Functional unit
The functional unit is the reference unit used to quantify the performance of a production system (ISO 14044 2006). The most commonly used functional unit in the twelve studies reviewed here is 1 ton of live fish at the farm gate (six studies; see Table 1). Two other studies also limit themselves to the farm gate, with Grönroos et al. (2006) adopting dead weight and Phong (2010) adopting 1 kcal alongside 1 kg as his two functional units. Four studies defined their functional unit in terms of edible yield, defined as the main part of the organism that was marketed (fillets, flesh, or tails).
The functional unit is the basis of comparison in comparative LCAs. The functional unit follows the goal of the study, since different goals may require different functional units. The goal of the study and the associated functional unit partially defines the system boundary of the inventory. For example, if frozen fillets in supermarkets are chosen as a functional unit, the system boundary needs to be defined so as to include processing, transportation, and distribution. The functional unit may, moreover, significantly influence comparative LCAs involving different species, since the edible portions and nutritional values of products can differ by an order of magnitude (Roy et al. 2009). Mussels and shrimp, for example, provide respectively, 13.6 and 140 kg of protein per ton of whole animals harvested (Mungkung 2005; Iribarren et al. 2010; www.nutraqua.com accessed 23-June-2010).
The choice of the functional unit is important for comparisons between species as well as across cultures, as the definition of edible will depend on cultural influenced consumer preferences. The choice of functional unit will also influence allocation decisions at the farm gate where more descriptive functional units, such as kilocalorie, may be more appropriate for comparisons between multi-output systems (Phong 2010). We therefore recommended to carefully choose a functional unit tailored to the goal and scope of the study.
System boundaries
The system boundary determines which unit processes will be included within an LCA study and which ones are to be excluded. With respect to the diverse goals of the reviewed studies, few have considered supply chain impacts beyond the farm gate (Pelletier and Tyedmers 2010). Iribarren et al. (2010), however, did include the whole production chain and found that a significant part of the emissions from mussel production is related to processing and marketing, with dispatch centers contributing significantly to the overall emissions from live mussel production. Infrastructure is also often excluded, due to the large amount of time that has to be invested in calculating the total input in relation to the small impact that is considered (Ayer and Tyedmers 2009). Where included and distinguished (Aubin et al. 2006; Ayer and Tyedmers 2009; Aubin et al. 2009; d’Orbcaster et al. 2009), however, infrastructure was found to contribute between 0% and 19.0% to the overall impacts in terms of global warming, eutrophication, and acidification indicators. Common cut-offs were based on the outcomes of previous studies, selection of impact categories, available data, and resource constraints (Mungkung 2005; Ellingsen and Aanondsen 2006; Grönroos et al. 2006; Pelletier et al. 2009; Pelletier and Tyedmers 2010).
The selection of the system boundary should be consistent with the goal of the study, and the criteria used to establish the system boundary should be identified and explained (ISO 14044 2006). In aquaculture systems, the length of the full production chain is largely dependent on the type of system (Fig. 1). For example, external inputs of feed and hatchery-reared juveniles may not be needed in extensive systems and if the product is sold fresh on the market it needs no processing (e.g., carp in China). Fish and seafood are also the most perishable of food products, and the level of processing will influence the longevity of the product, as well as the amounts wasted, hence the environmental impacts (Sonesson et al. 2005).
The problem in defining cut-offs for the quantification of inventories is a lack of readily accessible data, implying disproportionate expenditure of funds and efforts on data collection. Limitations of time, funds, or data access will inevitably lead to the exclusion of processes and to less complete and accurate results. Nowadays, it is, however, possible to handle the cut-off problem better, by estimating the environmental interventions associated with flows for which no readily accessible data is available using environmentally extended input–output analysis (EIOA) (Suh et al. 2004). For the purpose of consumer guidance, we recommend a more extensive system boundary at or beyond the market, as impacts may otherwise be underestimated (Iribarren et al. 2010). Further efforts should also be directed towards expanding current knowledge about the contributions from infrastructure, as this has been reported to have a larger influence on agriculture than on most other industrial processes (Frischknecht et al. 2007).
