Towards sustainable agriculture: modelling carbon emission prediction in crop production

<p dir="ltr">The challenge of mitigating carbon emissions in the Food Supply Chain (FSC) is critical, particularly within the agriculture sector, where emissions significantly impact the environment. This study identifies the gap in understanding Carbon Factor (CF) across diverse locations and crop types and aims to develop a comprehensive framework for standardising carbon factor estimation through Machine Learning (ML) techniques. The study identifies key features in agricultural activities that contribute to carbon emissions and further develops a methodology to transform agricultural datasets to represent these key standardised features. A major contribution of the thesis is this feature standardisation technique for training universal agricultural carbon factor estimation models. A robust methodology has been proposed that integrates various regression models, employing advanced feature engineering to process agricultural data from farms in Iran, Morocco, and Peru. The experiments are performed using models such as Random Forest (RF), Multiple Linear Regression, Lasso Regression, Neural Network Regression (NNR), K-Nearest Neighbours (KNN) Regression, and a hybrid model, with a focus on empirical validation of a universal carbon factor estimation model applicable across different regions and crops. Results demonstrate that the developed models achieve high estimation accuracy, with the Random Forest model yielding a R2 score of 0.9887 in single-crop scenarios and 0.9733 in multi-crop applications. Additionally, the Life Cycle Assessment (LCA)-based Cool Farm Tool (CFT) is used as a baseline to compare the outcomes of the machine learning models against traditional techniques. The findings confirm that machine learning models work better, and while feature significance may vary, a core set of features consistently influences model predictions. This research addresses the gaps in accurate carbon factor estimation using machine learning in agriculture, particularly focusing on effective methods and regional agricultural challenges. The implications of this research are substantial for sustainable agricultural practices, providing regional decision-makers with actionable insights to balance productivity and environmental responsibility while significantly informing carbon emission reduction strategies within food production.</p>