Mapping the impacts of farmed Scottish salmon from a life cycle perspective

The European Union is becoming ever more reliant on seafood imports to supply growing demand that European fisheries and aquaculture have failed to meet (Little et al. 2012). Politically motivated media reports have made claims that competing seafood imports may be associated with environmental and social concerns, and the EU should be seeking to import less, in favour of promoting growth in European aquaculture. This aligns with current consumer interests in consuming “local” food as a lifestyle choice (Janßen and Langen, 2016). However, as trade becomes more globalised, and competition for resources increases, raw materials for livestock production are increasingly sourced on a global basis, which inevitably result in environmental trade-offs (Ziegler et al. 2016).

Apart from smolts (juveniles), Scottish Atlantic salmon (Salmo salar) is produced exclusively in net pen systems in sheltered marine lochs or bays. Production started in the 1970s, increased quickly during the 1980s and 1990s but subsequently plateaued due to production challenges and increased competition for suitable sites with other stakeholders. Recently, production has begun to increase as the Scottish government has set a target to produce over 200,000 t of salmon by 2020 (http://​www.​gov.​scot/​Topics/​marine/​Fish-Shellfish accessed 5/7/2017). Scottish salmon is now exported to over 60 countries worldwide (SSPO 2015) and is the second most valuable UK food export after whisky. In some circumstances, salmon aquaculture has raised environmental concerns from a number of standpoints, both locally and globally.

LCA is an ISO-accredited (ISO 2006a, b) environmental impact accounting tool which measures the cumulative impact throughout a product’s value chain, covering a range of different environmental impacts. Several LCAs of salmon have been conducted in the past, notably Pelletier et al. (2009), Ayer and Tyedmers (2009), Torrisen et al. (2011) and McGrath et al. (2015). Some other LCAs have focussed on salmonid feeds such as Papatryphon et al. (2004) and Boissy et al. (2011). Of all the studies related to salmon, only Boissy et al. (2011) included water consumption and land use, and none of them attempted to map the various impacts throughout the value chain (VC), i.e. the connected and interrelated nodes of an industry that create and add value (Dey et al. 2015). Troell et al. (2014a) identified water footprints from a life cycle perspective as an important issue in the assessment of mariculture systems. In previous LCAs of salmon VCs, the vast majority of environmental impact has been associated with feed ingredient sourcing (e.g. Pelletier and Tyedmers 2007; Pelletier et al. 2009). Indeed, it has been estimated that around a third of agricultural land use may be devoted to providing livestock feed ingredients (Robinson et al. 2011) of which aquaculture represents an estimated 4% (Troell et al. 2014b).

Aquaculture feed formulators source ingredients from a global pool of resources such as marine ingredients from South America, which may represent as much as 40% of the diet by volume. Efforts to replace marine ingredients, which are regarded as being highly limited (Naylor et al. 2009; Ziegler et al. 2016), have resulted in greater substitutions of vegetable-based proteins, especially those derived from soybean which is critical for many European livestock feeds (FAO 2006), including for aquaculture (FAO 2006; Torrisen et al. 2011). The vast majority of European-sourced soybean meals and concentrates are also from South America and have been connected to environmental concerns around deforestation of sensitive habitats for agricultural expansion (FAO 2006; Robinson et al. 2011; Tritsch and Arvor 2016; da Costa et al. 2017). Some vegetable-based feed ingredients have also been shown to be highly energy intensive in their processing compared to the marine ingredients which they substitute for (Boissy et al. 2011; Pelletier et al. 2009). There are clearly trade-offs between different impacts with regard to water, land, energy use and their associated emissions and impacts, but also to how impacts are geographically distributed. This is particularly the case for comparing between marine ingredients with low demands, especially on land and agricultural ingredients with potentially high land and water demands. Many aquaculture LCAs have presented a comprehensive set of environmental impacts but have often lacked contextualisation of these impacts from a local or global perspective and the implication for improvements or decision making, and it can be argued that a lack of regionalised/localised impact assessment is common throughout LCA. Several studies have investigated regionalised impacts through LCA (e.g. Schlich and Fleissner 2005; Boulay et al. 2011; Patterson et al. 2011), and some efforts have been made to combine LCA with GIS methodology (e.g. Geyer et al. 2010; Gasol et al. 2011; Dresen and Jandewerth 2012; Mutel et al. 2012); however, these have not been frequently applied and only a few LCA studies have represented data geographically. Other challenges have arisen in how to measure the impacts of land use change related to the dependence of land required for agricultural expansion (Mila i Canals et al. 2007). Different approaches are being developed to measure the impacts of land use change (LUC) such as UNEP-SETAC initiatives, although consensus on regional factors (Nordborg et al. 2017) and effects on carbon sequestration (Nguyen et al. 2010) amongst others remain areas for debate. A summary of the challenges in characterising the impacts from land use and land use change are given by Nordborg et al. (2017). However, the authors of this paper regard LUC as more of an end-point issue and for that reason, as well as the complexities involved which we regard as out of scope, this article will focus only on land use rather than the effects of change.

