], where a persistent low flow and the formation of thermal stratification caused the proliferation of cyanobacteria.

^{−1}appear under the conditions of high TOC, pH, Chl–a, and cyanobacterial share.

_{4}–P, and SiO

_{2}have the greatest effect on the positive direction of PC1, while pH, DO, and Chl–a have the greatest effect on the negative direction. C.dominance has the largest effect on the PC2 axis, whereas APRC7 and Q7day have the largest effects on the PC1 and PC3 axes. The PC1 axis corresponds to conditions of large precipitation and flow, which mean inward flow of nonpoint pollutant sources; this explains the high TP, PO

_{4}–P, Fe, and SiO

_{2}concentrations. From the examination of the bi-plot on the PC1–PC2 plane (Figure 8c), the cyanobacterial share loading vector of HCW has the same direction as EC, Temp, and ΔT, and the opposite direction to the NO

_{3}–N and TN vectors. This explains the high C.dominance under the conditions of high EC, water temperature, and stratification strength. On the PC2–PC5 plane, the C.dominance vector has the same cluster direction as EC, ΔT, BOD, Chl–a, Temp, and NH

_{3}–N. The alarm data (corresponding to a cyanobacteria cell density higher than 10,000 cells mL

^{−1}) appeared when these variables were high.

_{2}, Q7day, APRCP7, PO

_{4}–P, and Fe have the greatest effect on the positive direction of PC1, while pH and DO have the greatest effect on the negative direction of PC1. On the PC2 axes, EC, C.dominance, Fe, TP, and NH

_{3}–N have the greatest effect on the positive direction, while NO

_{3}–N and TN have the greatest effect on the negative direction. In the bi-plots (Figure 8e,f), the C.dominance has the largest effect on the PC2 axis, whereas APRC7 and Q7day have the largest effects on the PC1 and PC4 axes. On the PC1–PC2 plane, the APRC7 and Q7day vectors form clusters in the same direction as PO

_{4}–P, SiO

_{2}, TP, and Fe, and in the opposite direction to pH, Chl–a, C.dominance, △T, BOD, and DO (Figure 8e). In the CHW bi-plots, the alarm data corresponding to a cyanobacterial cell density higher than 10,000 cells mL

^{−1}appeared for conditions where the cyanobacterial share, EC, △T, and Fe were high (Figure 8e,f).

#### 3.5. Integrated Analysis of Environmental and Control Variables for Cyanobacterial Dominance at Each Weir

_{3}–N concentration, and the NO

_{3}–N concentration is high in spring and autumn, when the base outflow is more dominant than in summer (when rainfall is concentrated). However, care should be taken in interpreting this result, because NO

_{3}–N is depleted in the algal growth stage and has a low concentration when cyanobacterial cell density is high.

## 4. Conclusions

^{−1}formed a cluster. Therefore, securing an appropriate flow rate and removing thermal stratification are critical measures for controlling algal blooms at every weir examined in this study, and various efforts including TP load reduction are required at DSW.

^{−1}). The TP concentration only limits the algal growth during certain periods when the algal biomass increases sharply.

