Influence of sediments on water quality in serially impounded river systems

Abstract

There is increasing number of researches related to influence of sediments on water quality of rivers. However, little researches have been conducted to evaluate characteristics of sediment and to predict oxygen demand when OM and nutrients are released from sediment of serially impounded river systems. The purpose of this study was to evaluate effects of dissolved organic matter (DOM) and inorganic phosphorus by fraction on water quality of rivers, which are released from sediments upstream of Gangcheon, Yeoju, Ipo weir in the Namhan River according to rainfall. In addition, a machine-learning-based model was constructed to predict sediment oxygen demand, which was amount of oxygen consumed by release of DOM and inorganic phosphorus, etc. In order to investigate change in properties of DOM released from sediment to water layer according to rainfall, incubation experiments were conducted in laboratory. To analyze spatial and temporal variation in DOM, sediment and bottom water samples were collected upstream of the Gangcheon, Yeoju and Ipo weirs of the Namhan River during rainy and non-rainy seasons. The collected surface sediments were incubated with bottom water for 0, 3 and 7 days. Initial dissolved organic carbon (DOC) was correlated with precipitation and residence time. The change in concentration of dissolved organic carbon are due to organic carbon released from sediment, which can cause water quality issues. Fluorescence analysis revealed that DOM contained higher levels of hydrophilic and low-molecular-weight (LMW) organic matter (OM) in non-rainy season and higher levels of hydrophobic and high-molecular-weight (HMW) OM in rainy season. Since the Namhan River is located upstream of Paldang Dam, main source of drinking water in the Seoul metropolitan area, seasonal variations of released DOM suggest that it may be help optimize the drinking water treatment process with season. Using statistical analysis, it was confirmed that nutrient content of pore-water and sediment can be used to estimate DOM release rate from sediment to water layer. Investigation of change of inorganic phosphorus fractions released from sediment according to rainfall was conducted under same location, period and conditions as incubation experiment for OM. Concentration of inorganic phosphorus for each fraction of sediment was analyzed before and after incubation. Concentrations of four inorganic phosphorus (Al-bounded P, Al-P; Fe-bounded P, Fe-P; reductant soluble P, Red-P; Ca-bounded P, Ca-P) were significantly higher in the non-rainy season than in the rainy season. The P fractions with highest concentrations, Red-P and Ca-P, were also higher in the non-rainy season than in the rainy season. From incubation experiment, concentration of inorganic phosphorus in sediment tended to decrease with incubation time in the non-rainy season resulting in significant correlation between changes in Al- and Fe-P in sediment and PO4-P concentration in overlying water layer. Inorganic phosphorus in sediment were more dominant from non-point source than from point source according to enrichment factor value. As a result of statistical analysis, release rate of Al-P was correlated with pH, and Fe and Ca-P release rate were correlated with OM. This result suggests that phosphorus released from sediment is affected by each fraction, rainfall, and physicochemical factors. A machine learning algorithm was used to build a model to predict oxygen demand when OM and nutrients are released from sediment of the river with weirs. Model was constructed using Deep Neural Network (DNN) and Random Forest (RF), which are useful for learning random data, and performance between the two models was compared. In addition, variables that have a highest importance on sediment oxygen demand (SOD) prediction were selected through RF model. As input data for model, experimental data, water quality data, and meteorological data were used. Performance of train, validation, and test data of DNN model were higher than 0.900 in