The LCA method aims to compare the environmental impacts of the production of dairy butter and creams with plant-based alternative products using a standard attributional approach as per the PAS 2050 (BSI 2012), aligning with the latest international standards for dairy products, published by the International Dairy Federation (IDF 2015) and the European Dairy Association (EDA 2016). This study is not intended for investigating a large-scale change of the two systems nor long-term consequences of a decision to switch from one system to another. For the butter (or Nordic dairy spreads) vs plant-based fat spreads comparison, the functional unit (FU) was 1 kg of product (fresh matter) for spreading, baking or shallow frying, at consumer level. For the dairy cream vs plant-based cream comparison, the FU was 1 kg product (fresh matter) for whipping or cooking, at consumer level. The choice of FU is discussed further in the sensitivity analysis section. To address the research questions above, we developed a regionalised LCA framework to consistently assess a large portfolio (228 plant-based spreads/creams and 40 diary alternatives (see the supplementary material Section 3 for the definition of terminology)) of product recipes sold in 21 countries based on primary data from Upfield (previously Unilever’s margarine business). The methodological framework is presented in Fig. 1, illustrating the main procedural steps, which is inherently iterative. It starts from goal and scope definition, which define the objectives, product systems, data quality requirement and cut-off criteria, as well as spatiotemporal context. In this study, the goal and scope define the overall data quality requirement using “minimal significance level” based on expert judgement for the difference of comparative study results to be considered as significant (see Electronic Supplementary Matrial-ESM_1.docx. Table S3 for minimal significance level definition). It further defines data quality requirement using pedigree scores See the supplementary material Table S6 for key processes (notably agricultural oilseeds LCI datasets) identified through the gap and prioritisation process, which further involved sensitivity analysis of choices related to allocation, models and assumptions together with parameter uncertainty analysis. Results obtained from each step are evaluated against the predefined data quality requirement. In terms of spatial scope definition, the regionalised LCA conducted was required at the country scale for key life cycle stages. It includes variations in product recipes, key agricultural ingredients’ country of origin and corresponding country-specific agricultural practices and embedded natural variations (such as fertilisation, tillage practice, irrigation, yield, climate, soil properties), production factories and energy mixes, as well as packaging designs, transportation distances and packaging materials’ end-of-life. More detailed descriptions of each step are provided in the ESM (ESM_1.docx. Section 1).
The following sections give further descriptions of product recipes, system boundaries, data collection, regionalisation of supply chain, spatial (archetype) LCI development, treatment of LUC and water flow modelling, allocation procedures, sensitivity analyses and parameter uncertainty assessment for climate change results.
Products studied
A total of 228 plant-based spreads/creams are assessed. Of them, 201 had no butter fat and 27 were blended with a small amount of butter fat (less than 18%). For products used mainly for spreading and for baking or shallow frying, we assessed 212 predominately plant-based spreads with different levels of fat and types of packaging, sold in 21 markets in Europe and North America. The plant-based spreads were compared with local butter substitute. Additionally, for Nordic countries, (Denmark, Finland and Sweden) the plant-based spreads were also compared to spreads with 40%, 60% and 75% dairy fat (containing no vegetable fat). Plant-based spreads are packaged in various tubs or wrappers of different shapes and volumes (dairy spread packaging is the same as plant-based spread tubs in Denmark, Finland and Sweden), whereas typical packaging for butter in Europe is aluminium foil laminated paper, or waxed paper in North America. For creams, used for whipping or cooking, we assessed 16 plant-based cream recipes and compared them with their dairy cream alternatives. Packaging formats used for plant-based creams are identical to that of dairy creams (polyethylene terephthalate (PET) bottle or liquid packaging board, depending on the market). The numbers of plant-based spreads/creams and their dairy substitute in each consumer markets are given in the ESM (ESM_1.docx. Table S1 and Table S2).
