The temporal variations in CO2 net atmospheric flux (NAF) in stratified reservoirs are controlled by both physical
and biological factors. However, research on the factors and processes affecting CO2 NAF variability over time is
insufficient, and as a result, there is considerable uncertainty in present estimations of global reservoir CO2 emis?sions. In the present study, we analyzed the effects of hydrodynamic and water quality factors on CO2 NAF var?iability in a stratified reservoir based on field studies and data modeling. Three empirical and four surface
renewal gas transfer models were used to characterize the effects of hydrodynamic factors on gas transfer rate
and CO2 NAF at the air?water interface. Buoyant turbulence notably affected CO2 NAF when the stratification
strength was reduced. As a result, the CO2 NAF (1485 mg-CO2 m
?2 day?1
) estimated using surface renewal
models that considered the effects of buoyant turbulence were twice greater than the NAFs estimated using em?pirical models that only considered wind force (724 mg-CO2 m
?2 day?1
). The best linear regression model ex?plained 81.6% of the temporal variation in CO2 NAF using water temperature (Tw), electrical conductivity (EC),
pH, chlorophyll a, total organic C (TOC), and alkalinity. The nonlinear parsimonious random forest model ex?plained 84.4% of the temporal change in CO2 NAF using only three independent variables (EC, dissolved oxygen,
and TOC). Principal component analysis revealed that the CO2 NAF tended to be large under low Tw, weak strat?ification, and low pH. These results indicate that the temporal variability of CO2 NAF in the stratified reservoir can
be predicted using data-driven modeling with minimal water quality variables and selection of an appropriate
gas exchange model. The findings improve the accuracy of estimates of CO2 emissions and monitoring activities
in stratified reservoirs.