## Supplementary Materials

## Author Contributions

## Funding

## Conflicts of Interest

## References

- Welch, E.B.; Jacoby, J.M.; Horner, R.R.; Seeley, M.R. Nuisance biomass levels of periphytic algae in streams. Hydrobiologia
**1988**, 157, 161–168. [Google Scholar] [CrossRef] - Peter, A.; Köster, O.; Schildknecht, A.; Von, G.U. Occurrence of dissolved and particle-bound taste and order compounds in Swiss lake waters. Water Res.
**2009**, 43, 2191–2200. [Google Scholar] [CrossRef] [PubMed] - Parinet, J.; Rodriguez, M.J.; Serodes, J. Influence of water quality on the presence of off-flavour compounds (geosmin and 2-methylisoborneol). Water Res.
**2010**, 44, 5847–5856. [Google Scholar] [CrossRef] [PubMed] - Paerl, H.W.; Otten, T.G. Harmful cyanobacterial blooms: Causes, consequences, and controls. Microb. Ecol.
**2013**, 65, 995–1010. [Google Scholar] [CrossRef] [PubMed] - Reynolds, C.S. Growth and buoyancy of Microcystis aeruginosa Kutz. Emend. Elenkin in a shallow Eutrophic Lake. Proc. Biol. Sci.
**1973**, 184, 29–50. [Google Scholar] [CrossRef] - Srivastava, A.; Ahn, C.Y.; Asthana, R.K.; Lee, H.G.; Oh, H.M. Status, alert system, and prediction of cyanobacterial bloom in South Korea. BioMed Res. Int.
**2015**, 2015, 1–8. [Google Scholar] [CrossRef] [PubMed] - Newcombe, G.; Chorus, I.; Falconer, I.; Lin, T.F. Cyanobacteria: Impacts of climate change on occurrence, toxicity and water quality management. Water Res.
**2012**, 46, 1347–1355. [Google Scholar] [CrossRef] [PubMed] - Reichwaldt, E.S.; Ghadouani, A. Effects of rainfall patterns on toxic cyanobacterial blooms in a changing climate: Between simplistic scenarios and complex dynamics. Water Res.
**2012**, 46, 1372–1393. [Google Scholar] [CrossRef] - Webster, I.T.; Jones, G.J.; Oliver, R.L.; Bormans, M.; Sherman, B.S. Control strategies for cyanobacterial blooms in weir pools. Final Gr. Rep. Natl. Res. Manag. Strateg. Tech. Rep.
**1995**, 119, M3116. [Google Scholar] - Sherman, B.S.; Webster, I.T. Transitions between Aulacoseira and Anabaena dominance in a turbid river weir pool. Limnol. Oceanogr.
**1998**, 43, 1902–1915. [Google Scholar] [CrossRef] - Bormans, M.; Maier, H.; Burch, M. Temperature stratification in the lower River Murray, Australia: Implication for cyanobacterial bloom development. Mar. Freshwater Res.
**1997**, 48, 647–654. [Google Scholar] [CrossRef] - Oliver, R.L.; Hart, B.T.; Olley, J.; Grace, M.; Rees, C.; Caitcheon, G. The Darling River: algal growth and the cycling and sources of nutrients. Murray Darling Basin Comm. Proj. M386. Final Rep.
**1999**, 201. Available online: https://publications.csiro.au/rpr/download?pid=procite:1edb0e8f-ba1b-46d5-b748-aa77850f0f03&dsid=DS1 (accessed on 3 June 2019). - Mitrovic, S.M.; Oliver, R.L.; Rees, C.; Bowling, R.; Buckney, T. Critical flow velocities for the growth and dominance of Anabaena circinalis in some turbid freshwater rivers. Freshwater Biol.
**2003**, 48, 164–174. [Google Scholar] [CrossRef] - Tsujimura, S.; Okubo, T. Development of Anabaena blooms in a small reservoir with dense sediment akinete population, with special reference to temperature and irradiance. J. Plankton Res.
**2003**, 25, 1059–1067. [Google Scholar] [CrossRef] - Liu, X.; Lu, X.; Chen, Y. The effects of temperature and nutrient ratios on Microcystis blooms in Lake Taihu, China: An 11-year investigation. Harmful Algae
**2010**, 10, 337–343. [Google Scholar] [CrossRef] - Reynolds, C.S.; Walsby, A.E. Water-blooms. Biol. Rev. Camb. Philos. Soc.
**1975**, 50, 437–481. [Google Scholar] [CrossRef] - Walsby, A.E. Stratification by cyanobacteria in lakes: A dynamic buoyancy model indicates size limitations met by Planktothrix rubescens filaments. New Phytol.
