Fluxes of N2O were measured from an area of intensively managed, grazed grassland (Easter Bush, Scotland, 55∘51′55.30′′ N, 3∘12′22.17′′ W) before and after a tillage event on 1 May 2012 and were compared with fluxes measured from an adjacent grassland which remained untilled (as described in Jones et al., 2011) (Fig. 1). The climate is temperate maritime, with an average annual rainfall of 921 mm and average annual air temperature of 9 ∘C (in the period 2001–2011). The two fields (each approximately 5.4 ha) have been managed for intensive livestock production for at least 20 years and since 2002 have been predominately grazed by sheep. The average stocking densities were 0.7 LSU ha-1 (livestock units) and average N fertiliser application rates have been approximately 200 kg N ha-1 yr-1. Mainly NH4NO3 or NPK compound fertilisers were applied in three split applications usually between March and July (Skiba et al., 2013).

The soil in the fields is a clay loam with a sand/silt/clay texture of 52/20/28 and 57/19/24 for the top 30 cm in the untilled and tilled fields, respectively, with a pH of approximately 5.1 (in H2O). They are classed as an imperfectly drained Macmerry soil of the Rowanhill association (Eutric Cambisol, FAO classification). A drainage system had been installed about 50 years prior to the tillage event, but is no longer functioning well, resulting in frequent occurrence of surface water during rainy periods. The fields had not been tilled for at least 20 years, and the farmer had reported reduced fertility and productivity. One field (also called the South Field in Jones et al., 2011) was therefore tilled in May 2012 (Table 1).

As standard practice, glycophosphate (1.5 L ha-1) was applied to kill the grass 3 days prior to ploughing on 27 April. The field was ploughed to a depth of 30 cm on 1 May 2012. Two days after ploughing, the field was harrowed, and then rolled and sown with ryegrass (Lolium perenne L.) on the third day after ploughing. The untilled field (also called the “North” field in Jones et al., 2011) was managed as usual and grazed by sheep (approximately 30 sheep ha-1). Fertilisation events continued as normal on the untilled field which received two ammonium nitrate (Nitram) fertiliser applications of 70 kg N ha-1, one on 28 May and the second on 9 August. The tilled field only received a 70 kg N ha-1 Nitram application on 9 August, approximately 4 months after the tillage event.

Biomass samples were collected from the South Field prior to tillage in order to estimate the grass biomass that would be tilled into the soil. Twenty soil cores (12 cm deep and 5.8 cm diameter) were extracted from the field. At these points, all above-ground biomass was harvested and dried in an oven at 80 ∘C to constant weight. Once dry, the above-ground biomass was weighed. The soil cores were broken up by hand and dried at 100 ∘C until constant weight. After drying, the root material was separated from the soil by hand and weighed. Subsamples of the dried plant materials were prepared for elemental analysis of total carbon and nitrogen contents (vario EL cube, Elemantar, Hanau, Germany).

Total (above- and below-ground) biomass on the tilled field before tillage averaged of 369 ± 310 g m-2, with a root-to-shoot ratio of ∼ 1.5. The nitrogen content was 2.5 %. Based on these measurements it is estimated that the tillage event added a total of 93.6 kg ha-1 of nitrogen to the field in the form of crop residues.

N2O fluxes were measured from both tilled and untilled fields over a 7-month period using three measurement methodologies: eddy covariance, static chamber and dynamic chamber techniques. The mixture of methods was used to try to obtain as many measurements as practically possible, both temporally and spatially, during the experiment. Eddy covariance was the primary measurement methodology used. However, due to unpredictable changes in wind direction at the site, it was necessary to deploy manual chamber methodology to ensure that both fields were measured periodically during the experiment. The dynamic chamber measurements were used as a cost-effective way to provide many (> 30) high-resolution N2O fluxes on the days immediately after tillage without the need for time-consuming gas chromatography lab analysis required by static chambers.

