Application of Bayesian statistics to estimate nitrous oxide emission factors of three nitrogen fertilisers on UK grasslands

Abstract

Trapezoidal integration by linear interpolation of data points is by far the most commonly used method of cumulative flux calculations of nitrous oxide (N 2 O) in studies that use flux chambers; however, this method is incapable of providing accurate uncertainty estimates. A Bayesian approach was used to calculate N 2 O emission factors (EFs) and their associated uncertainties from flux chamber measurements made after the application of nitrogen fertilisers, in the form of ammonium nitrate (AN), urea (Ur) and urea treated with Agrotain® urease inhibitor (UI) at four grassland sites in the UK. The comparison between the cumulative fluxes estimated using the Bayesian and linear interpolation methods were broadly similar (R 2 = 0.79); however, the Bayesian method was capable of providing realistic uncertainties when a limited number of data points is available. The study reports mean EF values (and 95% confidence intervals) of 0.60 ± 0.63, 0.29 ± 0.22 and 0.26 ± 0.17% of applied N emitted as N 2 O for the AN, Ur and UI treatments, respectively. There was no significant difference between N 2 O emissions from the Ur and UI treatments. In the case of the automatic chamber data collected at one site in this study, the data did not fit the log-normal model, implying that more complex models may be needed, particularly for measurement data with high temporal resolution.

Nitrous oxide emission factors of mineral fertilisers in the UK and Ireland: A Bayesian analysis of 20 years of experimental data

“…Data were either i) extracted from published studies in which one aim of the experimentation was to explicitly measure N 2 O and report EFs after a mineral fertiliser application, or ii) raw data were used from the Agricultural and Environmental Data Archive (AEDA). The raw data extracted from the AEDA was used to calculate EFs for 25 days after fertiliser application using the log-normal Bayesian approach as described in Levy et al, 2017 , Cowan et al, 2019a , which accounts for the log-normal spatial and temporal distribution of N 2 O fluxes after fertiliser events. Fertilisers treated with nitrification and urease inhibitors in these studies were included for comparison.…”

“…Some of the EFs in the published studies are calculated by taking a yearly average after several fertiliser applications, while others report emissions for a shorter period after the event (e.g. Skiba et al, 2013 , Cowan et al, 2019a , Cowan et al, 2019b ). The fluxes derived from the data taken from the AEDA archives report EFs for emissions up to 25 days after fertilisation.…”

Assessment of nitrous oxide emission factors for arable and grassland ecosystems

“…One way to reduce uncertainty when calculating cumulative emissions is to apply a Bayesian interpolation approach (Levy et al 2017;Cowan et al 2019Cowan et al , 2020. In this method, daily fluxes are interpolated by fitting values to a spatial log-normal distribution over time, i.e.…”

From research to policy: optimizing the design of a national monitoring system to mitigate soil nitrous oxide emissions

“…Empirical models are derived from measurement data using statistical methods and can be as simple as a single emission factor derived from the survey measurement data, or can be more complicated functions such as regression models [56][57][58][59]. Empirical models are useful in estimating emissions at regional and national scales, and in some cases are more accurate than more complex process-based models [60].…”