Scottish salmon production is based exclusively in net pens in marine sites off the West Coast mainland, and the islands of the Outer and Inner Hebrides, Orkney and Shetland. At the time of data collection (2012 to 2014), smolts (juveniles) were produced to 70 g average, in freshwater sites before transportation to sea by road and boat. The fish were kept in 90-m circumference net pens for a period of around 22 months, at which point they reached an average size of around 4.5 kg before slaughter. Fish were transported by well boat from the production site to a slaughter station which may be combined with primary processing (gutting and bleeding). At this point, the head-on gutted (HOG) salmon were sent for further processing, either domestically or abroad.

All data collection was anonymous due to confidentiality agreements between ourselves and the companies involved. Primary data were collected using structured surveys in face-to-face interviews at the company locations. Secondary data were used for the production of feed ingredients and horizontally averaged according to Henriksson et al. (2014a). Data on emissions from managed soils were calculated according to IPCC (2006). Data on electricity mixes for each country or region in Table 1 were collected from the International Energy Authority website (IEA accessed 24/2/2014), and associated emissions from energy provision and transport were taken from the Ecoinvent database 2.2.

Primary data were collected from a major Scottish-based aquafeed mill which also has representation in other major salmon-producing countries. This included individual ingredients and quantities for standard grow out feed for 1 year of supply (2012), all electricity, gas, water consumption and land area. The data provided the quantity of each ingredient in the formulation and the location from which it was sourced, although specificity of location was variable. The source of some ingredients could be located to the county level, but for others only to the broad geographical region (e.g. “continental Europe”). The individual feed ingredients and their origins have been included in Table 1, but quantities have been shown aggregated, due to the wishes of the company, into broader categories such as marine proteins, vegetable oils etc.

Feed ingredient production and processing, data were taken from literature sources which provided information on fertiliser application, fuel use in machinery, other energy use, yield per hectare and chemicals used for separation and refining (Table 1). Road and ocean transport distances were calculated from Google Maps. Of the crop production data, only a few sources included freshwater requirements for those crops. Data on fishmeal and fish oil from Peru were from Henriksson et al. (2014b) adapted from Avadí et al. (2014) whereas the Danish fishmeal and fish oil were adapted from the same process with the energy mix and transport adjusted accordingly.

Farm data was collected from six individual sites, although some shared land bases and some other resources, such as boat use. Data were collected on quantity of feed used, electricity use, diesel for boats and service barges, fish mortality, medications, road distances from the shore base to the feed supplier, land occupation of the shore base and total salmon production. The weighted mean economic feed conversion ratio (eFCR) between the six farms was calculated at 1.19 ± 0.10 kg of feed per kg of fish growth including fish losses. Medications for salmon production have not been widely characterised, and in this case, a pesticide process was used from the Ecoinvent database (Frischknecht et al. 2005). Data on the production of smolts were not collected in this instance; it was assumed that their contribution to overall impacts would be minimal, being only 1.5% of the size of the final product at stocking. Finally, data was collected on the primary processing of salmon, which included gutting and packing in ice within polystyrene boxes, prior to transport for further processing.

