Environmental performance of crop cultivation at different sites and nitrogen rates in Sweden

Environmental performance of crop cultivation at different sites and nitrogen rates in Sweden Nutr Cycl Agroecosyst (2019) 114:139–155 https://doi.org/10.1007/s10705-019-09997-w(0123456789().,-volV)(0123456789().,-volV) ORIGINAL ARTICLE Environmental performance of crop cultivation at different sites and nitrogen rates in Sweden . . . . Kajsa Henryson Per-Anders Hansson Thomas Ka ¨ tterer Pernilla Tida ˚ ker Cecilia Sundberg Received: 14 November 2018 / Accepted: 29 April 2019 / Published online: 8 May 2019 The Author(s) 2019 Abstract Nitrogen (N) fertilisation has positive and unit. The calculations were based on data from multi- negative effects on the environmental impact of crop site long-term field experiments in Sweden and site- cultivation. The mechanisms governing these effects dependent data and models for non-measured pro- are highly site-dependent, a factor often ignored in cesses. Cultivation at three N levels was evaluated, assessments of the environmental impact of crop where the highest N rate was close to current average cultivation. By assessing outputs of crop rotations practices and the lowest level corresponded to one- using a life cycle approach, this study explored how third of that. Site characteristics had a stronger greenhouse gas emissions and marine eutrophication influence on both greenhouse gas emissions and caused by crop cultivation (including upstream pro- marine eutrophication (variations of up to 330% and cesses such as production of farm inputs) depend on 490%, respectively, within N levels) than N level fertiliser rate and the site at which the cultivation (variations of up to 74% and 59%, respectively, within occurs. Cereal unit (CU) was used as the functional sites). Main sources of variation in greenhouse gas emissions were soil nitrous oxide emissions (58–810 g -1 CO CU ) and soil organic carbon changes 2eq Electronic supplementary material The online version of -1 (14–720 g CO CU ), while variations in marine 2eq this article (https://doi.org/10.1007/s10705-019-09997-w) contains supplementary material, which is available to autho- eutrophication were mainly explained by field-level rized users. -1 waterborne N losses (0.9–8.2 g N CU ). The large eq variation between sites highlights the importance of K. Henryson (&)  P.-A. Hansson  P. Tida˚ker C. Sundberg considering site characteristics when assessing the Department of Energy and Technology, Swedish environmental impact of crop cultivation and evalu- University of Agricultural Sciences (SLU), ating the environmental consequences of crop man- P.O. Box 7032, 750 07 Uppsala, Sweden agement practices. e-mail: kajsa.henryson@slu.se T. Ka¨tterer Keywords Environmental impact  Life cycle Department of Ecology, Swedish University of assessment  Eutrophication  Climate impact Agricultural Sciences (SLU), P.O. Box 7044, Greenhouse gas emissions  Site-dependent 750 07 Uppsala, Sweden C. Sundberg Department of Sustainable Development, Environmental Science and Engineering, KTH Royal Institute of Technology, Teknikringen 10B, 100 44 Stockholm, Sweden 123 140 Nutr Cycl Agroecosyst (2019) 114:139–155 Introduction dependent on site conditions (Goglio et al. 2018). Site can thus influence estimated GHG emissions even The introduction of mineral nitrogen (N) fertilisers has though the impact itself is site-independent. Unlike enabled yield increases and has been a crucial factor climate impact, eutrophication is a site-dependent for increasing global food security during the past impact that mainly occurs due to regional emissions century (Sutton et al. 2013). Increasing crop yields (Potting and Hauschild 2006). Eutrophication is a may also have several positive effects on the environ- particularly urgent environmental issue for Sweden, ment, such as increasing resource use efficiency, which is surrounded by the world’s largest anthro- avoiding land use change and increasing input of crop pogenic hypoxic zone, the Baltic Sea (Diaz and residues available for soil carbon sequestration, but Rosenberg 2008), where agriculture is one of the main may require increased inputs of e.g. water and contributors to this issue (Licker et al. 2010). The site- pesticides (Snyder et al. 2009; Tilman et al. 2002). dependent nature of the eutrophication cause-effect Furthermore, N fertiliser use itself causes negative chain means that it is necessary to consider where environmental impacts through fossil fuel use and emissions occur in order to make meaningful assess- losses of reactive N both during the production phase ments of eutrophication impact, especially when and after field application. It is therefore possible that comparing products produced at different sites (Page there are trade-offs between several environmental et al. 2014). objectives. Due to its multifaceted effects on agroe- A system perspective is necessary to account for the cosystems, N fertiliser rate in relation to yield often environmental impacts of different substance flows emerges as one of the most influential variables for the occurring during the process of producing crops. A magnitude of environmental impact of cropping commonly used tool for determining the environmen- systems (Van Stappen et al. 2017; Aoun et al. 2014; tal impact of a product is life cycle assessment (LCA), Charles et al. 2006). However, the relationship which is frequently used for agricultural products between N fertiliser rate and environmental impacts (Notarnicola et al. 2017). In LCA, emissions and is complex, since both yield and field emissions of resource use during the whole or part of the life cycle reactive N exhibit non-linear responses to increased N of a product or process are compiled and related to fertiliser rate (Delin and Stenberg 2014). Since both environmental impacts (ISO 2006). This approach yield and emissions are also affected by local culti- enables identification of possible trade-offs between vation conditions such as climate and soil character- products or production process alternatives, and istics (Delin et al. 2005; Rochette et al. 2018), it is thereby reveals potential burden shifts between impact possible that the environmental impact of cultivation categories or life-cycle stages. LCA of cropping under different fertiliser rates will depend on the site. systems is typically performed for either a site In that context, multi-site long-term field experiments representing average conditions or for one or a few are a valuable source of data. specific sites, due to lack of data. Different method- Nitrogen fertiliser use is closely related to both ologies and degrees of spatial specificity limit inter- greenhouse gas (GHG) emissions and nutrient losses comparability (Bessou et al. 2013), and it is often to watersheds, which have been identified as key areas unclear whether differences in results between studies for actions to decrease the environmental impact of can be attributed to spatial or methodological varia- agriculture in the EU (European Commission 2017). tions. The aim of this study was therefore to evaluate Previous studies have shown that decreasing fertiliser cradle-to-farm gate GHG emissions and marine input can reduce GHG emissions per unit produced eutrophication of crop cultivation at different N (e.g. Ashworth et al. 2015; Williams et al. 2010; fertiliser levels and different field sites in Sweden, Brentrup et al. 2004; Goglio et al. 2012). However, and to use the results to analyse the influence of these since N fertilisers promote higher yields, this outcome two factors on estimated environmental impact. Data may change if soil organic carbon (SOC) changes are from multi-site long-term field trials were used, included, which are also affected by site conditions together with site-dependent data and models for (Garcıa-Ruiz et al. 2019). Another potential source of non-measured processes. spatial variability in estimated GHG emissions is soil nitrous oxide (N O) emissions, which are highly 123 Nutr Cycl Agroecosyst (2019) 114:139–155 141 Methods Fields receiving no N were not included in the analysis, since this does not reflect realistic agricul- Experimental sites and set-up tural practices, but are included in Fig. 2 to show the whole yield response curve. Average annual N The assessment was based on data from nine sites in fertiliser rates were higher at the southern sites than the Swedish Soil Fertility Trials (described by Carl- at the central sites (Table 1), but the high N level was gren and Mattsson 2001); four in southern (S1–S4) and close to current average cereal N fertilisation rates in five in central Sweden (C1–C5) (Fig. 1). These sites each study region. Applied amounts of phosphorus represent varying cultivation conditions in terms of (P) and potassium (K) fertilisers were chosen to soil characteristics and microclimate (Carlgren and replace annual removal, but the P and K rates were Mattsson 2001). Selected site characteristics are also similar to current average practice in the two shown in Table SM1 in Supplementary Material. regions (Statistics Sweden 2016). A full list of The trials have been ongoing since the 1950s and fertiliser amounts is provided in Tables SM3–SM5. 1960s in southern and central Sweden, respectively. Hereafter, we use the term ‘treatment’ to refer to each The crop rotations included in the present study combination of site and N level (9 9 3=27 consisted of a four-year rotation in the southern region treatments). (Table 1), with two replicates at each site. The crop rotations included were exclusively given mineral Goal, scope, functional unit, system boundaries fertilisers, and all crop residues were left in the field and assumptions and incorporated into the soil. Three N levels were considered in the analysis: low, medium and high. Life cycle assessment was used to assess the GHG emissions and marine eutrophication impact over the whole crop rotation, under three different N fertiliser levels, at the nine sites in southern and central Sweden. The system boundaries were set at cradle-to-farm gate, which included emissions from: • Inputs (fertiliser and pesticide production) • Field operations (fuel consumption, production and maintenance of machinery) • Soil emissions (waterborne N and P compounds and airborne N O, ammonia (NH ) and NO ) 2 3 x • Crop drying • SOC changes. In LCA, a functional unit is needed to relate inputs and outputs to a common unit. The functional unit is a quantified description of the product system functions (ISO 2006), and should be chosen with respect to the goals of the study (Caffrey and Veal 2013). Dry or fresh matter mass of harvested crop, together with area occupied by the crop, are the most common functional units used for agricultural LCAs (Notarnicola et al. 2017). However, mass is not a fair representation of the crop function when different crop types are present, and therefore cereal unit (CU) was chosen as the functional unit in this study instead. Cereal unit was first introduced to LCAs by Brankatschk and Finkbeiner (2014), as a basis for co-product allocation Fig. 1 Map showing location of experimental sites in southern between grain and straw (ibid.) or between the crops in Sweden (sites S1–S4) and central Sweden (sites C1–C5) 123 142 Nutr Cycl Agroecosyst (2019) 114:139–155 Table 1 Crop rotations and Southern sites Central sites fertiliser rates in the (S1–S4) (C1–C5) southern and central a a cropping systems Crop rotation Spring barley Spring barley b c Spring oilseed rape Oats d b Winter wheat Spring oilseed rape e d Sugar beet Winter wheat Oats Hordeum vulgare Winter wheat Brassica napus/rapa Average fertiliser rate at low N level 50 40 Avena sativa Average fertiliser rate at medium N level 100 80 Triticum aestivum Average fertiliser rate at high N level 150 120 Beta vulgaris -1 -1 Fig. 2 Mean annual yields obtained at each site and N level, measured in cereal units (CU) ha (dark grey) and dry matter (DM) ha (light grey) a crop rotation (Brankatschk and Finkbeiner 2015). It combined with data on animal species composition has also been used as a functional unit in LCAs and species-specific metabolism. We used the CU comparing cropping systems (Prechsl et al. 2017). The values presented in the Supplementary Material of CU represents the animal feeding value of each crop Brankatschk and Finkbeiner (2014), where calculation related to a reference crop (barley), and thereby procedures are also presented in more detail. The enables crop yields to be represented by a unit that also relationship between harvested dry mass and CU considers crop function. One CU corresponds to 1 kg harvested from the trial plots during the study period of fresh matter barley (at 14% moisture content). The (1988–2009) is presented in Fig. 2. Subdivision of animal feeding value is calculated based on protein, inputs and emissions among the crops in the rotation lipid, fibre and carbohydrate content of the crop, was not necessary in this study, since the aim was to 123 Nutr Cycl Agroecosyst (2019) 114:139–155 143 assess the environmental impact over the whole crop Fuel consumption L ha rotation. Inventories were instead calculated for each ¼ 8:5 þ 0:32  clay contentðÞ % ð1Þ treatment and year in the study period, and then Emissions from machinery production were calcu- aggregated over all years in the study period and lated from the number of passes (see Table SM6 in divided by the amount of CU produced for each Supplementary Material), data on kg machinery treatment during the study period. Amount of land consumption per pass for each operation and emis- occupied by crop cultivation was also determined by sions data for production of agricultural machinery calculating the field area used per CU in each from the ecoinvent database (Table 2). treatment. Waterborne N emissions at field level (nitrate leaching) were calculated using a model implemented Activity and emissions data for life cycle inventory in the farm management tool VERA, developed by the Swedish Board of Agriculture, adapted to Swedish Most of the farm activity data were obtained from field conditions and mainly based on empirical data. The measurements at the trial sites. They included data on model is described by Aronsson and Torstensson harvested yield, soil carbon content, bulk density and (2004) and calculates N leaching using site- and soil texture. The study period chosen was 1988–2010, management-dependent typical leaching, and with since data were readily accessible and consistent crop correction factors for over- or under-fertilisation in rotations were used for all sites during this period. relation to the yield achieved. Further information on Yields from 4 years in each time series were missing the model structure and parameters used is provided in due to harvest failure, and these years were thus Supplementary Material to this paper. Phosphorus removed from the series at all sites in the two regions, losses at field level were assumed to be equal to the to ensure consistent number of occurrences of each region-, soil- and crop-specific average P losses, using crop in each of the two crop rotations (see Table SM2 data from Johnsson et al. (2016). in Supplementary Material). Data that were not The emission factors for field-level volatilisation available from the field experiments, such as field -1 were set to 0.016 and 0.033 kg NH kg N fertiliser operations, fertiliser types and pesticide amounts, for soils with pH \ 7 and pH [ 7, respectively, and were chosen to represent typical Swedish farming -1 0.04 kg NO kg N fertiliser, after recommendations practices and, as far as possible, region-specific data from EMEP/EEA (2016). Direct soil N O emissions were used. Specific data for all activities included are were calculated using the equation for cumulative presented in this section. N O emissions (Eq. 2) derived by Rochette et al. The chosen emissions inventory for production of (2018), who tested field measurements from 474 ammonium nitrate, triple superphosphate and potas- treatment-years in Canada against different predictors. sium chloride was taken from the GaBi database The equation utilises data on precipitation during the (Fertilizers Europe 2018a, b, c; Brentrup et al. 2016), growing season, applied mineral fertiliser, soil sand which is representative of mineral fertilisers produced content, soil pH and mean annual air temperature, and in Europe. Assumed amounts of pesticide used were is thereby a more site-dependent model than the differentiated based on crop and cultivation region, commonly used generic emission factors, such as Tier using data from Statistics Sweden (2011). Emissions I values from IPCC (De Klein et al. 2006). associated with pesticide production were taken from the ecoinvent database (Table 2). ^ N O ¼ e ð3:91 þ 0:0022  PðtÞþ 0:0069 2 direct Diesel consumption for ploughing was calculated MinNðtÞ 0:0032  sand  0:747  pH using Eq. 1 (Arvidsson and Keller 2011) and diesel þ 0:0097  T ðtÞÞ air consumption values for the other field operations were ð2Þ taken from Lindgren et al. (2002). Assumed numbers -1 of passes for each machine operation were differen- where N O (kg N O–N ha ) is annual cumula- 2 direct 2 tiated according to crop (Table SM6 in Supplementary tive soil N O emissions from N fertiliser application, Material). Energy density and emissions data from P (mm) is the precipitation during May to September, -1 production and use of diesel were taken from Gode MinN N O (kg N ha ) is the mineral N applied, 2 direct et al. (2011). 