An eddy covariance system was installed on 27 March on the field boundary (Fig. 1). An ultra-sonic anemometer (WindMaster Pro 3-axis, Gill, Lymington, UK) mounted at 2.4 m was used to measure fluctuations in 3-D wind components at a frequency of 10 Hz. Mixing ratios of N2O, H2O and CO2 were measured at 10 Hz by a quantum cascade laser (QCL) gas analyser (CW-QC-TILDAS-76-CS, Aerodyne Research Inc., Billerica, MA, USA), housed in a temperature-controlled cabin. The inlet line to the QCL was a 13.5 m length of Dekabon tubing (0.25 in. outer diameter), with a flow rate of approximately 13 L min-1. Fluxes were calculated at 30 min intervals using the EddyPro software (version 5.2.1) (Li-Cor, Lincoln, NE, USA), based on the covariance between the N2O concentration (χ) and vertical wind speed (w): Fχ=χ′w′‾ In the processing, we applied double coordinate rotation (vertical and crosswind), spike removal, block averaging, and time lag removal by covariance maximisation. Corrections for the frequency response of the system, both high and low-frequency losses, were made using the method of Moncrieff et al. (1997). Corrections for density fluctuations were applied on a half-hourly basis using the method of (Burba et al., 2012). The quality control scheme of Foken et al. (2005), was used to remove poor-quality flux measurements (their category 2). Initially, fluxes measured with a mean wind direction between 180 and 270∘ from north were classed as from the tilled field; those measured at greater than 330 and less than 100∘ were classed as from the untilled field. The remaining data were disregarded due to obstruction of the wind by the cabin and fence line.

Further footprint analysis was carried out in which we visually checked individual footprint plots of each 30 min flux (Fig. 2). Any flux footprints in which the majority of the contribution came from a distance less than 10 m from the mast or overlapped the two fields were removed from the dataset. Standard meteorological variables (rainfall, air temperature and soil temperature) were recorded by a tipping bucket, thermometers (2 m height and 10 cm depth) and time-domain reflectometry soil moisture probe at 10 cm depth. These measurements were made adjacent to the flux tower at the site.

N2O fluxes were also measured from both fields using static chamber and dynamic chamber techniques. The static chambers consisted of a cylindrical polyvinyl chloride (PVC) plastic pipe of 38 cm inner diameter (ID) and 22 cm height. These chambers were inserted 5 cm into the soil, giving a headspace of approximately 20.4 L. Chambers were closed for 40 min, during which time three 100 mL gas samples were collected via a syringe and a three-way tap fitted to the lid, at t= 0, 20 and 40 min. After each measurement, chamber height was measured at five points to estimate the chamber volume. Gas samples were stored in 20 mL glass vials which were flushed with 100 mL of air in the syringe using a double needle. Samples were analysed using a Hewlett Packard 5890 series II gas chromatograph (Agilent Technologies, Stockport, fitted with an electron capture detector) (Skiba et al., 2013).

Ten static chambers were positioned in each of the fields, within the estimated flux footprint of the eddy covariance system (10 to 200 m from the mast). Chambers in the fields were occasionally moved to prevent the effects of a micro-climate within the chambers that could bias measurements when compared to the surrounding field area, and also to allow access to farm vehicles during the different stages of the tillage operation. Manual chamber measurements were carried out between 09:00 and 15:00 GMT on the measurement dates. Fluxes were calculated as F=dCdt⋅ρVA, where F is the gas flux from the soil (nmol m-2 s-1), dC/dt is the rate of change in concentration with time in nmol mol-1 s-1 estimated by linear regression, ρ is the density of air in mol m-3, V is the volume of the chamber in cubic metres and A is the ground area enclosed by the chamber in square metres. Static chamber measurements were made over a longer period than shown in this paper and are discussed in relation to a second tillage event by Drewer et al. (2016).

Fluxes were also measured using the QCL in a closed, dynamic chamber system (Cowan et al., 2014a). A chamber (39 cm inner diameter, 22 cm high) was placed onto a stainless steel collar inserted several centimetres into the soil (on average 5 cm) at least 15 min prior to measurement. Two 30 m lengths of 3/8 in. ID Tygon^® tubing connected the chamber to the inlet of the QCL and the outlet of a vacuum pump (SH-110, Varian Inc., CA, USA) to form a closed system. This allowed a 30 m possible radius from the instrument cabin in which the chamber could be placed (Fig. 1). A flow rate of approximately 6 to 7 L min-1 was used, with a lag time of approximately 22 s between the chamber and analyser. Fluxes of N2O were calculated with 1 Hz data over 3 min, using both linear and non-linear asymptotic regression methods (Levy et al., 2011; Pedersen et al., 2010). Using a mixture of goodness-of-fit statistics and visual inspection, the regression method that provided the best fit for the time series of mixing ratios of N2O was chosen for each individual measurement. The detection limit of individual fluxes calculated by this method was approximately 0.04 compared to 0.4 nmol m-2 s-1 when using the static chambers (Cowan et al., 2014a, b).

