Modeling of denitrification rates in eutrophic wetlands by artificial neural networks

Abstract

Eutrophication can be controlled by denitrification which is complex microbial processes converting nitrate to nitrogen gas in water body. Various environmental factors such as oxygen, available carbon and pH are known to regulate denitrification rate. However, those controlling variables affect denitrification rates nonlinearly. Further, interactions between those factors hinder a good prediction on the rate. As such, conventional mechanistic modeling of denitrification often fails to fit with data collected from fields. In this study, we applied artificial neural networks to elucidate complex relationships between denitrification rate and environmental factors. A MultiLayer Perceptron network based on the back-propagation algorithm was utilized for prediction of denitrification rate. High predictability of denitrification was achieved with R=0.910 with the trained network. Subsequently, sensitivity analysis was carried out to evaluate factors predominantly controlling denitrification rates. A sensitivity analysis exhibited that DO exert a dominant controlling effect on denitrification rate over other environmental factors.