Gap-filling carbon dioxide, water, energy, and methane fluxes in challenging ecosystems: Comparing between methods, drivers, and gap-lengths

Summary

This paper evaluates random forest regression (RFR) and alternative gap-filling approaches for eddy covariance flux data across challenging agricultural and land-use-changed ecosystems. The authors demonstrate that RFR is a competent alternative to standard methods, particularly superior for filling gaps >30 days in carbon dioxide and for all gap lengths in other fluxes including methane. Critically, RFR preserved key environmental–flux relationships and reliably reconstructed cumulative fluxes for gaps exceeding three months, addressing a significant limitation in current ecosystem monitoring practice.

UK applicability

The findings are directly applicable to UK managed pastures and grassland-based farming systems, which are prominently featured in the study's challenging ecosystems. The demonstration that RFR effectively fills multi-month gaps using globally available reanalysis drivers is particularly relevant for UK farm-scale monitoring where instrument downtime and data gaps are common.

Key measures

Gap-filling performance metrics for CO₂, H₂O, energy, and CH₄ fluxes; cumulative flux accuracy for gap lengths ranging from <12 days to >3 months; reconstruction of environment–flux relationships

Outcomes reported

The study compared gap-filling methodologies (random forest regression, marginal distribution sampling, and others) for carbon dioxide, water, energy, and methane fluxes across managed pastures, converted peatlands, and drylands. Random forest regression was assessed for its ability to reliably fill missing data gaps of varying lengths in eddy covariance time-series.

Topic tags