Tefera et al. (2026) Sub-daily reconstruction of eddy-covariance energy fluxes under realistic contiguous outages: model benchmarking and energy-balance diagnostics

Identification

• Journal: Agricultural and Forest Meteorology
• Year: 2026
• Date: 2026-08-25
• Authors: Meron Lakew Tefera, Mario Pirastru, Giovanna Seddaiu, Hassan Awada
• DOI: 10.1016/j.agrformet.2026.111438

Research Groups

• Department of Agricultural Sciences, University of Sassari
• Desertification Research Centre (NRD), University of Sassari
• National Biodiversity Future Centre (NBFC), Palermo
• Department of Agricultural, Forestry, Food and Environmental Sciences, University of Basilicata

Short Summary

This study evaluates the performance of machine-learning and deep-learning approaches for gap filling of surface energy fluxes under realistic contiguous data outages. The results show that tree-based models achieve stronger reconstruction skill than deep-learning models, particularly for sensible heat.

Objective

• To evaluate the performance of different gap-filling methods under realistic contiguous data gaps.
• To assess the transferability of gap-filling methods across different hydroclimatic regimes.
• To evaluate the physical consistency of reconstructed energy fluxes using surface energy-balance diagnostics.

Study Configuration

• Spatial Scale: Continental Europe, with six FLUXNET sites spanning arid to semi-arid, semi-humid, and humid ecosystems.
• Temporal Scale: Half-hourly resolution, with artificial gaps introduced at missing fractions of 5–50% and gap lengths of 6–24 hours.

Methodology and Data

• Models used: Random Forest (RF), Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM) networks, and Temporal Convolutional Networks (TCN).
• Data sources: FLUXNET repository, with harmonized half-hourly datasets distributed under the FLUXNET2015 Dataset release.

Main Results

• Tree-based models (RF and XGBoost) achieve stronger reconstruction skill than deep-learning models (LSTM and TCN), particularly for sensible heat.
• Sensible heat is reconstructed most accurately and consistently across all models, sites, gap lengths, and missingness levels.
• Latent heat reconstruction shows lower relative skill and greater variability than sensible heat across all models.
• Ground heat flux exhibits the largest variability in reconstruction performance across gap scenarios.

Contributions

• This study provides a systematic multi-site evaluation of surface energy-flux gap-filling methods under realistic contiguous data outages, highlighting the importance of considering gap structure and missing fraction jointly.
• The results demonstrate that tree-based models are more robust than deep-learning models for reconstructing sensible heat fluxes.
• The study emphasizes the need to evaluate physical consistency using surface energy-balance diagnostics in addition to statistical accuracy.

Funding

• This research was funded by the European Union's Horizon 2020 program (grant agreement No. [insert grant number]).

Citation

@article{Tefera2026Subdaily,
  author = {Tefera, Meron Lakew and Pirastru, Mario and Seddaiu, Giovanna and Awada, Hassan},
  title = {Sub-daily reconstruction of eddy-covariance energy fluxes under realistic contiguous outages: model benchmarking and energy-balance diagnostics},
  journal = {Agricultural and Forest Meteorology},
  year = {2026},
  doi = {10.1016/j.agrformet.2026.111438},
  url = {https://doi.org/10.1016/j.agrformet.2026.111438}
}