A contribution analysis of processes and raw material inputs is shown in Fig. 1. Most impacts at the feed mill are from the embodied impacts of feed ingredient production. Marine ingredients contributed a relatively small amount to overall impacts from feed production, compared to their inclusion rate within the diet, for all categories except for eutrophication. The high eutrophication impact is explained by the nutrient-richness of effluent from fishmeal and fish oil production, which contains significant amounts of protein and lipid rich bloodwater (UNEP 2000). However, in terms of energy, requirements for fishmeal and fish oil production are low compared to that of some vegetable-based ingredients, particularly maize gluten and rapeseed oil, both of which feature within the diet formulation. It can be seen that most of the impacts at the feed mill stage are over represented by vegetable ingredient inclusion, especially oils which only form less than 5% of the diet, but contribute to 18.0% of GWP, 16.9% of AP, 12.7% of PCO and 11.3% of EP. In the case of both CWU and LU, vegetable ingredients collectively contribute to over 99.7% of each category. Therefore, with increased substitution of marine ingredients with refined vegetable ingredients, the overall impacts of feed production would be expected to increase.

Feed is by far the most important contributor to impacts throughout the value chain (Fig. 1), as the embodied impacts of feed production account for more than 80% of all impact categories, and this finding is supported in other salmon LCAs (e.g. Ayer and Tyedmers 2009; Pelletier et al. 2009; McGrath et al. 2015). An exception is EP, where the majority of impacts come from the metabolites of salmon growth, on the production unit. Apart from feed, notable impacts on ozone depletion potential occur from chemical usage at both the farm and at the processor. On the farm, this is connected to the production of medicinal treatments, whereas at the processor, it is the use of refrigerants which is of most concern. Minor but significant contributions to ozone depletion potential also occur through diesel used on the service barges and boats at the farm site. Notable contributions also occur from packaging materials at the processor, to PCO and GWP. However, these contributions are dwarfed by the direct and indirect contributions from feed production through to the salmon raw material at the processor. Contributions are often affected by the allocation procedure that is adopted (e.g. Pelletier and Tyedmers 2007), and so, a sensitivity analysis was conducted using mass allocation to test whether this had a significant influence on either the overall impact or the contributions to those impacts, particularly from different feed ingredients. The detailed results of this analysis can be found in the supporting information (Figs. S1 and S2, Tables S1 and S2, Electronic Supplementary Material), but generally, there were minor differences, both in the overall impact and in the contributions between the two allocation methods. Generally, mass allocation showed slightly lower overall impacts in all categories except land and consumptive water use which were higher using mass allocation. Consequently, the contributions to these two categories change most, from both the input processes and geographically, between the two allocation methods.

If we take these contributions from the process level and plot them geographically (Fig. 2), we can see how the emissions relate to the manufacture of the raw materials used within the salmon production unit and processor. If we consider that feed contributes to over 80% of emissions in each category, apart from EP, to the farm gate and over 80% of the emissions at the processor are accounted for by the embodied impacts in the salmon raw material in each category, the influence of feed ingredient supply can clearly be seen. Indeed feed indirectly accounts for over 80% of emissions in five of the seven categories studied up to the processor gate. Only for EP (22.4%) and ODP (66.1%) were the contributions less. A total of 77.4% of eutrophying emissions originate from the farm site from the excretions of the salmon with most other contributions coming from fishmeal and fish oil processing in Peru and Denmark. Small contributions to eutrophication were from agriculture to provide feed ingredients. Figure 2 demonstrates that far from being a local product, only 22.8% of feed ingredients were sourced from the UK with 44.1% of total normalised emissions, including 40.9% of total GWP and 59.9% of total ODP emissions. The largest contributor to both of these emissions in the UK relate to the production of rapeseed oil. Similarly, only 34.9% of CWU and 50.7% of land use was within the UK.