In the first few days after the tillage event, the wind direction was north-easterly, meaning that the eddy covariance system could not record fluxes from the tilled field (to the south-west). The dynamic chamber measurements were primarily used to fill this gap in the eddy covariance time series with high-precision chamber measurements.

Because the eddy covariance system was placed on the field boundary, observations could only be made on a single field at any given time. Furthermore, some data were missing because of instrument failure and some had to be rejected according to the quality control criteria used. In order to estimate cumulative fluxes from both fields, temporal interpolation of the missing data points was required. However, in the absence of a well-validated process-based model for N2O fluxes on which to base predictions, it is not obvious how this is best achieved. The most common approach is to linearly interpolate in time between flux measurements. In this study, a general additive model (GAM) was used as an alternative approach, which accounted for temporal patterns at a range of timescales and nonlinear responses to environmental variables, implemented using the mgcv package in the R software (Wood, 2006).

Fluxes measured by eddy covariance and both chamber methods from the tilled and untilled fields were fitted to two separate GAMs using the same environmental terms for both fields. The environmental terms included were air temperature, soil temperature, precipitation, and time. Additional terms for temperature and precipitation aggregated over longer intervals (1, 6, 12, 24 and 48 h preceding the flux measurement) were examined and included where they improved the fit. The GAM allows for non-linearity by fitting a smooth response with cubic splines. The degree of smoothing is optimised by the algorithm but was also adjusted subjectively, such that the model was not overfitting to noise in the data. Observations from eddy covariance and the two chamber methods were given equal weighting. Predictions from the GAM were used to fill gaps when observations were not available. Uncertainty in predictions was estimated by simulating 2000 replicate time series from the GAM, using the uncertainty in the fitted parameters, to estimate the posterior distribution. The quantiles of this posterior distribution provided the 95 % credibility interval at each predicted 30 min interval time step. To calculate cumulative fluxes, observed fluxes were used with their associated uncertainties (Finkelstein and Sims, 2001) when available; otherwise the GAM predictions were used.

A total of 1191 mm of rain was recorded in 2012, higher than the average annual rainfall of 921 mm (2001 to 2011) for the Easter Bush area (Fig. 3a). The annual variation in temperature was fairly typical of the field site (Fig. 3b). The wind direction at the field site is predominantly south-westerly (85 %). However, during the measurement campaign, the wind direction was split fairly evenly between the tilled and untilled fields (Fig. 4). This allowed a better basis for comparison of N2O fluxes from the two fields, although data coverage for each field was low, 34 and 24 % for tilled and untilled, respectively.

Before the tillage event, N2O fluxes were similar in the tilled and untilled fields. In both cases, around 90 % of measured fluxes were below 0.5 nmol m-2 s-1 (Fig. 5). All three fertilisation events (the two fertiliser events in the untilled filed and single fertiliser event in the tilled field) were characterised by an emission peak of 5–10 nmol m-2 s-1 lasting a few days, which declined over the following days and weeks, often with considerable variability and some apparent secondary peaks (Fig. 5). Fluxes had returned to background levels (< 0.5 nmol m-2 s-1) within 28 days of each of the fertilisation events. Fluxes measured by all methods agreed reasonably well in magnitude, and there is no strong evidence for a systematic bias, given the differences in the spatial and temporal sampling (for a more specific insight see, e.g., Cowan et al., 2014a).