123 144 Nutr Cycl Agroecosyst (2019) 114:139–155 Table 2 Names of ecoinvent processes used for inventory data in this study. All inventories were taken from ecoinvent version 3.3, allocation: cut-off by classification Ecoinvent process name Used in this study for: Pesticide production, unspecified, RER Pesticide production emissions Tillage, cultivating, chiselling, RoW Machinery required for stubble cultivation Tillage, ploughing, RoW Machinery required for ploughing Tillage, harrowing, by spring tine harrow, RoW Machinery required for harrowing Sowing, RoW Machinery required for sowing Fertilising, by broadcaster, RoW Machinery required for applying fertiliser Application of plant protection product, by field sprayer, RoW Machinery required for applying pesticides Combine harvesting, RoW Machinery required for threshing Harvesting, by complete harvester, beets, RoW Machinery required for harvest/baling of sugar beet Agricultural machinery production, unspecified, RoW Production emissions per unit machinery -1 sand (g kg ) is the sand content of the soil, pH is the in the topsoil were then estimated from the slope and soil pH, T (C) is the mean annual air temperature site-specific bulk density values. The choice to include air and t is the year. Data for P and T were obtained a longer time period for the SOC change estimation air from the Swedish Meteorological and Hydrological was made to avoid unfair allocation between years in Institute (2018) by choosing the weather station the study period and because there were too few closest to each site with available data for all years measurements during the study period. This should be in the study period. Indirect soil N O emissions were considered when interpreting calculated SOC calculated by applying IPCC default emission factors changes, and for this reason we report and discuss to the calculated volatilised and leached N (De Klein both total estimated GHG emissions and GHG emis- et al. 2006). sions from non-SOC processes. The cereals in the rotations were assumed to be Seed production was included by subtracting cor- harvested at 17% moisture content at sites S1–S4 and responding seed rates from yield. Assumed seed rate -1 19% moisture content at sites C1–C5, and dried to for barley was 170 kg ha , for wheat and oats -1 -1 14% moisture content. Oilseed rape was assumed to be 210 kg ha and for oilseed rape 6 kg ha (Ahlgren dried from 12 to 9% moisture content, while no drying et al. 2009). It was assumed that 60 kg sugar beet were or other post-harvest treatment of sugar beet was required to produce the 2.1 kg seeds needed per ha included, since this typically occurs after farm-gate. (Nemecek and Schnetzer 2011a, b). The energy consumption during crop drying was Liming rates in the experiments were very low (on -1 -1 assumed to be 0.15 L of heating oil per kg of water average 0.1–0.15 kg CaO ha year ) during the removed and 19 kWh of electricity per ton grain study period and it was therefore not considered (Edstrom et al. 2005). Life cycle emissions data for relevant to include impacts of liming. heating oil were taken from (Gode et al. 2011) and data on the Swedish electricity mix from (Brander et al. Life cycle impact assessment 2011). Organic carbon concentration in the topsoil The GHGs carbon dioxide (CO ), methane (CH ) and 2 4 (0–20 cm) was measured 6–10 times in every treat- N O were included in the assessment, and their climate ment during the whole trial lifetime (1956–2011), and impacts were calculated using the IPCC characterisa- we opted to use these data to derive mean SOC tion factors (including climate-carbon feedbacks) for changes. This was done by estimating the slope for GWP reported by Myhre et al. (2013). SOC concentration measurements over time for each The marine eutrophication impact of waterborne N treatment using linear regression analysis (see regres- and P losses at field level was calculated using the sion plots in Fig. SM1). Annual carbon stock changes method described by Henryson et al. (2018). This 123 Nutr Cycl Agroecosyst (2019) 114:139–155 145 method uses site-dependent characterisation factors, the mean impact, and that the mean impacts tested for which express the impact in N equivalents (N ) and in the ANOVA analysis were consistent with the mean eq account for retention of nutrients along their way to impacts according to Eq. 3. Since the normalisation marine environments, as well as the limiting nutrient generated two sets of normalised impacts, the effects in the marine recipient, while most other marine of N level and site were analysed by separate ANOVA eutrophication life cycle impact assessment (LCIA) analyses. The significance level was set to 0.05. methods only account for N additions. In contrast to Impact  Yield i;t t Normalised impact ¼ ð4Þ the majority of marine recipients, the sub-basins of the i;t;n Yield kðt;nÞ Baltic Sea surrounding Sweden can be growth limited by either N or P, or both (Swedish EPA 2006). Impacts where i is the impact category, t is the treatment (any of other N emissions were calculated using the combination of N level and site), n is the normalisation ReCiPe2008 LCIA method (Struijs et al. 2013). variable (i.e. site when the effect of site was analysed Waterborne P emissions during fertiliser production and N level when the effect of N level was analysed) were assumed to occur at Yara’s Siilinja¨rvi phosphate and k(t, n) is the treatment site or N level. The results mine in Eastern Finland, which is Europe’s only from the ANOVA analysis were followed up by phosphate mine and thereby the only source from Tukey’s HSD (Honestly Significant Difference) test which P-containing production effluent could end up for N level. in the Baltic Sea. This assumption therefore represents In addition to the ANOVA, a simple sensitivity the worst case in terms of the marine eutrophication analysis was performed for direct soil N O emissions impact of P emissions during fertiliser production. and waterborne soil N emissions, since these were Retention of 98% was assumed for P-containing modelled and expected to contribute substantially to effluent from Siilinja¨rvi (Huttunen et al. 2013). All the estimated impacts. The sensitivity analysis was other P emissions were assumed to be emitted at performed by changing the modelled direct soil N O locations where they would not reach a P-limited emissions and waterborne N emissions one at a time marine environment and were therefore not included. by 20%. Statistical analyses was performed using the soft- Statistical analyses ware R version 3.5.1. All results from statistical tests are provided in the Supplementary Material The mean impacts for each N level and site were (Tables SM9–SM12). calculated as the mean impacts per CU produced, which involved accounting for the fact that each treatment produced different outputs. The approach Results described in Eq. 3 was therefore used to calculate mean impacts: The results revealed large variations between sites and N levels in both impact categories. Estimated ðImpact  Yield Þ i;k;a k;a a¼1 Mean impact ¼ P ð3Þ total GHG emissions varied from 260 to 1280 g i;k Yield -1 k;a a¼1 CO CU between all treatments, and from 200 to 2eq -1 1040 g CO CU when SOC changes were 2eq where k is a site or N level, i is impact category and A is excluded (Fig. 3). Maximum difference in non-SOC all sites when k is an N level and all N levels when k is GHG emissions between sites within the N levels was a site. 170, 250 and 330% for low, medium and high N, One-way ANOVA analysis with a randomised respectively, while the maximum difference between block design was used to investigate whether N level N levels within sites was 13–74%. Thus, variation or site affected the mean for each impact category. The between sites was larger than variation between N impacts for each treatment were normalised for yield levels (Fig. 3). The highest N level gave the highest according to Eq. 4 before performing the ANOVA non-SOC GHG emissions for all sites, but the order for analysis, to keep consistency with calculation of the low and medium N levels varied between sites. SOC means. This procedure ensured that treatments were losses generally contributed to large GHG emissions fairly weighted according to how they contributed to 123 146 Nutr Cycl Agroecosyst (2019) 114:139–155 Fig. 3 Greenhouse gas (GHG) emissions and land area different nitrogen (N) fertiliser intensities, where high N is occupied by crop cultivation at the four southern Swedish sites closest to current common practice (S1–S4) and five central Swedish sites (C1–C5) under three -1 (2.6–65% of total GHG emissions) compared with sites varied between 290 and 1090 g CO CU for 2eq -1 other emission sources, and including SOC changes total GHG emissions, or 220 and 800 g CO CU 2eq therefore influenced which N level minimised GHG when SOC changes were excluded (Fig. 3). In contrast emissions. Unlike the combined impact of other to the differences between N level means, the differ- emission sources, the contribution of SOC changes ences between site means were highly statistically generally decreased with higher N level. The com- significant for both total GHG emissions and non-SOC bined effect of SOC changes and other emission GHG emissions (Table SM10). sources was that the optimum N level in terms of Marine eutrophication varied from 2.0 to 9.9 g -1 minimising total GHG emissions was low, medium or N CU (Fig. 5), with a maximum difference of eq high depending on site. 300%, 330% and 390% between sites at the low, Mean total GHG emissions were 620, 560 medium and high N levels, respectively, and a -1 and 600 g CO CU for the low, medium and maximum difference of 7–59% between N levels at 2eq high N level, respectively, or 380, 400 and 480 g different sites. Thus, similarly to the results for GHG -1 CO CU , respectively, when SOC changes were emissions, differences between N levels were gener- 2eq excluded (Fig. 3). However, differences between N ally smaller than differences between sites. The level means were not statistically significant for total medium N level gave the lowest marine eutrophication GHG emissions, and differences between low and at all central sites (C1–C5) and one of the southern medium N level means were not statistically signif- sites (S3), while the low N level gave the lowest icant when SOC changes were excluded marine eutrophication at the remaining three southern (Tables SM10–SM11). Mean GHG emissions for the sites. The high N level gave the highest impact at five 123 Nutr Cycl Agroecosyst (2019) 114:139–155 147 of the sites and the lowest N level at the remaining four where the low-yielding site C3 had the second lowest sites. Mean marine eutrophication was 4.4, 4.2 and non-SOC GHG emissions (Figs. 2 and 6). One of the -1 5.0 g N CU for the low, medium and high N level, reasons for this was the magnitude of yield differences eq respectively (Fig. 5), but only the difference between between low- and high-yielding sites (Fig. 2). Another medium and high N levels was statistically significant reason was that the high-yielding central sites (C1 and (Table SM11 in Supplementary Material). In contrast, C2) had relatively high soil N O emissions per differences between site means were highly statisti- hectare, which in turn was mainly due to their low cally significant (Table SM10) and varied between 2.1 sand content. There was no consistent difference in the -1 and 9.1 g N CU . magnitude of GHG emissions between southern and eq The sensitivity analysis showed that altering soil central sites, which indicates that soil characteristics N O emissions by 20% changed estimated non-SOC may be more important for GHG emissions than GHG emissions by 3.0–15% (average 8.3%) for the geographical location, at least within the ranges of N respective treatments. The effect on total GHG fertiliser rates considered here. emissions was slightly smaller (range 1.1–12%, aver- The major contributing factor to the variation in age 6.3%). Changing waterborne field N emissions by non-SOC GHGs was soil N O emissions. Field-level 20% had a smaller effect on both non-SOC GHG N O emissions have previously been shown to be the emissions (range 0.2–1.7%, average 1.3%) and total main contributor to GHG emissions from crop culti- GHG emissions (range 0.3–3.3%, average 0.8%). vation (Goglio et al. 2014), and this was confirmed by However, it had a larger effect on estimated marine our results to a certain extent, although the importance eutrophication, which changed by 6.7–19.6% (average of soil N O emissions differed between treatments 16%) for the respective treatments. Full results of the (26–78% of total non-SOC GHGs; Fig. 3). Differ- sensitivity analysis are available in Table SM12. ences in soil N O emissions were also larger between Land area occupied by crops was inversely propor- sites (up to 900%) than between N levels (up to 90%). tional to the amount of CU produced and therefore However, the most common approach for calculating declined with higher N level. However, the rate of soil N O emissions in LCA is to use default emission decline was smaller between medium and high N factors from IPCC Tier I methodology (De Klein et al. levels than between low and medium N levels (Fig. 3). 2006), which neglect all local factors such as soil Mean land occupation was 3.0, 2.3 and 2.1 m - characteristics and weather and instead assume that -1 year CU for low, medium and high N level, 1% of the applied mineral fertiliser N is emitted as respectively. Mean land occupation for the sites varied direct N O, regardless of other factors. Since site 2 -1 between 1.6 and 3.9 m year CU . characteristics have been shown to be highly impor- tant for N O emissions (Rochette et al. 2018), the IPCC Tier I emission factors have been shown to be a Discussion poor predictor of N O emissions at field scale (Goglio et al. 2018). Instead, we used a model derived from Greenhouse gas emissions meta-analysis of field measurements in Canada, where for example precipitation was identified as an even The estimated GHG emissions differed between sites, more important variable than N rate supplied (Ro- both in terms of differences between N levels and in chette et al. 2018). Despite not being verified on a terms of magnitude. A feature in common for the sites global scale, this model should give better estimates which had the lowest non-SOC GHG emissions at than global emissions factors of field-level N O medium N level (S3, C1 and C2) instead of low N level emissions in cold temperate regions like Sweden. was that they showed a relatively high difference in Applying the field data-based model to our treatments yield between the low and medium N level (Fig. 2). gave emissions factors of 0.4–6.3% (fraction of added This yield difference was thereby sufficiently large to N fertiliser emitted as direct N O), with variations outweigh the increase in emissions per unit area. between sites and N levels. The results highlight the Among the southern sites, high-yielding sites (S1 and importance of using site-dependent emission models S3) had lower GHG emissions than low-yielding sites, when comparing spatially dependent processes at whereas this was not observed for the central sites, different sites, as recommended by others (Peter et al. 123 148 Nutr Cycl Agroecosyst (2019) 114:139–155 2016) but still rarely used for estimating soil N O that there were no statistically significant differences emissions in LCAs. between the N level means when SOC was included All soils, regardless of N level and site, displayed a (Table SM10). These results illustrate the complex likely SOC loss during the study period. We base this interplay between different variables affecting GHG discussion section on the mean values derived in the emissions from cropping systems. regression analysis, but uncertainties in SOC change estimations were high (Fig. 4), which is discussed Marine eutrophication further in ‘‘Uncertainties’’ section. Seven of the sites had their smallest mean SOC loss per CU at the highest The contribution of soil emissions of N and P N level and their highest SOC loss per CU at the lowest dominated the marine eutrophication results (Fig. 5), N level. The decreasing SOC loss per CU at higher N and differences between sites were therefore mainly level was due to slightly lower mean SOC loss per ha explained by differences in estimated field emissions (Fig. SM2) and to higher yield output (Fig. 2) and thus and their characterisation in the LCIA. Field N losses a smaller fraction of per ha SOC loss being allocated to were in turn dominated by the waterborne component each CU produced. Since SOC loss contributed a large ([ 93% of the total impact of field N losses for all fraction of the GHG emissions, these results affected treatments), which means that total estimated marine the conclusions regarding which N level was optimal eutrophication