In Fig. 2, the geographic spread of emissions highlights disproportionate impacts from PCO in Brazil and from AP in continental Europe. The PCO contributions can be matched to vegetable protein production in Brazil which is attributable to land management and refinement of soy protein. High AP contributions in Europe are mostly linked to SO₂ emissions from high energy inputs required for the processing of wheat and corn gluten, even though it makes up a relatively small proportion of the feed formulation. Whereas GWP and ODP may be considered of global importance, AD and PCO are of more localised and regional significance. Europe especially has a history of problems from acid rain dating back to early industrialisation, since when emissions from Western and central European countries have been linked to acidification effects in Central and Eastern Europe, Scandinavia and further afield (Menz and Seip 2004). Although significant steps have been made to reduce SO₂ emissions in Europe, disproportionate contributions such as from gluten, which contribute 36.8% to AP at the feed mill stage, may be considered to be of concern. The effects of PCO from soybean cultivation and refinement are less clear, as the emissions are likely to be in more rural areas and dispersed more easily than in cities, and where it is less of a human health concern.

Of more concern in Brazil are perhaps the contributions to CWU and LU from soybean cultivation at 26.8 and 16.0%, respectively. Concerns over clearance of sensitive rainforest habitats for agriculture in South America have been longstanding. Despite Schmidt (2010) arguing that the main driver of deforestation in Brazil is logging, at over four times that of agricultural expansion, according to Roberts et al. (2015), Brazil had the largest increase in land area devoted to agriculture of any country, between 2008 and 2013, at 73,000 km², an area the size of the Republic of Ireland. The Ukraine, although smaller in size, had an expansion in agricultural land use of 16.6% in the same time period so that in 2013, 34.5% of the country was devoted to agriculture. Such rates of expansion should raise concerns, as the changes will undoubtedly have large impacts on biodiversity as well as pressures on water resources for irrigation in some circumstances. High water demands were identified in the UK (34.9%), for various feed ingredients, Brazil (26.8%), from soybean production, Ukraine (26.2%), from sunflower production, and continental Europe (12.1%) also for various feed ingredients. The significance of this is dependent on the local situation as certain parts of these countries are more water stressed than others. Pfister et al. (2009, 2011) identified global areas of water scarcity and the proportion of abstracted water used for agriculture. Parts of the UK, Ukraine and continental Europe, where some feed ingredients were sourced, had water extraction rates up to five times higher than recharge, and therefore, formulators should be wary of using some of the more demanding of these crops as shown in Fig. 3. The water scarcity data referred to in this article, from Pfister et al. (2009, 2011), is being further developed into the AWARE methodology (http://​www.​wulca-waterlca.​org/​aware.​html accessed 19/4/2017) and adopted into EcoInvent v3 (Pfister et al. 2016) along with measures to harmonise methodology with water foot-printing methodology (Pfister 2017). In addition, Boulay et al. (2011) have characterised water according to different quality characteristics which may be altered by human activities, and Boulay et al. (2015) further developed water use into end-point impact categories including human health. These methodologies may be further adopted to identify areas which may be highly impacted due to existing water scarcity issues, although the resolution may need to be increased to gauge the impact on a local scale. The sensitivity analysis using mass allocation, shown in detail in the supplementary materials, showed a higher consumptive water and land use demand, mostly as a result of the influence of sunflower seed cake in the diet. Consequently, the balance between vegetable oils and vegetable proteins in the contribution analysis shifted towards the protein and also geographically. The LU and CWU contributions from the Ukraine were raised significantly as a result (Table S2, Electronic Supplementary Material).

While this LCA study serves to highlight geographically disproportionate emissions and demands on resources, it is limited in detail to gauge the consequence of these impacts. The significance of emissions may be on a highly localised level, e.g. sensitive water body; regional level, e.g. land mass; or global level. For localised emissions, more contextual information may be necessary on the type of emission and the local conditions to assess whether the system can absorb those impacts without having an adverse effect on public health or wildlife. An example is if sea currents are adequate to disperse both dissolved and solid waste emitted from net pen sites or at a city level, how much PCO emissions are likely to cause public health issues. In order to do this, it would be necessary to normalise data to whatever level was necessary for the context of the emission and model its dispersion within that environmental system. These were considered end-point assessments, and while beyond the scope of this article, geographic information systems (GIS) could aid in identifying such issues. There have been a few efforts to integrate LCA and GIS in such a way that adds context to the ability of environmental systems to absorb localised/regionalised impacts or emissions, notably by Gasol et al. (2011), Dresen and Jandewerth (2012) and Mutel et al. (2012), but while offering interesting findings and setting out methodology and principles; so far, they have not been widely adopted.

