Simulating net ecosystem exchange, latent heat and sensible heat fluxes in rainfed and irrigated maize using artificial neural networks

The main purpose of this research was first, the development of multi-layer perceptron (MLP) neural networks using back-propagation (BP) algorithm in order to explore their capabilities to accomplish simulation and prediction of the net ecosystem exchange (NEE), latent heat (LE) and sensible heat (H) flux measured by eddy covariance in two maize fields (irrigated & rain-fed) near Mead, Nebraska using some related physiological, meteorological, and edaphic factors. Then, the function of two different artificial neural network (ANN) methods known as the MLP (BP) algorithm and the Radial Basis Function (RBF) in order to fill the missing NEE data for rain-fed maize was investigated and compared with the technique suggested by Suyker et al. (2003), and Verma et al. (2005) [V&S]; and the ANN method presented by Papale et al. (2003). For the first part, the results demonstrated a high correlation between actual and estimated data. The R² values for NEE, LE and H were greater than 0.9620, 0.9576, and 0.8001 for both sites, respectively. Furthermore, the RMSE values for LE and H ranged from 0.6580 to 0.0721 W/m², and for NEE, from 0.0681 to 0.0642 (mg m¯² s¯¹). In addition, the sensitivity of the fluxes with respect to each input was analyzed over the growth stages. The most powerful effects among the inputs for LE flux were identified net radiation, leaf area index, vapor pressure deficit, wind speed, for H flux net radiation, wind speed, air temperature, leaf area index and vapor pressure deficit and for NEE Rn and LAI for irrigated site and Rn, and VPD for rain-fed site. For the second part, the results showed that the RBF network was able to find better fits for missing values compared to the MLP (BP) network and S&V method. Moreover, data analysis indicated Papale’s approach gave better fits than the RBF and MLP (BP) methods