The tillage event also produced an increase in emissions, and although the peak was less clearly defined, the effect was more prolonged. Fluxes generally ranged from ∼ 0 to 1.0 nmol m-2 s-1 in the days before tillage and ∼ 0 to 8.8 nmol m-2 s-1 in the week immediately after tillage (Fig. 5b). Three exceptionally high individual chamber measurements measured in the days immediately after the second fertilisation event in the untilled field which are included in the data analysis (19.5, 34.8 and 50 nmol m-2 s-1) are not included in Figs. 5 or 7 in order to keep the scale manageable. Fluxes from the tilled field from mid- to late May were approximately 1 nmol m-2 s1 higher than from the untilled field (before the latter was fertilised). There followed an apparent increase in N2O fluxes lasting approximately 4 weeks from the tilled field from late May to late June, peaking mid-June (Fig. 5b). Unfortunately, data coverage was rather low during this period due to changes in wind direction and a 5-day period in which the QCL was not operational. Because the tilled field had not been fertilised since the previous year, we infer that the increased fluxes were a result of the tillage event. Fluxes in the tilled field returned to pre-tillage magnitude during July. By July, a new sward of grass had grown in the tilled field, but sheep were not re-introduced into the field until September.

The relatively high N2O fluxes measured from the tilled field in the weeks after tillage (May to July) occur in a similar timeframe to the fertilisation event in the untilled field (Fig. 5). Beyond the analytical footprint analysis, we wanted to check that the high N2O fluxes, which we attribute to the tillage, actually do come from the tilled field and are not influenced by N2O from fertilisation events on surrounding fields. The CO2 fluxes (measured by QCL instrument) provide a suitable tracer. We know that no significant photosynthesis took place on the tilled field between 1 May and 17 June, as there was no green foliage visible until after this period. Therefore, if the CO2 fluxes showed no daytime uptake on the tilled field, we can be reasonably certain that the measured N2O fluxes were also coming from the tilled field. Figure 6 shows that this was the case: in fluxes attributed to the tilled field, there was no daytime uptake of CO2; in fluxes attributed to the untilled field, the normal diurnal cycle in CO2 flux is seen. By inference, we can attribute the high N2O emissions after tillage to the tilled field.

The GAM method was used to gap-fill flux data to calculate cumulative fluxes for both fields separately using the fluxes measured from each (Fig. 7). The total numbers of individual eddy covariance, dynamic chamber and static chamber flux measurements used to fit the GAMs were 1563 : 273 : 234 and 1153 : 56 : 221 for the tilled and untilled fields, respectively. Cumulative N2O fluxes calculated for the tilled and untilled fields from 1 April to 16 September were 2.14 ± 0.18 and 1.65 ± 1.02 kg N2O-N ha-1, respectively (Fig. 8). Uncertainty in the GAM prediction is particularly large when no measurements are available in which to fit the model. There are sustained periods in which very few eddy covariance measurements were recorded from the untilled field due to the wind direction being predominantly south-westerly (Fig. 7a). The uncertainty in predicted flux becomes very large when compared to periods when measurement data are available, and these uncertainties propagate significantly in cumulative flux estimates (Fig. 8).

The comparison of pre-tillage and post-tillage fluxes from the tilled field suggests that the tillage event was directly responsible for an immediate increase in N2O fluxes (Figs. 5 and 8). N2O fluxes significantly larger than those measured pre-tillage were observed from the tilled field over two separate periods during which no changes in N2O fluxes were observed in the adjacent untilled field. The initial increase in N2O flux from the tilled field is a short-lived peak which occurs directly after the disturbance of the soil caused by ploughing and harrowing. The second is a sustained increase which is observed throughout May and June. In the 2-month period in which fluxes from the tilled field were elevated, a total of 1.26 ± 0.12 kg N2O-N ha-1 was estimated to have been released. Assuming fluxes in the tilled field had remained at approximately pre-tillage magnitude, had the tillage event not taken place (∼ 0.27 nmol m-2 s-1, based on an average of flux measurements before the tillage event), it can be concluded that the tillage event contributed to an additional 0.85 ± 0.11 kg N2O-N ha-1 emitted from the field over a 2-month period.

