Spatially differentiated midpoint indicator for marine eutrophication of waterborne emissions in Sweden

The characterisation model presented in this study derives midpoint CFs (kg N_eq/kg) for marine eutrophication, with focus on fate modelling. The following equation based on the adapted version of Seppälä et al. (2004) from Gallego et al. (2010) was used:

where transport factor _i,j (−) describes the transport of substance i from the point of emission j to a marine recipient, nutrient limitation factor _i,j (−) the sensitivity to substance i in the marine recipient corresponding to the point of emission j, and equivalency ratio _i (kg N_eq/kg) the algae growth potential of substance i. The marine recipient is given by the location of the point of emission.

Recent midpoint LCIA methods for emission-related impacts are designed using the concept CF = fate × exposure × effect (Pennington et al. 2004). A different terminology was chosen for this study, mainly due to the focus on substance transport from point of emission to recipient, and disregard of substance fate and effect in the recipient. However, the components of the CFs presented in this study can be related to this framework. The retention on the substance pathway, here represented by the transport factor, could be complemented by a component representing residence time in the recipient, to constitute a fate factor. This approach is adopted by Cosme and Hauschild (2017), who use watershed export fractions as a component of site-dependent marine eutrophication fate factors. The combination of our nutrient limitation factor and equivalency ratio could be seen as an exposure factor (following the terminology of e.g. Cosme et al. 2015), although it lacks a component to describe the recipients’ degree of sensitivity, but rather assumes that all nutrients that reach a recipient sensitive to that nutrient will cause eutrophication.

Globally or regionally weighted average CFs are commonly provided to allow for application of the method when spatial information on emissions is missing. In this study, a weighted country average CF was calculated and used for comparison with other LCIA methods. The local CFs derived were weighted according to the amount of agricultural land used in the particular area in 2005, based on data from SMED (2015b).

The purpose of the comparison was to indicate the quantitative level of agreement between the new CFs and commonly used midpoint CFs for marine eutrophication. Three established eutrophication methods stated to be applicable for emissions to soil were chosen for comparison: CML 2001 (Guinée 2002), EDIP 2003 (Hauschild and Potting 2005) and ReCiPe 2008 (Struijs et al. 2013). CML was chosen because it is still widely used, EDIP because it provides country-specific CFs and ReCiPe because it was recently recommended as the best available LCIA methodology for aquatic eutrophication (Hauschild et al. 2013). The emerging impact assessment methodology LC-IMPACT (http://​www.​lc-impact.​eu/) would also be an appropriate candidate for comparison due to its comprehensiveness, but it was excluded from the quantitative comparison since the marine eutrophication characterisation factors are not published in full yet. Instead, it is discussed qualitatively in the Discussion section. Marine eutrophication CFs for emissions to soil are currently not available for IMPACT World+ (http://​www.​impactworldplus.​org), and comparison was therefore not possible.

The CF for eutrophication in the CML 2001 baseline methodology (Guinée 2002) is equal to the equivalency ratio. CML 2001 does not discriminate between terrestrial, marine and freshwater eutrophication, so these values represent the total potential eutrophication impacts on all recipients. For EDIP 2003, the CF for Sweden was used. EDIP 2003 recommends not including P in the environmental impact category marine eutrophication (Hauschild and Potting 2005), but since a CF for P to marine environments is included in the EDIP 2003 report, it was included here for comparison. For ReCiPe 2008, the site-generic CF for fertiliser was used. Since the ReCiPe 2008 CF is developed to be applied to gross input of nutrients to the soil, it was corrected according to the correction factors provided in Struijs et al. (2013) to comply with the system boundary. The ReCiPe 2008 eutrophication methodology includes nutrient fate and considers marine recipients as strictly N-limited.

