The experimental site, Easter Bush, is located in south-east Scotland, 10 km south of Edinburgh (03∘02′ W, 55∘52′ N; 190 m a.s.l.), with a mean annual rainfall of 947 ± 234 mm and a mean annual temperature of 9.0 ± 0.4 ∘C (2002–2010). The field (5.424 ha) has been under permanent grassland management for more than 20 years before the start of the experiment with a species composition of > 99 % perennial ryegrass (Lolium perenne) and < 0.5 % white clover (Trifolium repens). The soil type is an imperfectly drained Macmerry soil series, Rowanhill soil association (Eutric Cambisol), with a pH of 5.1 (in H2O) and a clay fraction of 20–26 %. The ground water table was assumed to be at 0.85 m depth on average, and the main rooting zone extends down to 0.31 m below the soil surface.

The grassland was grazed rotationally throughout the experimental period by heifers in calf, ewes and lambs at different stocking densities (Table 1 and Fig. 1a). Livestock units used for heifers, ewes and lambs were 0.75, 0.10 and 0.04, respectively (1 livestock unit has a standard live weight of 600 kg head-1; SAC, 1995). Lambs were present on the field from April to September only. The grass was cut for silage only in the first 2 years, on the 1 June and 8 August 2002 and on the 29 May 2003. Ammonium nitrate fertiliser was applied to the field three to four times per year, usually between March and July (56 kg N ha-1 application-1 on average). In 2008 an additional fifth mineral N application was added, using urea instead of ammonium nitrate fertiliser. Organic manure was applied on 28 September 2004 and 27 March 2005 as cattle slurry, using a vacuum slurry spreader (Tables 3 and 4 and Fig. 1b and c).

We assessed the N and C budget by measuring or estimating the import and export of all relevant fluxes to and from the grassland field on an annual basis. Throughout the paper all fluxes are presented following the sign convention used in micrometeorology; fluxes from the ecosystem to the atmosphere are positive (exported from the field), while negative values indicate fluxes from the atmosphere to the ecosystem (imported to the field). We set the system boundary for inputs and exports of N and C by the field perimeters. The balance of all imports and exports results in the observed changes in N and C on this field over time.

The change in the N balance (ΔN) over time (Δt) of our grassland ecosystem can be written as ΔN/Δt=FNorgfert.+FNsyntfert.+FNfeed+FNN2(biol.fixation)+FNdep.+FNharvest+FNanimal+FNleaching+FNNH3,NOx(fert.,manure,animal)+(FNNOx(soil)+FNN2O+FNN2(denitr.). N imports include the addition of N from organic and inorganic fertiliser (FNorgfert.+ FNsyntfert.), cake concentrate fed to ewes during lactation (FNfeed), the fixation of N2 through biological fixation (FNN2(biol.fixation)), and the deposition of NH3, HNO3, NH4+, NO3- from dry and NH4+ and NO3- from wet deposition (summarised as FNdep.). Exports include the N lost from plant biomass at cuts for silage (FNharvest) the offtake of N in meat (including bones) and wool from animals (FNanimal), the loss of organic and inorganic dissolved N through leaching (FNleaching), the NH3 and NOx emissions from the volatilisation of inorganic and organic fertiliser spreading as well as from animal excretion (FNNH3,NOx(fert.,manure,animal)), the emission of NOx from the soil (FNNOx(soil)), the emission of N2O from the soil (FNN2O), and the loss of N2 from total denitrification (FNN2(denitr.)).

The change in the C balance (ΔC) over time equals the net biome production (NBP) and can be written for our site as ΔC/Δt=NBP=FCCO2+FCorgfert+FCanimal+FCCH4+FCleaching+FCharvest. FCCO2 represents the net ecosystem exchange (NEE) of CO2, and FCorgfert is the C input through manure application. Carbon input from animal excretion was not included in the budget as it was assumed to be recycled C from plant and soil uptake. FCanimal includes the C offtake through animal weight increase and wool production. As grazing cows were heifers in calf and ewes' milk was consumed by their lambs, there was no C offtake through milk to be considered. Methane emissions from enteric fermentation by ruminants, animal excretion and manure application as well as CH4 fluxes from the soil are included in FCCH4. FCleaching is the C lost through dissolved organic and inorganic C and dissolved CH4 leaching, and FCharvest represents the C lost from the system through plant biomass export from harvests (cut for silage). Carbon emissions from farm operations (i.e. tractor emissions) or off-farm emissions (i.e. fertiliser manufacture) are not included in the C budget.

Details of methods to quantify each N and C budget component, as listed in Eqs. (1) and (2), are described under Sect. 2.4 to 2.17 and in the Supplement. Some budget components were measured throughout the 9 years presented, while others were only measured in some years or not at all. Missing data were derived from the literature, models or averages from available data from other years.

The annual net GHG exchange (NGHGE) was calculated from annual NEE (FCCO2), CH4 (FCCH4) and N2O (FNN2O) fluxes using global warming potentials (GWPs) at the 100-year time horizon (1 for CO2, 298 for N2O and 25 for CH4, IPCC, 2013): NGHGE=(FCCO2)+FCCH4×kCH4+FNN2O×kN2O, where kCH4=9.09,since 1 kgCH4-C corresponds to9.09kgCO2-C,kN2O=127,since 1 kgN2O-N corresponds to127kgCO2-C. In addition the net annual greenhouse gas balance (NGHGB) was calculated by including the loss of C through animal meat and wool production, harvest offtake, C leaching, and input by organic fertiliser application: NGHGB=NGHGE+FCorgfert+FCanimal+FCleach+FCharvest.

Mineral fertiliser was applied by a spreader as either ammonium nitrate or urea. Six-month-old cattle slurry was spread by a vacuum slurry tanker. Data of application rates and N content were obtained from the farmer. Three samples from the slurry tank were taken at each application and analysed for ammoniacal nitrogen (NH3 and NH4+), dry-matter content, total N, total C, pH and nitrate. The total N and C import to the field by the slurry was calculated by the volume of the slurry applied and the N and C analyses of the slurry.

Ewes were given standard cake concentrate “Davidsons super ewe nuts” during lactation, which contained 20 % protein. They were given 1 kg of this additional feed per day for 7 weeks in spring. This resulted in an extra N intake of 0.032 kg N per ewe per day (assuming an N content of 16 % of protein; IPCC, 2006a).