System boundaries and cut-off
The LCA considered all identifiable activities across the product life cycle (cradle-to-grave) for all products in the 21 markets where they are sold (see Fig. 2). Capital goods (ingredient delivery by trucks and ships, buildings, equipment, etc.) were included wherever data was available, such as for crop production, oil extraction and transformation and dairy processing. Capital goods at the distribution centre and the point of retail were not included as the contributions of these processes to the total system’s environmental impacts were expected to be less than 1%. The capital equipment and infrastructure processes from the ecoinvent database (v3.3) were used in the background system (Wernet et al. 2016). The following processes were left out of the system boundaries, consistent with attributional LCA practices: labour, commuting of workers, administrative work, cattle insemination and disease control. Food loss and food waste can take place at any stage in the products’ life cycle. Statistical data at the national scale for specific product categories are not available and are therefore highly uncertain. At farm and processing level, losses are already accounted for in the processes’ efficiency; therefore, uncertainty remains regarding food losses and waste during distribution, at retail point and at the consumer’s home. There is no evidence showing different food losses and waste rates between plant-based spreads and butter (and between plant-based creams and dairy cream). Further, the PEFCR for Dairy Products (EDA 2016) does not require the inclusion of food waste in the assessment but rather suggests a waste rate of 7% for butterfat products, tested in a sensitivity analysis. Food loss and waste during distribution, at retail point and at the consumer’s home, is thus excluded from the scope of the study. Additional information is given in the ESM (ESM_1.docx. Table S4).
Environmental impact indicators considered
The assessment includes 15 environmental impact indicators from the European ILCD 2011 Midpoint+ v1.08 impact assessment method (JRC-IES 2011). Three additional indicators were included: land occupation (m2/year), which reflects the total area of land used over one year and is a proxy for biodiversity and ecosystem services (Nemecek et al. 2011; Milà i Canals et al. 2012), water consumption (m3), the total amount of fresh water consumed (ISO 14046), which includes, for example, evapotranspiration from irrigation water, and water scarcity footprint (m3 water equivalent (eq)) based on the AWARE approach that assesses the water deprivation potential considering spatial water scarcity differences (Boulay et al. 2018). Additional information is given in the ESM (ESM_1.docx. Table S3).
Data sources, supply chain regionalisation and spatial (archetype) LCI modelling
Primary data was collected from the manufacturer of the plant-based spreads and creams for all process stages within its control, namely recipe (i.e. ingredients and sourcing); oil processing where data is available e.g. from supplier or processing carried out by the manufacturer; product manufacturing; packaging materials weights and specifications; distribution transport distances from factories to markets. Secondary data was used to determine the bill of activities of other stages: crop production for oil crops and feed crops; raw milk production in each country; butter and cream production in each country; packaging materials and properties for butter and cream; distribution transport distances to point of sale (dairy products); storage at distribution centre and at point of sale; use stage; packaging end-of-life. Main data sources are summarised in the ESM (ESM_1.docx. Table S4). The detailed modelling steps are given below, following the described framework in Fig. 1.
Tracing agricultural commodity country of origin
Gap and prioritisation analysis of the preliminary LCA results indicates that the most important data to be improved are spatial differentiations of agricultural ingredients. The modelling of crop-country combinations for agricultural ingredients is described below:
- 1.
Identification of crop sources and vegetable oil refining activities. When primary data of sourcing of country of origins were unavailable or incomplete (e.g. countries or regions are known, but exact quantities are unknown), the sourcing mix was based on average historical (2006-2011) FAOSTAT data for import and domestic production (country of origin and % sourcing). The model assumes that the final sourcing mix is proportional to the total of domestic and imported production volumes. A list of datasets accounting for parameters representative of average cultivation practices for each crop-country combination in the supply chain was developed for this study.
- 2.
Gap assessment for spatially differentiated LCI data development. The availability and quality of spatially differentiated country-level LCI datasets were evaluated according to crop sourcing information and data quality requirements. A list of missing data for further development are identified.
- 3.
Gap assessment for spatially differentiated elementary flows for impact assessment. Regionalised inventory data was further examined to evaluate the consistency with the requirements of the impact assessment methods. As a result, a customised version of ecoinvent v3.3 was developed to consistently support the AWARE method for the water scarcity footprint. The assessment of the water scarcity footprint indicator requires particular attention to the consistent modelling of all life cycle inventory data, both in the foreground and background systems. In the present study, all foreground and background inventory data were adapted to ensure the following: Water flows in every process were properly balanced, which enabled calculation of the amount of water consumed as the difference between inputs and outputs. Water flows were all regionalised at country level as per the location where the withdrawals (inputs) and releases (outputs) were taking place, therefore enabling the association to the appropriate characterization factor.
- 4.
With key missing data identified, the sections below provide more details regarding generation of spatially differentiated (archetype) LCI datasets for plant-based and dairy products and the inclusion of GHG emissions from LUC.
Spatially differentiated LCI data generation for plant-based products
To conduct the gap assessment for plant-based spreads and plant-based creams, many of the regionalised LCI data were derived from the World Food LCA Database (WFLDB) v3.1 (Nemecek et al. 2015), which was updated with ecoinvent v3.3 data, system model “Allocation, cut-off by classification” (Weidema et al. 2013). The WFLDB was used as it provides unit process LCI data for many crops and countries, is representative of average production practices and includes data for dairy systems and processed food products.