To estimate the relative contribution of N and P to the marine eutrophication impact of typical Swedish crop cultivation, the CFs were applied to realistic values on nutrient leaching from spring barley and grass ley. This was done using leaching data from Appendices 2 and 4 in Johnsson et al. (2008), which list average leaching per crop and field characteristics for 22 agricultural regions in Sweden. These were combined with data on soil type, phosphorus content and field inclination for 8559 sub-catchments, corresponding to sub-areas of the catchment areas for which transport factors were derived, and encompassing all parts of Sweden with current crop production (SMED 2015a). The obtained emission inventories for each sub-catchment were then multiplied by the site-dependent CFs following Eq. (1).

The calculated midpoint CFs represent the transport of waterborne nutrients from point of emission to coast and the limiting nutrient in the recipient ecosystem. The site-dependent CFs covering all of Sweden are presented in Fig. 2 for N emissions and in Fig. 3 for P emissions. These CFs varied between 0.056 and 0.986 kg N_eq/kg N and between 0 and 7.23 kg N_eq/kg P (all values presented in Online Resource 1). The CFs were generally higher in areas closer to the coast, but there were several deviations from this pattern. The CFs for P also tended to be higher in the north and were zero for the catchments that drain into Öresund and Kattegat, since the nutrient limitation factor for P is zero in these areas.

The resulting area-weighted CF for Sweden and the corresponding CFs from CML 2001, EDIP 2003 and ReCiPe 2008 are presented in Table 2. As can be seen, the CFs for N and P were lower than the CML values. The CF for N was similar to the EDIP and ReCiPe values, but the CF for P was several-fold higher than the EDIP CF.

The marine eutrophication impacts calculated for 8559 sub-catchments in Sweden varied between 1.2 and 58 kg N_eq/ha for spring barley and between 0.31 and 32 kg N_eq/ha for grass ley (Fig. 4). Area-weighted total average was 21 kg N_eq/ha for spring barley and 7.5 kg N_eq/ha for grass ley. Nitrogen was the nutrient that accounted for the largest eutrophication impact (82% in spring barley cultivation and 75% in grass ley cultivation) under the simplified assumption that the same amount of land was cultivated in each sub-catchment. However, P had a larger impact than N in 2 and 15% of the sub-catchment for spring barley and grass ley, respectively.

The sensitivity analysis showed that assuming N and P limitation in all sub-basins had a minor impact on the results, increasing calculated average marine eutrophication impact from spring barley and grass ley production by less than 10% (Table 3). Assuming only N limitation in all sub-basins reduced the calculated marine eutrophication impact by up to 25%.

The variation in CFs within Sweden (Figs. 2 and 3) shows that using the country average as a site-dependent measure can mask potential differences in impacts from nutrient emissions at different sites. This means that country average CFs may give misleading LCIA results for site-dependent LCAs, supporting the claim by Bockstaller et al. (2008) that local impacts should ideally not be aggregated into higher scales than the actual impact.

A study on eutrophication CFs for airborne emissions within Sweden concluded that spatial differentiation did not add much variability (Finnveden and Nilsson 2005), so there may be a discrepancy in appropriate scale for airborne and waterborne nutrient emissions. However, Finnveden and Nilsson (2005) point out that distinguishing between environmental regions would be better than between countries. In the current study, the transport factor varied between 0.06 and 1 for N, and 0.007 and 1 for P, indicating that the transport pathway between the point of emission and the marine recipient is an important source of spatial variability in the CFs and therefore should not be neglected when deriving spatially differentiated fate factors. This was also highlighted by Basset-Mens et al. (2006) for nitrate transport in Brittany, France. Cosme and Hauschild (2017) used watershed export fractions with a spatial resolution of 5772 watersheds globally to model endpoint and damage CFs, and found soil-related fate factors to be responsible for most of the spatial differentiation of their damage model results. The scope of previous studies on site-dependent aquatic eutrophication assessment (e.g. Seppälä et al. (2004); Basset-Mens et al. (2006); Gallego et al. (2010) do not allow for direct comparison with the CFs obtained in the present study, but they generally support the conclusion that even country-specific CFs miss some of the local variation. Alternative spatial resolution for site-dependent eutrophication LCIA based on hydrological processes has been proposed by e.g. Basset-Mens et al. (2006) and Nitschelm et al. (2016). This is also in line with the approach in the EU Water Framework Directive (Directive 2000/60/EC), which considers water management at the river basin scale.