R2, and performance of train and test data of RF model were 0.919 and 0.802 in R2, respectively Both DNN and RF performed better than multiple linear regression model. The feature that had most influence on SOD prediction was BOD, and surface sediment TOC, chlorophyll a (chl-a), TP, surface sediment, Chl-a, NH3-N, and dissolved total nitrogen had also high importance. DNN and RF model for SOD prediction provide a better understanding for analyzing effect of sediments on water layer. According to results of study, it became clear that organic matter and phosphorus release occurred in river sediments with weir which could affect quality of overlying water layer. In addition, release rate and properties of DOM and inorganic phosphorus were different depending on rainfall. Therefore, it is necessary that continuous sediment monitoring is required in order to understand effects of sediments on water layer and reduce impact on water quality according to rainfall. Results of sediment monitoring can be used to SOD prediction by machine-learning-based models. SOD prediction can help understand mechanism of hypoxia at sediment-water interface. Furthermore, monitoring data and predicted SOD can be used for establishment of watershed management policies. Chapter II : In this study, changes in the properties of dissolved organic matter (DOM) released from sediments into water layers were investigated. To analyze the spatial and temporal variation in dissolved organic carbon (DOC), sediment and bottom water samples were collected upstream of the Gangcheon, Yeoju and Ipo weirs of the Namhan River during the rainy and non-rainy seasons. The initial DOC was correlated with precipitation (R = 0.543, p = 0.034) and residence time (R = 0.524, p = 0.040). The change of bottom water DOC concentration resulted from the DOC released from the sediments which may cause water quality issues in the bottom water. The Fluorescence analysis revealed that the DOM contained higher levels of hydrophilic and low-molecular-weight (LMW) organic matter in non-rainy season and higher levels of hydrophobic and high-molecular-weight (HMW) organic in rainy season. Since the Namhan River is the main resource of drinking water for Seoul metropolitan area, our results can help optimize drinking water treatment process reflecting the DOM characteristics varying with season. Furthermore, the statistical analysis confirmed that the nutrient content of pore-water and sediment can be used to estimate the DOM release rate from sediment to water layer. This result of this study provided a better understanding of the DOM movement in aquatic ecosystem and the influences of rainfall on the water quality of surface water body. Chapter III : In this study, we investigated the temporal and spatial variations of the inorganic P fractions and their release using the sediment of three weirs built along the Namhan River in South Korea. The sediment and water samples that were collected upstream of Gangcheon, Yeoju, Ipo weir in rainy and non-rainy seasons were incubated during 0, 3, and 7 days. The inorganic P fraction concentrations of the sediment sample were analyzed after incubation. The sum of the four inorganic P fractions were significantly higher in the non-rainy season than in the rainy season. The P fractions with the highest concentrations, reductant soluble P and Ca-bounded P, were also higher in the non-rainy season than in the rainy season. From the incubation experiment, it was observed that the concentration of inorganic P in the sediments tends to decrease with the incubation time in the non-rainy season resulting in the significant correlation between changes in Al- and Fe-bounded P in the sediment and PO4-P concentration in the overlying water layer. The inorganic phosphorus in the sediment were more dominant from non-point source than from the point source according to the enrichment factor value. The Al-bounded P release rate was positively correlated with the pH, while those of the Fe- and Ca-bounded P were correlated with the OM. The fractionation and release of P in the sediments was affected