Data and data quality
Although all of the studies reviewed here model relevant agriculture, fisheries and other related processes—to different extents—most of the inventory details of these modeling efforts remain unpublished. Articles that do extensively report the data mainly specify economic flows, with environmental flows often limited to nutrient balances. This may be due to the aquaculture-based background of most of the researchers for whom eutrophication has historically been a major concern. In various articles, it remains unclear whether background databases were used or whether real foreground data (site-samples) had been collected (ILCD 2010); neither is it always clear which processes that were included in the study. Consequently, reproducing their results is difficult or impossible. For example, Ellingsen and Aanondsen (2006) reported: “Data are generally collected from various sources by both literature surveys, a study of available data sources, telephone conversations, and meetings”. This, unfortunately, provides no clue as to which processes, data, or data sources were included in the study.
Background data used in the studies were derived from a wide range of databases, including some which were quite out-dated (ETH 1996 and BUWAL 250) (see Table 1). Some authors used combinations of different databases or did not clearly specify the precise database(s) consulted. Several studies, for example, reported that they had used SimaPro software, with all of the databases included in it. As SimaPro includes many different databases (e.g., Ecoinvent, US LCI database, US IO dbase, Danish IO dbase, Dutch IO dbase, LCA food dbase, Industry data, Japanese IO dbase, IVAM dbase; see http://www.pre.nl/simapro/inventory_databases.htm), the actual data sources used in these studies remain unclear. All studies, moreover, rely on European databases (commonly different versions of Ecoinvent), even though various studies dealt with aquaculture in non-European countries. Although the authors of several studies did invest much effort in adapting inventories to regional conditions, there remains a real need for databases representing technologies of developing countries.
The focus of the studies ranged from single farms (Aubin et al. 2009; Ayer and Tyedmers 2009) to small samples of each farming system (Phong 2010), to aggregated industry averages representing significant parts of national outputs (Pelletier et al. 2009; Pelletier and Tyedmers 2010). However, the quality of foreground data available for aquaculture systems often depends on the intensity of the system and the region of data collection. Highly intensive systems, such as land-based salmon systems, often keep more complete records of all inputs and outputs, while only general estimations are available for most extensive pond systems in rural areas. Accessibility to feed inventories may, moreover, be subject to the scale and nature of the feed mill, as exact mixtures of ingredients often are held confidential. Site-specific measurements are, moreover, dependent on the resources available. The articles offer limited reporting on other environmental flows beyond nutrient budgets (including methane, nitrous oxide emissions, copper-based anti-fouling agents, antibiotics, etc.).
The International Organization for Standardization (ISO 14044 2006) states that data quality requirements should be specified to enable the goal and scope of the LCA to be met, and also that the treatment of missing data should be documented, and that data sources as well as an assessment of the reproducibility of the study results by independent practitioners should be addressed as part of the data quality requirements. ISO does not, however, demand publication of all data used.
Transparency in the reporting of data and reproducibility of results are important for proper peer-reviewing and interpretation of background data, at least to the extent that this is possible with regards to sensitive industry inventory data. A good example of the way data can be published without compromising the focus of the article was given by Grönroos et al. (2006) and Pelletier et al. (2009), who both published supporting documents describing inventories (although with different coverage of environmental data; see above), core processes, assumptions, and calculations. Another solution to fitting large inventories to the often restricted format of scientific journals is to report which processes derived from a background database (e.g., ecoinvent) were included in the study without actually including the data of that process. Such processes could simply be reported using the process ID numbers, rather than the full process names. This kind of more open reporting of data is critical for developing specific LCA data sets for aquaculture-related processes, as much primary data currently are lost by aggregating results and by only presenting impacts, rather than inventories. It should, however, be pointed out that the data sourcing and reporting issues discussed here are not unique to aquaculture LCAs, but rather apply to the majority of LCA studies published, whether peer-reviewed or not.
