Ammonia fluxes in relation to cutting and fertilization of an intensively managed grassland derived from an inter-comparison of gradient measurements

Ammonia fluxes in relation to cutting and fertilization of an intensively managed grassland derived from an inter-comparison of gradient measurements C. Milford, M. R. Theobald, E. Nemitz, K. J. Hargreaves, L. Horvath, J. Raso, U. Dämmgen, A. Neftel, S. K.Jones, A. Hensen, B. Loubet, P. Cellier, and M. A. Sutton Centre for Ecology and Hydrology (Edinburgh Research Station), Bush Estate, Penicuik, Midlothian, EH26 0QB, UK Institute of Atmospheric Physics, P.O. Box 39, 1675, Budapest, Hungary Institut fur Agrarokologie, Bundesforschungsanstalt fur Landwirtschaft (FAL), Bundesallee 50, 38 116 Braunschweig, Germany Agroscope Reckenholz-Tänikon Research Station ART, Reckenholzstrasse 191, 8046 Zürich, Switzerland

poorer agreement on particular days.However, by using four systems, the inter-comparison was able to provide a robust mean estimate of continuous ammonia fluxes through the experiment.The observed fluxes were: a) small bi-directional fluxes prior to cutting (−64 to 42 ng NH 3 m −2 s −1 ), b) larger diurnally-varying emissions following cutting (−49 to 703 ng NH 3 m −2 s −1 ) and c) much larger emissions following fertilizer application (0 to 3820 ng NH 3 m −2 s −1 ).The results are a salutary reminder of the uncertainty in unreplicated ammonia flux measurements, while the replication of the present study provides a uniquely robust dataset for the evaluation of ammonia exchange processes.It is clear that consistently reliable determination of ammonia concentrations remains the major measurement challenge.al., 2004), including grasslands (e.g.Horvath 1982;Harper et al., 1983;Horvath et al., 2005).Initially, the motivation for measurement of ammonia was to quantify N losses from agricultural systems.The first measurements used techniques such as bubblers/acid traps (Denmead et al., 1974), filter packs (Harrison and Kitto, 1990) or denuder tubes (e.g.Ferm 1979).All these methods operated at a time resolution of at least 1 h and required significant operator effort for changing over samples and off-line analysis.As a consequence, most early measurements were for short fieldcampaigns only, of a few days or weeks (e.g.Sutton et al., 1993b;Duyzer 1994).
As interest in ammonia (NH 3 ) as an atmospheric pollutant has grown, so too has the range of measurement techniques available.Developments in continuous measurementtechniques enabled much longer time periods of measurements to be conducted.In particular, the continuous flow denuder (AMANDA) developed by Wyers et al. (1993), achieved the balance between reasonable cost, a wide operating concentration range (0.02-100 µg m −3 ) and on-line analysis with short time resolution (<5 min).The AMANDA utilises wet chemistry techniques, stripping ammonia from the air in an acidic capture solution.Other similar instruments have been developed which also operate via wet chemistry methods, for example the Wet Effluent Diffusion Denuder (WEDD) technique (Vecera and Dasgupta, 1991), as well as recent more automated implementations of the AMANDA technique (Trebs et al., 2006;Kruit et al., 2007).Although these techniques have been proven to measure NH 3 successfully, there are limitations; in particular they cannot be used for eddy correlation measurements where a sensor with a time response of typically at least 4 Hz is needed.Shaw et al. (1998), Famulari et al. (2004) and Whitehead et al. (2008) have reported eddy correlation measurements of NH 3 fluxes, using a tandem mass spectrometer and tunable diode laser technology, but substantial further developments (and cost reductions) are required before eddy correlation becomes more widely used for NH 3 flux measurement.
