Advanced statistical methods for gap-filling N2O emissions data

Agriculture is a significant contributor to global nitrous oxide (N₂O) emissions. Due to various constraints, emission data is typically not measured continuously. This leads to gaps in data that must be filled to better estimate annual emissions. The most common way of filling gaps in emission data is to do a simple linear interpolation between missing data points. However, when the length or number of gaps increases, this method becomes less effective due to the high variability in N₂O fluxes. Thus, the use of simplistic gap-filling methods can potentially lead to inaccurate annual estimates. Having accurate annual estimates is vital for determining the potential of different management strategies to reduce N₂O emissions. Advanced statistical methods utilizing auxiliary data, such as climate and soil measurements, present an opportunity to inform gap-filling estimates. These advanced methods, which potentially require substantial auxiliary data, have just started to be researched within the N₂O field. One of these gap-filling methods is a Feedforward Multi-Layer Perceptron (MLP) neural network, trained using backpropagation to predict N₂O emissions. Our model had three objectives; 1) assess performance of the MLP to predict N₂O emissions, 2) determine a minimum number of data points and examine auxiliary data needs, and 3) compare MLP performance against other gap-filling methods. Initial results suggest the MLP is a good (R²>0.7) method for estimating N₂O emissions when compared to field data. Further testing across a suite of sites found in the Global N₂O Database is being done to examine minimum data needs and comparison to other gap-filling methods. Our next step is working on developing a globally convergent algorithm that will allow researchers to use our model to analyze a site and compare various gap-filling estimates.