**2005**, 168, 365–376. [Google Scholar] [CrossRef] - Brookes, J.D.; Granf, G.G. Variations in the buoyancy response of Microcystis aeruginosa to nitrogen, phosphorus and light. J. Plankton Res.
**2001**, 23, 1399–1411. [Google Scholar] [CrossRef] - Bonnet, M.P.; Poulin, M. Numerical modelling of the planktonic succession in a nutrient-rich reservoir: environmental and physiological factors leading to Microcystis aeruginosa dominance. Ecol. Model.
**2002**, 156, 93–112. [Google Scholar] [CrossRef] - Chung, S.W.; Imberger, J.; Hipsey, M.R.; Lee, H.S. The influence of physical and physiological processes on the spatial heterogeneity of a Microcystis bloom in a stratified reservoir. Ecol. Model.
**2014**, 289, 133–149. [Google Scholar] [CrossRef] - Moss, B.; Balls, H. Phytoplankton distribution in a floodplain lake and river and river system: Ⅱ, Seasonal changes in the phytoplankton communities and their control by hydrology and nutrient availability. J. Plankton Res.
**1989**, 11, 839–867. [Google Scholar] [CrossRef] - Wu, N.; Schmalz, B.; Fohrer, N. Distribution of phytoplankton in a German Lowland river in relation to environmental factors. J. Plankton Res.
**2011**, 33, 807–820. [Google Scholar] [CrossRef] - Schindler, D.W. Evolution of phosphorus limitation in lakes. Science
**1977**, 195, 260–262. [Google Scholar] [CrossRef] [PubMed] - Schindler, D.W.; Hecky, R.E.; Findlay, D.L.; Stainton, M.P.; Parker, B.P.; Paterson, M.J.; Beaty, K.G.; Lyng, M.; Kasian, S. Eutrophication of lakes cannot be controlled by reducing nitrogen input: Results of a 37-year whole-ecosystem measurement. Proc. Natl. Acad. Sci. USA
**2008**, 105, 11254–11258. [Google Scholar] [CrossRef] - Ahn, C.Y.; Lee, J.Y.; Oh, H.M. Control of microalgal growth and competition by N:P ratio manipulation. KRIBB
**2013**, 31, 61–68. [Google Scholar] [CrossRef] - Bostrom, B.; Jansson, M.; Forsberg, C. Phosphorus release from lake sediments. Arch. Hydrobiol.
**1982**, 18, 5–59. [Google Scholar] [CrossRef] - Wu, Y.; Wen, Y.; Zhou, J.; Wu, Y. Phosphorus release from lake sediments: Effects of pH, temperature and dissolved oxygen. KSCE J. Civ. Eng.
**2014**, 18, 323–329. [Google Scholar] [CrossRef] - Health Canada. Guidelines for Canadian Drinking Water Quality: Supporting Documentation—Cyanobacterial Toxins-Microcystin-LR. Water Quality and Health Bureau, Healthy Environments and Consumer Safety Branch, Health Canada, Ottawa, Ontario. 2002. Available online: https://www.canada.ca/en/health-canada/services/publications/healthy-living/guidelines-canadian-drinking-water-quality-guideline-technical-document-cyanobacterial-toxins-document.html/ (accessed on 27 August 2018).
- Brock, T.D. Lower pH limit for the existence of blue-green algae: Evolutionary and ecological implications. Science
**1973**, 179, 480–483. [Google Scholar] [CrossRef] - Holm-Hansen, O. Ecology, physiology, and biochemistry of blue-green algae. Annu. Rev. Microbiol.
**1968**, 22, 47–70. [Google Scholar] [CrossRef] - Cho, H.J.; Na, J.E.; Jung, M.H.; Lee, H.Y. Relationship between phytoplankton community and water quality in lakes in Jeonnam using SOM. Korean J Limnol.
**2017**, 50, 148–156. [Google Scholar] [CrossRef] - Czerny, J.; Barcelos e Ramos, J.; Riebesell, U. Influence of elevated CO
_{2}concentrations on cell division and nitrogen fixation rates in the bloom-forming cyanobacterium Nodularia spumigena. Biogeosciences**2009**, 6, 1865–1875. [Google Scholar] [CrossRef] - Yamamoto, Y.; Nakahara, H. The formation and degradation of cyanobacterium Aphanizomenon flos-aquae blooms: the importance of pH, water temperature, and day length. Jpn. J Limnol.
