Abstract
Two models have been developed for simulating CO2 emissions from wheat farms: (1) an artificial neural network (ANN) model; and (2) a multiple linear regression model (MLR). Data were collected from 40 wheat farms in the Canterbury region of New Zealand. Investigation of more than 140 various factors enabled the selection of eight factors to be employed as the independent variables for final the ANN model. The results showed the final ANN developed can forecast CO2 emissions from wheat production areas under different conditions (proportion of wheat cultivated land on the farm, numbers of irrigation applications and numbers of cows), the condition of machinery (tractor power index (hp/ha) and age of fertilizer spreader) and N, P and insecticide inputs on the farms with an accuracy of ±11% (± 113 kg CO2/ha). The total CO2 emissions from farm inputs were estimated as 1032 kg CO2/ha for wheat production. On average, fertilizer use of 52% and fuel use of around 20% have the highest CO2 emissions for wheat cultivation. The results confirmed the ANN model forecast CO2 emissions much better than MLR model.
Highlights
Around the world wheat is used as one of the main food sources to provide a large proportion of the calories and protein needed by human beings [1]
The feed-forward multi-layered perception (MLP) paradigm consists of independent variables, hidden layers and an output layer trained by the back propagation (BP) learning method
The study revealed that an average of 1032 kg CO2/ha, was released from each wheat cultivation farm
Summary
Around the world wheat is used as one of the main food sources to provide a large proportion of the calories and protein needed by human beings [1]. A simple model with a high r2 can be developed through the use of sufficient numbers of samples and independent variables. Input variables are always maintained in the best model if the actual and predicted data are correlated with a p value of 0.05 [23]. ANN models are simple applications that can predict or classify different data to give with robust results. The feed-forward MLP paradigm consists of independent variables, hidden layers and an output layer trained by the back propagation (BP) learning method. The neurons associated with the first hidden layer analysis, the weighted independent variables, use a transfer function to lead to the results. Data reduction is useful if the number of input variables is high and the available sample size is limited [25]
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call