Investigating the application of machine learning models to improve the eddy covariance data gap-filling

Errors and uncertainties associated with gap filling of water, carbon, and energy flux data have been a challenge for the global network of micrometeorological tower sites that use the eddy covariance (EC) technique. This research aims to determine whether more sophisticated machine learning (ML) methods can improve the gap-filling of flux and meteorological data. Results showed that simple models performed almost as well as ML models for filling meteorological gaps. However, for the more complex EC gas flux variables, ML models outperformed simple linear models, with random forest resulting in the smallest gap-filling uncertainty for CO₂ flux. This work has improved on existing EC data gap-filling approaches by reducing uncertainty.