**2005**, 6, 1–6. [Google Scholar] [CrossRef] - Walsby, A.E.; Hayes, P.K.; Boje, R. The gas vesicles, buoyancy and vertical distribution of cyanobacteria in the Baltic Sea. Eur. J. Phycol.
**1995**, 30, 87–94. [Google Scholar] [CrossRef][Green Version] - Wetzel, R.G. Limnology: Lake and River Ecosystems; Academic Press: San Diego, CA, USA, 2001; 1006p, ISBN1 0127447601. ISBN2 9780127447605. [Google Scholar]
- Thomas, R.H.; Walsby, A.E. The effects of temperature on recovery of buoyancy by Microcystis. J. Gen. Microbiol.
**1986**, 132, 1665–1672. [Google Scholar] [CrossRef] - Carpenter, S.R.; Kitchell, J.R. Cascading trophic interactions and lake productivity. Bioscience
**1993**, 35, 634–639. [Google Scholar] [CrossRef] - Konopka, A.E.; Klemer, A.R.; Walsby, A.E.; Ibelings, B.W. Effects of macronutrients upon buoyancy regulation by metalimnetic Oscillatoria agardhii in Deming Lake. J. Plankton Res.
**1993**, 15, 1019–1034. [Google Scholar] [CrossRef] - Arnold, D.E. Ingestion, assimilation, survival, and reproduction by Daphnia pulex fed seven species of blue-green algae. Limnol. Oceanogr.
**1971**, 16, 906–920. [Google Scholar] [CrossRef] - Fulton, R.S.; Paerl, H.W. Toxic and inhibitory effects of the blue-green algae Microcystis aeruginosa on herbivorous zooplankton. J. Plankton Res.
**1987**, 9, 837–855. [Google Scholar] [CrossRef] - DeMott, W.R. Optimal foraging theory as a predictor of chemically mediated food selection by suspension-feeding copepods. Limnol. Oceanogr.
**1989**, 34, 140–154. [Google Scholar] [CrossRef] - Javier, M.G.; Saul, A. Formation and maintenance of nitrogen fixing cell patterns in filamentous cyanobacteria. Proc. Natl. Acad. Sci. USA
**2016**, 113, 6218–6223. [Google Scholar] [CrossRef] - Lee, J.K. Bacterial quorum sensing and quorum quenching for the inhibition of biofilm formation. Korean J. Microbiol. Biotechnol.
**2012**, 40, 83–91. [Google Scholar] [CrossRef] - Manefield, M.; Rasmussen, T.B.; Henzter, M.; Andersen, J.B.; Steinberg, P.; Kjelleberg, S.; Givskov, M. Halogenated furanones inhibit quorum sensing through accelerated LuxR turnover. Microbiology
**2002**, 148, 1119–1146. [Google Scholar] [CrossRef] [PubMed] - Rasmussen, T.B.; Givskov, M. Quorum sensing inhibitors a bargain of effects. Microbiology
**2006**, 152, 895–904. [Google Scholar] [CrossRef] - Recknagel, F.; French, M.; Harkonen, P.; Yabunaka, K. Artificial neural network approach for the modelling and prediction of algal blooms. Ecol. Model.
**1997**, 96, 11–28. [Google Scholar] [CrossRef] - Anderson, D.M.; Alpermann, T.J.; Cembella, A.D.; Collos, Y.; Masseret, E.; Montresor, E. The globally distributed genus Alexandrium: Multifaceted roles in marine ecosystem and impacts on human health. Harmful Algae
**2012**, 14, 10–35. [Google Scholar] [CrossRef] [PubMed] - Isles, P.D.F.; Rizzo, D.M.; Xu, Y.; Schroth, A.W. Modeling the drivers of interannual variability in cyanobacterial bloom severity using self-organizing maps and high-frequency data. Inland Water
**2017**, 7, 333–347. [Google Scholar] [CrossRef] - Tian, D.; Xie, G.; Tian, J.; Tseng, K.H.; Shum, C.K.; Lee, J.Y.; Liang, S. Spatiotemporal variability and environmental factors of harmful algal blooms (HABs) over western Lake Erie. PLoS ONE
**2017**, 12, 1932–6203. [Google Scholar] [CrossRef] [PubMed] - Horne, A.J.; Goldman, C.R. Limnology; McGraw-Hill: New York, NY, USA, 1994; p. 465. ISBN 9780070236738. [Google Scholar]
- Sze, P. A Biology of the Algae, 3rd ed.; McGraw-Hill: Boston, MA, USA, 1998; 278p, ISBN1 0697219100. ISBN2 9780697219107. [Google Scholar]