The farmer estimated a forage harvest of 15 t fresh weight (FW) per hectare at the first cut and 10 t FW ha-1 at the second cut of a year, based on the plant height at the field at the time of cutting and information from harvested plot experiments. As there were two cuts in 2002 and one cut in 2003 the estimated harvest was 25 t FW ha-1 a-1 for 2002 and 15 t FW ha-1 a-1 for 2003. A subsample of harvested vegetation was collected and dried at 80 ∘C for plant N and C analysis using a Carbo-Erba/400 automated N and C analyser.

It was estimated by the farmer that heifers increased in weight by 0.8 kg per day (starting weight of 450 kg). The ewe weight was assumed to be constant (60 kg). They were fed extra protein (standard cake concentrate) to reduce weight loss during lactation, whereas lambs were brought to the field at a weight of 5 kg and removed when they reached a weight of 45 kg. All animals were weighed before they came onto the field at the beginning of the season and again at the end of the season. The total meat export, which includes bones, was calculated from the daily weight increase in heifers and lambs multiplied by the animal number per day. To calculate the N and C export from meat, we assumed an N content of meat of 3.5 % and a C content of meat of 21 % (Flindt, 2002), an N content of bones of 7 % and a C content of bones of 20 % (Marchand, 1842), assuming a total bone content of 20 % for sheep (Lambe et al., 2007) and 14 % for heifers (Navajas et al., 2010) . Ewes were sheared annually in June, yielding an estimated 2.5 kg of wool per sheep. Wool N and C export was calculated from wool production multiplied by the average sheep number in June, assuming an N and C content of wool of 16.5 and 50 %, respectively (Roche, 1995).

Two sets of 10 glass suction cups (pore size < 1 µm, ecoTech, Bonn, Germany) for soil water and 4 Teflon suction cups (ecoTech, Bonn, Germany) for soil gas collection were installed in August 2006. One set was located on a slope, another in a hollow. For the budget calculations we only used results from the slope location as the hollow location was frequently water logged. Suction cups were installed horizontally from a soil pit beneath the A horizon (30 cm depth) and at 90 cm depth and connected to 2 L glass bottles in an insulated aluminium box placed into the soil pit. Samples were collected every 2–3 weeks. For further details and description of dissolved organic and inorganic C (DIC, DOC) and dissolved CH4 analysis see Kindler et al. (2011). Dissolved inorganic and organic N (DIN, DON) and total N (TN) concentrations in leachate water were analysed by colorimetric analysis (San++, automated wet chemistry analyser – continuous-flow analyser (CFA), Skalar, the Netherlands). Leachate C and N concentrations were measured from 1 October 2006 to 30 March 2008. Dissolved C and N were calculated by multiplying concentrations of DIC, DOC and dissolved CH4 or DIN and DON, respectively, with leachate volume. The latter was derived from a soil water model based on balancing daily precipitation and evaporation considering the water-holding capacity of the soil (Kindler et al., 2011). To calculate the annual leaching flux for 2007, data were gap filled using daily average values for missing data. For the remaining years N was simulated using the LandscapeDNDC model (Haas et al., 2013), with the model tested and validated with comprehensive measured data. For C leaching linear regression models describing the relationship between calculated C leaching fluxes and leachate volume for the measurement period (DOC: y=0.0186x-0.0695, R2= 0.8663; DIC: y=0.021x-0.0008, R2= 0.8056; and dissolved CH4: y=0.0019x-0.0135, R2= 0.7623) were used to extrapolate to the remaining years.

From June 2002 to July 2003 N2O fluxes were measured continuously by eddy covariance (EC) using an ultrasonic anemometer coupled with a tunable diode laser absorption spectrometer (TDL) at a frequency of 10 Hz. Details of the gap-filling method of the N2O–EC data are described in Jones et al. (2011). From August 2006 to November 2010, N2O fluxes were measured using manual closed static chambers (Clayton et al.,1994; Skiba et al., 2013). Samples were analysed for N2O using a Hewlett Packard 5890 series II gas chromatograph (GC) (Agilent Technologies, Stockport, UK), fitted with an electron capture detector (detection limit: N2O < 33 ppbV). Fluxes were measured weekly and more frequently during fertilisation. Cumulative fluxes were calculated by gap filling data for missing days using linear interpolation and summing up all gapfilled data over each calendar year. For the periods where no N2O fluxes were measured (January–May 2002, July 2003–March 2004, May 2004–July 2006) fluxes were simulated by LandscapeDNDC (Haas et al., 2013). LandscapeDNDC was tested in detail with available data on plant growth, soil temperature, moisture, inorganic soil N concentration, NO and N2O, which resulted in general good agreement of simulations and measurements. Results except for NO emissions are published in Molina et al. (2016).

NOx fluxes from the soil were only measured for a short period (June 2009–August 2010). The NOx fluxes were measured using an autochamber system described in detail by Butterbach-Bahl et al. (1997). Measurements were made four times per day, every 6 h for an 8 min period per chamber. We used simulated data from LandscapeDNDC for years where no NOx fluxes were measured.

The fraction of nitrogen that volatilises as NH3 and NOx from applied synthetic fertiliser or cattle slurry application and animal excretion was estimated to be 10 and 20 % of total N applied, respectively (IPCC, 2006b). The animal excretion amount was estimated in accordance with the IPCC guidelines (IPCC, 2006a); for details, please see the Supplement.

Dinitrogen (N2) emissions resulting from total denitrification in the soil was not measured in our experiment. We therefore used the N2 emission rates from LandscapeDNDC simulations.

The species composition was measured once in 2002 and at monthly intervals in 2003 by the visual estimation method (Braun-Blanquet, 1964), where 50 quadrats of 0.25 m2 were distributed across the field at random. As the legume fraction (Trifolium repens) was smaller than 0.5 % at each measuring point, we assumed the nitrogen fixation through plants to be zero.