An alternative micrometeorological flux measurement technique is the relaxed eddy accumulation method (REA).This method has the advantage of not requiring a fast response analyser.REA flux measurements of NH 3 have been conducted in recent years (e.g.Neftel et al., 1999;Nemitz et al., 2001a;Meyers et al., 2006) and constitute an area of ongoing research.
Ammonia has been recognised as contributing to eutrophication and acidification of ecosystems (Fangmeier et al., 1994;Krupa et al., 2003).To assess these effects, reliable models of ammonia deposition are needed (Hertel et al., 2006).Currently, there are only crude parameterisations of NH 3 exchange present in national and European deposition models and measurements are needed to improve these parameterisations.A European project GRAMINAE (GRassland AMmonia INteractions Across Europe) was therefore initiated to improve quantification and parameterisation of NH 3 exchange with grasslands across Europe (Sutton et al., 2001a(Sutton et al., , 2007)).
In addition to the interest in contrasting ecosystem types (e.g.Horvath et al., 2005;Walker et al., 2006;Kugler et al., 2008), large rates of NH 3 emission have been observed following cutting of intensively managed grassland, from the sward itself (Sutton et al., 1997(Sutton et al., , 2001a;;Milford et al., 2001b), with this effect being modelled by Riedo et al. (2002).There is, however, a shortage of studies investigating this emission source, while Bussink et al. (1996) observed the opposite, i.e. an apparent increase in deposition flux immediately after cutting.Emission of NH 3 from senescing and decomposing vegetation has also been investigated (Whitehead and Lockyer, 1989;Mannheim, 1997), but this is rather distinct from the enhanced emissions from a regrowing canopy following cutting noted above.Emission of volatile organic compounds (VOCs) has been detected from cut grassland and attributed to a "wounding" of the vegetation (Gouw et al., 1999).Other work has investigated the effect of cutting regimes on the health, yield and species diversity of swards (Smith et al., 1996a, b;Blum et al., 1997;Evans et al., 1998), but not the effect on NH 3 exchange.Much research has been conducted on NH 3 volatilisation following N fertilization applied to grassland (e.g.Pain et al., 1989;Thompson et al., 1990aThompson et al., , b, 1991;;van der Weerden and Jarvis, 1997;Sommer et al., 2004), but little research has been conducted to quantify the contributions from direct fertilizer emission of NH 3 and indirect NH 3 emission from the plants themselves.
To address some of these issues, the GRAMINAE Integrated Experiment was held in Braunschweig, Germany, over intensively managed grassland (Sutton et al., 2009b).In order to provide a robust dataset of NH 3 exchange with the vegetation, four independent continuous flux gradient systems were operated.Although there have been many intercomparisons of ammonia concentration measurements (e.g.Gras, 1984;Appel et al., 1988;Harrison and Kitto, 1990;Wiebe et al., 1990;Sutton et al., 2001b;Schwab et al., 2007) there have been much fewer inter-comparisons of ammonia flux measurements (Sutton et al., 2000;Whitehead et al., 2008).This paper presents the results of a major intercomparison of continuous NH 3 flux measurements over intensively managed grassland.The best estimates of NH 3 flux and concentration from this inter-comparison are subsequently used to investigate the effect of cutting and fertilizing on NH 3 exchange fluxes.These best estimates of NH 3 concentrations and fluxes were also provided to other participants in the experiment as a basis for further analyses, for example in quantifying advection fluxes (Loubet et al., 2009), assessing the relaxed eddy accumulation technique (Hensen et al., 2008), modelling the dynamics of ammonia fluxes (Burkhardt et al., 2009;Personne et al., 2009;Sutton et al., 2009a) and quantifying production of particulate ammonium aerosol (Nemitz et al., 2009a).