- An, K.G.; Jones, J.R. Factors regulating bluegreen dominance in a reservoir directly influenced by the Asian monsoon. Hydrobiologia
**2000**, 432, 37–48. [Google Scholar] [CrossRef] - Jung, S.Y.; Kim, I.K. Analysis of water quality factor and correlation between water quality and Chl-a in middle and downstream weir section of Nakdong River. J. Korean Soc. Environ. Eng.
**2017**, 39, 89–96. [Google Scholar] [CrossRef] - Water Resources Management Information System. Available online: http://www.wamis.go.kr/ (accessed on 27 December 2018).
- Water Environment Information System. Available online: http://water.nier.go.kr/ (accessed on 27 December 2018).
- Ministry of Environment. 2017. Available online: http://www.me.go.kr/ (accessed on 27 December 2018).
- Korea Meteorological Administration (KMA). Available online: http://data.kma.go.kr/ (accessed on 27 December 2018).
- K-Water. Available online: http://English.kwater.or.kr/ (accessed on 27 December 2018).
- Breiman, L.; Cutler, A. Breiman and Cutler’s Random Forests for Classification and Regression. 2015. Available online: http://cran.R-project.org/package=randomForest/ (accessed on 25 May 2018).
- Kuhn, M. Classification and Regression Training. 2011. Available online: https://github.com/topeop/caret/issues (accessed on 3 June 2019).
- Breiman, L. Random forests. Mach. Learn.
**2001**, 45, 5–32. [Google Scholar] [CrossRef] - Box, G.; Cox, D. An analysis of transformations. J. R. Stat. Soc. B
**1964**, 26, 211–252. [Google Scholar] [CrossRef] - Jung, S.J.; Lee, D.J.; Hwang, K.S.; Lee, K.H.; Choi, K.C.; Im, S.S.; Lee, Y.H.; Lee, J.Y.; Lim, B.J. Evaluation of pollutant characteristics in Yeongsan River using multivariate analysis. Korean J. Limnol.
**2012**, 45, 368–377. [Google Scholar] [CrossRef] - Tromas, N.; Fortin, N.; Bedrani, L.; Terrat, Y.; Cardoso, P.; Bird, D.; Greer, C.W.; Shapiro, B.J. Characterising and predicting cyanobacterial blooms in an 8-year amplicon sequencing time course. ISME J.
**2017**, 11, 1–18. [Google Scholar] [CrossRef] [PubMed] - Husson, F.; Le, S.; Pages, J. Exploratory Multivariate Analysis by Example Using R; Chapman and Hall: Boca Raton, FL, USA, 2010; ISBN 9780429190032. [Google Scholar]
- Ferreira, F.; Straus, N.A. Iron deprivation in cyanobacteria. J. Appl. Phycol.
**1994**, 6, 199–210. [Google Scholar] [CrossRef] - Liaw, A.; Wiener, M. Classification and Regression by Randomforest. R News
**2002**, 2, 18–22. [Google Scholar] - Johnson, T.C.; Odada, E.; Whittaker, K.T. The Limnology, Climatology and Paleoclimatology of the East African Lakes; Gordon and Breach Publishers: Amsterdam, The Netherlands, 1996; ISBN 2884492348. [Google Scholar]
- Moreira, S.; Schultze, M.; Rahn, K.; Boehrer, B. A practical approach to lake water density from electrical conductivity and temperature. Hydrol. Earth Syst. Sci.
**2016**, 20, 2975–2986. [Google Scholar] [CrossRef][Green Version] - Choi, A.R.; Oh, H.M.; Lee, J.A. Ecological study on the toxic Microcystis in downstream of the Nakdong River. Algae
**2002**, 17, 171–185. [Google Scholar] [CrossRef] - Paerl, H.W. Coastal eutrophication and harmful algal blooms: Importance of atmospheric deposition and groundwater as “new” nitrogen and other nutrient sources. Limnol. Oceanogr.
**1997**, 42, 1154–1165. [Google Scholar] [CrossRef] - Lee, S.H.; Kim, B.R.; Lee, H.W. A study on water quality after construction of the weirs in the middle area in Nakdong River. J. Korean Soc. Environ. Eng.
**2014**, 36, 258–264. [Google Scholar] [CrossRef]

