Developing a neural network to simplify durum wheat yield prediction in Italy

Crop models are frequently used in agronomy for simulating crop variables at a discrete time step. This paper describes the application of an artificial neural network in developing a model for yield forecasts in durum wheat, using back-propagation algorithms based on the mechanistic model AFRCWHEAT2 (Porter, 1993; Porter et al., 1993). Given the relevant number of inputs (16) required to operate AFRCWHEAT2, we have tried to develop a simpler model based on a neural network, in order to match AFRCWHEAT2 performance while substantially reducing the need of inputs.