impact is highly sensitive to the choice from a GHG emissions perspective. The lowest total of model for waterborne N emissions at field level. In GHG emissions were achieved at the highest N level LCAs, nitrate leaching is generally estimated using for two of the sites, at the medium N level for five of simple models that neglect spatial characteristics such the sites and at the low N level for two of the sites. The as soil type and climate (Nitschelm et al. 2017), results were thus very divergent and no general despite their significant contribution to leaching conclusions on the optimum N level for minimising magnitude observed here (Table SM8 in Supplemen- GHG emissions can be drawn, especially considering tary Material). Although P losses on average Fig. 4 Estimated annual emissions of CO CU from soil closest to current common practice. The emissions estimates organic carbon (SOC) changes at the four southern Swedish sites were derived through regression analysis of measured SOC (S1–S4) and five central Swedish sites (C1–C5) under three levels in the topsoil. The bars represent the confidence intervals different nitrogen (N) fertiliser intensities, where high N is at 95% confidence level 123 Nutr Cycl Agroecosyst (2019) 114:139–155 149 Fig. 5 Marine eutrophication impact from crop cultivation at the four southern Swedish sites (S1–S4) and five central Swedish sites (C1–C5) under three different nitrogen (N) fertiliser intensities, where high N is closest to current common practice contributed considerably less to the marine eutroph- Correlations and trade-offs between greenhouse ication impact than N, the contribution was 67% of the gas emissions and marine eutrophication marine eutrophication for one of the treatments (low N level at site C2), which in this case was explained by The GHG emissions varied by a factor of five between treatments, and the variation was similar for marine the high clay content in combination with a relatively high characterisation factor. Marine eutrophication eutrophication. However, results for the two impact categories generally exhibited different patterns, both LCIA models tend to focus on N emissions and neglect P emissions (see e.g. Struijs et al. 2013; Cosme and in terms of N level ranking at each site and in terms of Hauschild 2017), since a majority of global marine ranking of site means (Figs. 3, 5 and 6). This indicates recipients are N-limited. However, neglecting P can that although yield responses can have a large underestimate the impact when the recipient is P-lim- influence on the results, as previously shown by e.g. ited or co-limited in P and N, as is the case for some of Brentrup et al. (2004) and Wang et al. (2016), site- our sites (see values for e.g. site C2). Ignoring the specific emission profiles are also highly important. Mean impacts were lowest at the medium N level contribution of P to marine eutrophication in the present study would have identified C2 as a site with for both GHG emissions and marine eutrophication (Fig. 6), but only two individual sites (S3 and C1) had low marine eutrophication contribution due to its clayey soil (Table SM1), giving high P leaching but their lowest impact at the medium N level for both GHGs and marine eutrophication (Figs. 3 and 5). This low N leaching. The results in the present study clearly illustrate that the selection of emissions and LCIA outcome highlights the importance of choosing an models is important for assessment outcomes. appropriate spatial scale with respect to the question to be answered in order to attain relevant LCA results. 123 150 Nutr Cycl Agroecosyst (2019) 114:139–155 Fig. 6 Mean environmental impacts for nitrogen (N) fertiliser levels and sites (arranged by ascending non-soil organic carbon (SOC) greenhouse gas (GHG) emissions). Black dots represent total GHG emissions, blue circles represent non-SOC GHG emissions and orange triangles represent marine eutrophication. (Color figure online) Similarly, previous studies have found that impacts texture was not consistent for all sites, since both N O were smallest at the lowest (Ashworth et al. 2015; emissions and N leaching per unit produced are also Goglio et al. 2012; Brentrup et al. 2004), medium affected by other factors such as climate and yield (Ruan et al. 2016; Wang et al. 2016) or highest response. For example, the increase in both GHG (Charles et al. 2006) N fertiliser rate tested in each emissions and marine eutrophication from medium to study. While the methods, models and assumptions high N level at site S2 and site S4 can be explained by differ between these studies, site-related differences the weak yield response (Fig. 2), which in turn may be may also explain why they reach different an indication that factors other than N fertiliser limited conclusions. growth in those treatment. Despite climate impact gaining most of the atten- tion as the dominant environmental issues of our time, Uncertainties waterborne reactive N losses are estimated to impose greater costs in the EU, through damage to human Data from long-term field trials are a valuable resource health and ecosystems, than the climate impact caused for life cycle assessments, since the availability of by N empirical data decreases the dependence on models O emissions (Sutton et al. 2011). As Fig. 6 illustrates, there were trade-offs between GHG emis- and their associated uncertainties. However, using sions and marine eutrophication at some sites, which measured data also means that variations due to e.g. means that aspects that decreases GHG emissions does crop disease and extreme weather events are embodied not necessarily decrease marine eutrophication. in the results. There are also uncertainties connected to Although the major process contributions to both measuring methods, which could be one reason for the impact categories (soil N O emissions, SOC changes, large uncertainties in estimated SOC change (Fig. 4). waterborne N and P losses) are dependent on the same A similar pattern where field data exhibited larger factors (soil characteristics, climate, fertilisation, variability than modelled data was observed by Goglio yield), they were clearly influenced in different ways et al. (2018). Although the farming practices used in at different sites. For example, sand content is reported the Swedish long-term field trials represent typical to be a reducing factor for N O emissions, while a high farming practices in terms of crop rotation, field soil sand content increases the risk of N leaching operations and amount of inputs, a typical farmer (Rochette et al. 2018; Kyllmar et al. 2006). This effect would adapt e.g. crop choice and annual fertiliser is evident for the sandy soil at site C3, where soil N O amounts to yearly conditions and cumulative experi- emissions were low while N leaching was high ence, instead of leaving cropping system management (Figs. 3 and 5). However, the connection to soil unchanged over more than 60 years. For that reason, 123 Nutr Cycl Agroecosyst (2019) 114:139–155 151 the quantitative results from this study should not be LCA outcomes depends on the availability of emission interpreted as a benchmark for average Swedish crop models that can be applied with existing data. Our cultivation, or as representing the environmental sensitivity analysis showed that estimated GHG impact of an ideal cropping system at each individual emissions were highly sensitive to changes in soil site. On the other hand, consistent management at the N O emissions. Feasible models for estimating N O 2 2 different sites over time enabled differences arising emissions at field scale are scarce, mainly due to from site-dependent characteristics to be emissions being highly variable over time and space distinguished. (Butterbach-Bahl and Dannenmann 2011), which is Despite LCA being an ISO standardised method- why the crude Tier I model presented by IPCC (De ology (ISO 2006), performing an LCA requires Klein et al. 2006) is used in most crop LCAs. The making methodological choices, which could signif- model used to estimate soil N O emissions in this icantly affect the outcomes of the study. One such study was instead derived from a meta-analysis of choice is the functional unit, which is particularly Canadian field trials, and has not been verified for complicated for agricultural systems due to their Swedish conditions. While this is an important source multiple functions and outputs (Brankatschk and of uncertainty in the quantitative results, it should give Finkbeiner 2015; Notarnicola et al. 2017). Cereal unit a more realistic representation of differences in has been applied as the functional unit in at least one emissions between sites than using a site-independent other LCA of crop rotations, is used in some national emission factor. Similarly, marine eutrophication was agricultural statistics and accounts for the most highly sensitive to changes in nitrate leaching, since important nutritional functions of crops (Brankatschk this emission dominated the impacts assessed. The and Finkbeiner 2014; Prechsl et al. 2017). It was model used for estimating nitrate leaching is based on therefore deemed appropriate for the present study. data derived from national reporting, together with Drawbacks of using CU as the functional unit are that correction factors for fertiliser rates over or under the values are based on animal feeding value although optimum level originally developed for a national not all crops are used for animal feed, and that the data farmer’s advisory tool in Sweden. Considering the on livestock species composition are based on German large contribution of soil N O and nitrate emissions to conditions. However, since more Swedish cereals are these two important impact categories, harmonised used for animal feed than for human consumption models that are globally applicable but still account (Eklo¨f 2014), and since livestock species composition for spatial and management variations would improve in Germany and Sweden are similar (FAO 2016), this both the accuracy and inter-comparability of crop compromise was considered acceptable in the present LCAs. study. Charles et al. (2006) included a quality criterion The third largest contributor to the impacts assessed in their LCA of wheat cultivation, which is relevant was SOC changes, which are uncertain both in terms since fertiliser management affects protein content in of methodological choice (see Methods section) and in harvested crops, but difficult to apply when assessing terms of empirical data (Fig. 4). The confidence whole crop rotations, since different characteristics are intervals of both SOC change per ha (Fig. SM2) and valued in different crops. In addition, low-quality resulting CO emissions per CU (Fig. 4) exhibited crops are not always discarded but instead used for considerable overlap, which complicates interpreta- other purposes, e.g. low-protein wheat can be used as tion of the results. Modelling SOC dynamics instead of animal feed or for biofuel production. We therefore using measured values would be an option to achieve chose not to include any crop quality criteria in this more stable results, but would also introduce new study, although it should be noted that the quality uncertainties associated with the chosen model and aspect should be considered when evaluating appro- methodological choices such as temporal system priate fertiliser management in a decision-making boundary and choice of initial SOC level. Measured context. data were therefore chosen as the least biased way to Environmental impacts of agricultural systems are represent results, in accordance with recommenda- dominated by soil emissions, which in turn are site- tions by Goglio et al. (2015), but the large confidence dependent (Notarnicola et al. 2017). Since measured intervals mean that interpretation of observed impacts emission data are rarely available, the reliability of due to SOC changes is uncertain. 123 152 Nutr Cycl Agroecosyst (2019) 114:139–155 Implications and perspectives Stenberg 2014). However, further research is needed to identify factors that can be used to characterise sites While the medium N level had the lowest GHG in terms of their environmental impact profile. The emissions and marine eutrophication impact, it variations in environmental impact between sites in required more land per unit produced than the high this study illustrate how using emissions and impact N level (Figs. 3 and 6). Agricultural land is a limited assessment models operating at a relevant spatial scale resource, so reducing the area of land required for in relation to the research question improves the producing a unit of crops can potentially prevent land possibility of drawing relevant conclusions. use change, or free up space for environmental impact mitigation measures. Examples of these mitigation measures are constructed wetlands to retain nutrients Conclusions (Land et al. 2016) and producing biomass that promotes soil carbon sequestration and replaces fossil This study explored the influence of site and N fuels (Hammar et al. 2014; Prade et al. 2014). Several fertiliser rate on the GHG emissions and marine recent studies have indicated high climate mitigation eutrophication impact from crop cultivation in a life potential of reducing land requirements through cycle perspective. Results from a 20-year assessment agricultural intensification (Balmford et al. 2018; at nine sites and three N fertiliser levels revealed that Searchinger et al. 2018), while others claim that yield site affected the N level that gave the lowest impacts improvements do not necessarily mean that less and the impact level in general, and that results were cropland is actually used (Lambin and Meyfroidt also not consistent between impact categories. This 2011). These discrepancies highlight the importance outcome illustrates that general management plans for of considering the system beyond the field scale in decreasing the environmental impact of crop cultiva- order to make a fair assessment of the environmental tion will have difficulty succeeding without consider- consequences of different management practices. ing site characteristics and potential trade-offs However, the results presented in this study show that between different impacts. the field-level environmental performance response to Overall, the results showed that site influenced different fertiliser intensities is site-dependent, which GHG emissions and marine eutrophication more than is potentially also the case for other proposed inten- N level did, at least for the moderate fertiliser rates sification measures, such as altering tillage practices or studied here. The medium N level, which was lower introducing catch crops (see e.g. Doltra and Olesen than current average rate in the study regions, gave on 2013; Zaher et al. 2013). The highly site-dependent average the lowest total GHG emissions and marine nature of agricultural systems is therefore relevant to eutrophication. However, differences between mean consider when evaluating the mitigation potential of impacts at each N level were small (up to 10% for intensification strategies at larger scales. GHG emissions and 20% for marine eutrophication) Site-dependent effects of management practices and not statistically significant for total GHG emis- and management change have been reported previ- sions, and only significant between medium and high ously (Goglio et al. 2014). The difference in impact N level for marine eutrophication. In contrast to the magnitude and preferred N level for minimising moderate differences observed between N levels, impacts at sites located geographically close to each differences between mean impacts at the different other in this study (Figs. 1, 3 and 5), as well as other sites were large (up to 280% for GHG emissions and patterns in the results, indicate that soil texture was 340% for marine eutrophication) and statistically one of the most important variables. This outcome significant for both impact categories. These results indicates possibilities for decreasing the environmen- show that site-specific information can improve the tal impact by considering soil characteristics when accuracy of assessments of the environmental impact planning e.g. crop rotations and fertiliser strategy, as of crop cultivation and thereby generate better deci- farmers already do to maximise profit. Since soil sion support. texture can vary even within fields, it is also possible Acknowledgements The authors would like to thank Claudia that precision fertilisation could have a significant von Bromssen at the Department of Energy and Technology, effect on the overall environmental impact (Delin and 123 Nutr Cycl Agroecosyst (2019) 114:139–155 153 Swedish University of Agricultural Sciences, for providing Cycle Assess 18:24–36. https://doi.org/10.1007/s11367- guidance on the statistical analyses. Agricultural Sciences and 012-0457-0 Spatial Planning (Formas) [Grant number 229-2013-82]. This Brander M, Sood A, Wylie C, Haughton A, Lovell J (2011) work was funded by the Swedish Research Council for Electricity-specific emission factors for grid electricity. Environment. Econometrica. https://ecometrica.com/assets/Electricity- specific-emissionfactors-for-grid-electricity.pdf. Accessed 29 Oct 2018 Compliance with ethical standards Brankatschk G, Finkbeiner M (2014) Application of the Cereal Unit in a new allocation procedure for agricultural life Conflict of interest The authors declare that they have no cycle assessments. J Clean Prod 73:72–79. https://doi.org/ conflict of interest. 