**Figure 1.**Locations of study weirs and sampling stations; Gangjeong Goryeong weir (GGW), Dalseong weir (DSW), Hapcheon Changnyeong weir (HCW), and Changnyeong Haman weir (CHW).

**Figure 2.**Flowchart of statistical analyses and data mining processes employed in this study. SMLR - step-wise multiple linear regression, RFE-RF - recursive feature elimination using random forest models, DT- decision tree, PCA – principle component analysis.

**Figure 3.**Temporal variations of precipitation and cell density (by phytoplankton group) at the Changnyeong Haman weir (CHW) in the Nakdong River from May (month 5) to December (month 12) in; (

**a**) 2017 and (

**b**) 2018.

**Figure 4.**Correlations between; (

**a**) Log (Chl–a) (mg m

^{−3}) and Log (TP) (mg L

^{−1}), and (

**b**) Log(Chl–a) (mg m

^{−3}) and Log (TN) (mg L

^{−1}) at Gangjeong Goryeong weir (GGW), Dalseong weir (DSW), Hapcheon Changnyeong weir (HCW), and Changnyeong Haman weir (CHW).

**Figure 5.**Comparison of measured cyanobacterial dominance with simulated results using step-wise multiple linear regression (SMLR) and random forest (RF) models at; (a) Gangjeong Goryeong weir (GGW), (b) Dalseong weir (DSW), (c) Hapcheon Changnyeong weir (HCW), and (d) Changnyeong Haman weir (CHW).

**Figure 6.**Evaluation of environmental conditions for high cyanobacterial dominance using a decision tree model; (

**a**) Gangjeong Goryeong weir (GGW), (

**b**) Dalseong weir (DSW), (

**c**) Hapcheon Changnyeong weir (HCW), and (

**d**) Changnyeong Haman weir (CHW). Note: see the main text for definitions of abbreviated variables.

**Figure 7.**Correlation between electrical conductivity (EC) (µS cm

^{−1}) and flow rate (m

^{3}s

^{−}

^{1}) at; (

**a**) Gangjeong Goryeong weir (GGW), (

**b**) Dalseong weir (DSW), (

**c**) Hapcheon Changnyeong weir (HCW), and (

**d**) Changnyeong Haman weir (CHW).

**Figure 8.**Principle component analysis (PCA) bi-plots grouped by the harmful algal bloom (HAB) level. The larger symbols indicate the group mean point for each HAB level. (

**a**) Gangjeong Goryeong weir (GGW; PC1-PC2), (

**b**) Dalseong weir (DSW; PC1-PC2), (

**c**) Hapcheon Changnyeong weir (HCW; PC1-PC2), (

**d**) HCW (PC2-PC5), (

**e**) Changnyeong Haman weir (CHW: PC1-PC2), and (

**f**) CHW (PC2-PC5).

**Table 1.**Mean values (± standard deviation) of water quality data from the Gangjeong Goryeong weir (GGW), Dalseong weir (DSW), Hapcheon Changnyeong weir (HCW), and Changnyeong Haman weir (CHW). Note: see the main text for definitions of variables.

Variable | Unit | Weir | |||
---|---|---|---|---|---|

GGW | DSW | HCW | CHW | ||

Sample size | n | 108 | 108 | 96 | 93 |

Temperature | °C | 23.7 (±4.8) | 23.9 (±4.8) | 23.9 (±5.2) | 25.2 (±4.5) |

pH | - | 8.2 (±0.6) | 8.2 (±0.8) | 8.1 (±0.8) | 8.2 (±0.8) |

DO | mg L^{−1} | 9.01 (±1.79) | 9.20 (±1.90) | 9.07 (±2.15) | 8.63 (±1.74) |

EC | µS cm^{−1} | 237.5 (±59.0) | 318.4 (±89.6) | 312.7 (±107.1) | 259.3 (±64.6) |

SS | mg L^{−1} | 8.66 (±7.86) | 10.1 (±13.7) | 8.79 (±6.44) | 9.80 (±5.55) |

BOD | mg L^{−1} | 1.91 (±0.88) | 2.23 (±2.85) | 1.98 (±1.22) | 2.08 (±1.02) |

COD | mg L^{−1} | 4.85 (±0.89) | 6.16 (±5.34) | 5.59 (±0.90) | 5.04 (±0.81) |

TOC | mg L^{−1} | 4.30 (±1.14) | 4.45 (±1.16) | 4.36 (±0.96) | 4.01 (±0.78) |

TN | mg L^{−1} | 2.29 (±0.48) | 2.83 (±0.85) | 2.61 (±0.50) | 2.18 (±0.41) |

NH_{3}–N | mg L^{−1} | 0.121 (±0.094) | 0.144 (±0.102) | 0.102 (±0.076) | 0.107 (±0.085) |

NO_{3}–N | mg L^{−1} | 1.65 (±0.43) | 2.03 (±0.50) | 1.88 (±0.50) | 1.48 (±0.46) |

TP | mg L^{−1} | 0.068 (±0.036) | 0.087 (±0.100) | 0.072 (±0.039) | 0.078 (±0.063) |

PO_{4}–P | mg L^{−1} | 0.037 (±0.069) | 0.035 (±0.033) | 0.027 (±0.021) | 0.028 (±0.022) |