Allocation
Some of the main differences amongst the studies reviewed here are related to allocation. While all of the most commonly applied procedures for allocation (including mass, economic value, gross energetic content, and system expansion) have been applied to aquaculture LCAs, economic value and gross nutritional energy content have more frequently been used in the more recent publications (see Table 1). This is also the main methodological difference between the two main publishing institutions, with INRA-IFREMER applying economic allocation, while researchers at Dalhousie University commonly prefer gross energy content as the basis for allocation (see Table 1).
Four publications applied system expansion to certain allocation situations (Ayer and Tyedmers 2009; Pelletier et al. 2009; Pelletier and Tyedmers 2010; Iribarren et al. 2010). Iribarren et al. (2010), for example, used system expansion for Spanish mussel production (with mussel as the main product and shells as a co-product) with the assumption that mussel shells could be used to replace conventional calcium carbonate production. Grönroos et al. (2006) restricted their analysis to whole fish at the farm gate to avoid allocation in the processing phase, while mass allocation was used for feed inputs. Some authors did not report their allocation decisions in their articles.
Most industrial processes yield more than one product, and some recycle expanded products as raw materials. As a result, the materials and energy flows, as well as the associated environmental releases, have to be allocated to the different products according to clearly stated and justified procedures. In aquaculture, many of the feed inputs are co-produced in other production systems (e.g., rice bran, fisheries by-catch, and co-products from livestock processing), and co-products also occur in the processing phase.
It is our belief that the multi-functionality problem is an artefact of the desire to isolate one function out of many and as artefacts can only be resolved in an artificial way, there will always be more than one way of solving the multi-functionality problem. This is illustrated by the debate on methods to deal with the multi-functionality problem over the last two to three decades which still has not provided a generally accepted method. Depending on the application (e.g., policy or scientific publications), using alternative allocation methods could be seen as an opportunity to produce more realistic ranges of results and provide stronger conclusions. There are, however, certain requirements that need to be addressed when dealing with allocation issues, such as that the solution should be consistent, well justified and in-line with main methodological principles (Guinée et al. 2004; ILCD 2010). It is also important to always report on the allocation method(s) applied and perform a sensitivity analysis, as allocation plays a pivotal role in the performance of a production system (ISO 14044 2006).
Life cycle impact assessment methods
All reviewed studies applied one or more life-cycle impact assessment methods. The major impact assessment methodology used for characterization was the midpoint CML baseline method (Guinée et al. 2002) with only Ellingsen and Aanondsen (2006) applying an endpoint approach (eco-indicator 99 method; Goedkoop and Spriensma 1999). Grönroos et al. (2006) choose to use region-specific characterization factors for eutrophication and acidification, while making a distinction between aquatic and terrestrial emissions. Only climate change, acidification, and eutrophication were adopted as impact categories by all studies. In addition, a few novel methods were introduced for biotic resource use, water dependency, and land (surface) use (Table 2).
Table 2 Frequency of applying different impact categories in LCA studies on aquaculture and the impact assessment method used.
As regards climate change, the characterization factors suggested by the international panel on climate change (IPCC; Houghton et al. 2001) were the basis for all reviewed studies. This therefore enables for valid conclusions to be drawn amongst the studies, e.g., the great importance of feed inputs for aquaculture systems.
As regards acidification, all but three studies adopted the approach developed by Huijbregts (1999a). Apart from Ellingsen and Aanondsen (2006) and Grönroos et al. (2006), Phong (2010) chose alternative characterization factors, in this case, the older Heijungs et al. (1992) acidification method.
As regards eutrophication, similar differences are found as for acidification, while Grönroos et al. (2006) chose to separate terrestrial and aquatic emissions due to their distinct association to feed production and feed application, respectively. Phong (2010), again, refers to an older alternative publication, Weidema et al. (1996).
