Uncertainty in aquatic greenhouse gas flux estimates arises from subjective processing of floating chamber time series

Camille Minaudo, Miguel Cabrera-Brufau, Jorge Juan Montes-Pérez, Katrin Attermeyer, Antonio Camacho, Alba Camacho-Santamans, Constantin Cazacu, Relu Giuca, Ana I. Lillebø, Benjamin Misteli, Daniel Morant, Biel Obrador, Bruna R.F. Oliveira, Jolita Petkuvienė, Antonio Picazo, Carlos Rochera, Daniel von Schiller

Accurate quantification of greenhouse gas (GHG) fluxes from aquatic systems is essential for constraining regional and global carbon budgets. Closed floating chambers are widely used to measure carbon dioxide (CO₂) and methane (CH₄) fluxes at the water–air interface, yet large uncertainties persist due to subjective processing of chamber time series. In particular, the treatment of non-linear patterns and abrupt events such as ebullition often relies on expert judgement, which may strongly influence flux estimates. We present the first quantitative assessment of bias arising from expert subjectivity in the processing of floating chamber measurements. Seventeen researchers, all of whom participated in field sampling, independently evaluated a common dataset of 794 incubations from 36 European coastal wetlands. Each expert visually selected valid data segments and flagged abnormal time series prior to flux calculation. In total, 2,679 manual inspections were compared with fully automated flux estimates based on untrimmed time series. Experts showed substantial variability in handling non-linear or irregular concentration patterns. While flux variability was generally low for most CO₂ and CH₄ incubations, disagreements exceeded 100% for curved or abrupt time series. For CH₄, ebullition was a major source of divergence, but marked variability also occurred when ebullition contributed only marginally to total fluxes. This methodological experiment demonstrates that expert judgement introduces significant, previously unquantified uncertainty into aquatic GHG flux estimates. We advocate for transparent, standardised, and reproducible data-processing workflows, including automated tools for objective identification and treatment of non-linearities.