Identifying a relevant scale for LCIA models is not primarily about using the highest resolution, since factors such as uncertainty and possibility to interpret the results should also be considered (Mutel et al. 2012). The variability found within catchments in this study indicates that there are significant spatial variations even before the nutrients reach the marine recipient. This means that high-resolution fate factors can add valuable information to the assessment and that the appropriate scale for marine eutrophication impact assessment is likely to be small.

The weighted average CFs were lower than the CML 2001 CFs, which was to be expected since the eutrophication impacts calculated by the CML method represent the maximum potential eutrophication impact on all recipients, including freshwater and terrestrial ecosystems. While the weighted average CF for N agreed well with the EDIP 2003 and ReCiPe CFs (Table 2), the average CF for P did not. A possible reason for this is that EDIP 2003 defines the inventory as the amount of nutrients available for leaving the topsoil after plant uptake and binding (Hauschild and Potting 2005), whereas we defined it as the nutrients leaving the root zone. This means that the system boundary between the technosphere and the environment was defined differently in the studies. The nutrients retained in the root zone would then be included in the inventory following EDIP 2003, while according to our definition, they would not be considered an emission until they had actually left the topsoil. This could make a significant difference, since P, unlike N, binds strongly to soil particles and therefore can accumulate in the topsoil (Hooda et al. 2001). Another factor possibly contributing to the discrepancy in CF values is that the model used for EDIP 2003 assumes fixed removal rates in soil and freshwater (Hauschild and Potting 2005). Fixed removal rates are based on the assumption that the fate of P is directly correlated with fertiliser application and that it is equal for all soils and freshwater systems. In reality, P leaching and retention are highly variable and depend on factors such as soil properties, soil management and previous P loads (Bergström et al. 2015). In contrast, the retention data for our transport factors are calibrated against field measurements (Brandt et al. 2009), so it appears that Swedish conditions deviate from the generic removal rates assumed for EDIP 2003 CFs. This highlights the importance of considering local characteristics when deriving regional CFs and somewhat contradicts use of generic models to derive regional CFs, since some local factors which may have a significant effect on the actual impact are difficult to capture in these models.

The combination of this study’s nutrient limitation factor and equivalency ratio (i.e. 1 kg N_eq/kg N) can be compared with the exposure factors presented by (Cosme et al. 2015), which for the Baltic Sea equates to 0.807 kg N_eq/kg N (recalculated using the Redfield ratio to match our units). Even though this comparison shows that our assumed exposure factor gives a similar result as the model by Cosme et al. (2015) which includes more processes, it should be noted that the Baltic Sea has the highest exposure factors among LMEs (Cosme et al. 2015), and thus, this assumption may be less valid for other LMEs (the lowest exposure factor equated to approximately 0.023). However, we want to highlight the fact that the results in this study show that it may be important to also include P limitation in marine eutrophication assessment and that differentiation in the marine environment on a smaller spatial scale than LMEs may be called for.

Accurate impact assessment is becoming more important as the role of LCA in environmental assessment of agricultural products increases, and the potential of LCA as a decision support tool for farmers has been highlighted (Bartl et al. 2012; Fedele et al. 2014). As demonstrated by the variability in CFs in Figs. 2 and 3, higher spatial resolution should improve the correlation between LCIA results and actual impacts.