Micrometeorological theory
The NH 3 flux measurements were made using the aerodynamic gradient method, following the approaches described in Fowler and Duyzer (1989), Monteith and Unsworth (1990) and Sutton et al. (1993).It should be noted that recent research (Flesch et al., 2002;McNaughton, 2006) has questioned the validity and accuracy of the aerodynamic gradient method.The flux (F z ) is calculated from the friction velocity (u * ) and concentration scaling parameter (χ * ): with emission fluxes denoted as being positive.χ * is calculated as: where k is von Karman's constant (=0.41), z is height above the surface, d is zero plane displacement, χ is NH 3 concentration and H is the integrated stability correction function for scalar properties, calculated from the Monin-Obukhov length (L) according to the description of Sutton et al. (1993).Although u * can be calculated from profiles of windspeed in an equation analogous to Eq. ( 2), in this instance it was obtained from ultrasonic anemometry using eddy covariance (Moncrieff et al., 1997).u * and all other meteorological parameters were measured by several institutes in the experiment.These estimates have been compared and evaluated by Nemitz et al. (2009b), who provided a consensus dataset for use in the flux calculations here.
The basic calculations of NH 3 fluxes derived here assume stationarity and homogeneity of the atmospheric conditions, such that the vertical flux results entirely from turbulent transfer in the vertical and that errors related to local advection (horizontal heterogeneity), storage (temporal instationarity) and chemical production or consumption in the surface layer are ignored.Strictly, this assumption cannot always be maintained, with the result that the vertical flux is not constant with height: In the case of ammonia, where exchange fluxes are relatively large in proportion to concentration, the flux divergence due to storage (∂χ a /∂t) can generally be ignored as trivial (Sutton et al., 1993a).However, the resulting difference in the vertical flux due to storage (the storage error, F z,sto ) may be found from: which is approximately equal to: The horizontal flux divergence ∂F x /∂x may be more significant for NH 3 , particularly where large NH 3 sources exist in the vicinity, causing large local horizontal concentration gradients in the direction of the wind (∂χ a /∂x) (Loubet et al., 2001(Loubet et al., , 2006;;Milford et al., 2001a).Estimates of the difference in the vertical flux due to advection (the advection error, F z,adv ) are usually not made, due to lack of information on ∂χ a /∂x.This issue was, however, given special attention in the Braunschweig Experiment, with both measurements and modelling of ∂χ a /∂x being conducted (Loubet et al., 2009).On this basis, where relevant, derived F z,adv were applied to correct the measured fluxes at 1 m (F z (1 m)) and provide estimates of the fluxes at the canopy surface (F z (z o )).
Chemical production or consumption (Q chem ) may also be significant for ammonia where either evaporation of ammonium containing aerosol, or production of aerosol, respectively, occur in the surface layer.The difference in the vertical flux due to these effects ( F z,che ) is assessed separately by Nemitz et al., 2009a.

Ammonia measurement techniques and implementation
Ammonia concentrations were determined in gradient configuration at four locations using two different measurement techniques.At three out of the four sites the continuous flow wet denuder system "AMANDA" (Ammonia Measurement by ANnular Denuder sampling with online Analysis) was deployed (Wyers et al., 1993).This technique captures gaseous ammonia in a continuous-flow horizontal annular denuder using a stripping solution of 3.6 mM sodium hydrogen sulphate (NaHSO 4 ) and determines the aqueous ammonium concentration online by conductivity analysis.The time resolution of this method can be set to 1 min, however, in this instance, concentrations were measured sequentially for 150 s at each of three heights (including liquid-flow delay loops), resulting in a full profile measurement every 450 s.These concentrations were averaged to 15 min periods for flux calculation.The air-flow rate of the AMANDAs was approximately 25 l min −1 and the liquid flow rate through the denuders was approximately 1 ml min −1 ; the detection limit was about 0.02 µg NH 3 m −3 .The heights of the concentration measurements were varied throughout the measurement campaign according to the canopy height, but the maximum height and minimum heights above ground were 2.37 m and 0.32 m respectively.A second continuous ammonia measurement technique utilised miniaturised Wet Effluent Denuders (mini-WEDD), which are silica-coated glass tubes (length 125 mm) positioned vertically, with continuous flow of a stripping solution in a membrane tube analysed online by a four-channel fluorescent analyser (e.g.Neftel et al., 1998;Vecera and Dasgupta, 1991).The mini-WEDDs were maintained by ART the former Swiss Federal Research Station for Agroecology and Agriculture (FAL-CH) and placed at four heights (0.15 m, 0.3 m, 0.6 m and 1.2 m above ground).However, the lowest mini-WEDD concentration was not used in the flux calculations as it was judged to be too close to the canopy.An air-flow rate of 600 ml min −1 and a liquid flow rate of 0.12 ml min −1 were used.The detection limit was 0.1 µg NH 3 m −3 .Calibration of both systems was conducted with aqueous standards of 0, 50 and 500 µg kg −1 NH + 4 .A summary of the different NH 3 measurement systems applied and acronyms used in the intercomparison is provided in Table 1.
During the experiment, aqueous calibration standards were prepared centrally and distributed amongst the ammonia analyser operators.In addition, unknown quality control standards were distributed on 3 occasions to test the accuracy of the analysers.On each occasion, 2 unknown standards were distributed and the difference in concentration measured by each analyser was compared with the actual concentration difference.