Chl–a | mg m^{−3} | 14.7 (±11.3) | 36.5 (±162.7) | 24.0 (±21.0) | 26.2 (±15.6) |

Cyano | cells mL^{−1} | 3398 (±9,663) | 9236 (±67,068) | 10,761 (±48,555) | 7,323 (±25,767) |

Green | cells mL^{−1} | 3618 (±7,654) | 3239 (±3,037) | 3150 (±4035) | 3,250 (±3472) |

Diatom | cells mL^{−1} | 2582 (±2,395) | 3491 (±3,513) | 4139 (±3971) | 5,312 (±3847) |

Outflow7 | m^{3} s ^{−1} | 173.91 (±171.43) | 172.83 (±144.54) | 180.19 (±143.73) | 287.08 (±191.96) |

APRCP7 | mm | 29.8 (±41.7) | 17.6 (±23.4) | 19.5 (±20.7) | 22.7 (±22.4) |

△T | °C | 2.2 (±1.7) | 1.5 (±1.1) | 1.0 (±1.2) | 0.6 (±0.8) |

Fe | mg L^{−1} | 0.10 (±0.06) | 0.01 (±0.01) | 0.08 (±0.05) | 0.07 (±0.04) |

SiO_{2} | mg L^{−1} | 4.93 (±3.43) | 4.17 (±3.29) | 3.59 (±2.71) | 2.81 (±2.80) |

**Table 2.**Subset of regression variables that best-matched performance criterion in step-wise multiple linear regression (SMLR) analysis for Gangjeong Goryeong weir (GGW), Dalseong weir (DSW), Hapcheon Changnyeong weir (HCW), and Changnyeong Haman weir (CHW). Note: see the main text for definitions of abbreviated variables.

Weir | Variables | Adj.R^{2} | RMSE | C_{P} | AIC |
---|---|---|---|---|---|

GGW | EC, NO_{3}–N, TN, ΔT | 0.476 | 0.195 | 4.4 | −43.9 |

DSW | NO_{3}–N, EC, TN, Temp, NH_{3}–N, Q7day | 0.521 | 0.182 | 4.5 | −59.7 |

HCW | EC, NO_{3}–N, Temp, TN, Q7day, Chl–a, APRCP7, Fe | 0.729 | 0.155 | 7.8 | −86.8 |

CHW | NO_{3}–N, EC, APRCP7, Q7day, TN | 0.538 | 0.182 | 2.8 | −52.6 |

_{P}: Mallow’s C

_{P}, a smaller value indicates a higher-precision model; AIC: Akaike information criterion, a smaller value indicates a higher-precision model.

**Table 3.**Variables selected by recursive feature elimination (RFE) and order of importance for each weir; Gangjeong Goryeong weir (GGW), Dalseong weir (DSW), Hapcheon Changnyeong weir (HCW), and Changnyeong Haman weir (CHW). Note: see the main text for definitions of abbreviated variables.

Weir | Order of variable importance | RMSE (%) |
---|---|---|

GGW | EC > Temp > ΔT > Q7day > TN > PO_{4}–P > APRCP7 | 0.162 |

DSW | EC > TOC > Temp > TP > TN > Q7day > ΔT | 0.156 |

HCW | EC > ΔT > Temp > Q7day > APRCP7 > TOC > Fe > PO_{4}-P | 0.145 |

CHW | EC > TN > APRCP7 > ΔT > Q7day | 0.160 |

**Table 4.**Contributions of variables to principal components; Gangjeong Goryeong weir (GGW), Dalseong weir (DSW), Hapcheon Changnyeong weir (HCW), and Changnyeong Haman weir (CHW). Note: see the main text for definitions of abbreviated variables.

