Estimating the Agricultural Environmental Burden As Part of a Holistic Life Cycle Assessment of Food

Abstract

A precise calculation of the environmental burden of food products is a prerequisite for creating food eco-labeling as a strategy for environmental impact mitigation. Life cycle assessment (LCA) is widely used for this purpose, and proxy data is traditionally used due to the shortage of data. Uncertainties are introduced in this process since food products contain a variety of origins. In this study, data from the United States Department of Agriculture (USDA) is used to examine the temporal and geographic variability of the global warming potential (GWP) of seven kinds of field crops. Artificial neural network (ANN) models are then used to predict the GWP of these products at both product and category levels based on temporal and spatial variables such as soil properties, climate, latitude and elevation. The results show that temporally, a monotonic GWP trend was found in corn, soybean and winter wheat. The average geographic variability is more than 27% and is larger than temporal variability. ANN was proven to be a good prediction tool at the product level, with a coefficient of correlation (CC) of at least 0.78 in the simplest model and higher CCs when the number of neurons increases. Predictions with ANN at the category level shows that the selected variables cannot fully encompass all temporal and geographical variability.