Comparison of two gap-filling techniques for nitrous oxide fluxes from agricultural soil

Abstract

Micrometeorological methods are ideally suited for continuous measurements of N2O fluxes, but gaps in the time series occur due to low-turbulence conditions, power failures, and adverse weather conditions. Two gap-filling methods including linear interpolation and artificial neural networks (ANN) were utilized to reconstruct missing N2O flux data from a corn–soybean–wheat rotation and evaluate the impact on annual N2O emissions from 2001 to 2006 at the Elora Research Station, ON, Canada. The single-year ANN method is recommended because this method captured flux variability better than the linear interpolation method (average R2 of 0.41 vs. 0.34). Annual N2O emission and annual bias resulting from linear and single-year ANN were compatible with each other when there were few and short gaps (i.e., percentage of missing values <30%). However, with longer gaps (>20 d), the bias error in annual fluxes varied between 0.082 and 0.344 kg N2O-N ha−1 for linear and 0.069 and 0.109 kg N2O-N ha−1 for single-year ANN. Hence, the single-year ANN with lower annual bias and stable approach over various years is recommended, if the appropriate driving inputs (i.e., soil temperature, soil water content, precipitation, N mineral content, and snow depth) needed for the ANN model are available.

Global Research Alliance N2O chamber methodology guidelines: Guidelines for gap‐filling missing measurements

“…Neural networks are a fundamental building block of modern machine learning and data science and have become one of the strongest approaches for data‐driven predictive modeling in many fields. Initial results using NN within N 2 O flux data have been successful, with NN providing higher R 2 values than linear interpolations when used for gap‐filling agricultural datasets (Bigaignon et al., 2020; Taki et al., 2018). A NN tries to predict an output variable (N 2 O flux in this case) by determining the interconnected relationships between the covariate data (e.g., temperature, moisture, inorganic N, etc.)…”

“…Future research is needed for a comprehensive treatment of gap‐filling that can account for issues related to diurnality in fluxes (Grace et al., 2020) and the choice of flux calculation method (Venterea et al., 2020). The methods described in this paper have also been used for gap‐filling non‐chamber‐based measurements (Taki et al., 2018).…”

“…Although soil N 2 O emissions are inherently more difficult to measure and predict than soil carbon dioxide (CO 2 ), the state of gap‐filling methods in N 2 O emissions trail efforts for other ecosystem exchange measures, like that used in eddy covariance CO 2 fluxes (Barba et al., 2018; Hoffmann et al., 2015; Moffat et al., 2007; Wutzler et al., 2018) or methane (CH 4 ) emissions (Dengel et al., 2013; Kim et al., 2019). Examination of some of these methods within soil N 2 O emissions (Bayesian estimates of emissions [Cowan et al., 2020], autoregressive integrated moving averages [ARIMA, De Rosa et al., 2016; De Rosa et al., 2018], generalized additive model [GAMs; Cowan et al., 2019; De Rosa et al., 2020; Webb et al., 2019], and artificial neural networks [NNs; Bigaignon et al., 2020; Taki et al., 2018]) has begun, showing promise in their ability to improve estimates. However, there is little understanding about conditions under which these models can successfully be implemented, essential data requirements, and sampling strategies and limitations.…”

“…Beyond the already‐mentioned challenge of underestimating emissions due to missing peak emission events, sampling strategies that do not sample outside of the growing season present issues, as they may be difficult to extrapolate to an annual value or accurate EF (Scheer et al., 2017; Wagner‐Riddle et al., 2017). Although some studies have examined the influence of sampling frequency from chamber methods on reported annual emissions (Barton et al., 2015; Mishurov & Kiely, 2011; Savage, Phillips, & Davidson, 2014), research into determining and testing gap‐filling methods has been limited (Bigaignon, Fieuzal, Delon, & Tallec, 2020; Cowan et al., 2019; De Rosa et al., 2018; Taki, Wagner‐Riddle, Parkin, Gordon, & VanderZaag, 2018; Webb et al., 2019). Thus, there is a need for updating methodologies within soil N 2 O research to improve gap‐filling and develop estimation methods.…”

Global Research Alliance N2O chamber methodology guidelines: Summary of modeling approaches

“…Further methods, such as look‐up tables or regional‐level estimates, have been used to estimate annual emissions as well (Berdanier & Conant, 2012; Mishurov & Kiely, 2011). Recently, advanced methods, such as artificial neural networks (ANN), have started to be tested and show promise to improve estimates beyond that of simple linear interpolations (Taki, Wagner‐Riddle, Parkin, Gordon, & VanderZaag, 2018).…”