Variable | GGW | DSW | HCW | CHW | ||||
---|---|---|---|---|---|---|---|---|

Component | Component | Component | Component | |||||

1 | 2 | 1 | 2 | 1 | 2 | 1 | 2 | |

C.dominance | 6.75 | 6.00 | 3.99 | 5.61 | 0.42 | 19.26 | 2.10 | 16.77 |

APRCP7 | 3.69 | 16.42 | - | 9.29 | 0.12 | 11.40 | 0.58 | |

Q7day | 8.97 | 11.53 | - | 12.24 | 0.39 | 14.26 | 1.54 | |

Temp | 0.39 | 13.41 | - | 0.57 | 14.55 | - | - | |

ΔT | 6.56 | 1.20 | - | 1.65 | 10.31 | 2.79 | 0.91 | |

DO | - | - | 8.21 | 0.54 | 5.87 | 2.98 | ||

pH | 5.40 | 5.07 | 5.15 | 16.98 | 17.04 | 0.03 | 12.23 | 1.81 |

EC | 14.66 | 2.92 | - | 0.03 | 13.27 | 0.10 | 19.43 | |

BOD | - | 15.01 | 22.48 | 4.21 | 9.75 | 2.38 | 0.03 | |

COD | 7.59 | 4.76 | 7.51 | 35.81 | 0.05 | 0.19 | 0.01 | 2.60 |

TOC | 5.65 | 2.87 | 6.19 | 7.36 | 3.62 | 0.53 | - | - |

TP | 8.90 | 1.54 | 23.80 | 0.46 | 3.70 | 3.25 | 4.15 | 5.01 |

PO_{4}–P | 2.76 | 0.02 | - | 8.25 | 0.24 | 11.49 | 1.97 | |

TN | 8.93 | 13.07 | 20.99 | 0.62 | 1.09 | 5.35 | 3.32 | 14.21 |

NH_{3}–N | - | - | - | - | 2.15 | 1.28 | 1.28 | 7.73 |

NO_{3}–N | 8.36 | 12.49 | - | 0.60 | 14.54 | 3.42 | 17.72 | |

Fe | 1.60 | 1.68 | - | 12.88 | 0.92 | 6.42 | 5.89 | |

SiO_{2} | 9.79 | 7.01 | - | 6.98 | 0.47 | 15.55 | 0.52 | |

Chl–a | - | 17.36 | 10.68 | 7.01 | 3.63 | 3.23 | 0.29 |

**Table 5.**Integration of statistical analysis and data mining results for comprehensive interpretation; Gangjeong Goryeong weir (GGW), Dalseong weir (DSW), Hapcheon Changnyeong weir (HCW), and Changnyeong Haman weir (CHW). Note: see the main text for definitions of abbreviated variables.

Weir | Correlation Analysis | Recursive Feature Elimination | Decision Tree | PCA | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|

(r > ∣0.5∣) | (Variable importance rank) | (C.dominance > 50% conditions) | (Clustering) | ||||||||||

1st | 2nd | 3rd | 4th | 1st | 2nd | 3rd | 4th | 5th | 1st | 2nd | Positive | Negative | |

GGW | Chl–a | TOC | ΔT | - | EC | ΔT | TN | Temp | Q7day | EC ≥ 325 μS cm^{−1} | - | EC, ΔT, Temp | TOC NO _{3}–NTN |

DSW | Chl–a | SS | TN | TP | EC | TOC | Temp | ΔT | TP | EC ≥ 336 μS cm^{−1}TP ≥ 0.082 mg L ^{−1} | EC ≥ 336 μS cm^{−1}TP < 0.082 mg L ^{−1}TN < 2.4 mg L ^{−1} | pH, TOC Chl–a | - |

HCW | ΔT BOD | - | - | - | EC | ΔT | Temp | Q7day | Fe | EC ≥ 443 μS cm^{−1} | 325 μS cm^{−1} ≤ EC < 443 μS cm^{−1}ΔT ≥ 0.7 °C | EC Temp ΔT BOD NH _{3}–NChl-a | NO_{3}–N TN |

CHW | - | - | - | - | EC | ΔT | TN | Q7day | APRCP7 | EC ≥ 385 μS cm^{−1} | EC < 385 μS cm^{−1}TN < 2 mg L ^{−1}Q7day < 167 m ^{3} s^{−1} | EC ΔT Fe | NO_{3}–N TN |

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Kim, S.; Chung, S.; Park, H.; Cho, Y.; Lee, H. Analysis of Environmental Factors Associated with Cyanobacterial Dominance after River Weir Installation. *Water* **2019**, *11*, 1163.
https://doi.org/10.3390/w11061163

**AMA Style**

Kim S, Chung S, Park H, Cho Y, Lee H. Analysis of Environmental Factors Associated with Cyanobacterial Dominance after River Weir Installation. *Water*. 2019; 11(6):1163.
https://doi.org/10.3390/w11061163

**Chicago/Turabian Style**

Kim, Sungjin, Sewoong Chung, Hyungseok Park, Youngcheol Cho, and Heesuk Lee. 2019. "Analysis of Environmental Factors Associated with Cyanobacterial Dominance after River Weir Installation" *Water* 11, no. 6: 1163.
https://doi.org/10.3390/w11061163