Determination and modelling of energy consumption in wheat production using neural networks : "A case study in Canterbury Province, New Zealand"
New Zealand farmers practice a form of 'industrialised‘ agriculture that relies on relatively high inputs of fossil fuels, not only to power machinery directly but also for manufacturing artificial fertilizers and agrichemicals (Wells, 2001). Consequently, New Zealand is one of the countries with the highest energy input per unit weight (in agriculture) in the world (Conforti & Giampietro, 1997). Furthermore, in terms of shipping, the influence of increasing global fuel costs is greater on New Zealand farming than in other countries. The main aim of this study was to estimate energy consumption in wheat production. Energy determination can give a clear picture of farms in order to compare different farming systems and energy inputs. The second main target of this study was to develop a neural network model to simulate and predict energy use in wheat production under different conditions incorporating social, geographical, and technical factors. Additionally, the interaction effects between different factors were examined in this study. This study was conducted on irrigated and dryland wheat fields in Canterbury, New Zealand, in the 2007-2008 harvest year. Canterbury represents 87% of the wheat area and 66% of the arable area harvested in New Zealand. Energy consumption here is defined as the energy used for the production of wheat until it leaves the farm. The data were collected from three different sources: questionnaire, literature review, and field measurements. The energy inputs estimated in this study are those that go into on-farm production systems before the post-harvest processes. The study considered only the energy used in wheat production, without taking into account the natural sources of energy (radiation, wind, rain, etc). A survey was conducted to collect the most important data and to identify farmers‘ attitudes and opinions about energy consumption. In this study, 40 arable farms were selected randomly, as far as possible. From the initial analysis, it was found that 30 farms were irrigated and the rest were dryland farms. Irrigated farms were irrigated between one to ten times annually depending on the rainfall. Some irrigated farms have also been converted to dryland farms, or vice versa, in different years. The data for a large number of farming factors were gathered in the survey. Average energy consumption for wheat production was estimated at around 22,600 MJ/ha. On average, fertilizer and electricity (mostly for irrigation) were used more than other energy sources, at around 10,654 MJ/ha (47%) and 4,870 MJ/ha (22%), respectively. The average energy consumption for wheat production in irrigated farming systems and dryland farming systems was estimated at 25,600 and 17,458 MJ/ha, respectively. This study is the first to create an appropriate Artificial Neural Network (ANN) model to predict energy consumption in wheat production with optimum variables. This study would be the first to investigate the factors related to the efficient use of energy in agricultural production. A careful study of all factors was first made to find trends and correlations and their relationship to energy consumption. A two step approach to input reduction involving correlation and Principal Component Analysis (PCA) revealed five highly relevant inputs for predicting energy consumption. After testing different learning algorithms, neuron activation functions and network structures using genetic algorithm optimization, a modular network with two hidden layers was developed using Quick Prop learning method. The final model can predict energy consumption based on farm conditions (size of crop area), farmers' social considerations (level of education), and energy inputs (amount of N and P used and irrigation frequency). It predicts energy use in Canterbury arable farms with an error margin of ± 2972 MJ/ha (12%) and this size of an error in agricultural studies with several uncontrolled factors and as an initial investigation is acceptable. Furthermore, comparisons between the ANN model and a Multiple Linear Regression (MLR) model showed that the ANN model can predict energy consumption better than the MLR model. As part of conclusions, this thesis provides extensive suggestions for future research and recommendations for reducing energy consumption in wheat production with minimum income loss.... [Show full abstract]