Field site and instrument locations
A full site description and diagram are given in Sutton et al. (2009b).The field site was intensively managed grassland of approximately 12 ha and the principal micrometeorological measurement location (Site 1) was 380 m from the western edge of the field, this being the main wind direction during the experiment.There was a second micrometeorological measurement location (Site 2), which was 210 m east of Site 1 and 36 m from the eastern edge of the field.Measurements were made at Site 2 in order that any advection of NH 3 emitted from a farm located 610 m west of Site 1 (Hensen et al., 2009) could be identified and quantified (see Sect. 2.5).Two gradient AMANDAs (each consisting of 3 denuder inlets linked to a common ammonium detector) were deployed at Site 1 (FAL-D, FRI), as well as the mini-WEDD system (FAL-CH), while one gradient AMANDA (CEH) was deployed at Site 2. The field was cut for silage on the morning of 29 May 2000 (starting at 06:00 GMT), with grass removed from the field on the morning of 31 May 2000.The field was fertilized with 108 kg N ha −1 calcium ammonium nitrate on the morning of 5 June 2000 (06:00-07:00 GMT).

Data processing procedures
The complexities of data processing are increased by the availability of the 4 independent estimates of NH 3 concentration profiles.The following procedure was applied: i) Any periods of calibration or obvious malfunctioning of each instrument were removed from the dataset of measured concentrations.
ii) Fluxes F z (1 m) of NH 3 and concentrations at 1 m (χ(1 m)) were calculated according to Sect.2.1.Flux measurements were rejected during periods when the fetch was obstructed by other equipment or by the edge of the field (see below for details).
iii) Once F z (1 m) and χ(1 m) estimates were available for the different systems, these were compared to identify any further periods where one system had malfunctioned or underperformed which had not yet been identified.These data were then removed.
iv) Given the different estimates of the 4 systems, gaps in the data create an artificial change in the mean estimate when one system goes offline or comes back online.To avoid this artefact, gaps of <6 h in each instrument were filled for F z (1 m) and χ(1 m) according to the technique described below.
v) As the flux measurements were made at two sites, different vertical flux divergence will apply due to horizontal advection where local sources are present.The advection corrections for 1 m above d were calculated by  Loubet et al. (2009) for the Site 1 and 2 measurements and applied to the measured fluxes, resulting in an estimate of the fluxes at the canopy surface (F z (z o )) (see Sect. 3.3).No corrections to the data for the potential effect of chemical production or consumption are made here, as this effect is assessed by Nemitz et al. (2009a).
vi) The "mean gradient estimate" (subscript mg) was calculated for (F z (z o )) and χ (1 m), as the arithmetic mean of all the remaining individual measurements.At the same stage, corrections for storage errors were also applied.
vii) The data were filtered according to the passing or failing of a set of defined micrometeorological criteria (see below).Data failing these criteria were retained in the dataset, but distinguished as being of lower reliability.
There was a substantial amount of measurement equipment at Site 1, concentrated on a N-S axis, as well as three mobile laboratories in the N direction (see site description in Sutton et al., 2009b).As a result, flux measurements from the FAL-D and FRI gradient systems were rejected for wind directions from both the N and S direction (0 • to 20 • and 180 • to 190 • rejected), whilst flux measurements were rejected for winds from the NNE direction (10 • to 45 • ) for the FAL-CH gradient system.Due to the close proximity of the edge of the field to the east of Site 2, flux measurements were rejected at this site for wind direction 0 • to 170 • .
The gapfilling technique applied for each instrument involved calculating the ratio of the individual flux measurement to the mean estimate at the start and the end of the gap and then interpolating this ratio.This interpolated ratio was then multiplied by the available mean estimate to fill in missing data.This method propagates the deviations from the mean present at the start and end of the gap, and limits the occurrence of step changes in the flux when individual analysers fail or are restored in the dataset.Only gaps of <6 h were filled.
Finally, the data were filtered according to micrometeorological criteria to identify periods where the fluxes are estimated with less certainty.These micrometeorological criteria were: u (1 m) <0.8 m s −1 , |L|<5 m and cumulative normalised footprint function (CNF) <67%.The cumulative normalised footprint function was calculated using the Kormann and Meixner (2001) formulation as described in Nemitz et al. (2009b).