10.1016/j.jclepro.2014.02.005 Brankatschk G, Finkbeiner M (2015) Modeling crop rotation in Open Access This article is distributed under the terms of the agricultural LCAs—challenges and potential solutions. Creative Commons Attribution 4.0 International License (http:// Agric Syst 138:66–76. https://doi.org/10.1016/j.agsy.2015. creativecommons.org/licenses/by/4.0/), which permits unre- 05.008 stricted use, distribution, and reproduction in any medium, Brentrup F, Kusters J, Lammel J, Barraclough P, Kuhlmann H provided you give appropriate credit to the original (2004) Environmental impact assessment of agricultural author(s) and the source, provide a link to the Creative Com- production systems using the life cycle assessment (LCA) mons license, and indicate if changes were made. methodology—II. The application to N fertilizer use in winter wheat production systems. Eur J Agron 20:265–279. https://doi.org/10.1016/s1161-0301(03)00039-x Brentrup F, Hoxha A, Christensen B (2016) Carbon footprint References analysis of mineral fertilizer production in Europe and other world regions. In: Paper presented at the LCA Food 2016, Dublin Ahlgren S, Hansson P-A, Kimming M, Aronsson P, Lundkvist H Butterbach-Bahl K, Dannenmann M (2011) Denitrification and (2009) Greenhouse gas emissions from cultivation of associated soil N O emissions due to agricultural activities agricultural crops for biofuels and production of biogas in a changing climate. Curr Opin Environ Sustain from manure—implementation of the Directive of the 3:389–395. https://doi.org/10.1016/j.cosust.2011.08.004 European Parliament and of the Council on the promotion Caffrey KR, Veal MW (2013) Conducting an agricultural life of the use of energy from renewable sources. In: Revised cycle assessment: challenges and perspectives. Sci World according to instructions for interpretation of the Directive J. https://doi.org/10.1155/2013/472431 from the European Commission Carlgren K, Mattsson L (2001) Swedish soil fertility experi- Aoun WB, El-Akkari M, Gabrielle B, Flenet F (2014) Consid- ments. Acta Agric Scand Sect B Soil Plant Sci 51:49–76. ering the variability of farming practices improves the LCA https://doi.org/10.1080/090647101753483787 of biodiesel from oilseed rape. In: Proceedings of the 9th Charles R, Jolliet O, Gaillard G, Pellet D (2006) Environmental international conference on life cycle assessment in the analysis of intensity level in wheat crop production using agri-food sector. American Center for Life Cycle Assess- life cycle assessment. Agric Ecosyst Environ 113:216–225. ment, San Francisco https://doi.org/10.1016/j.agee.2005.09.014 Aronsson H, Torstensson G (2004) Berakning av olika odling- Cosme N, Hauschild MZ (2017) Characterization of waterborne ˚ ¨ ˚ ¨ satgarders inverkan pa kvaveutlakningen. Beskrivning av nitrogen emissions for marine eutrophication modelling in ett pedagogiskt verktyg fo¨r bera¨kning av kva¨veutlakning life cycle impact assessment at the damage level and global fra˚n enskilda fa¨lt och ga˚rdar [Swedish]. Swedish Univer- scale. Int J Life Cycle Assess 22:1558–1570. https://doi. sity of Agricultural Sciences, Uppsala org/10.1007/s11367-017-1271-5 Arvidsson J, Keller T (2011) Comparing penetrometer and shear De Klein C, Novoa RSA, Ogle S, Smith KA, Rochette P, Wirth vane measurements with measured and predicted mould- TC, McConkey BG, Mosier A, Rypdal K, Walsh M, Wil- board plough draught in a range of Swedish soils. Soil liams SA (2006) Chapter 11: N O emissions from managed Tillage Res 111:219–223. https://doi.org/10.1016/j.still. soils, and CO emissions from lime and urea application. 2010.10.005 In: Gytarsky M, Hiraishi T, Irving W, Krug T, Penman J Ashworth AJ, Taylor AM, Reed DL, Allen FL, Keyser PD, (eds) 2006 IPCC guidelines for national greenhouse gas Tyler DD (2015) Environmental impact assessment of inventories. IPCC International Panel on Climate Change, regional switchgrass feedstock production comparing Geneva, pp 11.1–11.54 nitrogen input scenarios and legume-intercropping sys- Delin S, Stenberg M (2014) Effect of nitrogen fertilization on tems. J Clean Prod 87:227–234. https://doi.org/10.1016/j. nitrate leaching in relation to grain yield response on loamy jclepro.2014.10.002 sand in Sweden. Eur J Agron 52:291–296. https://doi.org/ Balmford A et al (2018) The environmental costs and benefits of 10.1016/j.eja.2013.08.007 high-yield farming. Nat Sustain 1:477–485. https://doi.org/ Delin S, Linde´n B, Berglund K (2005) Yield and protein 10.1038/s41893-018-0138-5 response to fertilizer nitrogen in different parts of a cereal Bessou C, Lehuger S, Gabrielle B, Mary B (2013) Using a crop field: potential of site-specific fertilization. Eur J Agron model to account for the effects of local factors on the LCA 22:325–336. https://doi.org/10.1016/j.eja.2004.05.001 of sugar beet ethanol in Picardy region, France. Int J Life 123 154 Nutr Cycl Agroecosyst (2019) 114:139–155 Diaz RJ, Rosenberg R (2008) Spreading dead zones and con- (LCA): a review. J Clean Prod 104:23–39. https://doi.org/ sequences for marine ecosystems. Science 321:926–929. 10.1016/j.jclepro.2015.05.040 https://doi.org/10.1126/science.1156401 Goglio P et al (2018) A comparison of methods to quantify Doltra J, Olesen JE (2013) The role of catch crops in the eco- greenhouse gas emissions of cropping systems in LCA. logical intensification of spring cereals in organic farming J Clean Prod 172:4010–4017. https://doi.org/10.1016/j. under Nordic climate. Eur J Agron 44:98–108. https://doi. jclepro.2017.03.133 org/10.1016/j.eja.2012.03.006 Hammar T, Ericsson N, Sundberg C, Hansson PA (2014) Cli- Edstro¨m M, Pettersson O, Nilsson L, Ho¨rndahl T (2005) Jord- mate impact of willow grown for bioenergy in Sweden. brukssektorns energianva¨ndning [Swedish]. Swedish Bioenergy Res 7:1529–1540. https://doi.org/10.1007/ Institute of Agricultural and Environmental Engineering s12155-014-9490-0 (JTI), Uppsala Henryson K, Hansson P-A, Sundberg C (2018) Spatially dif- Eklo¨f P (2014) Marknadso¨versikt – Spannma˚l. Jordbruksverket ferentiated midpoint indicator for marine eutrophication of (Swedish Board of Agriculture) waterborne emissions in Sweden. Int J Life Cycle Assess EMEP/EEA (2016) EMEP/EEA air pollutant emission inven- 23:70–81. https://doi.org/10.1007/s11367-017-1298-7 tory guidebook 2016. In: Technical guidance to prepare Huttunen M, Vehvila¨inen B, Huttunen I (2013) Typen, fosforin national emission inventories, vol EEA report no 21/2016. ja kiintoaineksen pida¨ttyminen vesisto¨issa¨–WSFS-Ve- European Environment Agency, Luxembourg mala-mallin arvio. Finnish Environment Institute, Helsinki European Commission (2017) Modernising and simplifying the ISO (2006) ISO 14040: environmental management-life cycle cap. In: Background document: climate and environmental assessment-principles and framework, 2. utg. edn. Inter- challenges facing EU agriculture and rural areas. European national Organization for Standardization (ISO), Geneva Commission. Directorate-General for Agriculture and Johnsson H, Ma˚rtensson K, Lindsjo¨ A, Persson K, Rangel YA, Rural Development, Brussels Blomba¨ck K (2016) La¨ckage av na¨ringsa¨mnen fran svensk FAO (2016) FAOSTAT statistics database: live animals, by akermark. Bera¨kningar av normalla¨ckage av kva¨ve och 1058 country. http://www.fao.org/faostat/en/#data/QA. fosfor for 2013 vol SMED Rapport Nr 189 2016. Sveriges Accessed 22 Feb 2019 Meteorologiska och Hydrologiska Institut, Norrkoping Fertilizers Europe (2018a) Ammonium nitrate (AN, 33.5% N); Kyllmar K, Carlsson C, Gustafson A, Ulen B, Johnsson H from ammonia and nitric acid, including primary produc- (2006) Nutrient discharge from small agricultural catch- tion; production mix, at plant. UUID: c2c4ebba-358f-493e- ments in Sweden characterisation and trends. Agric Eco- 83ba-71bffe847e9a (version 00.00.001). thinkstep, GaBi syst Environ 115:15–26. https://doi.org/10.1016/j.agee. database 2005.12.004 Fertilizers Europe (2018b) Potassium chloride (KCl/MOP, 60% Lambin EF, Meyfroidt P (2011) Global land use change, eco- K O); shaft mining and beneficiation, including primary nomic globalization, and the looming land scarcity. Proc production; production mix, at plant. UUID: a2a8695e- Natl Acad Sci 108:3465. https://doi.org/10.1073/pnas. 968c-4341-922c-a007c8a8c56d (version: 00.00.001). 1100480108 thinkstep, GaBi database Land M, Grane´li W, Grimvall A, Hoffmann CC, Mitsch WJ, Fertilizers Europe (2018c) Triple superphosphate (TSP, 46% Tonderski KS, Verhoeven JTA (2016) How effective are P2O5); rock phosphate acidulation with phosphoric acid, created or restored freshwater wetlands for nitrogen and including primary production; production mix, at plant. phosphorus removal? A systematic review. Environ Evid UUID: 8d0007f0-9ad8-43b0-86e4-ebe6e9f9d0e6 (version 5:9. https://doi.org/10.1186/s13750-016-0060-0 00.00.001). thinkstep, GaBi database Licker R, Johnston M, Foley JA, Barford C, Kucharik CJ, Garcı´a-Ruiz R, Carranza-Gallego G, Aguilera E, De Molina Monfreda C, Ramankutty N (2010) Mind the gap: how do MG, Guzma´n GI (2019) C and N mineralisation of straw of climate and agricultural management explain the ‘yield traditional and modern wheat varieties in soils of con- gap’ of croplands around the world? Glob Ecol Biogeogr trasting fertility. Nutr Cycl Agroecosyst 113:167–179. 19:769–782. https://doi.org/10.1111/j.1466-8238.2010. https://doi.org/10.1007/s10705-019-09973-4 00563.x Gode J, Martinsson F, Hagberg L, Oman A, Ho¨glund J, Palm D Lindgren M, Pettersson O, Hansson P-A, Noren O (2002) (2011) Miljo¨faktaboken 2011. Estimated emission factors Engine load pattern and engine exhaust gas emissions from for fuels, electricity, heat and transport in sweden. off-road vehicles and methods to reduce fuel-consumption Va¨rmeforsk, Stockholm and engine exhaust gas emissions [Swedish]. Swedish Goglio P, Bonari E, Mazzoncini M (2012) LCA of cropping Institute of Agricultural and Environmental Engineering systems with different external input levels for energetic (JTI), Uppsala purposes. Biomass Bioenergy 42:33–42. https://doi.org/10. Myhre G et al (2013) Anthropogenic and natural radiative 1016/j.biombioe.2012.03.021 forcing. In: Stocker TF, Qin D, Plattner G-K, Tignor M, Goglio P, Grant BB, Smith WN, Desjardins RL, Worth DE, Allen SK, Boschung J, Nauels A, Xia Y, Bex V, Midgley Zentner R, Malhi SS (2014) Impact of management PM (eds) Climate change 2013: the physical science basis. strategies on the global warming potential at the cropping Contribution of working group I to the fifth assessment system level. Sci Total Environ 490:921–933. https://doi. report of the intergovernmental panel on climate change. org/10.1016/j.scitotenv.2014.05.070 Cambridge University Press, Cambridge Goglio P, Smith WN, Grant BB, Desjardins RL, McConkey BG, Nemecek T, Schnetzer J (2011a) Data collection of inputs and Campbell CA, Nemecek T (2015) Accounting for soil yields in LCIs of agricultural production systems in carbon changes in agricultural life cycle assessment 123 Nutr Cycl Agroecosyst (2019) 114:139–155 155 Switzerland and other European countries. Data v3.0. and fertilizer management effects. Agric Ecosyst Environ Agroscope, Zu¨rich 133:247–266. https://doi.org/10.1016/j.agee.2009.04.021 Nemecek T, Schnetzer J (2011b) Data collection of inputs and Statistics Sweden (2011) Va¨xtskyddsmedel i jord-och yields in LCIs of agricultural seed production systems and tra¨dga˚rdsbruket 2010. Anva¨ndning i gro¨dor. [Swedish]. seed processing-Data v3.0. Agroscope, Zu¨rich Plant protection products in agriculture and horticulture. Nitschelm L et al (2017) Improving estimates of nitrogen Use in crops. MI 31 SM 1101, Statistiska centralbyra˚n emissions for life cycle assessment of cropping systems at (SCB) the scale of an agricultural territory. Environ Sci Technol. Statistics Sweden (2016) SCB statistikdatabasen: [Elektronic https://doi.org/10.1021/acs.est.6b03865 resource] Statistical database. Statistiska centralbyra˚n, Notarnicola B, Sala S, Anton A, McLaren SJ, Saouter E, Stockholm Sonesson U (2017) The role of life cycle assessment in Struijs J, Beusen A, van Jaarsveld H, Huijbregts MAJ (2013) supporting sustainable agri-food systems: a review of the Eutrophication. In: Goedkoop M, Heijungs R, Huijbregts challenges. J Clean Prod Part 140(Part 2):399–409. https:// M, De Schryver A, Struijs J, van Zelm R (eds) ReCiPe doi.org/10.1016/j.jclepro.2016.06.071 2008. Report I: characterisation. First edition (version Page G, Ridoutt B, Bellotti B (2014) Location and technology 1.08) options to reduce environmental impacts from agriculture. Sutton MA, Oenema O, Erisman JW, Leip A, van Grinsven H, J Clean Prod 81:130–136. https://doi.org/10.1016/j.jclepro. Winiwarter W (2011) Too much of a good thing. Nature 2014.06.055 472:159. https://doi.org/10.1038/472159a Peter C, Fiore A, Hagemann U, Nendel C, Xiloyannis C (2016) Sutton MA et al (2013) Our nutrient world: the challenge to Improving the accounting of field emissions in the carbon produce more food and energy with less pollution. Centre footprint of agricultural products: a comparison of default for Ecology and Hydrology, Edinburgh on behalf of the IPCC methods with readily available medium-effort Global Partnership on Nutrient Management and the modeling approaches. Int J Life Cycle Assess 21:791–805. International Nitrogen Initiative https://doi.org/10.1007/s11367-016-1056-2 Swedish EPA (2006) Eutrophication of Swedish seas: final Potting J, Hauschild M (2006) Spatial differentiation in life report. Report 5509. Swedish Environmental Protection cycle impact assessment: a decade of method development Agency, Stockholm to increase the environmental realism of LCIA. Int J Life Swedish Meteorological and Hydrological Institute (2018) Cycle Assess 11:11–13. https://doi.org/10.1065/lca2006. SMHI Oppna data. Meteorologiska observationer. https:// 04.005 opendata-download-metobs.smhi.se/explore/. Accessed 16 Prade T, Svensson SE, Bjo¨rnsson L (2014) Introduction of May 2018 grass-clover crops as biogas feedstock in cereal-dominated Tilman D, Cassman KG, Matson PA, Naylor R, Polasky S crop rotations. Part I: effects on soil organic carbon and (2002) Agricultural sustainability and intensive production food production. In: Schenck R, Huizenga D (eds) Pro- practices. Nature 418:671–677. https://doi.org/10.1038/ ceedings of the 9th international conference on life cycle nature01014 assessment in the agri-food sector (LCA Food 2014), 8–10 Van Stappen F et al (2017) Sensitive parameters in local agri- October 2014, San Francisco. ACLCA, Vashon, cultural life cycle assessments: the illustrative case of pp 1032–1041 cereal production in Wallonia, Belgium. Int J Life Cycle Prechsl UE, Wittwer R, van der Heijden MGA, Lu¨scher G, Assess. https://doi.org/10.1007/s11367-017-1325-8 Jeanneret P, Nemecek T (2017) Assessing the environ- Wang ZB, Zhang HL, Lu XH, Wang M, Chu QQ, Wen XY, mental impacts of cropping systems and cover crops: life Chen F (2016) Lowering carbon footprint of winter wheat cycle assessment of FAST, a long-term arable farming field by improving management practices in North China Plain. experiment. Agric Syst 157:39–50. https://doi.org/10. J Clean Prod 112:149–157. https://doi.org/10.1016/j. 1016/j.agsy.2017.06.011 jclepro.2015.06.084 Rochette P et al (2018) Soil nitrous oxide emissions from Williams AG, Audsley E, Sandars DL (2010) Environmental agricultural soils in Canada: exploring relationships with burdens of producing bread wheat, oilseed rape and pota- soil, crop and climatic variables. Agric Ecosyst Environ toes in England and Wales using simulation and system 254:69–81. https://doi.org/10.1016/j.agee.2017.10.021 modelling. Int J Life Cycle Assessment 15:855–868. Ruan LL, Bhardwaj AK, Hamilton SK, Robertson GP (2016) https://doi.org/10.1007/s11367-010-0212-3 Nitrogen fertilization challenges the climate benefit of Zaher U, Stockle C, Painter K, Higgins S (2013) Life cycle cellulosic biofuels. Environ Res Lett 11:8. https://doi.org/ assessment of the potential carbon credit from no- and 10.1088/1748-9326/11/6/064007 reduced-tillage winter wheat-based cropping systems in Searchinger TD, Wirsenius S, Beringer T, Dumas P (2018) Eastern Washington State. Agric Syst 122:73–78. https:// Assessing the efficiency of changes in land use for miti- doi.org/10.1016/j.agsy.2013.08.004 gating climate change. Nature 564:249–253. https://doi. org/10.1038/s41586-018-0757-z Publisher’s Note Springer Nature remains neutral with Snyder CS, Bruulsema TW, Jensen TL, Fixen PE (2009) Review regard to jurisdictional claims in published maps and of greenhouse gas emissions from crop production systems institutional affiliations. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Nutrient Cycling in Agroecosystems Springer Journals

Environmental performance of crop cultivation at different sites and nitrogen rates in Sweden

Loading next page...
