The use of artificial neural networks (ANNs) to simulate N2O emissions from a temperate grassland ecosystem

An artificial neural network (ANN) was used to simulate nitrous oxide (N2O) emissions from an intensive grassland ecosystem in New Zealand. Daily N2O emitted was simulated as a function of six input variables of daily rainfall, soil moisture content and temperature, soil nitrate (NO3−), ammonium (NH4+) and total inorganic nitrogen content. Results showed that the ANN was able to calibrate itself to within ±0.77% of measured N2O values in the training data set, and within ±2.0% of values used in the validation data set. This was well within the range of the calculated uncertainties (CV=10–43%) of the measured N2O emissions in the field, and demonstrated that ANNs are a viable tool for simulating complex and highly variable biological systems.