NEE was measured continuously from 1 January 2002 till 31 December 2010 by an eddy covariance system consisting of a fast-response 3-D ultrasonic anemometer (Metek USA-1, Metek GmbH, Elsmhorn, Germany) and a fast closed-path CO2–H2O analyser (LI-COR 7000 infrared gas analyser (IRGA), LI-COR, Lincoln, NE, USA). NEE is the arithmetic sum of the gross primary production (GPP) and total ecosystem respiration (TER). Flux partitioning of measured NEE into GPP and TER was calculated by the same online tool used for gap filling (http://www.bgc-jena.mpg.de/~MDIwork/eddyproc/upload.php, Reichstein et al., 2005). In this flux-partitioning approach, daytime TER is obtained by the extrapolation of a nighttime parameterisation of NEE on air temperature and GPP is the difference between ecosystem respiration and NEE. Contrary to unmanaged ecosystem, TER at our site also includes the respiratory loss of CO2 by grazing animals. Net primary production (NPP), which represents the annual plant growth (difference between GPP and autotrophic respiration) was calculated as 50 % of GPP (Amthor, 2000; Zhang et al., 2009).

Methane fluxes from the soil were measured with closed static chambers simultaneously with the N2O measurements (see Sect. 2.10). The same GC was fitted with a flame injection detector (detection limit: CH4 < 70 ppbV). The minimal detectable flux was 17 ng CH4–C m-2 s-1. Fluxes were measured weekly and more frequently at fertilisation events. As measured soil CH4 fluxes were close to zero and did not vary significantly between months, we calculated CH4 for months where no CH4 fluxes were measured (January–May 2002, July 2003–March 2004, May 2004–July 2006) as an average monthly cumulative flux from other years.

Methane emissions from grazing animals, i.e. animal excretion and enteric fermentation, were estimated following the IPCC Tier 2 methodology (IPCC, 2006a; Stewart et al., 2009). For details, see the Supplement. Methane emissions from slurry applications were assumed to be small. As no chamber measurements were conducted at the time of slurry spreading, the emissions were estimated as 0.07 % of the slurry applied assuming that emissions were comparable to those in a related study (Jones et al., 2006), where CH4 was measured from chambers after slurry application on a nearby field in 2002 and 2003.

Fluxes of non-methane volatile organic compounds (VOCs) were not measured. We assumed similar VOC emissions to those reported by Davison et al. (2008) for an intensively managed grassland in Switzerland, where the daily average flux of methanol, acetaldehyde and acetone over 3 days after cutting was 21.1, 5.1 and 2.6 nmol m-2 s-1, respectively. Based on those values, annual VOC emissions from our field were estimated to be in the order of 0.03 % of the annual C offtake in harvest and 0.08 % of annual C offtake by grazing animals. We therefore assumed VOC emissions to be negligible and did not account for them in the C balance.

Total N and C content of the soil were measured in May 2004 and May 2011. One hundred soil cores with an inner diameter of 8.7 (2004) and 8.3 cm (2009; both corers from Eijkelkamp Agrisearch Equipment BV, Giesbeek, the Netherlands) were collected along a regular grid with a distance of 10 m between sampling points on both occasions. The soil sampling grid covered the main footprint area of the site, not the entire field. Cores were separated into layers of 0–5, 5–10, 10–20, 20–30, 30–40, 40–50 and 50–60 cm. Coarse stones of a diameter > 4 mm and roots of a diameter > 1 mm were removed from the samples prior to drying at 40 ∘C. Stone and root samples were air-dried separately. Then, soil samples were sieved to < 2 mm. Particles > 2 mm were combined with the coarse stones. Dry weights of roots and combined stone fractions were determined. Total N and C concentrations in < 2 mm soil separates were determined using dry combustion (VarioMax, Elementar Analysensysteme GmbH, Hanau, Germany). As the site contains no carbonates, total C was assumed to equal organic C. As bulk density varies spatially and over time (e.g. through compaction by livestock), the soil N and C content per unit ground area to a fixed depth will also change, without any change in the mass fraction of N and C in dry soil. Therefore, total N and C stocks were calculated on an equivalent soil mass (ESM) basis, so that comparisons between years were valid (see Gifford and Roderick, 2003; Wendt and Hauser, 2013). A cubic polynomial was fitted to the data to predict cumulative N and C with cumulative soil mass in the profile. A soil mass of 800 kg m-2 was used (Table 6), which corresponds to approximately 60 cm depth, which was the depth of the corer. Uncertainty in the estimates of stock change was based on the prediction intervals in the cubic polynomial at a soil mass of 800 kg m-2.

Soil temperature and volumetric soil moisture were continuously recorded at four depths (3.5, 7.5, 15 and 30 cm) by temperature probes (temperature probe 107, Campbell Scientific, Loughborough, UK) and time domain reflectometry (TDR) probes (TDR 100, Campbell Scientific, Loughborough, UK), the latter installed in June 2002. Rain was measured by a tipping-bucket rain gauge, while air temperature and relative humidity were measured by an integrated humidity and temperature transmitter (HUMITTER^®, Vaisala Ltd, Suffolk, UK).

Random error was determined as 2σ standard error (95 % confidence) of the overall mean according to Gaussian statistics. The confidence intervals for group means were used to establish whether or not differences were significantly different from zero. Linear correlations between C and N inputs and outputs were calculated by calendar year. For systematic errors the uncertainty range of measurements as well as of parameterisations and literature-based estimates was estimated according to expert judgment. To calculate the combined effect of systematic uncertainties of each flux component on the C and N budget, simple Gaussian error propagation rules were used; details are provided in Table S1 (Supplement). Confidence intervals are given at the 95 % confidence level.

Additional details of the methods are provided in the Supplement.

The meteorological conditions exhibited substantial inter-annual variability in the study period 2002–2010 (Table 2). Annual rainfall ranged from 575 to 1238 mm with highest monthly rainfalls of 280 mm month-1 in September 2002. Lowest annual reported rainfall was in 2010; this low value was caused by a gap in data from January to March due to snowfall. Average annual air temperature ranged from 8.3 to 9.6 ∘C, with highest daily air temperatures of 30.4 ∘C in August 2005 and lowest in December 2010 at -10.3 ∘C. Highest average monthly air temperatures were measured in July 2006 at 17 ∘C and lowest monthly average air temperatures at 2 ∘C in November 2009. In 2003 the highest average annual temperature (9.6 ∘C) and lowest annual rainfall (680 mm) were measured with a correspondingly low annual soil water content of 31 %. The duration of the growing season was defined per calendar year as the period bounded by the first and last 5 consecutive days with a mean daily air temperature ≥ 5 ∘C. The length of the growing season (LGS) varied between 151 days (2006) and 242 days (2009; Table 2).