 
/lp/springer-journals/environmental-performance-of-crop-cultivation-at-different-sites-and-OB1wa0y07P
Publisher
Springer Journals
Copyright
Copyright © 2019 by The Author(s)
Subject
Life Sciences; Agriculture
ISSN
1385-1314
eISSN
1573-0867
DOI
10.1007/s10705-019-09997-w
Publisher site
See Article on Publisher Site

Abstract

Nutr Cycl Agroecosyst (2019) 114:139–155 https://doi.org/10.1007/s10705-019-09997-w(0123456789().,-volV)(0123456789().,-volV) ORIGINAL ARTICLE Environmental performance of crop cultivation at different sites and nitrogen rates in Sweden . . . . Kajsa Henryson Per-Anders Hansson Thomas Ka ¨ tterer Pernilla Tida ˚ ker Cecilia Sundberg Received: 14 November 2018 / Accepted: 29 April 2019 / Published online: 8 May 2019 The Author(s) 2019 Abstract Nitrogen (N) fertilisation has positive and unit. The calculations were based on data from multi- negative effects on the environmental impact of crop site long-term field experiments in Sweden and site- cultivation. The mechanisms governing these effects dependent data and models for non-measured pro- are highly site-dependent, a factor often ignored in cesses. Cultivation at three N levels was evaluated, assessments of the environmental impact of crop where the highest N rate was close to current average cultivation. By assessing outputs of crop rotations practices and the lowest level corresponded to one- using a life cycle approach, this study explored how third of that. Site characteristics had a stronger greenhouse gas emissions and marine eutrophication influence on both greenhouse gas emissions and caused by crop cultivation (including upstream pro- marine eutrophication (variations of up to 330% and cesses such as production of farm inputs) depend on 490%, respectively, within N levels) than N level fertiliser rate and the site at which the cultivation (variations of up to 74% and 59%, respectively, within occurs. Cereal unit (CU) was used as the functional sites). Main sources of variation in greenhouse gas emissions were soil nitrous oxide emissions (58–810 g -1 CO CU ) and soil organic carbon changes 2eq Electronic supplementary material The online version of -1 (14–720 g CO CU ), while variations in marine 2eq this article (https://doi.org/10.1007/s10705-019-09997-w) contains supplementary material, which is available to autho- eutrophication were mainly explained by field-level rized users. -1 waterborne N losses (0.9–8.2 g N CU ). The large eq variation between sites highlights the importance of K. Henryson (&)  P.-A. Hansson  P. Tida˚ker C. Sundberg considering site characteristics when assessing the Department of Energy and Technology, Swedish environmental impact of crop cultivation and evalu- University of Agricultural Sciences (SLU), ating the environmental consequences of crop man- P.O. Box 7032, 750 07 Uppsala, Sweden agement practices. e-mail: kajsa.henryson@slu.se T. Ka¨tterer Keywords Environmental impact  Life cycle Department of Ecology, Swedish University of assessment  Eutrophication  Climate impact Agricultural Sciences (SLU), P.O. Box 7044, Greenhouse gas emissions  Site-dependent 750 07 Uppsala, Sweden C. Sundberg Department of Sustainable Development, Environmental Science and Engineering, KTH Royal Institute of Technology, Teknikringen 10B, 100 44 Stockholm, Sweden 123 140 Nutr Cycl Agroecosyst (2019) 114:139–155 Introduction dependent on site conditions (Goglio et al. 2018). Site can thus influence estimated GHG emissions even The introduction of mineral nitrogen (N) fertilisers has though the impact itself is site-independent. Unlike enabled yield increases and has been a crucial factor climate impact, eutrophication is a site-dependent for increasing global food security during the past impact that mainly occurs due to regional emissions century (Sutton et al. 2013). Increasing crop yields (Potting and Hauschild 2006). Eutrophication is a may also have several positive effects on the environ- particularly urgent environmental issue for Sweden, ment, such as increasing resource use efficiency, which is surrounded by the world’s largest anthro- avoiding land use change and increasing input of crop pogenic hypoxic zone, the Baltic Sea (Diaz and residues available for soil carbon sequestration, but Rosenberg 2008), where agriculture is one of the main may require increased inputs of e.g. water and contributors to this issue (Licker et al. 2010). The site- pesticides (Snyder et al. 2009; Tilman et al. 2002). dependent nature of the eutrophication cause-effect Furthermore, N fertiliser use itself causes negative chain means that it is necessary to consider where environmental impacts through fossil fuel use and emissions occur in order to make meaningful assess- losses of reactive N both during the production phase ments of eutrophication impact, especially when and after field application. It is therefore possible that comparing products produced at different sites (Page there are trade-offs between several environmental et al. 2014). objectives. Due to its multifaceted effects on agroe- A system perspective is necessary to account for the cosystems, N fertiliser rate in relation to yield often environmental impacts of different substance flows emerges as one of the most influential variables for the occurring during the process of producing crops. A magnitude of environmental impact of cropping commonly used tool for determining the environmen- systems (Van Stappen et al. 2017; Aoun et al. 2014; tal impact of a product is life cycle assessment (LCA), Charles et al. 2006). However, the relationship which is frequently used for agricultural products between N fertiliser rate and environmental impacts (Notarnicola et al. 2017). In LCA, emissions and is complex, since both yield and field emissions of resource use during the whole or part of the life cycle reactive N exhibit non-linear responses to increased N of a product or process are compiled and related to fertiliser rate (Delin and Stenberg 2014). Since both environmental impacts (ISO 2006). This approach yield and emissions are also affected by local culti- enables identification of possible trade-offs between vation conditions such as climate and soil character- products or production process alternatives, and istics (Delin et al. 2005; Rochette et al. 2018), it is thereby reveals potential burden shifts between impact possible that the environmental impact of cultivation categories or life-cycle stages. LCA of cropping under different fertiliser rates will depend on the site. systems is typically performed for either a site In that context, multi-site long-term field experiments representing average conditions or for one or a few are a valuable source of data. specific sites, due to lack of data. Different method- Nitrogen fertiliser use is closely related to both ologies and degrees of spatial specificity limit inter- greenhouse gas (GHG) emissions and nutrient losses comparability (Bessou et al. 2013), and it is often to watersheds, which have been identified as key areas unclear whether differences in results between studies for actions to decrease the environmental impact of can be attributed to spatial or methodological varia- agriculture in the EU (European Commission 2017). tions. The aim of this study was therefore to evaluate Previous studies have shown that decreasing fertiliser cradle-to-farm gate GHG emissions and marine input can reduce GHG emissions per unit produced eutrophication of crop cultivation at different N (e.g. Ashworth et al. 2015; Williams et al. 2010; fertiliser levels and different field sites in Sweden, Brentrup et al. 2004; Goglio et al. 2012). However, and to use the results to analyse the influence of these since N fertilisers promote higher yields, this outcome two factors on estimated environmental impact. Data may change if soil organic carbon (SOC) changes are from multi-site long-term field trials were used, included, which are also affected by site conditions together with site-dependent data and models for (Garcıa-Ruiz et al. 2019). Another potential source of non-measured processes. spatial variability in estimated GHG emissions is soil nitrous oxide (N O) emissions, which are highly 123 Nutr Cycl Agroecosyst (2019) 114:139–155 141 Methods Fields receiving no N were not included in the analysis, since this does not reflect realistic agricul- Experimental sites and set-up tural practices, but are included in Fig. 2 to show the whole yield response curve. Average annual N The assessment was based on data from nine sites in fertiliser rates were higher at the southern sites than the Swedish Soil Fertility Trials (described by Carl- at the central sites (Table 1), but the high N level was gren and Mattsson 2001); four in southern (S1–S4) and close to current average cereal N fertilisation rates in five in central Sweden (C1–C5) (Fig. 1). These sites each study region. Applied amounts of phosphorus represent varying cultivation conditions in terms of (P) and potassium (K) fertilisers were chosen to soil characteristics and microclimate (Carlgren and replace annual removal, but the P and K rates were Mattsson 2001). Selected site characteristics are also similar to current average practice in the two shown in Table SM1 in Supplementary Material. regions (Statistics Sweden 2016). A full list of The trials have been ongoing since the 1950s and fertiliser amounts is provided in Tables SM3–SM5. 1960s in southern and central Sweden, respectively. Hereafter, we use the term ‘treatment’ to refer to each The crop rotations included in the present study combination of site and N level (9 9 3=27 consisted of a four-year rotation in the southern region treatments). (Table 1), with two replicates at each site. The crop rotations included were exclusively given mineral Goal, scope, functional unit, system boundaries fertilisers, and all crop residues were left in the field and assumptions and incorporated into the soil. Three N levels were considered in the analysis: low, medium and high. Life cycle assessment was used to assess the GHG emissions and marine eutrophication impact over the whole crop rotation, under three different N fertiliser levels, at the nine sites in southern and central Sweden. The system boundaries were set at cradle-to-farm gate, which included emissions from: • Inputs (fertiliser and pesticide production) • Field operations (fuel consumption, production and maintenance of machinery) • Soil emissions (waterborne N and P compounds and airborne N O, ammonia (NH ) and NO ) 2 3 x • Crop drying • SOC changes. In LCA, a functional unit is needed to relate inputs and outputs to a common unit. The functional unit is a quantified description of the product system functions (ISO 2006), and should be chosen with respect to the goals of the study (Caffrey and Veal 2013). Dry or fresh matter mass of harvested crop, together with area occupied by the crop, are the most common functional units used for agricultural LCAs (Notarnicola et al. 2017). However, mass is not a fair representation of the crop function when different crop types are present, and therefore cereal unit (CU) was chosen as the functional unit in this study instead. Cereal unit was first introduced to LCAs by Brankatschk and Finkbeiner (2014), as a basis for co-product allocation Fig. 1 Map showing location of experimental sites in southern between grain and straw (ibid.) or between the crops in Sweden (sites S1–S4) and central Sweden (sites C1–C5) 123 142 Nutr Cycl Agroecosyst (2019) 114:139–155 Table 1 Crop rotations and Southern sites Central sites fertiliser rates in the (S1–S4) (C1–C5) southern and central a a cropping systems Crop rotation Spring barley Spring barley b c Spring oilseed rape Oats d b Winter wheat Spring oilseed rape e d Sugar beet Winter wheat Oats Hordeum vulgare Winter wheat Brassica napus/rapa Average fertiliser rate at low N level 50 40 Avena sativa Average fertiliser rate at medium N level 100 80 Triticum aestivum Average fertiliser rate at high N level 150 120 Beta vulgaris -1 -1 Fig. 2 Mean annual yields obtained at each site and N level, measured in cereal units (CU) ha (dark grey) and dry matter (DM) ha (light grey) a crop rotation (Brankatschk and Finkbeiner 2015). It combined with data on animal species composition has also been used as a functional unit in LCAs and species-specific metabolism. We used the CU comparing cropping systems (Prechsl et al. 2017). The values presented in the Supplementary Material of CU represents the animal feeding value of each crop Brankatschk and Finkbeiner (2014), where calculation related to a reference crop (barley), and thereby procedures are also presented in more detail. The enables crop yields to be represented by a unit that also relationship between harvested dry mass and CU considers crop function. One CU corresponds to 1 kg harvested from the trial plots during the study period of fresh matter barley (at 14% moisture content). The (1988–2009) is presented in Fig. 2. Subdivision of animal feeding value is calculated based on protein, inputs and emissions among the crops in the rotation lipid, fibre and carbohydrate content of the crop, was not necessary in this study, since the aim was to 123 Nutr Cycl Agroecosyst (2019) 114:139–155 143 assess the environmental impact over the whole crop Fuel consumption L ha rotation. Inventories were instead calculated for each ¼ 8:5 þ 0:32  clay contentðÞ % ð1Þ treatment and year in the study period, and then Emissions from machinery production were calcu- aggregated over all years in the study period and lated from the number of passes (see Table SM6 in divided by the amount of CU produced for each Supplementary Material), data on kg machinery treatment during the study period. Amount of land consumption per pass for each operation and emis- occupied by crop cultivation was also determined by sions data for production of agricultural machinery calculating the field area used per CU in each from the ecoinvent database (Table 2). treatment. Waterborne N emissions at field level (nitrate leaching) were calculated using a model implemented Activity and emissions data for life cycle inventory in the farm management tool VERA, developed by the Swedish Board of Agriculture, adapted to Swedish Most of the farm activity data were obtained from field conditions and mainly based on empirical data. The measurements at the trial sites. They included data on model is described by Aronsson and Torstensson harvested yield, soil carbon content, bulk density and (2004) and calculates N leaching using site- and soil texture. The study period chosen was 1988–2010, management-dependent typical leaching, and with since data were readily accessible and consistent crop correction factors for over- or under-fertilisation in rotations were used for all sites during this period. relation to the yield achieved. Further information on Yields from 4 years in each time series were missing the model structure and parameters used is provided in due to harvest failure, and these years were thus Supplementary Material to this paper. Phosphorus removed from the series at all sites in the two regions, losses at field level were assumed to be equal to the to ensure consistent number of occurrences of each region-, soil- and crop-specific average P losses, using crop in each of the two crop rotations (see Table SM2 data from Johnsson et al. (2016). in Supplementary Material). Data that were not The emission factors for field-level volatilisation available from the field experiments, such as field -1 were set to 0.016 and 0.033 kg NH kg N fertiliser operations, fertiliser types and pesticide amounts, for soils with pH \ 7 and pH [ 7, respectively, and were chosen to represent typical Swedish farming -1 0.04 kg NO kg N fertiliser, after recommendations practices and, as far as possible, region-specific data from EMEP/EEA (2016). Direct soil N O emissions were used. Specific data for all activities included are were calculated using the equation for cumulative presented in this section. N O emissions (Eq. 2) derived by Rochette et al. The chosen emissions inventory for production of (2018), who tested field measurements from 474 ammonium nitrate, triple superphosphate and potas- treatment-years in Canada against different predictors. sium chloride was taken from the GaBi database The equation utilises data on precipitation during the (Fertilizers Europe 2018a, b, c; Brentrup et al. 2016), growing season, applied mineral fertiliser, soil sand which is representative of mineral fertilisers produced content, soil pH and mean annual air temperature, and in Europe. Assumed amounts of pesticide used were is thereby a more site-dependent model than the differentiated based on crop and cultivation region, commonly used generic emission factors, such as Tier using data from Statistics Sweden (2011). Emissions I values from IPCC (De Klein et al. 2006). associated with pesticide production were taken from the ecoinvent database (Table 2). ^ N O ¼ e ð3:91 þ 0:0022  PðtÞþ 0:0069 2 direct Diesel consumption for ploughing was calculated MinNðtÞ 0:0032  sand  0:747  pH using Eq. 1 (Arvidsson and Keller 2011) and diesel þ 0:0097  T ðtÞÞ air consumption values for the other field operations were ð2Þ taken from Lindgren et al. (2002). Assumed numbers -1 of passes for each machine operation were differen- where N O (kg N O–N ha ) is annual cumula- 2 direct 2 tiated according to crop (Table SM6 in Supplementary tive soil N O emissions from N fertiliser application, Material). Energy density and emissions data from P (mm) is the precipitation during May to September, -1 production and use of diesel were taken from Gode MinN N O (kg N ha ) is the mineral N applied, 2 direct et al. (2011). 