by the rainfall and the physiochemical factors. The results of this study will enhance our understanding of the water quality as well as the impoundment and rainfall on sediment, thereby aiding in an efficient river environment management. Chapter IV : This study used machine learning to predict sediment oxygen demand (SOD), which is one of method to evaluating effect of sediment on the water layer according to the release of organic matter (OM) and nutrient. A comparative analysis was conducted a performance test of the deep neural network (DNN) and random forest (RF). Factors that have more influence on SOD prediction were selected as RF model. The data for machine learning model was experimental data, water quality data, and meteorological data. Performance of train, validation, and test dataset of DNN model was greater than 0.9 in R2. As result of RF model, performance of train and test data set was 0.919 and 0.802 in R2, respectively. Both DNN and RF model outperformed better than multiple linear regression (MLR) model to predict trend of SOD. The feature that had the most influence on SOD prediction was biological oxygen demand (BOD), and surface sediment total organic carbon (TOC), Chlorophyll a (Chl-a), total phosphorus (TP), surface sediment Chl-a, NH3-N, and dissolved total nitrogen (DTN) had also high importance. Among the variables, OM and nutrients were selected to be main factors affecting SOD. Since these features are easy to acquire data and relatively simple to measure, SOD could be easily predicted using machine learning. This study using DNN and RF models during machine learning for SOD prediction will provide a better understanding for analyzing effects of sediments on the water layer.;보가 설치된 하천 퇴적물이 수층에 미치는 영향에 대한 연구가 증가하고 있지만, 강우시기에 따른 퇴적물 용출 특성을 평가하기 위한 구체적인 연구와 체계적인 예측 모델 개발은 거의 이루어지지 않았다. 따라서 이 연구에서는 남한강에 설치된 강천보, 여주보, 이포보 상류 퇴적물에서 용출되는 용존 유기물과 존재형태별 무기 인이 하천의 수질에 미치는 영향을 강우에 따라 평가하는 것을 목표로 하였다. 또한, 퇴적물이 수층에 미치는 복합적인 영향을 평가하는데 사용되는 SOD를 예측하는 machine learning 기반 모델을 구축하였다. 퇴적물에서 수층으로 용출되는 용존 유기물의 특성 변화를 조사하기 위해 실험실에서 배양실험을 진행하였다. 용존 유기물의 공간적, 시간적 변화 패턴을 분석하기 위해 남한강에 설치된 강천보, 여주보, 이포보 상류에서 강우기와 비강우기에 퇴적물 시료를 채취하였다. 채취한 표층 퇴적물은 저층수와 함께 0일, 3일, 7일 동안 배양하였다. 초기 용존 유기탄소는 강우량 및 체류시간과 유의한 상관관계가 있었다. DOC의 농도 변화는 퇴적물에서 용출된 유기탄소에 의한 것으로, 이는 수질에 영향을 미칠 수 있다. 형광 분석을 실시한 결과 DOM은 비강우기에 더 높은 수준의 친수성 저분자량의 유기물을 함유하고 있었고, 강우기에는 소수성 고분자량의 유기물을 더 많이 함유하고 있는 것으로 나타났다. 남한강은 수도권의 주요 식수원인 팔당댐의 상류에 있으므로 계절에 따라 변화하는 DOM의 특성은 식수처리 공정 최적화에 도움을 줄 수 있음을 시사한다. 또한 통계 분석을 통해 공극수 및 퇴적물의 유기물 함량이 퇴적물에서 수층으로 용출되는 DOM의 용출률에 영향을 미치는 것으로 나타났다. 강우에 따라 퇴적물에서 용출되는 존재형태별 무기인의 변화에 대한 조사는 유기물 용출 실험과 같은 장소, 기간, 조건으로 실시하였다. 퇴적물 시료의 존재형태별 무기인의 농도는 배양 전과 배양 3일, 7일 후에 측정하였다. 4가지의 무기인 (Al-P, Fe-P, Red-P, Ca-P)의 농도는 비강우기에 강우기보다 유의하게 높게 나타났으며, 무기인 중 농도가 높게 나타난 형태는 Red-P와 Ca-P였다. 용출 실험결과는 비강우기에서 용출 시간에 따라 퇴적물 내 무기인의 농도가 감소하는 경향이 있었고, 특히 존재형태 중 Al-P와 Fe-P의 감소는 수층의 인산염 인 농도 증가와 유의한 상관관계가 있음을 제공하였다. 존재형태별 인의 농도를 이용해 퇴적물의 기원을 추정하는 Enrichment factor 값에 따르면 점오염원보다 비점오염원에서 유래한 무기인이 남한강 보 상류 퇴적물에서 우세하였다. 통계분석 결과 Al-P의 용출률은 pH, Fe와 Ca-P의 용출률은 유기물과 상관관계가 높았다. 이 결과는 퇴적물에서 용출되는 인은 존재형태와, 강우, 주변 환경의 물리화학적 요인의 영향을 받음을 시사한다. 보가 설치된 하천의 퇴적물에서 유기물과 영양염이 수층에 미치는 영향을 평가하기 위해 퇴적물 산소소모율(SOD)를 예측하는 모델을 머신러닝 알고리즘을 이용하여 구축하였다. 랜덤한 자료 학습에 유용하게 활용되는 Deep Neural Network (DNN)와 Random Forest (RF)을 이용하여 모델을 구축하였고, 두 모델간 성능을 비교하였다. 또한 구축한 모델을 통해 SOD 예측에 더 많이 영향을 미치는 변수를 선정하였다. Machine learning 기반 모델의 구축을 위한 입력 자료로는 실측자료, 수질측정망 자료, 기상 자료를 사용하였다. DNN 모델의 train, validation, test data의 R2은 0.9 이상이었고, RF 모델의 train data와 test data의 R2은 각각 0.919과 0.802로 두 모델 모두 훌륭한 성능을 갖춘 것을 확인하였다. DNN과 RF 모두 다중선형회귀 모델보다 더 나은 성능을 보였다. SOD를 예측할 때 가장 큰 영향을 미친 인자는 BOD이었으며, TOC, 클로로필a, TP, 암모니아성 질소, 용존 총 질소 등이 중요도가 높은 인자로 선정되었다. SOD 예측을 위한 DNN과 RF 모델 구축을 통해 퇴적물이 수층에 미치는 영향을 확인하고 주요 영향 인자를 도출할 수 있었다. 본 연구를 통해 하천에 설치된 보 상류의 퇴적물에서 발생하는 유기물과 인의 용출은 수질에 영향을 미칠 수 있으며, 강우에 의해 용출률과 용출 특성이 다르게 나타나는 것을 확인할 수 있었다. 따라서 강우시기에 따라 퇴적물이 수층에 미치는 영향을 파악하고, 수질에 미치는 영향을 줄이기 위해서는 지속적인 퇴적물 모니터링이 필요하다. 퇴적물 모니터링의 결과는 machine learning 기반 모델을 이용해 SOD를 예측하는 데 활용할 수 있으며, 이는 수층 경계의 빈산소 발생 기작에 대한 이해를 도울 수 있다. 더 나아가 모니터링 자료와 예측된 SOD는 유역관리를 위한 정책수립을 위한 정보로 활용될 수 있다.