Automated versus expert processing of in situ CO2 and CH4 chamber measurements in complex and heterogeneous aquatic ecosystems

Measuring reliable greenhouse gases (GHGs) fluxes at the interface between water bodies and the atmosphere in inland waters is crucial in the context of climate change but remains highly challenging.  GHG fluxes can be measured directly in-situ with chambers placed at the water-atmosphere interface and connected to portable gas analysers providing high-frequency timeseries of GHGs partial pressures inside the chamber. Fluxes are usually assumed constant over the time of incubation, but varying GHG sources and changing environmental conditions and/or ebullition from the sediment produce non-linear patterns and breakpoints in the timeseries, not mentioning the possibility for poor manipulation of the device, disturbance of the sampling site by the operator, or malfunctioning sensors. Accordingly, it is common procedure to visualize and select part of the measurements manually for each incubation before proceeding with fluxes computation. In the ongoing Horizon Europe project RESTORE4Cs, we have performed CO2 and CH4 chamber measurements in 36 different sites located in 6 major coastal wetlands across Europe, including intertidal saltmarshes and seagrass beds, freshwater and brackish ponds and marshes, and coastal lagoons. Between October 2023 and August 2024, we have gathered a database of 822 floating chamber incubations, collected by multiple operators and with 3 different gas analysers.  

Here we focus on the data processing part and assess to what extent we need expert evaluation of the time series to produce reliable flux estimates. We have developed an automated data processing script able to compute fluxes estimates for all incubations. Timeseries are fitted with both a linear and non-linear models. The script identifies potential bubbling in CH4 measurements and estimates the diffusive versus ebullitive components based on the statistical characteristics of the first derivative of pCH4. All incubations were manually inspected by 16 members of our team, all experts in GHG chamber measurements with various levels of experience. About half of these timeseries were inspected independently by at least 3 experts, enabling to compare if and why experts disagree. 

For both CO2 and CH4, non-linear fitting performed better than linear models for 69% incubations, indicating a substantial number of non-linear patterns in the dataset; however, the difference between the two models was less than 10% for 86% of the incubations. The ebullition pathway was the dominant CH4 flux in less than 10% of the incubations. Experts disagreed substantially on the data selection in 34% of the incubations, which produced uncertainties in flux estimates larger than 10% of the inter-expert flux average. The highest discrepancies were related to suspicious or non-linear features in the time series. To avoid subjectivity and ensure robustness and repeatability of flux estimates, we present guidelines on how CO2 and CH4 incubation time series should be processed, regardless of whether they are processed automatically or after an expert manual inspection.