Cumulative primary fossil energy demand was the fourth most commonly included impact category amongst the studies and showed a large overlap with abiotic resource depletion (Ayer and Tyedmers 2009). Strikingly, six studies adopted and quantified a novel impact category, biotic resource use. Its use aims to capture the ultimate carbon-based energy stemming from biological systems that support fed aquaculture production, although a standardized protocol for this impact category still remains to be developed (Pelletier et al. 2007). Marine exotoxicity, an impact category for which the existing characterization methods have been widely debated within the LCA community (Pettersen and Hertwich 2008; Gloria et al. 2006), was adopted and quantified in four studies. A range of other toxicity related impact categories were less frequently adopted, along with abiotic resource depletion and ozone depletion. Water dependency and land use were represented in only two studies each, using own methodology. Little consideration was, however, given to the type of water use (e.g., marine or freshwater, degradative or consumptive; Bayart et al. 2010) on either the input or the output side, nor were emissions relating to land use and transformation considered (ILCD 2010). Other concerns not covered by the LCA methodologies reported in our review include impacts on the seafloor from capture fisheries, the introduction of invasive species, the spread of diseases, genetic pollution, and socio-economic concerns (Pelletier et al. 2007).
In summary, the current review of aquaculture LCAs shows that impact assessment methodologies have been applied to all studies reviewed. The range of impact categories covered is, however, limited, and the methods adopted for the various categories differ, hampering comparisons of study results. Some authors used old characterization factors, while others developed their own quantification methods. Future harmonization with the developments within the LCA community is therefore advised, focusing on the standardization efforts promoted by ILCD (the European Commission’s Join Research Centre) and UNEP-SETAC (the United Nations Environment Programme and the Society of Environmental Toxicology and Chemistry) including the ILCD handbook (lct.jrc.ec.europa.eu), USEtox™ (www.usetox.org), land-use (lcinitiative.unep.fr, accessed: 17-Oct-2010) and freshwater use (Bayart et al. 2010) (for a complete overview of the life cycle impact assessment methods adopted by the different studies reviewed here, please see the Online Resource (ESM) to this paper).
Interpretation methods
Although all studies performed a dominance or contribution analysis, many did not perform a complete set of sensitivity analyses, as is required by the current ISO standards. Ayer and Tyedmers (2009), however, conducted an extensive set of sensitivity analyses, one of which highlighted the importance of electricity sourcing. Another study by Pelletier and Tyedmers (2007) concluded that allocation factors strongly influence the impact of different feed inputs. Both d’Orbcaster et al. (2009) and Pelletier et al. (2009) drew a parallel between food conversion ratios (FCRs, defined as kilogram dry feed/kilogram live fish) and GHG (greenhouse gas) emissions, while Mungkung (2005) supported her conclusions by performing a sensitivity analysis on data assumptions for fishing practices as well as for different impact assessment methods. Ellingsen and Aanondsen (2006) also used two alternative impact assessment methods to strengthen their conclusions. Pelletier et al. (2009) evaluated the range of nitrous oxide emissions from nitrogen fertilizers, compared to the default value indicated by the IPCC. Only Phong (2010) applied statistical tools to different farming practices, in the form of one-way ANOVA (analysis of variance).
According to ISO (2006), the life cycle interpretation phase of an LCA comprises the identification of the significant issues based on the results of the LCI (life cycle inventory) and LCIA stages, an evaluation involving completeness, sensitivity, and consistency checks, and finally the formulation of conclusions, limitations, and recommendations. It is an important phase of any LCA study, where any weaknesses should be highlighted and results critically tested.
Irregularities at temporal and spatial scales give rise to deviations in inventories of aquaculture production. Underlying models, moreover, rely on assumptions and methodological choices influence the results. Statistical tools and sensitivity analyses are therefore important to strengthen the arguments and conclusions in aquaculture LCAs. Treating farms individually, rather than as averages, would here allow for more extensive statistical comparisons to be made between farms. Known pivotal factors identified in the articles reviewed here include various inventory choices (feeds, raw materials, infrastructure, etc.), GHG emissions from agricultural fields and aquatic systems, nitrogen and phosphorus emissions, allocation factors, and characterization factors. Further efforts are therefore needed to account for the many degrees of freedom, using more extensive sensitivity analyses and implementing, e.g., Monte Carlo analysis.