Gap‐filling approaches for eddy covariance methane fluxes: A comparison of three machine learning algorithms and a traditional method with principal component analysis

Abstract

Methane flux (FCH4) measurements using the eddy covariance technique have increased over the past decade. FCH4 measurements commonly include data gaps, as is the case with CO2 and energy fluxes. However, gap‐filling FCH4 data are more challenging than other fluxes due to its unique characteristics including multidriver dependency, variabilities across multiple timescales, nonstationarity, spatial heterogeneity of flux footprints, and lagged influence of biophysical drivers. Some researchers have applied a marginal distribution sampling (MDS) algorithm, a standard gap‐filling method for other fluxes, to FCH4 datasets, and others have applied artificial neural networks (ANN) to resolve the challenging characteristics of FCH4. However, there is still no consensus regarding FCH4 gap‐filling methods due to limited comparative research. We are not aware of the applications of machine learning (ML) algorithms beyond ANN to FCH4 datasets. Here, we compare the performance of MDS and three ML algorithms (ANN, random forest [RF], and support vector machine [SVM]) using multiple combinations of ancillary variables. In addition, we applied principal component analysis (PCA) as an input to the algorithms to address multidriver dependency of FCH4 and reduce the internal complexity of the algorithmic structures. We applied this approach to five benchmark FCH4 datasets from both natural and managed systems located in temperate and tropical wetlands and rice paddies. Results indicate that PCA improved the performance of MDS compared to traditional inputs. ML algorithms performed better when using all available biophysical variables compared to using PCA‐derived inputs. Overall, RF was found to outperform other techniques for all sites. We found gap‐filling uncertainty is much larger than measurement uncertainty in accumulated CH4 budget. Therefore, the approach used for FCH4 gap filling can have important implications for characterizing annual ecosystem‐scale methane budgets, the accuracy of which is important for evaluating natural and managed systems and their interactions with global change processes.

Methane emissions reduce the radiative cooling effect of a subtropical estuarine mangrove wetland by half

“…used another machine learning algorithm ANN to address the nonlinearity problem. Both random forest and ANN are capable of handling complex processes and overfitting issues, but the tree ensemble of random forest model can be interpreted more easily in comparison with the black box problem of ANN(Banerjee, Ding, & Noone, 2012) Kim et al (2020). found that random forest performed better than ANN in gap-filling F CH4 Tramontana et al (2016).…”

Decadal variation in CO<sub>2</sub> fluxes and its budget in a wheat and maize rotation cropland over the North China Plain

“…The cropland has been reported as carbon neutral to the atmosphere (e.g., Ciais et al, 2010), as a carbon source (e.g., Anthoni et al, 2004a;Verma et al, 2005;Kutsch et al, 2010;Wang et al, 2015;Eichelmann et al, 2016) and also as a carbon sink (e.g., Kutsch et al, 2010). Such inconsistency probably results from the different crop types and management practices (residue removal, the use of organic manure, etc.…”

“…Field management practices (e.g., irrigation, fertilization and residue removal, etc.) impact the cropland CO 2 fluxes (Baker and Griffis, 2005;Béziat et al, 2009;Ceschia et al, 2010;Eugster et al, 2010;Drewniak et al, 2015;de la Motte et al, 2016;Hunt et al, 2016;Vick et al, 2016), but their relative importance in determining the cropland CO 2 budget remain unclear because of limited field observations (Kutsch et al, 2010), motivating comprehensive CO 2 budget assessments across different cropland management styles.…”

“…Differentiating among these components is a prerequisite for understanding the response of terrestrial ecosystems to changing environment (Heimann and Reichstein, 2008), thus the carbon budget evaluations have been reported for a few croplands (e.g., Moureaux et al, 2008;Ceschia et al, 2010;Wang et al, 2015;Demyan et al, 2016;Gao et al, 2017). In particular, to account for the literal carbon export, the net biome productivity (NBP) is often estimated by combining the eddy covariance technique and field carbon measurements associated with harvests and residue treatments (Ceschia et al, 2010;Kutsch et al, 2010). As a detailed CO 2 budget might facilitate better predictions of agroecosystems' responses to climate change, CO 2 budget evaluations in different croplands remain necessary.…”

A Remote Sensing Technique to Upscale Methane Emission Flux in a Subtropical Peatland

“…For example, Peltola et al (2019) applied a Random Forest machine learning approach to upscaling eddy covariance CH 4 flux observations in northern wetlands using 500-to 1,000-m Moderate Resolution Imaging Spectroradiometer (MODIS) products, wetland maps, and other gridded data sets. Machine learning-based models have proven valuable in gap filling of eddy covariance CH 4 flux data (e.g., Dengel et al, 2013;Kim et al, 2020;Morin et al, 2017) and upscaling eddy covariance observations in northern wetlands (Peltola et al, 2019). However, machine learning models have been barely applied for upscaling point-based CH 4 flux measurements using relatively high-resolution Landsat products for local/regional emission mapping.…”