Data processing
The data processing procedures ensured that high data coverage at individual sites was achieved even after periods of malfunctioning and obstructed wind sectors were removed (Table 2).Having four estimates of the flux lead to a "mean gradient estimate" of the flux with an overall data coverage of 98%.This highlights the advantage of having a number of independent systems to achieve a robust estimate of the flux.windspeed, stable conditions) were the main limitation to obtaining a complete flux dataset.Data on the gapfilling procedure revealed that FAL-CH had the greatest number of gaps filled and that the median gap length varied from 15 min (FAL-D) to 90 min (CEH).

Temporal inter-comparison of gradient measurements
The range of NH 3 concentrations at 1 m and fluxes from the 4 different systems are shown for example days from the precutting, post-cutting and post-fertilizing periods (Figs. 1 and  2).These figures show that there were periods of close agreement (e.g. 6 and 7 June 2000) and periods of substantial divergence (e.g. 8 June 2000).It can be seen that for certain periods (31 May to 2 June 2000) there were consistent concentration differences between the different instruments with FAL-D generally reading higher concentrations than CEH and FAL-CH.
The blind testing of the aqueous ammonium Quality Control standards by the different analysers did indicate periods of significant concentration differences (Table 3).In particular, the result of FAL-D over-reading by 41% on 31 May 2000 is consistent with Fig. 1b.However, the small number of aqueous quality tests meant that it was not possible to adjust the concentrations in an objective manner and so the quality tests were used for interpretation rather than adjustment.
The flux intercomparison highlights the changing pattern of NH 3 exchange during the experiment; before the cutting of the grass the flux was predominantly deposition to the surface.After cutting of the grass, the NH 3 exchange changed to predominantly emission with emission fluxes of up to 760 ng m −2 s −1 , whilst after fertilization the fluxes according to individual analyzers increased up to 6000 ng m −2 s −1 .The emission fluxes peaked in the daytime and were generally close to zero during nighttime.As with the comparison of ammonia air concentrations, there were periods of close agreement (e.g. 6 and 7 June 2000) and periods of substantial disagreement (e.g. 1 and 8 June 2000).However, gen-

Assessment of advection corrections
Estimates of the difference in the vertical flux due to advection (the advection error, F z,adv ) were derived from both measurements and modelling by Loubet et al. (2009).The modelled estimates of F z,adv were applied to correct the flux measurements (F z (1 m)) in order to provide an estimate of the fluxes at the canopy surface (F z (z o )).Examples of the magnitude of the advection errors in relation to the flux measurements for particular periods are given in Loubet et al. (2009).The modelled advection errors during the pre-cut period due to the farm 610 m from Site 1 estimated by Loubet et al. (2009), ranged between 0 to 27 ng m −2 s −1 .However, relatively few periods of advection from the farm were observed due to winds occurring directly from the W occurring for only a small period of the time, and accounted on average for 32% of the measured flux.As well as advection due to the farm emissions, advection errors also occurred due to NH 3 emissions from the field itself, which were largest following fertilization of the field.Although these were larger in absolute terms (−209 to 13 ng m −2 s −1 ), they represented a smaller percentage of the measured fluxes at 1 to 2%.For the week following 29 May 2000 (post-cut period), the farm and field advection errors amounted on average to +3% and −4% of the measured fluxes, respectively (Loubet et al., 2009).

