A comparison of gap-filling algorithms for eddy covariance fluxes and their drivers

Abstract

Abstract. The errors and uncertainties associated with gap-filling algorithms of\nwater, carbon, and energy fluxes data have always been one of the main\nchallenges of the global network of microclimatological tower sites that use the eddy covariance\xa0(EC) technique. To address these concerns and find more efficient gap-filling algorithms, we reviewed eight algorithms to estimate missing values of environmental drivers and nine algorithms for\nthe three major fluxes typically found in EC time series. We then examined\nthe algorithms performance for different gap-filling scenarios utilising\nthe data from five EC\xa0towers during\xa02013. This research s objectives were (a)\xa0to evaluate the impact of the gap lengths on the performance of each\nalgorithm and (b)\xa0to compare the performance of traditional and new\ngap-filling techniques for the EC data, for fluxes, and separately for their corresponding meteorological drivers. The algorithms performance was\nevaluated by generating nine gap windows with different lengths, ranging\nfrom a day to 365\u2009d. In each scenario, a gap period was chosen randomly,\nand the data were removed from the dataset accordingly. After running each\nscenario, a variety of statistical metrics were used to evaluate the\nalgorithms performance. The algorithms showed different levels of\nsensitivity to the gap lengths; the Prophet Forecast Model\xa0(FBP) revealed\nthe most sensitivity, whilst the performance of artificial neural networks\xa0(ANNs), for instance, did not vary as much by changing the gap length. The algorithms performance generally decreased with increasing the gap length, yet the differences were not significant for windows smaller than 30\u2009d. No significant differences between the algorithms were recognised for the meteorological and environmental drivers. However, the linear algorithms showed slight superiority over those of machine learning\xa0(ML), except the random forest\xa0(RF) algorithm estimating the ground heat flux (root mean square errors – RMSEs – of\xa028.91 and\xa033.92 for\xa0RF and classic linear regression\xa0– CLR, respectively). However, for the major fluxes, ML algorithms and the MDS showed superiority over the other algorithms. Even though ANNs, random forest\xa0(RF), and eXtreme Gradient Boost\xa0(XGB) showed comparable performance in gap-filling of the major fluxes, RF\xa0provided more consistent results with slightly less bias against the other ML algorithms. The results indicated no single algorithm that outperforms in all situations, but the RF\xa0is a potential alternative for the MDS and ANNs as regards flux gap-filling.\n