123 144 Nutr Cycl Agroecosyst (2019) 114:139–155 Table 2 Names of ecoinvent processes used for inventory data in this study. All inventories were taken from ecoinvent version 3.3, allocation: cut-off by classification Ecoinvent process name Used in this study for: Pesticide production, unspecified, RER Pesticide production emissions Tillage, cultivating, chiselling, RoW Machinery required for stubble cultivation Tillage, ploughing, RoW Machinery required for ploughing Tillage, harrowing, by spring tine harrow, RoW Machinery required for harrowing Sowing, RoW Machinery required for sowing Fertilising, by broadcaster, RoW Machinery required for applying fertiliser Application of plant protection product, by field sprayer, RoW Machinery required for applying pesticides Combine harvesting, RoW Machinery required for threshing Harvesting, by complete harvester, beets, RoW Machinery required for harvest/baling of sugar beet Agricultural machinery production, unspecified, RoW Production emissions per unit machinery -1 sand (g kg ) is the sand content of the soil, pH is the in the topsoil were then estimated from the slope and soil pH, T (C) is the mean annual air temperature site-specific bulk density values. The choice to include air and t is the year. Data for P and T were obtained a longer time period for the SOC change estimation air from the Swedish Meteorological and Hydrological was made to avoid unfair allocation between years in Institute (2018) by choosing the weather station the study period and because there were too few closest to each site with available data for all years measurements during the study period. This should be in the study period. Indirect soil N O emissions were considered when interpreting calculated SOC calculated by applying IPCC default emission factors changes, and for this reason we report and discuss to the calculated volatilised and leached N (De Klein both total estimated GHG emissions and GHG emis- et al. 2006). sions from non-SOC processes. The cereals in the rotations were assumed to be Seed production was included by subtracting cor- harvested at 17% moisture content at sites S1–S4 and responding seed rates from yield. Assumed seed rate -1 19% moisture content at sites C1–C5, and dried to for barley was 170 kg ha , for wheat and oats -1 -1 14% moisture content. Oilseed rape was assumed to be 210 kg ha and for oilseed rape 6 kg ha (Ahlgren dried from 12 to 9% moisture content, while no drying et al. 2009). It was assumed that 60 kg sugar beet were or other post-harvest treatment of sugar beet was required to produce the 2.1 kg seeds needed per ha included, since this typically occurs after farm-gate. (Nemecek and Schnetzer 2011a, b). The energy consumption during crop drying was Liming rates in the experiments were very low (on -1 -1 assumed to be 0.15 L of heating oil per kg of water average 0.1–0.15 kg CaO ha year ) during the removed and 19 kWh of electricity per ton grain study period and it was therefore not considered (Edstrom et al. 2005). Life cycle emissions data for relevant to include impacts of liming. heating oil were taken from (Gode et al. 2011) and data on the Swedish electricity mix from (Brander et al. Life cycle impact assessment 2011). Organic carbon concentration in the topsoil The GHGs carbon dioxide (CO ), methane (CH ) and 2 4 (0–20 cm) was measured 6–10 times in every treat- N O were included in the assessment, and their climate ment during the whole trial lifetime (1956–2011), and impacts were calculated using the IPCC characterisa- we opted to use these data to derive mean SOC tion factors (including climate-carbon feedbacks) for changes. This was done by estimating the slope for GWP reported by Myhre et al. (2013). SOC concentration measurements over time for each The marine eutrophication impact of waterborne N treatment using linear regression analysis (see regres- and P losses at field level was calculated using the sion plots in Fig. SM1). Annual carbon stock changes method described by Henryson et al. (2018). This 123 Nutr Cycl Agroecosyst (2019) 114:139–155 145 method uses site-dependent characterisation factors, the mean impact, and that the mean impacts tested for which express the impact in N equivalents (N ) and in the ANOVA analysis were consistent with the mean eq account for retention of nutrients along their way to impacts according to Eq. 3. Since the normalisation marine environments, as well as the limiting nutrient generated two sets of normalised impacts, the effects in the marine recipient, while most other marine of N level and site were analysed by separate ANOVA eutrophication life cycle impact assessment (LCIA) analyses. The significance level was set to 0.05. methods only account for N additions. In contrast to Impact  Yield i;t t Normalised impact ¼ ð4Þ the majority of marine recipients, the sub-basins of the i;t;n Yield kðt;nÞ Baltic Sea surrounding Sweden can be growth limited by either N or P, or both (Swedish EPA 2006). Impacts where i is the impact category, t is the treatment (any of other N emissions were calculated using the combination of N level and site), n is the normalisation ReCiPe2008 LCIA method (Struijs et al. 2013). variable (i.e. site when the effect of site was analysed Waterborne P emissions during fertiliser production and N level when the effect of N level was analysed) were assumed to occur at Yara’s Siilinja¨rvi phosphate and k(t, n) is the treatment site or N level. The results mine in Eastern Finland, which is Europe’s only from the ANOVA analysis were followed up by phosphate mine and thereby the only source from Tukey’s HSD (Honestly Significant Difference) test which P-containing production effluent could end up for N level. in the Baltic Sea. This assumption therefore represents In addition to the ANOVA, a simple sensitivity the worst case in terms of the marine eutrophication analysis was performed for direct soil N O emissions impact of P emissions during fertiliser production. and waterborne soil N emissions, since these were Retention of 98% was assumed for P-containing modelled and expected to contribute substantially to effluent from Siilinja¨rvi (Huttunen et al. 2013). All the estimated impacts. The sensitivity analysis was other P emissions were assumed to be emitted at performed by changing the modelled direct soil N O locations where they would not reach a P-limited emissions and waterborne N emissions one at a time marine environment and were therefore not included. by 20%. Statistical analyses was performed using the soft- Statistical analyses ware R version 3.5.1. All results from statistical tests are provided in the Supplementary Material The mean impacts for each N level and site were (Tables SM9–SM12). calculated as the mean impacts per CU produced, which involved accounting for the fact that each treatment produced different outputs. The approach Results described in Eq. 3 was therefore used to calculate mean impacts: The results revealed large variations between sites and N levels in both impact categories. Estimated ðImpact  Yield Þ i;k;a k;a a¼1 Mean impact ¼ P ð3Þ total GHG emissions varied from 260 to 1280 g i;k Yield -1 k;a a¼1 CO CU between all treatments, and from 200 to 2eq -1 1040 g CO CU when SOC changes were 2eq where k is a site or N level, i is impact category and A is excluded (Fig. 3). Maximum difference in non-SOC all sites when k is an N level and all N levels when k is GHG emissions between sites within the N levels was a site. 170, 250 and 330% for low, medium and high N, One-way ANOVA analysis with a randomised respectively, while the maximum difference between block design was used to investigate whether N level N levels within sites was 13–74%. Thus, variation or site affected the mean for each impact category. The between sites was larger than variation between N impacts for each treatment were normalised for yield levels (Fig. 3). The highest N level gave the highest according to Eq. 4 before performing the ANOVA non-SOC GHG emissions for all sites, but the order for analysis, to keep consistency with calculation of the low and medium N levels varied between sites. SOC means. This procedure ensured that treatments were losses generally contributed to large GHG emissions fairly weighted according to how they contributed to 123 146 Nutr Cycl Agroecosyst (2019) 114:139–155 Fig. 3 Greenhouse gas (GHG) emissions and land area different nitrogen (N) fertiliser intensities, where high N is occupied by crop cultivation at the four southern Swedish sites closest to current common practice (S1–S4) and five central Swedish sites (C1–C5) under three -1 (2.6–65% of total GHG emissions) compared with sites varied between 290 and 1090 g CO CU for 2eq -1 other emission sources, and including SOC changes total GHG emissions, or 220 and 800 g CO CU 2eq therefore influenced which N level minimised GHG when SOC changes were excluded (Fig. 3). In contrast emissions. Unlike the combined impact of other to the differences between N level means, the differ- emission sources, the contribution of SOC changes ences between site means were highly statistically generally decreased with higher N level. The com- significant for both total GHG emissions and non-SOC bined effect of SOC changes and other emission GHG emissions (Table SM10). sources was that the optimum N level in terms of Marine eutrophication varied from 2.0 to 9.9 g -1 minimising total GHG emissions was low, medium or N CU (Fig. 5), with a maximum difference of eq high depending on site. 300%, 330% and 390% between sites at the low, Mean total GHG emissions were 620, 560 medium and high N levels, respectively, and a -1 and 600 g CO CU for the low, medium and maximum difference of 7–59% between N levels at 2eq high N level, respectively, or 380, 400 and 480 g different sites. Thus, similarly to the results for GHG -1 CO CU , respectively, when SOC changes were emissions, differences between N levels were gener- 2eq excluded (Fig. 3). However, differences between N ally smaller than differences between sites. The level means were not statistically significant for total medium N level gave the lowest marine eutrophication GHG emissions, and differences between low and at all central sites (C1–C5) and one of the southern medium N level means were not statistically signif- sites (S3), while the low N level gave the lowest icant when SOC changes were excluded marine eutrophication at the remaining three southern (Tables SM10–SM11). Mean GHG emissions for the sites. The high N level gave the highest impact at five 123 Nutr Cycl Agroecosyst (2019) 114:139–155 147 of the sites and the lowest N level at the remaining four where the low-yielding site C3 had the second lowest sites. Mean marine eutrophication was 4.4, 4.2 and non-SOC GHG emissions (Figs. 2 and 6). One of the -1 5.0 g N CU for the low, medium and high N level, reasons for this was the magnitude of yield differences eq respectively (Fig. 5), but only the difference between between low- and high-yielding sites (Fig. 2). Another medium and high N levels was statistically significant reason was that the high-yielding central sites (C1 and (Table SM11 in Supplementary Material). In contrast, C2) had relatively high soil N O emissions per differences between site means were highly statisti- hectare, which in turn was mainly due to their low cally significant (Table SM10) and varied between 2.1 sand content. There was no consistent difference in the -1 and 9.1 g N CU . magnitude of GHG emissions between southern and eq The sensitivity analysis showed that altering soil central sites, which indicates that soil characteristics N O emissions by 20% changed estimated non-SOC may be more important for GHG emissions than GHG emissions by 3.0–15% (average 8.3%) for the geographical location, at least within the ranges of N respective treatments. The effect on total GHG fertiliser rates considered here. emissions was slightly smaller (range 1.1–12%, aver- The major contributing factor to the variation in age 6.3%). Changing waterborne field N emissions by non-SOC GHGs was soil N O emissions. Field-level 20% had a smaller effect on both non-SOC GHG N O emissions have previously been shown to be the emissions (range 0.2–1.7%, average 1.3%) and total main contributor to GHG emissions from crop culti- GHG emissions (range 0.3–3.3%, average 0.8%). vation (Goglio et al. 2014), and this was confirmed by However, it had a larger effect on estimated marine our results to a certain extent, although the importance eutrophication, which changed by 6.7–19.6% (average of soil N O emissions differed between treatments 16%) for the respective treatments. Full results of the (26–78% of total non-SOC GHGs; Fig. 3). Differ- sensitivity analysis are available in Table SM12. ences in soil N O emissions were also larger between Land area occupied by crops was inversely propor- sites (up to 900%) than between N levels (up to 90%). tional to the amount of CU produced and therefore However, the most common approach for calculating declined with higher N level. However, the rate of soil N O emissions in LCA is to use default emission decline was smaller between medium and high N factors from IPCC Tier I methodology (De Klein et al. levels than between low and medium N levels (Fig. 3). 2006), which neglect all local factors such as soil Mean land occupation was 3.0, 2.3 and 2.1 m - characteristics and weather and instead assume that -1 year CU for low, medium and high N level, 1% of the applied mineral fertiliser N is emitted as respectively. Mean land occupation for the sites varied direct N O, regardless of other factors. Since site 2 -1 between 1.6 and 3.9 m year CU . characteristics have been shown to be highly impor- tant for N O emissions (Rochette et al. 2018), the IPCC Tier I emission factors have been shown to be a Discussion poor predictor of N O emissions at field scale (Goglio et al. 2018). Instead, we used a model derived from Greenhouse gas emissions meta-analysis of field measurements in Canada, where for example precipitation was identified as an even The estimated GHG emissions differed between sites, more important variable than N rate supplied (Ro- both in terms of differences between N levels and in chette et al. 2018). Despite not being verified on a terms of magnitude. A feature in common for the sites global scale, this model should give better estimates which had the lowest non-SOC GHG emissions at than global emissions factors of field-level N O medium N level (S3, C1 and C2) instead of low N level emissions in cold temperate regions like Sweden. was that they showed a relatively high difference in Applying the field data-based model to our treatments yield between the low and medium N level (Fig. 2). gave emissions factors of 0.4–6.3% (fraction of added This yield difference was thereby sufficiently large to N fertiliser emitted as direct N O), with variations outweigh the increase in emissions per unit area. between sites and N levels. The results highlight the Among the southern sites, high-yielding sites (S1 and importance of using site-dependent emission models S3) had lower GHG emissions than low-yielding sites, when comparing spatially dependent processes at whereas this was not observed for the central sites, different sites, as recommended by others (Peter et al. 123 148 Nutr Cycl Agroecosyst (2019) 114:139–155 2016) but still rarely used for estimating soil N O that there were no statistically significant differences emissions in LCAs. between the N level means when SOC was included All soils, regardless of N level and site, displayed a (Table SM10). These results illustrate the complex likely SOC loss during the study period. We base this interplay between different variables affecting GHG discussion section on the mean values derived in the emissions from cropping systems. regression analysis, but uncertainties in SOC change estimations were high (Fig. 4), which is discussed Marine eutrophication further in ‘‘Uncertainties’’ section. Seven of the sites had their smallest mean SOC loss per CU at the highest The contribution of soil emissions of N and P N level and their highest SOC loss per CU at the lowest dominated the marine eutrophication results (Fig. 5), N level. The decreasing SOC loss per CU at higher N and differences between sites were therefore mainly level was due to slightly lower mean SOC loss per ha explained by differences in estimated field emissions (Fig. SM2) and to higher yield output (Fig. 2) and thus and their characterisation in the LCIA. Field N losses a smaller fraction of per ha SOC loss being allocated to were in turn dominated by the waterborne component each CU produced. Since SOC loss contributed a large ([ 93% of the total impact of field N losses for all fraction of the GHG emissions, these results affected treatments), which means that total estimated marine the conclusions regarding