Comparisons of χ(1 m) and F z (z o ) against FAL-CH
FAL-CH was chosen as the reference for a preliminary analysis because it was present at Site 1 and also because it helps to illustrate the variation in the response of the FAL-D analyzer.Comparisons of χ (1 m) and F z (z o ) for each instrument versus FAL-CH were conducted (FAL-CH data not available for the pre-cutting period).FAL-D, χ(1 m) and F z (z o ) agreed well with FAL-CH across the data range for some of the time (Fig. 3).However, there were a considerable number of data points which greatly overestimated the concentration and flux compared with FAL-CH.The fact that this was not evident for the whole period suggests that there was some variation in the accuracy of the FAL-D analyser throughout the measurement period.For example, this could be due to variation in the accuracy of the calibration.Inaccuracies in the measurement could be exacerbated in the large concentration range, where the calibration is no longer as robust.In addition, temperatures inside the analysers reached 40

Regressions of χ(1 m) and F z (z o ) against mean gradient estimate
After the data processing procedures were conducted as detailed in Sect.2.4 the "mean gradient estimate" was calculated for F z (z o ) and χ(1 m) (F z (z o ) mg and χ(1 m) mg ), this is the arithmetic mean of all available individual measurements remaining in the dataset.These data include the corrections for advection.A comparison of χ(1 m) mg versus the individual systems was conducted (Fig. 4a-d).The regression results are presented in Table 4, these include results calculated from the simple linear function and also using log transformed data.Data from 3, 8, 9 and 10 June 2000 were not included in the regression because on these days there was significant disagreement between the systems with FAL-D and FRI giving higher estimates compared with the CEH and FAL-CH estimates.To include the FAL-D and FRI estimates for these days in the regression would bias the regression towards FAL-D and FRI and might give a false impression of the overall dataset.
With data from 3, 8, 9 and 10 June removed, Fig. 4a-d demonstrate that there was close agreement of χ (1 m) between each individual system and the mean estimate across the full concentration range.As indicated by the temporal graphs, FAL-D and FRI showed slightly higher concentrations than the best estimate, whilst FAL-CH and CEH showed slightly lower.The r 2 value for all the linear regressions was high (>0.93)(Table 4) which gives confidence in the 4 systems and the values of χ(1 m) mg .
The regression results for the log transformed data show lower values of r 2 with respect to the linear results.This is due to increased scatter of small concentrations near the detection limit in the log transformed dataset, resulting in a greater variance in the dataset for low concentrations.By contrast, in the simple linear plot (Fig. 4) the variance in y is similar for the full range of concentrations, and therefore this is more appropriate for the comparison of the data than the log transformed data.
Such a log transformation for the fluxes data is less suited, due to the existence of negative fluxes.The most negative flux, recorded for FAL-D was −315 ng m −2 s −1 and a transformation of (log 10 [F z (z o ) individ + 320]=e log 10 [F z (z o ) mg + 320] + d) was tested.However, this did not succeed in normalizing the data, while the variance was again larger at smaller values (due to scatter at smaller fluxes).For this reason, only the linear relationships are shown for the fluxes.
The regression of F z (z o ) mg versus F z (z o ) from the individual systems (Fig. 5a-d, Table 5) demonstrates that once the four uncertain days (3, 8, 9 and 10 June) were removed from the regression then FAL-D tended to underestimate the flux compared with the mean gradient estimate, as did CEH.By contrast, FRI and FAl-CH both overestimated the mean gradient flux by about 10% compared with the mean gradient estimate.
As a result of the disagreement between systems on the 3, 8, 9 and 10 June an "alternative gradient estimate" for χ (1 m) and F z (z o ) was proposed for these days.The alternative gradient estimate (subscript ag) consisted of the mean of the two systems (CEH and FAL-CH).It was not considered that there was sufficient justification to remove the high measurements from the mean dataset.However, it was suspected that on these days the two high systems might not have been  operating correctly.Therefore, this alternative estimate was also provided to other end-users of the data.