Livestock stocking density exhibited both intra- and inter-annual variability. The average annual stocking densities were lowest in 2002 and 2003 at 0.27 LSU ha-1 and 0.54 LSU ha-1, respectively (Table 1), which were the years where the grass was cut for silage and no lambs were present in the field. In 2007, 2008, 2009 and 2010, no heifers were present in the field. Highest annual average stocking densities occurred in 2004 and 2007 at 0.99 and 0.91 LSU ha-1, respectively. Maximum monthly stocking density occurred in September 2006 at 13.8 LSU ha-1, while interim periods with no grazing at all were observed in all years (Fig. 1a). Mineral N fertiliser was applied split into three to five applications per year, ranging from 2.5 to 9.6 g N m-2 per application (Fig. 1b). Organic manure was applied in 2004 and 2005 as cattle slurry, spread at a rate of 6.9 and 15.8 g N m-2 per application, respectively, which resulted in a C input of 55.4 and 171.8 g C m-2 per application, respectively (Fig. 1b and c). The grass was only cut in 2002 and 2003. Harvested biomass in 2002 and 2003 ranged from 2.60 to 3.75 t DW ha-1 per cut, which resulted in an N offtake ranging from 1.7 to 4.7 g N m-2 per cut and a C removal from the field ranging from 113.1 to 169.5 g C m-2 per cut (Fig. 1b and c).

In our grassland system the N balance is the difference between the N input through fertiliser and atmospheric deposition and the N output through harvest, animal export, leaching and gaseous emissions. The total resulting balance over the 9 years, derived from flux calculations and estimations, showed that N was stored at an average rate of -6.0 ± 5.9 g N m-2 a-1 (p < 0.05). From 2003 to 2010, N was stored at a rate of -1.2 to -18.2 g N m-2 a-1, whilst in 2002 N was lost at a rate of 13.2 g N m-2 a-1 (Table 3). The major N input consisted of inorganic fertiliser, ranging from -11 to -25.9 g N m-2 a-1, averaging at -19.4 g N m-2 a-1, while N deposition represented only between 1.9 and 5.9 % of the total N input. N input through standard cake feed given to ewes during lactation ranged between 1.1 and 1.9 g N m-2 a-1. During the years when N was stored, a significant positive correlation between total N input from fertiliser and N storage was observed (R2= 0.55). The largest losses resulted from leaching at an average rate of 5.34 ± 3.1 g N m-2 a-1 and were estimated to be the highest in 2002 at 14.9 g N m-2 a-1 and the lowest in 2003 at 0.09 g N m2 a-1. The total N offtake through meat and wool ranged from 0.15 to 3.12 g N m-2 a-1, while the total annual N offtake from harvest was 11.8 in 2002 and 10.4 g N m-2 a-1 in 2003. Amongst gaseous exchanges, the highest losses were estimated from N2 emissions, averaging at 2.9 g N m-2 a-1 with maximum losses of 4.12 g N m-2 a-1 in 2009, although in 2004 and 2005 losses from NOx and NH3 volatilisation from excretion and organic fertilisation exceeded losses from N2 emissions. Losses through NOx from the soil were always less than 1 % of the total N exchange (0.2 g N m-2 yr-1). Nitrous oxide emissions ranged from 0.11 to 1.27 g N m-2 a-1, representing 1.3–8.4 % of the total N export. Annual N2O emissions showed no correlation with precipitation, livestock density or total N input. N2O emission factors (percentage of N lost from total N inputs by mineral and organic fertiliser), ranged between 0.6 and 7.5 % (Table 5).

To investigate the influence of different managements on the N and C budget, we separated experimental years into harvested and grazed (2002 and 2003) and grazed-only years (2004–2010, Figs. 2 and 3). During the harvested and grazed years, the main loss of N from the system occurred from the export through harvest (57.6 % of total N inputs), followed by leaching (39.2 %), while the export from animals (meat and wool) accounted for less than 2 % of total N losses (Fig. 2a). The main loss to the atmosphere resulted from total denitrification (N2; 15.4 %), followed by NOx and NH3 volatilisation from inorganic N fertiliser applications (9.5 %), while N2O emissions accounted for 3.3 %, NOx/NH3 volatilisation from excretion for 3.3 % and NOx from soil for less than 1 %. On average, N was lost from the grassland system during 2002 and 2003; The surplus of 31.3 % (all outputs added up in Fig. 2a result in 131.3 %) represents the N lost from the soil as well as the error in the budget. When grazed-only years were considered (Fig. 2b), N was stored in the grassland system; the residual part, which includes the net accumulation of soil organic nitrogen and all the error in the budget, was 37.7 %. Losses through leaching (18.7 %) and N2 (11.3 %) were lower in grazed years compared to harvested years, while the export through grazing animals was considerably higher at 18.8 % (sum of N loss through meat, wool and NOx and NH3 volatilisation from excretion). An additional loss occurred in grazed years through the volatilisation of NOx and NH3 from organic fertiliser applications in 2004 and 2005 (2.6 %). Losses through N2O and NOx and NH3 from inorganic fertiliser were comparable to harvested years at 2.4 and 7.9 %, respectively.

Cumulative soil N stocks were derived from soil core measurements taken in May 2004 and May 2011 (Table 6). In 2004 N stocks were 840.86 (±11.89) g N m-2 and in 2011 they were 870.02 (±14.14) g N m-2. Nitrogen storage over the 7 years was calculated from the cumulative equivalent soil mass (ESM) for the soil mass increment of 800 kg m-2, which corresponds to approximately 60 cm depth. The estimated N storage over the 7 years was -4.51 ± 2.64 g N m-2 a-1 and was a significant N accumulation to the soil (p < 0.01). The estimated N storage derived from flux calculations between 2004 and 2010 was -9.40 ± 4.14 g N m-2 a-1, which is almost 2 times more than that calculated by sequential soil analysis; however, values were not significantly different from each other.

