Modeling Carbon Dioxide Flux over a Cropland Using Artificial Neural Networks

The goal of this research is the simulation of meteorological and physiological impacts on CO2 flux using artificial neural networks over a determinate timescale. Dynamic of CO2 flux apart from its relationships to genetic of cultivar, physiological, ecological and edaphic factors, strongly depends on meteorological conditions. Moreover, the nonlinearity relationship between CO2 flux and the factors that mentioned above limits the capabilities of CO2 models in order to predict flux dynamics accurately. Recently, the application of Artificial Neural Networks (ANNs) has developed into a powerful tool that can compute most complicated equations and numerical analyses to the best approximation. In this study, a multi-layer perceptron with back-propagation algorithm applied to simulate CO2 flux above two different sits,rain-fed and irrigated maize, in Mead, Nebraska. Leaf area index, soil water content, soil temperature, air temperature, vapor pressure deficit, precipitation, net radiation, and wind speed were used to train the networks and predict CO2 flux. Diurnal hourly data from June 18th through end of August for 2001, 2003, 2005, 2007, and 2009 were randomly used to make different networks in order to train and test and finally gain the optimum predicted values. The results showed that R² and RMSE values for rain-fed and irrigated sites were 0.96, 0.069 and 0.95, 0.93 respectively.