NH 3 concentration and flux in relation to management activities
The resulting mean gradient concentrations of NH 3 at 1 m (χ(1 m) mg ) and flux (F z (z o ) mg ) for the whole period (Fig. 6) demonstrate clearly the effect of the management activities (cutting and fertilizing) on the concentration and flux.In addition to the mean gradient estimate, Fig. 6 also shows the alternative gradient estimate for χ(1 m) and F z (z o ) on 3, 8, 9 and 10 June.Statistics for the mean gradient estimate of χ(1 m) and F z (z o ) were calculated for the three periods: i) pre-cutting; ii) post-cutting/pre-fertilizing and iii) post-fertilizing (Tables 6 and 7).These data have been corrected for both advection and storage errors.

Pre-cutting
Prior to cutting of the grass, the flux was predominantly deposition to the surface (see Fig. 2 for typical diurnal course).
The mean flux of the pre-cutting period was −5.8 ng m −2 s −1 if all data were included and −6.0 ng m −2 s −1 if only data which passed the micrometeorological criteria were included.Any emission which was observed was generally small, the maximum emission observed over the period was 42 ng m −2 s −1 .The intensive grassland was generally acting as a sink for NH 3 during this period.

Post-cutting, pre-fertilizing
Immediately after cutting (29 May 2000, 06:00 GMT) the NH 3 flux switched to emission.The emission had a diurnal pattern with very small fluxes during night-time and emission fluxes increasing during the daytime; daily peak emission values were 135 to 700 ng m −2 s −1 .The mean flux during the cutting period was 100 ng m −2 s −1 (all data) or 138 ng m −2 s −1 (only data which passes the micromet criteria).These values are equivalent to 71 and 98 g N ha −1 day −1 , respectively.

Post-fertilizing
There was a rapid increase in NH 3 flux observed following the fertilization with N (5 June 2000, 06:00 GMT), with values peaking at 3820 ng m −2 s −1 .During the first two nights after fertilization (5 June and 6 June) there were mean nocturnal emissions of 1050 ng m −2 s −1 and 150 ng m −2 s −1 , respectively.The mean flux over the whole post-fertilizing period was 474 ng m −2 s −1 (all data) or 559 ng m −2 s −1 (only data which passed the micrometeorological criteria).These values are equivalent to 337 and 398 g N ha −1 day −1 , respectively.The emission flux decreased on 11 June 2000, but was still up to 565 ng m −2 s −1 13 days after cutting.The accumulated flux was −0.03, 0.50 and 3.41 kg N ha −1 for the pre-cutting, post cutting and post fertilizing periods, respectively.This accumulated flux up to 10 days after the fertilization represented 3.2% of the N applied, or an equivalent of 3.6% if the post-cutting emissions were included.

Inter-instrument differences for χ (1 m) and F z (z o ) for main management periods
It is relevant to consider the statistics of χ(1 m) and F z (z o ) between the different ammonia analyzers for the main management periods.With longer-averaging periods the uncertainties between the different systems decreased (Table 8).For the four different measurement systems, the relative standard error in χ(1 m) was similar for the three measurement periods at 15-20%.By contrast, the relative standard error in F z (z o ) varied substantially between the three measurement periods.This was as expected, with better agreement between the different ammonia analyzers being found for the periods with larger fluxes.The standard error (SE) of the mean 15 min values was calculated as σ n−1 / √ n, where σ n−1 is the sample standard deviation and n is number of estimates available for a given 15 min period.Hence the magnitude of the SE depended on both the level of agreement of the denuders and the number of denuders operating at a given time.The time course of the SE is presented for F z (z o ) and χ(1 m) (Fig. 7a and b).The SE of the 15 min values in the pre-cut period with substantial bi-directional exchange was typically around 60%, with absolute SE values of 10 to 20 ng m −2 s −1 (Fig. 7a).Following cutting, the errors differed diurnally with daytime SE in the flux typically 40%.The % SE was smallest following fertilization with daytime values typically 15%, increasing to typically 30% from 10 June.The overall median uncertainty in the 15 min estimates of χ(1 m) was 18%, with values mostly in the range 5 to 50% (Fig. 7b).Regarding the storage and advection calculations, for much of the period these were rather small, but they were important in specific instances.The storage errors were generally less than 10% of the measured flux, although individual values occurred in the range 25-125% during periods of rapidly changing χ (1 m) (Fig. 7c).The advection errors (Fig. 7d) were a similar magnitude to those for storage, with most values being less than 10% of F z .Although larger absolute advection errors were observed post cutting and post fertilizing, the values as a percentage of the measured flux were smaller for these periods than for the pre cutting period.