Annual C inputs through photosynthesis (GPP) varied between -982.1 and -2162.9 g C m-2, and losses through autotrophic and heterotrophic respiration (TER) varied between 972.1 and 2183.2 g C m-2, both considerably larger than any other C fluxes (Table 4). If only the NEE was considered (difference between GPP and TER), the grassland acted as a sink for CO2 at an average of 218 ± 155 g C m-2 a-1, and the CO2 uptake was significantly different from zero (p < 0.05). The sink strength ranged from -10 (2006) to -606 g C m-2 a-1 (2009); only in 2004, the grassland was a small source of CO2 (72 g C m-2 a-1). Taking into account all C inputs and outputs (NBP), C was sequestered on average at 163 ± 140 g C m-2 a-1 over the 9 years, although the storage was not significantly different from zero (p < 0.05). In 2004 and 2006 C was lost from the ecosystem. The major C import resulted from NEE in all years apart from 2005, when the C input from manure application was larger. The highest C export occurred from harvest in 2002 and 2003 (270.6 and 169.5 g C m-2 a-1, respectively), while the second-largest export in 2002 and 2003 and the largest exports in other years was leaching (6.8 to 25.1 g C m-2 a-1). The measured C leaching value for 2007 (17.0 g C m-2 a-1, Table 5) differs from the leaching value published for Easter Bush by Kindler et al. (2011), as we only used values of one of the two measured sites in this paper (slope, not hollow, as the hollow location was frequently water logged). The third largest C loss consisted of C export from meat in 2004–2010, ranging from 6.4 to 15.8 g C m-2 a-1. In 2002 and 2003, when no lambs were present in the field, C export from meat was exceeded by CH4 losses from enteric fermentation. Carbon export from wool ranged from 0.5 to 2.1 g C m-2 a-1. CH4 emissions from organic fertilisation, soil processes and animal excretion were always less than 1 % of the total C losses. CH4 losses from enteric fermentation ranged from 1.5 to 5.7 g C m-2 a-1, corresponding to 0.5–22.5 % of all C losses from the ecosystem. The NBP was dominated by the NEE. A high livestock density tended to reduce the net sink strength. A significant negative correlation of NEE with stocking density could be seen (R2= 0.47). The NBP correlated positively with rainfall (R2= 0.48), and there was only a weak correlation between NEE and rainfall (R2= 0.38).

The net primary production (NPP) in years when grass was harvested and grazed (2002 and 2003) and grazed only (2004–2010) are presented in Fig. 3. In both management types, most C was lost through ecosystem respiration, (67 and 65 % of NPP, respectively). Harvest export represented 21 % of NPP. Leaching accounted for 1.5 % of NPP during harvested years and 2.1 % in grazed-only years. Animal export (meat and wool) consisted of 1.6 % of NPP in grazed-only years and was 0.2 % of NPP in grazed and harvested years. The sum of all CH4 emissions (from organic fertilisation, excretion, enteric fermentation and soil) was less than 1 % of the NPP. The residual part, which includes the C storage in the soil as well as the uncertainty of the budget, was estimated at 10 and 30.8 % of NPP in harvested and grazed or grazed years, respectively.

The C content for the cumulative soil mass increment 0–800 kg m-2 (∼ 0–60 cm) was lower in 2011 (12026.05 ± 190.19 g C m-2) compared to 2004 (11 824.87 ± 187.84 g C m-2), resulting in a C loss of 29.08 ± 38.19 g C m-2 (Table 7). In comparison, based on flux calculations C was stored at 179.8 ± 180 g C m-2 yr-1 over the 7 years. However, neither C loss calculated by sequential soil analysis nor C storage estimated from flux calculations were significantly different from zero.

Average greenhouse gas fluxes, NGHGE and NGHGB for 2002–2010 are shown in Fig. 4. The CO2 storage from the NEE provided the largest term in the annual GHG budget. Carbon dioxide (NEE) was sequestered over the 9 years at a rate of -799 ± 567 g CO2 m-2 a-1. This storage was significantly different from zero (p < 0.05). On average, NGHGE was highly correlated with annual NEE (R2= 0.96). On average the grassland was a source of the GHGs CH4 and N2O at a rate of 148 ± 30 and 285 ± 131 g CO2 m-2 a-1, respectively, both being significantly different from zero (p < 0.001 and p < 0.01, respectively). Nitrous oxide losses ranged from 52 (2004) to 588 g CO2 eq. m-2 a-1 (2007; data for each year not shown). Methane from soil processes, manure input and animal excretion accounted for less than 5 % of total CH4 emissions. Methane emissions from enteric fermentation ranged from 53 g CO2 eq. m-2 a-1 (2002) to 199 g CO2 eq. m-2 a-1 (2004). Annual total CH4 emissions correlated positively with annual livestock density (R2= 0.99). The CH4 emissions, which were predominately (> 97 %) of ruminant origin weakened the sink strength of NEE by 18 %. If both CH4 and N2O were considered the total trade-off of NEE was a substantial 54 % and increased to a total of 67 % if only grazed years were considered. On average the grassland represented a GHG sink of -366 ± 601 g CO2 m-2 a-1 if only NEE, CH4 and N2O were included (NGHGE). If all C components (FCorg.fert, FCanimal, FCleaching, FCharvest) are included, the sink strength of the grassland decreased to -182 ± 560 g CO2 m-2 a-1 (NGHGB). This represents a weakening of the sink strength of NGHGE by 50 %, mainly due to the export of harvest. However, it has to be noted that in harvested years the return of the manure, resulting from the grass fed to livestock off-site, would reduce the GHG balance. If only grazed years were considered the sink strength increased slightly by 5.4 %, due to the C input from manure in 2004 and 2005. Both NGHGE and NGHGB were not significantly different from zero.

The main N inputs in our study were from inorganic and organic fertiliser additions. The amount of N added through fertiliser was determined by national recommendations (SAC, 2013). Atmospheric N deposition (wet and dry) accounted only for a small fraction of the total N input to our managed grassland. This is in contrast to semi-natural systems, where atmospheric N deposition and biological fixation represents the main N input (Phoenix et al., 2006; Bleeker et al., 2011). Due to high N fertilisation rates and low soil pH the legume fraction was less than 1 % and biological N2 fixation therefore a negligible source of N.

