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

Summary

This paper extends existing work on eddy covariance gap-filling methodologies by evaluating random forest regression alongside traditional approaches across multiple challenging ecosystems and flux types. The study finds that RFR is a competent alternative to standard algorithms, with MDS preferred for short CO₂ gaps (<12 days) but RFR superior for longer gaps (>30 days) and non-CO₂ fluxes including methane. Crucially, RFR reliably filled cumulative fluxes for gaps exceeding three months whilst preserving key environmental-flux relationships that alternative methods did not maintain.

UK applicability

The findings are directly applicable to UK managed grasslands and pasture systems, where eddy covariance monitoring is increasingly used for carbon and nutrient flux quantification. The RFR methodology's effectiveness with reanalysis climate drivers when measured data are unavailable is particularly relevant for UK research stations with incomplete instrumental coverage.

Key measures

Gap-filling accuracy for CO₂, H₂O, energy (sensible heat and latent energy), and CH₄ fluxes; performance across gap lengths (<12 days, >30 days, >3 months); preservation of environment-flux responses in gap-filled data

Outcomes reported

The study evaluated random forest regression (RFR) and marginal distribution sampling (MDS) methods for gap-filling carbon dioxide, water, energy, and methane flux data across challenging ecosystems including managed pastures, converted peatlands, and drylands. Performance was assessed across different gap lengths and ecosystem types to determine optimal approaches for each flux type and data availability scenario.

Topic tags