Freshwater is reported as an important carbon source in the atmosphere; however, at present, there is a lack of clear methodology for assessing greenhouse gas emissions from freshwater with a high amount of uncertainty involved in its estimation. Moreover, research on the analysis of carbon mass balance and carbon emissions in domestic freshwater remains limited.
In this study, two years of field experiments were carried out to quantitatively evaluate the CO2 net atmospheric flux (NAF) at the atmosphere-water interface in Daecheong Reservoir, a temperate reservoir located in the Asian monsoon climate region. In addition, machine learning and numerical modeling techniques were applied to analyze the causes of temporal and spatial fluctuations in CO2 NAF in the stratified reservoir. The partial pressure of CO2 (pCO2) in water, the most important factor in estimating the CO2 NAF, were accurately calculated by theoretical estimation method using thermodynamic equilibrium theory (CO2SYS) and measurement method using non-dispersive infrared detector (NDIR). The comparison of pCO2 values obtained using the two methods showed that the pCO2 value estimated by CO2SYS was 49.4% lower than the average pCO2 measured by the NDIR detector. Furthermore, as the depth of the water increased, greater deviation in pCO2 values was observed due to the limitations of the thermodynamic equilibrium theory.
Seven different gas exchange models, including three empirical models (Cole and Caraco, Crusius, Vachon) and four surface renewal models (Heiskanen, Maclntyre, Read, and Soloviev) were used and compared for the estimation of the gas exchange coefficient and NAF. The NAF values estimated by the empirical models (?1246.0 to 6510.3 mg-CO2/m2/d) were smaller than those estimated by the surface renewal models (-1436.1 to 8485.7 mg-CO2/m2/d). Moreover, the MacIntyre and Heiskanen models that take into account the effects of buoyancy flux and turbulence mixing have been shown to be more suitable for the estimation of CO2 NAF in the stratified reservoir, rather than an empirical model using only the wind speed functions,
Data-driven models, such as multiple linear regression (MLR) and random forest (RF) models, were developed to predict CO2 NAF using water quality, hydraulic, and meteorological data. The principal factors that influence CO2 NAF predictions by the developed models were determined using principal component analysis (PCA). The important variables influencing the prediction of CO2 NAF were Electrical Conductivity (EC), Dissolved oxygen (DO), and total organic carbon (TOC) for RF models and Temperature (Tw), EC, pH, chlorophyll a (Chl-a), and Schmidt stability number (Sc) for MLR models. PCA results showed that CO2 NAF tends to be larger under conditions of lower water temperature, weaker stratification, and lower pH.
Numerical modeling techniques were used to comprehensively analyze the temporal and spatial variations in the carbon cycle mechanism in the stratified reservoir. Although the results highlighted spatially different emission characteristics, the Daecheong Reservoir was overall found to be a pCO2 supersaturated and heterotrophic system that releases CO2 into the atmosphere. The inlet and transition zone of the reservoir has a long uptake period of atmospheric CO2, whereas the lacustrine zone showed a long-term release of CO2. A quantitative assessment of CO2 emissions from the reservoir revealed that in 2017 and 2018, approximately 1,810 tons and 2,113 tons were released into the atmosphere, respectively. Moreover, the amount of carbon released from the entire carbon sink into the atmosphere was about 10%. Greater CO2 emissions were observed at night because of higher convective mixing and lesser photosynthesis with the highest emissions observed in winter. The CO2 NAF estimation and evaluation methodology developed in this study is expected to be used as a protocol for carbon cycle analysis in domestic freshwater. Further research should focus on accounting for CH4 in the carbon cycle and a more comprehensive sediment diagenesis model. In addition, it is necessary to analyze the impact of dam reservoirs during the carbon mass balance analysis by integrating watershed, reservoirs, and rivers.