Improving gap-filling performance for CH4 fluxes of eddy covariance data by combining marginal distribution sampling and machine learning algorithm over paddy fields

• Published in: Journal of cleaner production Vol. 510; p. 145615
• Main Authors: Ma, Linhua, Yu, Qianan, Cui, Yuanlai, Liu, Bo, Tian, Wenxiang, Liao, Bin, Liu, Luguang, Dong, Wei, Huang, Jie
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 10.06.2025
• Subjects: CH4 fluxes; Eddy covariance; Gap-filling; Machine learning; Marginal distribution sampling; Paddy fields; Paddy fields; Eddy covariance; CH4 fluxes; Gap-filling; Machine learning; Marginal distribution sampling
• ISSN: 0959-6526
• DOI: 10.1016/j.jclepro.2025.145615

Accurate gap-filling CH4 fluxes from eddy covariance measurements over paddy fields is important in assessing agriculture carbon balance and greenhouse gas emission and therefore contribute to agriculture cleaner production. However, as a type of highly managed wetland, the mechanisms of CH4 emission from paddy field are different from those from natural wetlands. This study improved the methods to process CH4 fluxes data from paddy fields. Adding gross primary production (GPP) into the Moving Point test (MP) method enables it to calculate the threshold of friction velocity (Uc∗) in the paddy fields. The order of calculated Uc∗ for all sites is: TWT (0.27 m s−1) > MSE (0.20 m s−1) > CRK (0.12 m s−1) > HRC (0.11 m s−1) > RIP (0.10 m s−1) > HRA (0.09 m s−1) > NC (0.08 m s−1) > CAS (0.05 m s−1). GPP is 2.5 h ahead of CH4 fluxes, incorporating the asynchrony phenomenon improves the gap-filling performance at all 8 sites, with the R2 increases by 6.3 %–16.1 %, RMSE decreases by 8.0 %–26.8 %, MAE decrease by 10.5 %–25.2 %. Replacing soil moisture condition with the ‘on-off switch’ effect of water table only improves the gap-filling performance at CRK, HRA, HRC, MSE and NC sites, with the R2 increase by 7.2 %, 8.1 %, 10.4 %, 3.8 %, 7.1 %, and the improvement of gap-filling accuracy was positively correlated with times of alternate wetting and drying cycles during the growing season. The gap-filling effect of six machine learning algorithms were in order: Adaboost > XGBoost > RF > BPNN > SVM and KNN, with the MDS method worse than the RF. By solving the uneven distribution of CH4 fluxes, integrating the MDS method into machine learning algorithm could improve the performance of gap-filling, with R2 increased by 0.02–0.06, MAE decreased by 0.13 %–7.88 %. Overall, the study is benefit for data processing of CH4 fluxes over paddy fields. [Display omitted] •A modified procedure for CH4 fluxes gap-filling over paddy field was proposed.•The two-factor MP method adding GPP as grading criterion can better capture Uc∗ threshold.•The MDS method was improved by considering GPP asynchrony and ‘on-off switch’ effect of water table.•Adaboost performed best among six machine learning algorithms in gap-filling CH4 fluxes over paddy fields.•The shortcoming of Adaboost can be compensated by integrating the MDS method.