For datasets with missing or low-quality data, additional LCI datasets were modelled using the Agricultural Life Cycle Inventory Generator (ALCIG) (Quantis 2016) consistent with the WFLDB approach for modelling the life cycle inventory of agricultural products (Nemecek et al. 2015). ALCIG calculates direct emissions at the farm, based on a number of customisable parameters such as input fertilisers and pesticides, soil type, climate conditions and farming practices (e.g. tillage). It integrates default values for most variables, based on statistical data from FAOSTAT, that can be used when specific data are not available. The ecoinvent database (v3.3) was used as a background database. Oil extraction and refining from agricultural oilseeds or crops are modelled based on data from Blonk Agri-footprint (2015) and Schau et al. (2016); separate LCI datasets were derived for crude oil extraction and refined oil production. Allocation of co-products is further discussed in Section 2.5.
Spatial archetype LCI for dairy products
The spatial archetype-based approach was introduced to account for the variability of key parameters influencing the environmental footprint of raw milk, such as herd size, breed, feed composition, intensity (i.e. degree of mechanisation) and manure management systems (MMS) in different countries. These parameters, except for the latter, influence the yield (i.e. kg raw milk per cow per year), the quality of the milk (i.e. fat and protein content) and direct emissions (through enteric fermentation and grazing) as well as the amount of manure to be managed. The dairy systems vary significantly between and within countries and therefore the approach applied by the WFLDB methodology guideline (Nemecek et al. 2015) was used to generate datasets representative of average raw milk production at a national scale. The country average dairy milk datasets are constructed in the following steps: firstly, 23 archetypes (or typologies) of milk production systems were modelled, based on the IFCN “typical farms” (FAO, IDF, IFCN 2014), and also specific studies for USA (Thoma et al. 2013) and Canada (DFC 2012). They describe how cows are fed and tended to at the farm, representing a selection of the diversity of dairy systems considered in the study. Production systems were characterised by their size (i.e. number of lactating cows) and different feeding patterns (i.e. grazing or non-grazing; proportions of hay, grains and compound feed in rations). To be consistent with prior modelling approaches, emission models for different manure management systems were created based on IPCC (2006) emission factors for methane (CH4), nitrous oxide (N2O) and ammonia (NH3). Six manure management systems are represented with up to three climate conditions (cool, temperate, warm). Each country has its own mix of manure management systems for dairy farming, as per FAO (2010a). Secondly, archetypes of typical dairy farms and MMS are combined in different proportions as to represent the typical dairy system mix in different countries. These mixes are mainly based on qualitative information retrieved from IDF and IFCN (FAO, IDF, IFCN 2014) and Eurostat 2013 data. All dairy farming modules generate milk as the main product, as well as live animals for slaughter or further fattening (i.e. male calves and culled cows) as co-products. The amount of milk produced is then corrected to a standard of 4% fat and 3.3% protein equivalent, according to the International Dairy Federation (IDF 2015) formula for fat and protein-corrected milk (FPCM). Additional detailed illustration and data are given in the ESM (ESM_1.docx. Section 2). Butter and cream processing data are based on EDA (2016), which provides typical data that can be used to represent average processing of dairy products. According to EDA (2016), the technology used in different countries is quite homogeneous, although higher variations are observable between large dairy farms and Small and medium farms. WFLDB datasets combine these data with complementary information from literature (Nemecek et al. 2015; Djekic et al. 2014, Flysjö 2012) to generate comprehensive LCI datasets. To regionalise the processing step, the national milk mix and national electricity consumption mix are used. Butter processing results in two other co-products: skimmed milk and buttermilk. Allocation among these co-products is discussed in Section 2.5.