The data obtained from our budget were used to calculate the nitrogen use efficiency (NUE) expressed as the ratio between N in crop and animal products (in this case either the crop harvest or the sum of meat, wool and milk) and the total N inputs to the system (fertiliser, imported manure, standard cake concentrate fed to lactating ewes). The NUE of herbage in cut years (2002 and 2003) of 57.6 % (Fig. 2a) is comparable to reported N efficiencies of 55–80 % in harvested herbage from managed temperate grasslands (Ball and Ryden, 1984; Ammann et al., 2009). It has been shown that the NUE in crops is significantly higher compared to the NUE in animal production (Galloway and Cowling, 2002). The NUE of animal products on our grassland system ranged from 5 to 16.6 % in grazed-only years (2004–2010), with an average of 9.9 %. This is in agreement with the NUE reported for sheep of 6.2 % by Van der Hoek (1998) and beef production systems, which reported that N efficiencies range from 6 to 12 % (Whitehead et al., 1986; Tyson et al., 1992) and 5–20 % (Ball and Ryden, 1984). Approximately 85 % of total harvested and imported N in the EU is used to feed livestock (Sutton et al., 2011; Leip et al., 2011). A measure to reduce N pollution and GHG emissions could therefore be the reduction of meat consumption (Smith et al., 2013).

Nitrogen was lost from our grassland to the environment through different pathways. Nitrogen leaches from grassland soils in the form of nitrate (NO3-), ammonium (NH4+) and dissolved organic N (DON). Overall, leaching from our field (5.3 ± 3.1 g N m-2 a-1) was comparable to values measured at intensively grazed pastures in Ireland (1.8–6.4 g N m-2 a-1, Watson et al., 2007) and England (3.8–13.3 g N m-2 a-1, Scholefield et al., 1993) or croplands (e.g. Bechmann et al., 1998, max. leaching losses of 10.4 g N m-2 a-1). However, leaching from our study was high compared to the Swiss NitroEurope site, where a maximum loss of 0.35 g N m-2 a-1 was estimated from an ungrazed grass–clover sward, despite comparable annual rainfall and N inputs from inorganic and organic fertiliser (Ammann et al., 2009). This difference can be explained by the different management of the two sites. Grazed grasslands tend to have higher N leaching rates than cut grasslands since herbage that is not cut and exported from the field in grazed grasslands but eaten by the grazing animals and returned as urinary N adds N to the system. Furthermore, the uneven distribution of excreted organic N further enhances leaching due to the formation of N hotspots, which has been observed at outdoor pig farms (e.g. Eriksen, 2001). Ryden et al. (1984a) showed a 5.6 times higher leaching loss from grazed compared to cut grassland with 36 % of total N inputs lost from grazed compared to 6 % lost from cut grassland. On our site 18.7 % of total inputs was lost as leaching in grazed years, compared to 39 % in the cut years. However, the higher value in cut years was due to the high rainfall in 2002.

Due to its high background in the atmosphere, N2 fluxes cannot be measured directly in the field. There are different methods to measure N2 fluxes indirectly, which have been summarised by Groffman et al. (2006). In our study, we estimated N2 losses using the process-based biogeochemical model LandscapeDNDC (Haas et al., 2013; Molina-Herrera et al., 2016). These losses represented the highest gaseous N losses from our grassland in most years, with an average of 12.6 % of total N inputs and 14 % of inorganic fertiliser N inputs. This is comparable with an average N2 loss of 12.5 % from inorganic N applications measured by the acetylene inhibition method from a fertilised and cut but ungrazed grassland in Switzerland (Rudaz et al., 1999). Using the same method, Van der Salm et al. (2007) reported a higher loss of 22 % of total N input from a cattle-grazed pasture on a heavy clay soil in the Netherlands. In addition to the impact of the heavy clay soil, which could have enhanced denitrification due to reduced oxygen concentrations, grazing is likely to have enhanced denitrification rates in Van der Salm's study. Grazing not only enhances denitrification through soil compaction caused by trampling animals but also through the formation of N hot spots resulting from unevenly distributed soil N from excretion. In our study N2 losses simulated by LandscapeDNDC were based on average (per hectare) changes in the soil N pool instead of the more uneven distribution of soil N in hot spots like urine patches. Therefore, it is likely that N2 losses in our study have been underestimated, and better estimation would contribute to a significant reduction in the uncertainties associated with the overall N budget.

Annual N2O emissions measured in our study (0.1 to 1.3 g N m-2 a-1) are within the range of literature values from reported grazed as well as ungrazed European grasslands (Leahy et al., 2004; Flechard et al., 2007). Generally N2O losses are higher from grazed grassland compared to cut, ungrazed pasture (Velthof and Oenema, 1995; Luo et al., 1999) due to a more anaerobic environment as a consequence of soil compaction caused by animal treading and the influence of N and C from the deposition of animal excreta to the soil. We did not observe any correlations between annual N2O emissions and total N input. We found a relationship between the cumulative precipitation 1 week before and 3 weeks after fertilisation with N2O emissions (R2= 0.53; Skiba et al., 2013). This relationship, together with the influence of stocking density and the type of N applied (e.g. Jones et al., 2007; Flechard et al., 2007) needs to be considered when developing Tier 2 N2O emission factors. In our study emission factors (EFs) were above the uncertainty range (0.3–3 %) given by IPCC Tier 1 guidelines (IPCC, 2006b) in 4 out of 9 years. However, it has been shown that the N2O emission factor from managed grassland can be higher, especially under wet conditions and with a high soil C content as this is the case for Scottish soils (Jones et al., 2007; Dobbie et al., 1999; Buckingham et al., 2013).

In grazed pastures NH3 volatilises from urine patches, decomposing dung as well as from fertilisers containing urea and NH4+ (Twigg et al., 2011). Increased rates of NH3 losses have been associated with high stocking densities under a rotational grazing system by Ryden and Mc Neill (1984). In our study, N volatilised as NH3 and NOx from inorganic and organic fertiliser and animal excretion accounted for a considerable amount of total N, with losses of 13 % in cut and grazed years (2002, 2003) and 19 % in grazed-only years. In contrast, soil NOx emissions from our grassland were estimated to be negligible, accounting for less than 1 % of the total budget. Soil NOx emissions result predominantly from microbial nitrification of either added N fertilisers or following the mineralisation of soil organic matter, animal excretions or added manure. Emissions tend to be linked with aerobic soil conditions (Davidson, 1991).