Highly notable in this article is the difference in performance between marine and vegetable ingredients. Marine ingredients out-performed vegetable ingredients in most categories as shown in Fig. 4 and clearly seen in Figs. 1 and 2. However, the impacts on biodiversity have not been presented or discussed in this article. The trade-offs between impacts on marine vs terrestrial biodiversity are complex and challenging to measure in a single indicator. However, the impacts of reduction fisheries on marine ecosystems have been widely documented (e.g. Naylor et al. 2009; Alder et al. 2008), and although the majority of fishmeal comes from well-managed stocks, concerns are still held over its production. Fishmeal is widely considered to be at the limit of responsible production and the marine ingredients industry has been undergoing a period of change as increasing amounts of raw materials are sourced from the by-products of fisheries and aquaculture production so that now according to IFFO (The Marine Ingredients Organisation) as much as 35% of global supply originates from by-product. This trend looks likely to continue as the demand for fishmeal increases and the incentive to add value to by-products becomes greater. A report by IFFO (Jackson and Newton 2016) recently estimated that nearly 12 million tonnes of fishery and aquaculture processing by-products were being wasted that could provide in excess of a further 2 million tonnes of fishmeal of various grades. However, responsible sources of marine ingredients will remain limited and other ingredients will continue to be required. Far from advocating that alternatives to fishmeal should not be sought, it is suggested that formulators should endeavour to find the best environmentally and socially responsible ingredients, as well as cost-effective, from the increasing range available. Key ingredients, such as fishmeals and fish oils, should be targeted to use in feeds for certain species and life stages where their use is most efficient, for example the use of finishing diets to improve omega-3 fatty acid levels prior to harvest (Codabaccus et al. 2013). Although fishmeal is a globally traded commodity, inclusion level is often determined by cost and availability, and according to Henriksson et al. (2014b), some tilapia diets in Thailand contained up to 15% fishmeal from various sources. From a global life cycle perspective, this could perhaps be better utilised. New technology in marine ingredient processing, such as advanced enzymatic hydrolysis techniques, is enabling more efficiency and synergistic effects with other ingredients, through providing more advanced nutritional profiles than raw ingredients are able (Martínez-Alvarez et al. 2016). This has undoubted opportunities for the salmon aquaculture industry to improve its by-product utilisation strategy. Although there is virtually no wastage from UK processing activities, much is sent to fishmeal reduction as mixed by-product where better separation and utilisation could potentially improve the value and add to the range of ingredients available for livestock nutrition. Viscera from primary processing are already hydrolysed into quality protein ingredients for tropical shrimp feeds, which has added value and ended a waste disposal issue for the salmon industry (Newton et al. 2014). Considering the dynamic situation regarding fishery and aquaculture by-products, this is potentially a fertile area for further LCA investigations.

In terms of overall improvements, rather than substituting small amounts of feed ingredients within diets, more substantial impact reduction can be made improving feed efficiency overall. Although there may be seasonal variability due to different challenges, improved feed efficiency has been the trend in most fin-fish aquaculture as demonstrated by Tacon and Metian (2015) and more improvements are to be expected, aided by such innovations in advanced ingredient development mentioned above. Further improvements have and could be made by improved automated feed delivery technology, genetic improvements, maturation control, better stock management and reduced mortality. Mortality is included when calculating eFCR, and improved survival can therefore substantially improve environmental performance, depending on when the majority of mortality occurs. Weighted mean mortality for the six farm sites in this study was 4.0% by weight, which is below the industry average estimated at between 5 and 10% (SEPA 2004, Statistics Norway 2009) although the lowest mortality rate was just 1.0%. Mortality continues to be a challenge for the industry particularly from sea lice which not only cause losses at the farm, but also at the processor, as consignments are rejected because of poor quality if lice counts exceed a specified number. As this problem is tackled, it is likely that the dynamics and trade-offs between different impacts may change. The lowest eFCR out of the six sites was 1.04 which is a 14.4% improvement on the average, although this site did not have the lowest mortality, suggesting that further improvements could be made. This would translate into substantial environmental impact abatement, if all farms were to achieve this benchmark of feed efficiency.