which N level was optimal eutrophication impact is highly sensitive to the choice from a GHG emissions perspective. The lowest total of model for waterborne N emissions at field level. In GHG emissions were achieved at the highest N level LCAs, nitrate leaching is generally estimated using for two of the sites, at the medium N level for five of simple models that neglect spatial characteristics such the sites and at the low N level for two of the sites. The as soil type and climate (Nitschelm et al. 2017), results were thus very divergent and no general despite their significant contribution to leaching conclusions on the optimum N level for minimising magnitude observed here (Table SM8 in Supplemen- GHG emissions can be drawn, especially considering tary Material). Although P losses on average Fig. 4 Estimated annual emissions of CO CU from soil closest to current common practice. The emissions estimates organic carbon (SOC) changes at the four southern Swedish sites were derived through regression analysis of measured SOC (S1–S4) and five central Swedish sites (C1–C5) under three levels in the topsoil. The bars represent the confidence intervals different nitrogen (N) fertiliser intensities, where high N is at 95% confidence level 123 Nutr Cycl Agroecosyst (2019) 114:139–155 149 Fig. 5 Marine eutrophication impact from crop cultivation at the four southern Swedish sites (S1–S4) and five central Swedish sites (C1–C5) under three different nitrogen (N) fertiliser intensities, where high N is closest to current common practice contributed considerably less to the marine eutroph- Correlations and trade-offs between greenhouse ication impact than N, the contribution was 67% of the gas emissions and marine eutrophication marine eutrophication for one of the treatments (low N level at site C2), which in this case was explained by The GHG emissions varied by a factor of five between treatments, and the variation was similar for marine the high clay content in combination with a relatively high characterisation factor. Marine eutrophication eutrophication. However, results for the two impact categories generally exhibited different patterns, both LCIA models tend to focus on N emissions and neglect P emissions (see e.g. Struijs et al. 2013; Cosme and in terms of N level ranking at each site and in terms of Hauschild 2017), since a majority of global marine ranking of site means (Figs. 3, 5 and 6). This indicates recipients are N-limited. However, neglecting P can that although yield responses can have a large underestimate the impact when the recipient is P-lim- influence on the results, as previously shown by e.g. ited or co-limited in P and N, as is the case for some of Brentrup et al. (2004) and Wang et al. (2016), site- our sites (see values for e.g. site C2). Ignoring the specific emission profiles are also highly important. Mean impacts were lowest at the medium N level contribution of P to marine eutrophication in the present study would have identified C2 as a site with for both GHG emissions and marine eutrophication (Fig. 6), but only two individual sites (S3 and C1) had low marine eutrophication contribution due to its clayey soil (Table SM1), giving high P leaching but their lowest impact at the medium N level for both GHGs and marine eutrophication (Figs. 3 and 5). This low N leaching. The results in the present study clearly illustrate that the selection of emissions and LCIA outcome highlights the importance of choosing an models is important for assessment outcomes. appropriate spatial scale with respect to the question to be answered in order to attain relevant LCA results. 123 150 Nutr Cycl Agroecosyst (2019) 114:139–155 Fig. 6 Mean environmental impacts for nitrogen (N) fertiliser levels and sites (arranged by ascending non-soil organic carbon (SOC) greenhouse gas (GHG) emissions). Black dots represent total GHG emissions, blue circles represent non-SOC GHG emissions and orange triangles represent marine eutrophication. (Color figure online) Similarly, previous studies have found that impacts texture was not consistent for all sites, since both N O were smallest at the lowest (Ashworth et al. 2015; emissions and N leaching per unit produced are also Goglio et al. 2012; Brentrup et al. 2004), medium affected by other factors such as climate and yield (Ruan et al. 2016; Wang et al. 2016) or highest response. For example, the increase in both GHG (Charles et al. 2006) N fertiliser rate tested in each emissions and marine eutrophication from medium to study. While the methods, models and assumptions high N level at site S2 and site S4 can be explained by differ between these studies, site-related differences the weak yield response (Fig. 2), which in turn may be may also explain why they reach different an indication that factors other than N fertiliser limited conclusions. growth in those treatment. Despite climate impact gaining most of the atten- tion as the dominant environmental issues of our time, Uncertainties waterborne reactive N losses are estimated to impose greater costs in the EU, through damage to human Data from long-term field trials are a valuable resource health and ecosystems, than the climate impact caused for life cycle assessments, since the availability of by N empirical data decreases the dependence on models O emissions (Sutton et al. 2011). As Fig. 6 illustrates, there were trade-offs between GHG emis- and their associated uncertainties. However, using sions and marine eutrophication at some sites, which measured data also means that variations due to e.g. means that aspects that decreases GHG emissions does crop disease and extreme weather events are embodied not necessarily decrease marine eutrophication. in the results. There are also uncertainties connected to Although the major process contributions to both measuring methods, which could be one reason for the impact categories (soil N O emissions, SOC changes, large uncertainties in estimated SOC change (Fig. 4). waterborne N and P losses) are dependent on the same A similar pattern where field data exhibited larger factors (soil characteristics, climate, fertilisation, variability than modelled data was observed by Goglio yield), they were clearly influenced in different ways et al. (2018). Although the farming practices used in at different sites. For example, sand content is reported the Swedish long-term field trials represent typical to be a reducing factor for N O emissions, while a high farming practices in terms of crop rotation, field soil sand content increases the risk of N leaching operations and amount of inputs, a typical farmer (Rochette et al. 2018; Kyllmar et al. 2006). This effect would adapt e.g. crop choice and annual fertiliser is evident for the sandy soil at site C3, where soil N O amounts to yearly conditions and cumulative experi- emissions were low while N leaching was high ence, instead of leaving cropping system management (Figs. 3 and 5). However, the connection to soil unchanged over more than 60 years. For that reason, 123 Nutr Cycl Agroecosyst (2019) 114:139–155 151 the quantitative results from this study should not be LCA outcomes depends on the availability of emission interpreted as a benchmark for average Swedish crop models that can be applied with existing data. Our cultivation, or as representing the environmental sensitivity analysis showed that estimated GHG impact of an ideal cropping system at each individual emissions were highly sensitive to changes in soil site. On the other hand, consistent management at the N O emissions. Feasible models for estimating N O 2 2 different sites over time enabled differences arising emissions at field scale are scarce, mainly due to from site-dependent characteristics to be emissions being highly variable over time and space distinguished. (Butterbach-Bahl and Dannenmann 2011), which is Despite LCA being an ISO standardised method- why the crude Tier I model presented by IPCC (De ology (ISO 2006), performing an LCA requires Klein et al. 2006) is used in most crop LCAs. The making methodological choices, which could signif- model used to estimate soil N O emissions in this icantly affect the outcomes of the study. One such study was instead derived from a meta-analysis of choice is the functional unit, which is particularly Canadian field trials, and has not been verified for complicated for agricultural systems due to their Swedish conditions. While this is an important source multiple functions and outputs (Brankatschk and of uncertainty in the quantitative results, it should give Finkbeiner 2015; Notarnicola et al. 2017). Cereal unit a more realistic representation of differences in has been applied as the functional unit in at least one emissions between sites than using a site-independent other LCA of crop rotations, is used in some national emission factor. Similarly, marine eutrophication was agricultural statistics and accounts for the most highly sensitive to changes in nitrate leaching, since important nutritional functions of crops (Brankatschk this emission dominated the impacts assessed. The and Finkbeiner 2014; Prechsl et al. 2017). It was model used for estimating nitrate leaching is based on therefore deemed appropriate for the present study. data derived from national reporting, together with Drawbacks of using CU as the functional unit are that correction factors for fertiliser rates over or under the values are based on animal feeding value although optimum level originally developed for a national not all crops are used for animal feed, and that the data farmer’s advisory tool in Sweden. Considering the on livestock species composition are based on German large contribution of soil N O and nitrate emissions to conditions. However, since more Swedish cereals are these two important impact categories, harmonised used for animal feed than for human consumption models that are globally applicable but still account (Eklo¨f 2014), and since livestock species composition for spatial and management variations would improve in Germany and Sweden are similar (FAO 2016), this both the accuracy and inter-comparability of crop compromise was considered acceptable in the present LCAs. study. Charles et al. (2006) included a quality criterion The third largest contributor to the impacts assessed in their LCA of wheat cultivation, which is relevant was SOC changes, which are uncertain both in terms since fertiliser management affects protein content in of methodological choice (see Methods section) and in harvested crops, but difficult to apply when assessing terms of empirical data (Fig. 4). The confidence whole crop rotations, since different characteristics are intervals of both SOC change per ha (Fig. SM2) and valued in different crops. In addition, low-quality resulting CO emissions per CU (Fig. 4) exhibited crops are not always discarded but instead used for considerable overlap, which complicates interpreta- other purposes, e.g. low-protein wheat can be used as tion of the results. Modelling SOC dynamics instead of animal feed or for biofuel production. We therefore using measured values would be an option to achieve chose not to include any crop quality criteria in this more stable results, but would also introduce new study, although it should be noted that the quality uncertainties associated with the chosen model and aspect should be considered when evaluating appro- methodological choices such as temporal system priate fertiliser management in a decision-making boundary and choice of initial SOC level. Measured context. data were therefore chosen as the least biased way to Environmental impacts of agricultural systems are represent results, in accordance with recommenda- dominated by soil emissions, which in turn are site- tions by Goglio et al. (2015), but the large confidence dependent (Notarnicola et al. 2017). Since measured intervals mean that interpretation of observed impacts emission data are rarely available, the reliability of due to SOC changes is uncertain. 123 152 Nutr Cycl Agroecosyst (2019) 114:139–155 Implications and perspectives Stenberg 2014). However, further research is needed to identify factors that can be used to characterise sites While the medium N level had the lowest GHG in terms of their environmental impact profile. The emissions and marine eutrophication impact, it variations in environmental impact between sites in required more land per unit produced than the high this study illustrate how using emissions and impact N level (Figs. 3 and 6). Agricultural land is a limited assessment models operating at a relevant spatial scale resource, so reducing the area of land required for in relation to the research question improves the producing a unit of crops can potentially prevent land possibility of drawing relevant conclusions. use change, or free up space for environmental impact mitigation measures. Examples of these mitigation measures are constructed wetlands to retain nutrients Conclusions (Land et al. 2016) and producing biomass that promotes soil carbon sequestration and replaces fossil This study explored the influence of site and N fuels (Hammar et al. 2014; Prade et al. 2014). Several fertiliser rate on the GHG emissions and marine recent studies have indicated high climate mitigation eutrophication impact from crop cultivation in a life potential of reducing land requirements through cycle perspective. Results from a 20-year assessment agricultural intensification (Balmford et al. 2018; at nine sites and three N fertiliser levels revealed that Searchinger et al. 2018), while others claim that yield site affected the N level that gave the lowest impacts improvements do not necessarily mean that less and the impact level in general, and that results were cropland is actually used (Lambin and Meyfroidt also not consistent between impact categories. This 2011). These discrepancies highlight the importance outcome illustrates that general management plans for of considering the system beyond the field scale in decreasing the environmental impact of crop cultiva- order to make a fair assessment of the environmental tion will have difficulty succeeding without consider- consequences of different management practices. ing site characteristics and potential trade-offs However, the results presented in this study show that between different impacts. the field-level environmental performance response to Overall, the results showed that site influenced different fertiliser intensities is site-dependent, which GHG emissions and marine eutrophication more than is potentially also the case for other proposed inten- N level did, at least for the moderate fertiliser rates sification measures, such as altering tillage practices or studied here. The medium N level, which was lower introducing catch crops (see e.g. Doltra and Olesen than current average rate in the study regions, gave on 2013; Zaher et al. 2013). The highly site-dependent average the lowest total GHG emissions and marine nature of agricultural systems is therefore relevant to eutrophication. However, differences between mean consider when evaluating the mitigation potential of impacts at each N level were small (up to 10% for intensification strategies at larger scales. GHG emissions and 20% for marine eutrophication) Site-dependent effects of management practices and not statistically significant for total GHG emis- and management change have been reported previ- sions, and only significant between medium and high ously (Goglio et al. 2014). The difference in impact N level for marine eutrophication. In contrast to the magnitude and preferred N level for minimising moderate differences observed between N levels, impacts at sites located geographically close to each differences between mean impacts at the different other in this study (Figs. 1, 3 and 5), as well as other sites were large (up to 280% for GHG emissions and patterns in the results, indicate that soil texture was 340% for marine eutrophication) and statistically one of the most important variables. This outcome significant for both impact categories. These results indicates possibilities for decreasing the environmen- show that site-specific information can improve the tal impact by considering soil characteristics when accuracy of assessments of the environmental impact planning e.g. crop rotations and fertiliser strategy, as of crop cultivation and thereby generate better deci- farmers already do to maximise profit. Since soil sion support. texture can vary even within fields, it is also possible Acknowledgements The authors would like to thank Claudia that precision fertilisation could have a significant von Bromssen at the Department of Energy and Technology, effect on the overall environmental impact (Delin and 123 Nutr Cycl Agroecosyst (2019) 114:139–155 153 Swedish University of Agricultural Sciences, for providing Cycle Assess 18:24–36. https://doi.org/10.1007/s11367- guidance on the statistical analyses. Agricultural Sciences and 012-0457-0 Spatial Planning (Formas) [Grant number 229-2013-82]. This Brander M, Sood A, Wylie C, Haughton A, Lovell J (2011) work was funded by the Swedish Research Council for Electricity-specific emission factors for grid electricity. Environment. Econometrica. https://ecometrica.com/assets/Electricity- specific-emissionfactors-for-grid-electricity.pdf. Accessed 29 Oct 2018 Compliance with ethical standards Brankatschk G, Finkbeiner M (2014) Application of the Cereal Unit in a new allocation procedure for agricultural life Conflict of interest The authors declare that they have no cycle assessments. J Clean Prod 73:72–79. https://doi.org/ conflict of interest. 