A comparison of methods for gap-filling sensible and latent heat fluxes in different climatic conditions

Eddy covariance flux measurements need to be gap-filled when utilising the data for the calculation of annual balances. The measurement technique itself is prone to errors and technical failures may also lead to gaps of various lengths. Gap-filling of the flux time series is typically based on estimating statistically representative values based on various environmental variables through linear regression, lookup tables or machine learning methods.

A large number of methods for the imputation of energy fluxes have been applied and compared in recent literature (Zhu et al. 2022; Mahabbati 2022; Khan, Jeon, and Jeong 2021; Foltýnová, Fischer, and McGloin 2019). Both latent and sensible heat fluxes are strongly driven by the incoming solar radiation, and it is usually used as an independent variable in gap-filling models. Vekuri et al. showed that a widely used method for gap-filling carbon dioxide fluxes creates a systematic bias in northern ecosystems, where the distribution of incoming radiation is highly skewed.

Here, we assess if a similar bias error emerges for sensible and latent heat fluxes after gap-filling with the standard methods or suggested alternatives. We use global data from openly available flux measurement databases and compare the bias and other metrics between different latitudes. We assume that the errors in total energy balances are not as significant as in carbon budgets, but the results could still indicate which methods should be preferred when complete time series of energy flux data are needed.

 

References

Foltýnová, L., M. Fischer, and R.P. McGloin, ‘Recommendations for Gap-Filling Eddy Covariance Latent Heat Flux Measurements Using Marginal Distribution Sampling’, Theoretical and Applied Climatology, Vol. 139, No. 1–2, September 11, 2019, pp. 677–688.

Khan, M.S., S.B. Jeon, and M.-H. Jeong, ‘Gap-Filling Eddy Covariance Latent Heat Flux: Inter-Comparison of Four Machine Learning Model Predictions and Uncertainties in Forest Ecosystem’, Remote Sensing, Vol. 13, No. 24, January 2021, p. 4976.

Mahabbati, A., ‘Investigating the Application of Machine Learning Models to Improve the Eddy Covariance Data Gap- Filling’, The University of Western Australia, 2022.

Vekuri, H., J.-P. Tuovinen, L. Kulmala, D. Papale, P. Kolari, M. Aurela, T. Laurila, J. Liski, and A. Lohila, ‘A Widely-Used Eddy Covariance Gap-Filling Method Creates Systematic Bias in Carbon Balance Estimates’, Scientific Reports, forthcomig.

Zhu, S., R. Clement, J. McCalmont, C.A. Davies, and T. Hill, ‘Stable Gap-Filling for Longer Eddy Covariance Data Gaps: A Globally Validated Machine-Learning Approach for Carbon Dioxide, Water, and Energy Fluxes’, Agricultural and Forest Meteorology, Vol. 314, March 1, 2022, p. 108777.