Increases in N2O flux lasting up to 2 months after grassland tillage events have been observed before in other studies using both static chamber and eddy covariance measurements (Chatskikh and Olesen, 2007; Merbold et al., 2014). Reported fluxes can be relatively high over a sustained period of time (several days or weeks) and similar in magnitude to those recorded after fertilisation events. The mechanisms driving these large sustained fluxes are believed to be partly due to the mineralisation of organic materials in the soils (decaying grass materials from the previous sward in tilled grasslands) (Baggs et al., 2003; Hellebrand, 1998; Pimentel et al., 2015). The large quantities of decaying organic matter ploughed into the soils would have provided a gradual release of carbon and nitrogen into the soils, which provide substrate for the microbial processes of nitrification and denitrification (Pimentel et al., 2015; Seastedt et al., 1992). According to IPCC estimates, 1 % of N added to soils in the form of crop residues can be expected to be released as N2O (IPCC, 2006). Based on our pre-tillage biomass measurements made prior to tillage (93.6 kg N ha-1),we would expect to see N2O fluxes of approximately 0.94 kg N2O-N ha-1 from the field. This estimated value is within the range of uncertainty of our calculated cumulative fluxes in this study (0.85 ± 0.11 kg N2O-N ha-1). High emissions from crop residues tilled into arable crops have been recorded in similar wet soils with high clay content (Ball, 1999) which may indicate a similar process is occurring under these conditions at other field sites in the area.

Large N2O fluxes (> 0.5 nmol m-2 s-1) are observed from both fields after fertilisation events. Elevated fluxes recorded from the fields after fertilisation typically last 3 to 4 weeks with an occasional large spike lasting 24 to 48 h before returning to pre-fertilisation levels. This month-long period in which the majority of large fluxes occur after fertilisation is also generally observed by other similar studies from the local area (Skiba et al., 2013; Smith et al., 2012). Assuming the majority of N2O emitted after a fertilisation event occurs within a 28-day period after the fertiliser application, the 28-day cumulative flux emissions associated with the fertilisation events on 28 May and 9 August on the untilled field were 0.55 ± 0.05 and 0.76 ± 0.24 kg N2O-N ha-1, respectively. This equates to 0.79 and 1.09 % of the total nitrogen applied, respectively. The 28-day cumulative flux emissions associated with the fertilisation event on the tilled field was 0.77 ± 0.34 kg N2O-N ha-1, or 1.10 % of the total nitrogen applied. Assuming the 28-day periods account well for the emission factors of the fertiliser events, these results are well within the range of uncertainty of the generic 1 (0.3 to 3.0) % value reported by the IPCC for N fertiliser events (IPCC, 2014).

Gap-filling N2O flux measurements is challenging due to the lack of reliable process-based models on which to base predictions. N2O fluxes are believed to be driven primarily by the availability of nitrogen compounds in the soils (ammonium and nitrate) (Davidson et al., 2000) as well as physical properties of the soil such as WFPS, aerobic extent, soil type, temperature and compaction (Ball et al., 2008; Butterbach-Bahl et al., 2013; Choudhary et al., 2002; Davidson et al., 2000; Turner et al., 2008). The collection of these data on a temporal/spatial scale which would allow these models to be applied is not often logistically possible or affordable. The GAM method used in this study incorporates readily available meteorological data with the temporal pattern in the data in order to provide an empirical but practical means of temporal interpolation, which makes use of more information than simple linear interpolation. Although the GAM method has proved useful, we would also emphasise the dangers of extrapolating to conditions beyond those to which the model was fitted. For example, as we have not measured fluxes during the cold months in winter, the GAM is unable to reliably predict fluxes in temperatures lower than those measured during the study. The method deals appropriately with the large uncertainties where measurement data are unavailable, contributing considerably to the total uncertainty in cumulative flux estimates.

In this study, spatial variability was not explicitly accounted for in the cumulative flux uncertainty, and this remains a potentially large error if extrapolating to areas larger than the measurement footprint. Eddy covariance is able to integrate over a large area of the field (several hundred square metres) (Eugster and Merbold, 2015), but these measurements are still subject to an element of spatial variability which is difficult to fully account for given the spatially heterogeneous nature of N2O fluxes. Any study which plans to report cumulative flux estimates should consider how to minimise the uncertainties which arise when interpolating and/or extrapolating measurements to larger temporal and spatial scales (e.g. from occasional chamber measurements to annual field-scale emissions). Further studies may require more complex statistical analysis, using methods such as Bayesian statistics, to properly quantify the uncertainty in estimates of cumulative fluxes over large areas.