10.1016/j.jclepro.2014.02.005 Brankatschk G, Finkbeiner M (2015) Modeling crop rotation in Open Access This article is distributed under the terms of the agricultural LCAs—challenges and potential solutions. Creative Commons Attribution 4.0 International License (http:// Agric Syst 138:66–76. https://doi.org/10.1016/j.agsy.2015. creativecommons.org/licenses/by/4.0/), which permits unre- 05.008 stricted use, distribution, and reproduction in any medium, Brentrup F, Kusters J, Lammel J, Barraclough P, Kuhlmann H provided you give appropriate credit to the original (2004) Environmental impact assessment of agricultural author(s) and the source, provide a link to the Creative Com- production systems using the life cycle assessment (LCA) mons license, and indicate if changes were made. methodology—II. The application to N fertilizer use in winter wheat production systems. Eur J Agron 20:265–279. https://doi.org/10.1016/s1161-0301(03)00039-x Brentrup F, Hoxha A, Christensen B (2016) Carbon footprint References analysis of mineral fertilizer production in Europe and other world regions. In: Paper presented at the LCA Food 2016, Dublin Ahlgren S, Hansson P-A, Kimming M, Aronsson P, Lundkvist H Butterbach-Bahl K, Dannenmann M (2011) Denitrification and (2009) Greenhouse gas emissions from cultivation of associated soil N O emissions due to agricultural activities agricultural crops for biofuels and production of biogas in a changing climate. Curr Opin Environ Sustain from manure—implementation of the Directive of the 3:389–395. https://doi.org/10.1016/j.cosust.2011.08.004 European Parliament and of the Council on the promotion Caffrey KR, Veal MW (2013) Conducting an agricultural life of the use of energy from renewable sources. In: Revised cycle assessment: challenges and perspectives. Sci World according to instructions for interpretation of the Directive J. https://doi.org/10.1155/2013/472431 from the European Commission Carlgren K, Mattsson L (2001) Swedish soil fertility experi- Aoun WB, El-Akkari M, Gabrielle B, Flenet F (2014) Consid- ments. Acta Agric Scand Sect B Soil Plant Sci 51:49–76. ering the variability of farming practices improves the LCA https://doi.org/10.1080/090647101753483787 of biodiesel from oilseed rape. In: Proceedings of the 9th Charles R, Jolliet O, Gaillard G, Pellet D (2006) Environmental international conference on life cycle assessment in the analysis of intensity level in wheat crop production using agri-food sector. American Center for Life Cycle Assess- life cycle assessment. Agric Ecosyst Environ 113:216–225. ment, San Francisco https://doi.org/10.1016/j.agee.2005.09.014 Aronsson H, Torstensson G (2004) Berakning av olika odling- Cosme N, Hauschild MZ (2017) Characterization of waterborne ˚ ¨ ˚ ¨ satgarders inverkan pa kvaveutlakningen. Beskrivning av nitrogen emissions for marine eutrophication modelling in ett pedagogiskt verktyg fo¨r bera¨kning av kva¨veutlakning life cycle impact assessment at the damage level and global fra˚n enskilda fa¨lt och ga˚rdar [Swedish]. Swedish Univer- scale. Int J Life Cycle Assess 22:1558–1570. https://doi. sity of Agricultural Sciences, Uppsala org/10.1007/s11367-017-1271-5 Arvidsson J, Keller T (2011) Comparing penetrometer and shear De Klein C, Novoa RSA, Ogle S, Smith KA, Rochette P, Wirth vane measurements with measured and predicted mould- TC, McConkey BG, Mosier A, Rypdal K, Walsh M, Wil- board plough draught in a range of Swedish soils. Soil liams SA (2006) Chapter 11: N O emissions from managed Tillage Res 111:219–223. https://doi.org/10.1016/j.still. soils, and CO emissions from lime and urea application. 2010.10.005 In: Gytarsky M, Hiraishi T, Irving W, Krug T, Penman J Ashworth AJ, Taylor AM, Reed DL, Allen FL, Keyser PD, (eds) 2006 IPCC guidelines for national greenhouse gas Tyler DD (2015) Environmental impact assessment of inventories. IPCC International Panel on Climate Change, regional switchgrass feedstock production comparing Geneva, pp 11.1–11.54 nitrogen input scenarios and legume-intercropping sys- Delin S, Stenberg M (2014) Effect of nitrogen fertilization on tems. J Clean Prod 87:227–234. https://doi.org/10.1016/j. nitrate leaching in relation to grain yield response on loamy jclepro.2014.10.002 sand in Sweden. Eur J Agron 52:291–296. https://doi.org/ Balmford A et al (2018) The environmental costs and benefits of 10.1016/j.eja.2013.08.007 high-yield farming. Nat Sustain 1:477–485. https://doi.org/ Delin S, Linde´n B, Berglund K (2005) Yield and protein 10.1038/s41893-018-0138-5 response to fertilizer nitrogen in different parts of a cereal Bessou C, Lehuger S, Gabrielle B, Mary B (2013) Using a crop field: potential of site-specific fertilization. Eur J Agron model to account for the effects of local factors on the LCA 22:325–336. https://doi.org/10.1016/j.eja.2004.05.001 of sugar beet ethanol in Picardy region, France. Int J Life 123 154 Nutr Cycl Agroecosyst (2019) 114:139–155 Diaz RJ, Rosenberg R (2008) Spreading dead zones and con- (LCA): a review. J Clean Prod 104:23–39. https://doi.org/ sequences for marine ecosystems. Science 321:926–929. 10.1016/j.jclepro.2015.05.040 https://doi.org/10.1126/science.1156401 Goglio P et al (2018) A comparison of methods to quantify Doltra J, Olesen JE (2013) The role of catch crops in the eco- greenhouse gas emissions of cropping systems in LCA. logical intensification of spring cereals in organic farming J Clean Prod 172:4010–4017. https://doi.org/10.1016/j. under Nordic climate. Eur J Agron 44:98–108. https://doi. jclepro.2017.03.133 org/10.1016/j.eja.2012.03.006 Hammar T, Ericsson N, Sundberg C, Hansson PA (2014) Cli- Edstro¨m M, Pettersson O, Nilsson L, Ho¨rndahl T (2005) Jord- mate impact of willow grown for bioenergy in Sweden. brukssektorns energianva¨ndning [Swedish]. Swedish Bioenergy Res 7:1529–1540. https://doi.org/10.1007/ Institute of Agricultural and Environmental Engineering s12155-014-9490-0 (JTI), Uppsala Henryson K, Hansson P-A, Sundberg C (2018) Spatially dif- Eklo¨f P (2014) Marknadso¨versikt – Spannma˚l. Jordbruksverket ferentiated midpoint indicator for marine eutrophication of (Swedish Board of Agriculture) waterborne emissions in Sweden. Int J Life Cycle Assess EMEP/EEA (2016) EMEP/EEA air pollutant emission inven- 23:70–81. https://doi.org/10.1007/s11367-017-1298-7 tory guidebook 2016. In: Technical guidance to prepare Huttunen M, Vehvila¨inen B, Huttunen I (2013) Typen, fosforin national emission inventories, vol EEA report no 21/2016. ja kiintoaineksen pida¨ttyminen vesisto¨issa¨–WSFS-Ve- European Environment Agency, Luxembourg mala-mallin arvio. Finnish Environment Institute, Helsinki European Commission (2017) Modernising and simplifying the ISO (2006) ISO 14040: environmental management-life cycle cap. In: Background document: climate and environmental assessment-principles and framework, 2. utg. edn. Inter- challenges facing EU agriculture and rural areas. European national Organization for Standardization (ISO), Geneva Commission. Directorate-General for Agriculture and Johnsson H, Ma˚rtensson K, Lindsjo¨ A, Persson K, Rangel YA, Rural Development, Brussels Blomba¨ck K (2016) La¨ckage av na¨ringsa¨mnen fran svensk FAO (2016) FAOSTAT statistics database: live animals, by akermark. Bera¨kningar av normalla¨ckage av kva¨ve och 1058 country. http://www.fao.org/faostat/en/#data/QA. fosfor for 2013 vol SMED Rapport Nr 189 2016. Sveriges Accessed 22 Feb 2019 Meteorologiska och Hydrologiska Institut, Norrkoping Fertilizers Europe (2018a) Ammonium nitrate (AN, 33.5% N); Kyllmar K, Carlsson C, Gustafson A, Ulen B, Johnsson H from ammonia and nitric acid, including primary produc- (2006) Nutrient discharge from small agricultural catch- tion; production mix, at plant. UUID: c2c4ebba-358f-493e- ments in Sweden characterisation and trends. Agric Eco- 83ba-71bffe847e9a (version 00.00.001). thinkstep, GaBi syst Environ 115:15–26. https://doi.org/10.1016/j.agee. database 2005.12.004 Fertilizers Europe (2018b) Potassium chloride (KCl/MOP, 60% Lambin EF, Meyfroidt P (2011) Global land use change, eco- K O); shaft mining and beneficiation, including primary nomic globalization, and the looming land scarcity. Proc production; production mix, at plant. UUID: a2a8695e- Natl Acad Sci 108:3465. https://doi.org/10.1073/pnas. 968c-4341-922c-a007c8a8c56d (version: 00.00.001). 1100480108 thinkstep, GaBi database Land M, Grane´li W, Grimvall A, Hoffmann CC, Mitsch WJ, Fertilizers Europe (2018c) Triple superphosphate (TSP, 46% Tonderski KS, Verhoeven JTA (2016) How effective are P2O5); rock phosphate acidulation with phosphoric acid, created or restored freshwater wetlands for nitrogen and including primary production; production mix, at plant. phosphorus removal? A systematic review. Environ Evid UUID: 8d0007f0-9ad8-43b0-86e4-ebe6e9f9d0e6 (version 5:9. https://doi.org/10.1186/s13750-016-0060-0 00.00.001). thinkstep, GaBi database Licker R, Johnston M, Foley JA, Barford C, Kucharik CJ, Garcı´a-Ruiz R, Carranza-Gallego G, Aguilera E, De Molina Monfreda C, Ramankutty N (2010) Mind the gap: how do MG, Guzma´n GI (2019) C and N mineralisation of straw of climate and agricultural management explain the ‘yield traditional and modern wheat varieties in soils of con- gap’ of croplands around the world? Glob Ecol Biogeogr trasting fertility. Nutr Cycl Agroecosyst 113:167–179. 19:769–782. https://doi.org/10.1111/j.1466-8238.2010. https://doi.org/10.1007/s10705-019-09973-4 00563.x Gode J, Martinsson F, Hagberg L, Oman A, Ho¨glund J, Palm D Lindgren M, Pettersson O, Hansson P-A, Noren O (2002) (2011) Miljo¨faktaboken 2011. Estimated emission factors Engine load pattern and engine exhaust gas emissions from for fuels, electricity, heat and transport in sweden. off-road vehicles and methods to reduce fuel-consumption Va¨rmeforsk, Stockholm and engine exhaust gas emissions [Swedish]. Swedish Goglio P, Bonari E, Mazzoncini M (2012) LCA of cropping Institute of Agricultural and Environmental Engineering systems with different external input levels for energetic (JTI), Uppsala purposes. Biomass Bioenergy 42:33–42. https://doi.org/10. Myhre G et al (2013) Anthropogenic and natural radiative 1016/j.biombioe.2012.03.021 forcing. In: Stocker TF, Qin D, Plattner G-K, Tignor M, Goglio P, Grant BB, Smith WN, Desjardins RL, Worth DE, Allen SK, Boschung J, Nauels A, Xia Y, Bex V, Midgley Zentner R, Malhi SS (2014) Impact of management PM (eds) Climate change 2013: the physical science basis. strategies on the global warming potential at the cropping Contribution of working group I to the fifth assessment system level. Sci Total Environ 490:921–933. https://doi. report of the intergovernmental panel on climate change. org/10.1016/j.scitotenv.2014.05.070 Cambridge University Press, Cambridge Goglio P, Smith WN, Grant BB, Desjardins RL, McConkey BG, Nemecek T, Schnetzer J (2011a) Data collection of inputs and Campbell CA, Nemecek T (2015) Accounting for soil yields in LCIs of agricultural production systems in carbon changes in agricultural life cycle assessment 123 Nutr Cycl Agroecosyst (2019) 114:139–155 155 Switzerland and other European countries. Data v3.0. and fertilizer management effects. Agric Ecosyst Environ Agroscope, Zu¨rich 133:247–266. https://doi.org/10.1016/j.agee.2009.04.021 Nemecek T, Schnetzer J (2011b) Data collection of inputs and Statistics Sweden (2011) Va¨xtskyddsmedel i jord-och yields in LCIs of agricultural seed production systems and tra¨dga˚rdsbruket 2010. Anva¨ndning i gro¨dor. [Swedish]. seed processing-Data v3.0. Agroscope, Zu¨rich Plant protection products in agriculture and horticulture. Nitschelm L et al (2017) Improving estimates of nitrogen Use in crops. MI 31 SM 1101, Statistiska centralbyra˚n emissions for life cycle assessment of cropping systems at (SCB) the scale of an agricultural territory. Environ Sci Technol. Statistics Sweden (2016) SCB statistikdatabasen: [Elektronic https://doi.org/10.1021/acs.est.6b03865 resource] Statistical database. Statistiska centralbyra˚n, Notarnicola B, Sala S, Anton A, McLaren SJ, Saouter E, Stockholm Sonesson U (2017) The role of life cycle assessment in Struijs J, Beusen A, van Jaarsveld H, Huijbregts MAJ (2013) supporting sustainable agri-food systems: a review of the Eutrophication. In: Goedkoop M, Heijungs R, Huijbregts challenges. J Clean Prod Part 140(Part 2):399–409. https:// M, De Schryver A, Struijs J, van Zelm R (eds) ReCiPe doi.org/10.1016/j.jclepro.2016.06.071 2008. Report I: characterisation. First edition (version Page G, Ridoutt B, Bellotti B (2014) Location and technology 1.08) options to reduce environmental impacts from agriculture. Sutton MA, Oenema O, Erisman JW, Leip A, van Grinsven H, J Clean Prod 81:130–136. https://doi.org/10.1016/j.jclepro. Winiwarter W (2011) Too much of a good thing. Nature 2014.06.055 472:159. https://doi.org/10.1038/472159a Peter C, Fiore A, Hagemann U, Nendel C, Xiloyannis C (2016) Sutton MA et al (2013) Our nutrient world: the challenge to Improving the accounting of field emissions in the carbon produce more food and energy with less pollution. Centre footprint of agricultural products: a comparison of default for Ecology and Hydrology, Edinburgh on behalf of the IPCC methods with readily available medium-effort Global Partnership on Nutrient Management and the modeling approaches. Int J Life Cycle Assess 21:791–805. International Nitrogen Initiative https://doi.org/10.1007/s11367-016-1056-2 Swedish EPA (2006) Eutrophication of Swedish seas: final Potting J, Hauschild M (2006) Spatial differentiation in life report. Report 5509. Swedish Environmental Protection cycle impact assessment: a decade of method development Agency, Stockholm to increase the environmental realism of LCIA. Int J Life Swedish Meteorological and Hydrological Institute (2018) Cycle Assess 11:11–13. https://doi.org/10.1065/lca2006. SMHI Oppna data. Meteorologiska observationer. https:// 04.005 opendata-download-metobs.smhi.se/explore/. Accessed 16 Prade T, Svensson SE, Bjo¨rnsson L (2014) Introduction of May 2018 grass-clover crops as biogas feedstock in cereal-dominated Tilman D, Cassman KG, Matson PA, Naylor R, Polasky S crop rotations. Part I: effects on soil organic carbon and (2002) Agricultural sustainability and intensive production food production. In: Schenck R, Huizenga D (eds) Pro- practices. Nature 418:671–677. https://doi.org/10.1038/ ceedings of the 9th international conference on life cycle nature01014 assessment in the agri-food sector (LCA Food 2014), 8–10 Van Stappen F et al (2017) Sensitive parameters in local agri- October 2014, San Francisco. ACLCA, Vashon, cultural life cycle assessments: the illustrative case of pp 1032–1041 cereal production in Wallonia, Belgium. Int J Life Cycle Prechsl UE, Wittwer R, van der Heijden MGA, Lu¨scher G, Assess. https://doi.org/10.1007/s11367-017-1325-8 Jeanneret P, Nemecek T (2017) Assessing the environ- Wang ZB, Zhang HL, Lu XH, Wang M, Chu QQ, Wen XY, mental impacts of cropping systems and cover crops: life Chen F (2016) Lowering carbon footprint of winter wheat cycle assessment of FAST, a long-term arable farming field by improving management practices in North China Plain. experiment. Agric Syst 157:39–50. https://doi.org/10. J Clean Prod 112:149–157. https://doi.org/10.1016/j. 1016/j.agsy.2017.06.011 jclepro.2015.06.084 Rochette P et al (2018) Soil nitrous oxide emissions from Williams AG, Audsley E, Sandars DL (2010) Environmental agricultural soils in Canada: exploring relationships with burdens of producing bread wheat, oilseed rape and pota- soil, crop and climatic variables. Agric Ecosyst Environ toes in England and Wales using simulation and system 254:69–81. https://doi.org/10.1016/j.agee.2017.10.021 modelling. Int J Life Cycle Assessment 15:855–868. Ruan LL, Bhardwaj AK, Hamilton SK, Robertson GP (2016) https://doi.org/10.1007/s11367-010-0212-3 Nitrogen fertilization challenges the climate benefit of Zaher U, Stockle C, Painter K, Higgins S (2013) Life cycle cellulosic biofuels. Environ Res Lett 11:8. https://doi.org/ assessment of the potential carbon credit from no- and 10.1088/1748-9326/11/6/064007 reduced-tillage winter wheat-based cropping systems in Searchinger TD, Wirsenius S, Beringer T, Dumas P (2018) Eastern Washington State. Agric Syst 122:73–78. https:// Assessing the efficiency of changes in land use for miti- doi.org/10.1016/j.agsy.2013.08.004 gating climate change. Nature 564:249–253. https://doi. org/10.1038/s41586-018-0757-z Publisher’s Note Springer Nature remains neutral with Snyder CS, Bruulsema TW, Jensen TL, Fixen PE (2009) Review regard to jurisdictional claims in published maps and of greenhouse gas emissions from crop production systems institutional affiliations.

Journal

Nutrient Cycling in AgroecosystemsSpringer Journals

Published: May 8, 2019

References

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create folders to
organize your research

Export folders, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off