Artificial neural network approach for prediction of ammonia emission from field-applied manure and relative significance assessment of ammonia emission factors

This article presents a systematic method for enhancing the estimation accuracy of ammonia emission from field-applied manure and for assessing the relative significance of ammonia emission factors, using the feedforward-backpropagation artificial neural network (ANN) approach.

The multivariate linear regression (MLR) method well describes the ammonia emission tendency with the emission factor variation. However, ammonia emission from manure slurry is too complex to be captured in a linear regression model. This necessitates a model which can describe complex nonlinear effects between the ammonia emission variables such as soil and manure states, climate and agronomic factors. In the present study, a principle component analysis (PCA) based preprocessing and weight partitioning method (WPM) based postprocessing ANN approach (called the PWA approach) is proposed to account for the complex nonlinear effects.

The ammonia emission is predicted with precision by the 11 emission factors, using the nonlinear ANN approach. The relative importance among the 11 emission factors is identified using the elasticity analysis in the MLR method and using the WPM in the ANN approach. The relative significance obtained quantitatively by the PWA approach in the present study gives an excellent explanation of the most important processes controlling NH₃ emission.