Results from soil analysis taken in May 2004 and May 2011 indicate that our field has stored N (-4.51 ± 2.64 g N m-2 a-1). The N budget assessed from the net N flux balance showed that N was stored in the soil of our grassland over the same 7 years at a higher rate (-9.40 ± 4.14 g N m-2 a-1), although values were not significantly different from each other. The slight shifts in measurement periods (May 2004–May 2011) for the soil stock calculations and the period for flux budget calculations (January 2004–December 2010) is presumed to be insignificant in this comparison. Results from both methods are within the range of literature values. Neeteson and Hassink (1997) found an N accumulation in soil organic matter (SOM) of 0–25 g N m-2 a-1 from two cattle-grazed farms in the Netherlands, while Watson et al. (2007) reported an N storage in grazed Irish grasslands ranging from 10 to 15.2 g N m-2 a-1, depending on N inputs. Soil N storage assessed from soil measurements from a cut grassland close to our field, where plots were treated with cattle slurry, stored N over 6 years at a rate of -2.17 g N m-2 a-1 in the top 10 cm, while, in the same experiment, an N loss was observed from mineral N and urea treatments (4.5 and 8.3 g N m-1 a-1, respectively; Jones et al., 2007). In contrast, Schipper et al. (2007) reported an average loss of 9.1 g N m-2 a-1 in the top 100 cm from managed grasslands over 20 years in New Zealand. The reason for the small difference between methods (flux measurements vs. sequential soil sampling) in our study might lie in a possible underestimation of losses from flux measurements. Uncertainties of our estimates are high, especially those for N losses. The largest absolute systematic uncertainty for the N balance was attributed to N leaching as for most years values were modelled using data to evaluate the model from only one location. The uncertainty of the leaching estimate would therefore be reduced if the model could be validated with data measured from several locations. The second-highest systematic uncertainty was attributed to losses through N2, followed by NOx and NH3 emission from excretion and NOx and NH3 emission from inorganic fertilisation. Combined uncertainties from all components lead to a total systematic uncertainty in the N balance of 2.2 g N m-2 a-1 (2004–2010).

On an annual basis our grassland site was a sink for atmospheric CO2 in most years. NEE was only positive in 2004, which was likely to be due to a high livestock density. Generally, grazing causes a very gradual impact on the CO2 uptake as a part of the field is defoliated each day. The reduced leaf area index (LAI) then leads to a reduced CO2 uptake by plants. In addition to the reduced LAI, grazing presents a source of CO2 from animal respiration, thereby reducing the CO2 sink of the grassland within the field. The maximum uptake of CO2 measured in our study is close to the upper range of NEE reported for temperate grasslands (100 to 600 g C m-2 a-1, IPCC, 1996). On average over the 9 years, the magnitude of the NEE on our grassland (-218.0 ± 154.5 g C m-2 a-1) was close to the average NEE measured in a comparison of nine European grasslands over 2 years (240 ± 70 g C m-2 a-1) by Soussana et al. (2007) and comparable to the CO2 sink capacity of managed Irish grasslands measured by Byrne et al. (2007) (290 ± 50 g C m-2 a-1) and Leahy et al. (2004) (257 g C m-2 a-1). Despite high variability over the 9 years, the average NEE value was significantly different from zero (p < 0.05). The range of the calculated annual gross primary production (GPP; -982 to -2163 g C m-2 a-1) and terrestrial ecosystem respiration (TER; 972 to 2183 g C m-2 a-1) from our field was within reported values for other managed grasslands. Gilmanov et al. (2007) reported the GPP of 18 intensively managed European grasslands ranging from 467 to 1874 g C m-2 a-1 and TER ranging from 493 to 1541 g C m-2 a-1, while Mudge et al. (2011) reported values of 2000 g C m-2 a-1 for GPP and TER from an intensively grazed dairy pasture in New Zealand.

When all components of C import and export were included in addition to the CO2 exchange (NBP), C was stored in our grassland over the 9 years. However, due to the high variability between years, NBP was not significantly different from zero (p= 0.05), suggesting that our site is carbon neutral. The average C storage value on our site (163 ± 140 g C m-2 a-1) is higher than most estimates reported in the literature, but due to the high annual variation, still within the range of reported values; Soussana et al. (2007) reported C storage estimates from European grazed and cut grasslands of 104 ± 73 g C m-2 a-1, and Mudge et al. (2011) reported fluxes of 59 ± 56 and 90 ± 56 g C m-2 a-1 in 2 consecutive years for a grazed and cut grassland in New Zealand. NBP estimates from a Swiss grassland cut for silage was shown to sequester C at a rate of 147 ± 130 g C m-2 a-1 (Ammann et al., 2007), while estimates from a cut grassland in Germany were shown to vary from being a sink (-28 g C m-2 a-1) to being a source of C (+25 g C m-2 a-1), depending on years (Prescher et al., 2010). The inclusion of all C imports and exports leads to a weakening of the C sink strength assessed from NEE measurements in 5 years and even changed the grassland from being a sink to being a source in 2006. Due to the C export from harvest, as well as the stimulation of primary production through grazing, C sequestration tends to be lower in cut compared to grazed systems (Soussana et al., 2004). This is represented in our study in the lower residual value of NPP in cut years compared to the residual value from grazed-only years (Fig. 3), where the residual value represents the C storage in the soil as well as the uncertainty of the budget. However, it has to be kept in mind that the herbage yielded from cuts will end up as animal feed; C will be digested and respired off-site, releasing CO2 and CH4 to the atmosphere as well as being returned to the grassland as manure.

Results from soil analysis indicate that our grassland has lost C from 2004–2010 (29 ± 38 g C m-2 a-1, Table 6). In the literature, losses as well as storage of C at various rates have been reported for managed grasslands assessed from soil stock measurements. Soil stock change measurements from our field are comparable with values found in the literature. Depending on the study, managed grasslands in Belgium were shown to either lose (90 g C m-2 a-1, Lettens et al., 2005a) or sequester carbon (4.4 g C m-2 a-1 in 0–30 cm, Goidts and Van Wesemael, 2007; 22.5 g C m-2 a-1 in 0–30 cm, Lettens et al., 2005b). Schipper et al. (2007) reported losses of C from pastures in New Zealand at an average rate of 106 g C m-2 a-1 (top 100 cm) over 20 years, but these losses were a result of an earlier land use change from forestry. Schuman, et al. (2002) measured a C sequestration of 10–30 g C m-2 a-1 from US rangelands (0–60 cm), while Watson et al. (2007) measured a C storage at 112–145 g C m-2 a-1 in the top 15 cm soil layer from a grazed Irish grassland. Bellamy et al. (2005) showed no evidence of increased C in the topsoil of grasslands in England and Hopkins et al. (2009) found no significant change in SOC over time in two long-term UK experiments. The above-mentioned results are contrasting and inconclusive because observed C sinks in grasslands are the effect of land management or land use change prior to the beginning of the C stock change measurement. Soussana et al. (2014) concluded in a theoretical study that grassland SOC sequestration has a strong potential to partly mitigate the GHG balance of ruminant production systems at low grazing intensities but not with intensive systems. Smith (2014) examined evidence from repeated soil surveys, long-term grassland experiments and simple mass balance calculations and concluded that, although grasslands can act as C sinks, they cannot act as a perpetual C sink and thus could not be used as an offset for GHG emissions.