Eutrophication assessment in LCA using CFs with high spatial resolution would be especially informative when systems that include cultivation at different sites are compared, as in e.g. Korsaeth et al. (2014) and Lehuger et al. (2009), or for determining how to optimise the geographic placement or distribution of agricultural activities to minimise environmental impacts, as proposed by Nitschelm et al. (2016) and Page et al. (2014). Several case studies, e.g. Bessou et al. (2013) and Korsaeth et al. (2014), have attempted to spatialise LCA of crop production by using local emissions data, but are not using site-dependent impact CFs for eutrophication. Doing this may give a false impression of spatial differentiation, while in fact, significant spatial differences in environmental impacts are omitted. This indicates that site-dependent eutrophication CFs such as those presented in this study would be useful, but must then be available for the specific region. In addition, spatialised LCAs require systematic ways to handle the large amount of data in order to be a feasible approach for LCA practitioners, but methods to facilitate the application of regionalised impact assessment have been demonstrated (e.g. Mutel and Hellweg 2009; Mutel et al. 2012). There are also cases where CFs for a larger region, such as country or global scale, would be more suitable than regional CFs, e.g. LCAs of an average product or process from the region. It would then result in more accurate impact assessments if CFs at finer spatial scale were aggregated to represent average CFs at appropriate scale for the assessment, instead of CFs derived by site-generic methods. When these aggregated CFs are used, it should be made clear that they represent an average impact in the region, rather than representing the impact at all sites in the region.

On a global level, the displacement of N and P has been cited as one of the planetary boundaries where the safe operating space has been exceeded (Steffen et al. 2015), so there is a need for appropriate representation of these effects in environmental assessment tools such as LCA. For Sweden, climate change and lifestyle changes are expected to increase nutrient loads to the Baltic Sea (Hägg et al. 2014), giving further reason for action and increasing the need to prioritise between load-reducing measures. As better modelling of the consequences of different actions on watershed quality can help support policy improvements (Merot et al. 2009), recognising spatial variations in the relationship between nutrient loads and eutrophication effect can help improve the effectiveness of eutrophication-reducing measures and policies (Tysmans et al. 2013). For example, spatial differentiation was necessary to source-appoint the nutrient loads to the Baltic Sea in PLC-5 (Brandt et al. 2009). This spatial differentiation should also be considered when LCAs are used to connect the environmental pressure of agricultural products to the production location, as in e.g. Röös et al. (2011). In this context, our analysis of the relative importance of N and P highlights the applicability for single-site assessments when several eutrophying substances are involved, which is often the case for crop production where both N and P are emitted. Our analysis showed that spatialised CFs for eutrophication assessment could help identifying which nutrients to target when investing in measures that reduce waterborne nutrient emissions. Judging from the results, the eutrophication pressure on the seas surrounding Sweden is higher from N pollution than that from P pollution from spring barley and grass ley cultivation at most sites in Sweden.

The sensitivity analysis (Table 3) showed that varying the assumed limiting nutrient in the Baltic Sea had a moderate effect on CFs and an even smaller effect on the average potential impacts of cultivation. This is because the calculated impact of N generally was larger than that of P for the crop cultivation situations assessed here. Although the sensitivity to P limitation was of minor importance for the country average impacts in the cases tested here, assumptions made on the limiting nutrient may still be an important factor for emissions in the specific areas where P impacts were more prominent than N impacts (2 and 15% of the sub-catchments for grass ley and spring barley, respectively) and in cases where the N:P ratio of the emissions are lower than for the average cultivation situations studied here.

For practical reasons, sub-basin resolution is not always a viable alternative for LCIA. However, nutrient limitation at the same level as the spatial differentiation of the CFs should be considered when regionalised LCIA methods are being developed. In the case of waterborne emissions from Sweden to the Baltic Sea, the results presented in this study (Table 3) suggest that assuming the entire Baltic Sea as N- and P-limited would give more accurate results than assuming only N limitation. Although N seems to be the primary limiting nutrient in most marine ecosystems, the Baltic Sea is not the only exception. Other studies report marine environments where both N and P limitation occurs and thus where P impacts could be important to include in marine eutrophication LCIA. These cases include tidal systems (Barba-Brioso et al. 2010) and estuaries such as Chesapeake Bay (Fisher et al. 1999) and the rias in Galicia, Spain (Gallego et al. 2010).