Estimating the environmental impacts of 57,000 food products

Abstract

Understanding and communicating the environmental impacts of food products is key to enabling transitions to environmentally sustainable food systems [El Bilali and Allahyari, Inf. Process. Agric. 5, 456-464 (2018)]. While previous analyses compared the impacts of food commodities such as fruits, wheat, and beef [Poore and Nemecek, Science 360, 987-992 (2018)], most food products contain numerous ingredients. However, because the amount of each ingredient in a product is often known only by the manufacturer, it has been difficult to assess their environmental impacts. Here, we develop an approach to overcome this limitation. It uses prior knowledge from ingredient lists to infer the composition of each ingredient, and then pairs this with environmental databases [Poore and Nemecek Science 360, 987-992 (2018); Gephart et al., Nature 597, 360-365 (2021)] to derive estimates of a food product's environmental impact across four indicators: greenhouse gas emissions, land use, water stress, and eutrophication potential. Using the approach on 57,000 products in the United Kingdom and Ireland shows food types have low (e.g., sugary beverages, fruits, breads), to intermediate (e.g., many desserts, pastries), to high environmental impacts (e.g., meat, fish, cheese). Incorporating NutriScore reveals more nutritious products are often more environmentally sustainable but there are exceptions to this trend, and foods consumers may view as substitutable can have markedly different impacts. Sensitivity analyses indicate the approach is robust to uncertainty in ingredient composition and in most cases sourcing. This approach provides a step toward enabling consumers, retailers, and policy makers to make informed decisions on the environmental impacts of food products.

Keywords: ecolabelling; environmental impact of food; food system sustainability.

Fig. 1.

Approach used to estimate the environmental impact score for each food product. See Methods and SI Appendix, Supplementary Information Text, for more information.

Fig. 2.

Accuracy of estimated environmental impact scores. Panels show (A) the distribution of the accuracy when the percent composition of no ingredients is known, how accuracy varies with (B) the number of similar products when the composition of no ingredients is known, (C) the number of ingredients in the product when the percent composition of no ingredients is known, and (D) the percentage of the product’s ingredient composition is known. In (A) the x-axis is cut off at −50% and 50% for visibility; an additional 31 data points outside these limits, or 2.0% of the data sample, are available in SI Appendix, Dataset S3. Accuracy is the difference between the known and estimated environmental impact score (in percent), and is calculated as estimated environmental impact score/(known environmental impact score – estimated environmental impact score). The known environmental impact score is calculated using information in the ingredient list, while the estimated environmental impact score is calculated using randomly selected subsets of composition information for each product. Boxplots in B–D indicate, from bottom to top, the 25th percentile – 1.5*IQR (interquartile range), 25th percentile, median, 75th percentile, and 75th percentile + 1.5*IQR.

Fig. 3.

Environmental impact scores per 100 g of products in Tesco Aisles. Points indicates mean impact of all products categorized to the Aisle, and error bars indicate ±1 SEM. Aisles are colored by food type. Food types are shown from lowest median environmental impact on the left to highest median environmental impact at the right. Aisles within food types are ordered from lowest to highest mean environmental impact score. When plotting, Aisles containing similar products were condensed for visibility and clarity (see SI Appendix, Supplementary Information Text ). For instance, the Aisles “Fresh Vegetables” and “Frozen Vegetables” were condensed into “Vegetables.”

Fig. 4.

Environmental impact score and nutrition impact score per 100 g of multi-ingredient food products. Data were limited to products available for purchase from Tesco, and were categorized into Aisles using information from Tesco’s website. Colors indicate food types. Points indicate the environmental impact and nutrition impact scores of Aisles not denoted by a text label. When plotting, Aisles containing similar products were condensed for visibility and clarity (see SI Appendix, Supplementary Information Text ). For instance, the Aisles “Fresh Vegetables” and “Frozen Vegetables” were condensed into “Vegetables.” Labels were jittered to avoid overlap. See SI Appendix, Fig. S10 for the impacts of Aisles within each food type were separated into different panels, and SI Appendix, Fig. S11 for data from Tesco in which Aisles were not condensed for visibility and for data from the eight other food retailers in the analysis.

Fig. 5.

Radar plots indicating the variation in impacts per 100 g of product within each retail Aisle. Solid vertical and horizontal lines separate radar plots into one of eight food types, which are further differentiated by color. Within each food type, radar plots are ordered such that the Aisle with the lowest mean environmental impact score and nutrition impact score are in the upper left, and the Aisle with the highest environmental and nutrition impact are in the bottom right. In each radar plot, the black line indicates the mean impact of products in the Aisle while the shaded area indicates the 5th to 95th percentile impacts of the products within that Aisle. Inner and outer bounds of the radar plot correspond with 5th and 95th percentile impact of all food products available at Tesco, respectively. For the four individual nutrition indicators, the values used for plotting are the number of points assigned to each product for that food component using the NutriScore algorithm. For the four environmental indicators, the values used for plotting are the environmental impact per 100 g of product for the environmental indicator. When plotting, Aisles containing similar products were condensed for visibility and clarity (see SI Appendix, Supplementary Information Text). For instance, the Aisles “Fresh Vegetables” and “Frozen Vegetables” were condensed into “Vegetables.”

Fig. 6.

Variation in environmental and nutritional impact of sausages. Each point indicates a single food product, is colored to indicate different food types, and partially transparent to show overlaps of food products with similar environmental impact and nutrition impact scores. In (A), point shape indicates whether the pesto sauce contains (circles) or does not contain nuts (crosses). Products were identified based on the retail Aisle and Shelf they were categorized in and their product name. Data sample contains 503 sausages.