Biases in greenhouse gases static chambers measurements in stabilization ponds: Comparison of flux estimation using linear and non-linear models

The closed static chamber technique is widely used to quantify greenhouse gases (GHG) i.e. CH4, CO2 and N2O from aquatic and wastewater treatment systems. However, chamber-measured fluxes over air-water interfaces appear to be subject to considerable uncertainty, depending on the chamber design, lack of air mixing in the chamber, concentration gradient changes during the deployment, and irregular eruptions of gas accumulated in the sediment. In this study, the closed static chamber technique was tested in an anaerobic pond operating under tropical conditions. The closed static chambers were found to be reliable to measure GHG, but an intrinsic limitation of using closed static chambers is that not all the data for gas concentrations measured within a chamber headspace can be used to estimate the flux due to gradient concentration curves with non-plausible and physical explanations. Based on the total data set, the percentage of curves accepted was 93.6, 87.2, and 73% for CH4, CO2 and N2O, respectively. The statistical analyses demonstrated that only considering linear regression was inappropriate (i.e. approximately 40% of the data for CH4, CO2 and N2O were best fitted to a non-linear regression) for the determination of GHG flux from stabilization ponds by the closed static chamber technique. In this work, it is clear that when R2adj-non-lin > R2adj-lin, the application of linear regression models is not recommended, as it leads to an underestimation of GHG fluxes by 10-50%. This suggests that adopting only or mostly linear regression models will affect the GHG inventories obtained by using closed static chambers. According to our results, the misuse of the usual R2 parameter and only the linear regression model to estimate the fluxes will lead to reporting erroneous information on the real contribution of GHG emissions from wastewater. Therefore, the R2adj and non-linear regression model analysis should be used to reduce the biases in flux estimation by the inappropriate application of only linear regression models.