Modelling GHG emissions from land use change
In crop production, global land transformation impacts are mainly driven by deforestation of primary forests. However, land use change (LUC) from deforestation of secondary forest or conversion from other types of land (grassland, perennial or annual crops) to arable land are also addressed. In agricultural systems, LUC can be an important contributor to GHG emissions (Poore et al. 2018). In this study, country-specific GHG emissions due to land use and LUC are assessed for each relevant vegetable oil ingredient and dairy feed input. The LUC impact assessment follows the framework defined in ecoinvent v3 (Nemecek et al. 2014), which is based on IPCC (2006) methodology. Land inventory data are obtained at the national level per crop and per type of land use based on FAO data (FAOSTAT 2012, FAO 2010b). Land use changes are calculated over the period 1990–2010. The LUC modelling approach builds on the Direct Land Use Change Assessment Tool Version 2013.1 (Blonk Consultants 2013) and is compliant with PAS 2050-1 protocol (BSI 2012). The amortization of GHG emissions is 20 years, which is aligned with PAS 2050-1 (BSI 2012) and FAO guidelines for feed supply chains (LEAP 2015). It accounts for all carbon pools i.e. above-ground biomass (AGB), below-ground biomass (BGB), dead organic matter (DOM) and soil organic carbon (SOC) (Further data is provided in the ESM (ESM_1.docx. Table S5). The values for the relevant carbon pools were taken from the IPCC Agriculture, Forestry and Other Land Use report (IPCC 2006) and FAO (2010b), Annex 3, Table 11. Country climates and soil types were taken from the European Soil Data Centre (ESDAC 2010).
In this study, three major modifications were made to the original tool (Blonk Consultants 2013): (i) addition of the SOC-related emissions from peat drainage per hectare and year for pasture areas, using IPCC (2013) for emissions calculations, based on Joosten (2009) for the surface of forest grown on peatland in each country and emissions from peat degradation reported at the national scale for all countries in 2008; this adjustment is added because pasture is not included as a crop type and the degradation of drained peatland is not considered in the original Blonk tool; (ii) inclusion of carbon capture in vegetation when relevant (e.g. when grassland is transformed into perennial cropland); (iii) addition of N2O emissions related to SOC degradation according to IPCC (2006).
For climate change impacts from LUC, two allocation schemes corresponding to different “value systems” are considered: the “crop-specific” and “shared responsibility”. The default allocation scheme used in this study is “crop-specific”, while the “shared responsibility” approach is assessed in a sensitivity analysis.
Crop specific: LUC is allocated to all crops and activities for which production area expanded over the last 20 years in a given country, according to their respective area increase.
Shared responsibility approach: LUC during the last 20 years is evenly distributed among all crops and activities in the country, based on current area occupied.
Allocation procedures
A common methodological decision in LCA occurs when the system being studied produces co-products, such as vegetable oil and meal from oil extraction, or milk and meat from dairy farming. When systems are linked in this manner, the boundaries of the system of interest must be widened to include the system using all co-products, or the environmental impacts of producing the linked product must be attributed to the different co-products in the systems.
In this study, based on the Methodological Guidelines for Agricultural Products (Nemecek et al. 2015), economic allocation was used by default for crop co-products at the farm and also processed oil seeds ingredients. For dairy milk, upstream burdens and activities were allocated to the raw fat and protein-corrected milk (FPCM), using the IDF formula (IDF 2015) and live animals based on biophysical criteria following the ISO hierarchy of allocation procedure (ISO 2006a, 2006b). For dairy butter and cream processing, the allocation of the upstream burden embodied in the raw milk as well as other inputs (energy, water, refrigerants) and outputs (wastewater, etc) is based on the dry weight (i.e. dry matter content) of butter and cream and its co-products, following the IDF (2015) and the European PEF category rules for Dairy products EDA (2016). All transport was assumed to be weight-limited due to the high density of the ingredients (oils and raw milk) and final products. For all packaging recycling processes, in alignment with ecoinvent methodology, the “cut-off by classification” approach was used to allocate recycled content and recycling at end-of-life (Ekvall and Tillman 1997). The allocation method used for background processes depends on the approach applied in the ecoinvent database. More details of allocation procedures and data are further elaborated in the ESM (ESM_1.docx. Section 4).
Sensitivity and uncertainty analysis
To ensure robustness of the LCA results, various sensitivity analyses were conducted in this project on the following key aspects: functional unit, LUC allocation approach, vegetable oils extraction allocation approach, worse case scenarios for supplying country of origins of main vegetable oils, packaging types and electricity production mix. To further improve robustness of climate change results, an uncertainty assessment has also been performed. Each product system is considered to include uncertainty with respect to (1) reference flows and (2) emission factors that are used to determine the LCI based on the reference flows. The parameter uncertainty is assessed with the Pedigree approach (Weidema et al. 2013). The total uncertainty of climate change results for butter and dairy cream is performed in SimaPro version 8.3 by running a Monte Carlo simulation of 1000 times. To assess the results’ uncertainty of 228 plant-based spreads and plant-based creams, the analytical uncertainty propagation approach based on Taylor series expansion was used by adapting the uncertainty assessment method introduced by the GHG Protocol (2011). Results of sensitivity and uncertainty analysis are presented and discussed below in Section 3.