Also of note, is the minor contribution of transportation to any of the impact categories apart from ODP perhaps, to which ocean transport contributes 8.1%. Such issues raise further questions as to whether so-called local production is “better” than food imports. When less than 25% of the raw materials are considered local, can the final product be considered as “locally produced”? Therefore, “food miles” should not be of concern to stakeholders from an environmental perspective, more the efficiency and responsibility of production within those countries prior to transport. That is, from an environmental impact perspective, it is preferable to source materials produced efficiently and cleanly from further afield than local materials produced inefficiently and with polluting practices. For example, Boissy et al. (2011) demonstrated that palm oil from Malaysia was preferable to German rapeseed oil in many impact categories. Although there may be genuine concern over the environmental impacts of global value chains in some circumstances, these may be unfounded and must be looked at in more detail on an individual basis.

A conscientious consumer may wish to buy locally produced food on the basis that they know it is responsibly produced from a food safety, environmental and social impact perspective. EU food law (EC 2002) and traceability generally ensure the safety of food produced within and imported to the EU, but other responsibility issues are generally only covered by third party certification such as the Aquaculture Stewardship Council (ASC 2017). However, such certification schemes tend to focus on the production unit, with limited attention to whole value chain (Bush et al. 2013). This has led to issues elsewhere in the value chain being under-investigated. A recent example was the supply of fishmeal in Thai shrimp value chains, which was criticised for poor labour practices (https://​www.​theguardian.​com/​global-development/​2014/​jun/​10/​supermarket-prawns-thailand-produced-slave-labour, accessed 10/7/17). Whereas producers engaged with certification schemes that monitor shrimp farms, at the time, the monitoring did not extend to social practices throughout the VC and failed to pick up the problem. However, aquaculture has also been credited with improving the livelihoods of many vulnerable and/or remote communities, including rural populations on the West Coast of Scotland which has beneficial knock on effects throughout the value chain and associated communities (SSPO 2015). Such benefits to the local economy may out-weigh other considerations, especially when the global impacts of imported alternatives may be regarded as similar or more difficult to gauge. The choices made regarding food production, supply of materials, the knock-on environmental and social impacts and trade-offs are highly nuanced. However, retailers who advertise their produce as “local” have a duty to ensure and communicate the responsible credentials of their product from a value chain perspective, if consumers are purchasing the product on the basis that local production is ethically superior.

From a broader policy perspective, this study identifies the inadequacies of assessing the sustainability of food systems from highly localised, national standpoints alone. The study also highlights the need for balance between use of marine and terrestrial ingredients, and their resulting environmental impacts, in driving feed formulation for commodity food products such as farmed Atlantic salmon. The implications for global trade in food ingredients on policies supporting environmental restoration and conservation in both producer and consumer countries are also important, as has been noted for the telecoupled Brazil-China soybean trade (Torres et al. 2017)

This study shows that life cycle impact assessments can be strengthened by including more geographically specific data which enable contextualisation. To produce a fully integrated LCA/GIS study, a large dataset is required that details the characteristics of the environment where the emissions occur at each point in the value chain, so that the dispersion and assimilative capacity may be calculated. Such a dataset may not be practical to collect in the short term. However, efforts are being made by Global Feed LCA Institute (GFLI), constituted by a consortium of European feed associations, to build a feed ingredient LCA database using a standardised methodology developed by the FAO LEAP project (FAO 2016). Such a dataset may provide the necessary detail to facilitate GIS integration consistently. However, in this article, we have presented a practicable geographically contextualised result set which can be used to quickly highlight impacts of geographical concern which may then be investigated in more detail.

The study shows that Scottish salmon is not truly a “local” product but supports a vast global value chain. Consumers may be unaware that their support for local production, for environmental or other responsible production motivations such as socio-economic, food safety or animal welfare, is in fact supporting production on a much broader basis, and contributing to the associated impact that this may have or indeed benefits to those regions.