The comparison of the C storage calculated from the net C flux balance with soil C stock changes shows that the flux balance estimated a C sequestration, while, based on C stock changes, C was lost, although neither value was significantly different from zero (Table 6). A literature search by Soussana et al. (2010) showed that generally C sequestration calculations on grassland were lower if derived from SOC stock changes (average -5 ± 30 g C m-2 a-1) compared to C flux balances (average -22 ± 56 g C m-2 a-1), although these estimates were not significantly different from each other. However, in none of those reviewed studies were C flux and C stock change measured in the same field experiment. A reason for the discrepancy between estimation methods in our study might lie in a possible underestimation of C exports in the flux balance calculation, leading to an overestimation of C storage in the soil. One underestimated flux could be the export of DIC and DOC. Carbon leaching from managed grasslands has not been reported in many studies. Kindler et al. (2011) reported C leaching from various European ecosystems, where the measured data (2007) from our experimental field were part of the study. Our data (30.0 g C m-1 a-1, average of two locations as published in Kindler et al., 2011) were close to the average value (29.4 g C m-1 a-1) of the reported European grasslands, which showed a range of C losses of 6.5–42.5 g C m-1 a-1. Higher losses have been observed by McTiernan et al. (2001), who measured DOC export from grassland lysimeter plots treated with N fertiliser and slurry over 2 months. Up-scaled to 1 year, they measured DOC loss between 25.2 and 70.8 g C m-2 a-1, all above what we measured in our study. Important factors controlling the magnitude of C leaching have been shown to be drainage, the topsoil C / N ratio and the saturation of the subsoil's sorption capacity for organic C (Kindler et al., 2011; McTiernan et al., 2001). In waterlogged soils the SOM decomposition and groundwater recharge tend to be reduced and thus the amount of C prone to leaching compared to that under more aerobic conditions associated with drainage. Although our field was drained more than 50 years ago, the drainage system does not operate very well, resulting in large puddles of standing water during prolonged periods of rain. The measured data used for the budget were taken at one sampling point, which was not in a waterlogged area. The spatial heterogeneity within the grassland field caused by uneven water management as well as faeces and urine patches requires us to sample at more points in order to obtain a representative leaching value. Therefore our leaching estimates are highly uncertain and could be significantly lower and C exports overestimated. Furthermore, leaching was only measured in 1 year (2007), while values for remaining years were estimated using a simple regression model with an attributed high uncertainty of 32 % (5.3 g C m-2 a-1 of average fluxes). Indeed, Siemens (2003) hypothesised that the underestimation of C leaching from soils can explain a large part of the difference between atmosphere- and land-based estimates of the C uptake of European terrestrial ecosystems. Another underestimated flux could be the loss of CO2 in the NEE measurements. Gap filling can introduce uncertainties in the NEE data especially for years with low data capture. Furthermore, CO2 losses from animal respiration could be underestimated at times due to the animals moving out of the footprint of the EC mast. Using animal respiration values from chamber experiments of 12.1 for cows and 11.7 g CO2 kg-1 live weight d-1 for sheep and lambs (S. Troy, SRUC, personal communication, 2014), we estimated a maximum CO2 loss from animal respiration of 53 (2002–2010) or 59 g C m-2 a-1 (2004–2010). So if we assume that all animal respiration has been missed by eddy covariance measurements then the C sink estimated from NEE measurements would be reduced by 24 (2002–2010) or 33 % (2004–2010). This theoretical maximum 33 % reduction would reduce the net carbon balance to ∼ 122 g C m-2 a-1 (2004–2010).

In addition to uncertainties in the flux budget calculations, uncertainties are also attributed to soil C and N stock measurements. Soil inventory data in our study indicated a loss of C and a storage of N over 7 years, which seems contradictory, although C loss was not significantly different from zero. The uncertainty of soil C and N stock measurements arises from the variability of soil C and N concentrations due to errors from laboratory and to their high spatial variability as well as from the variability in the rock fragment content (Goidts et al., 2009).

In the overall N and C, budget N2O and CH4 emissions were negligible in terms of N and C losses from the system (1–8 % of total N losses and 0.6–4.5 % of total C losses). However, in terms of CO2 equivalents, N2O emissions as well as CH4 emissions strongly affected the GHG budget. Indeed, the sink strength of the NEE was weakened by N2O emissions by 29 % over all years. Methane emissions from soil processes, manure input and animal excretion were negligible in terms of the C budget as well as in the GHG budget. In contrast, enteric fermentation proved to be an important GHG source. Methane emissions were also measured by eddy covariance technique over several months in 2010 on the same field (Dengel et al., 2011). By dividing CH4 fluxes by the number of sheep in the field each day, Dengel et al. (2011) calculated CH4 emissions per head of livestock as 7.4 kg CH4 head-1 a-1 for sheep, which is close to the emission factor used in our budget of 7.6 kg CH4 head-1 a-1 for ewes, showing that our estimates were realistic. The positive correlation of CH4 emissions with the stock density indicates that any changes in animal production will have a major impact on the global CH4 budget. The weakening of the GHG sink strength of the NEE by N2O and CH4 emissions, show the importance of those two gases in terms of global warming. Thus, adapting the management of grasslands by adding fertiliser or manure to increase plant growth and thus improve C sequestration in soils may increase N2O emissions, while changing land use from cropland to pasture in the attempt to reduce C losses from soils might lead to increased CH4 losses from grazing animals.