Inter-comparison of gradient measurements
Overall, the inter-comparison of gradient measurements was encouraging, with three of the flux measurements being within 11% of the mean estimate and the other within 32% (Table 5).There was also close agreement achieved across the concentration range, except for some days where overestimation of concentration is suspected to have occurred in some of the systems, possibly due to high operating temperatures and inaccuracies in the measurement which could be exacerbated in the large concentration range, where the calibration is no longer as robust.The inter-comparison highlighted the need for regular calibration of flux gradient systems and regular quality standard checks.As concluded by Harrison and Kitto (1990), operator differences can induce the same amount of variation in NH 3 measurements as different measuring techniques and, although techniques such as AMANDA and the WEDD have been shown to be reliable in measuring NH 3 , operators have to be vigilant in their running of these systems.A reliable clean deionised water supply, regular changing of pump tubing and regulation of instrument operating temperature are all essential to maintain the reliability of these systems.
Having four independent systems did result in a robust final dataset with an extremely high data coverage of NH 3 concentrations and fluxes of 98%.The main restriction on estimates was a filter according to strict micrometeorological criteria, which reduced the flux data coverage to 69% if these filter criteria were applied.With typically 50-70% data coverage of concentration and 30-50% for flux being more common in other NH 3 studies (e.g., Erisman et al., 1998;Sutton et al., 2001b;Horvath et al., 2005) and with the continuous record of uncertainty from the replicate systems, it is clear that this dataset is unique in coverage and robustness.As such it provides an important resource for interpretation of ammonia fluxes in relation to bioassays, models and in-teractions with other atmospheric components (Sutton et al., 2009a).
The availability of replicate instruments in the present experiment also highlight the need for caution when interpreting results from the more usual situation of a single ammonia flux detection system.Table 8 shows that the different instruments used here agreed to within 20% (standard error) for ammonia concentrations in the three management periods.By contrast, while the flux by the three systems had a standard error of 20% for the post-fertilization period (when fluxes were largest and easiest to detect), the uncertainties were larger for the pre-fertilization periods.This indicated an uncertainty of 33% for the post-cut period and 76% for the pre-cut period.As demonstrated by Fig. 2, the uncertainties can be even larger for 15 min measurements (even leading to uncertainties in flux direction).As a result, great caution is needed when interpreting such NH 3 flux data when only one system is available.In this situation, it is thus essential to perform regular calibration and also zero-concentration tests (switching off air supply and checking the zero of the analyzer) and zero-flux tests (e.g. by bringing inlets to a common height and demonstrating that the measured flux is zero, Sutton et al., 1993a).Intercomparisons of measured ammonia fluxes with process models must similarly recognize the substantial uncertainty in the measured fluxes.

Influence of management activities on NH 3 flux
The present measurements support previous findings of enhanced emissions following grass cutting (Sutton et al., 1997(Sutton et al., , 2001a;;Milford et al., 2001), with a mean flux of 98 g N ha −1 day −1 after cutting compared with measurements of predominantly deposition before cutting.These emissions appear to be a consequence of altered N processing in the remaining sward rather than the lying cut grass (Sutton et al., 1997;David et al., 2009;Mattsson et al., 2008).The modelling of Sutton et al. (2001) and Riedo et al. (2002) explained this on the basis of a larger ammonia compensation point of the remaining plant (Riedo et al., 2002) and these interactions are further investigated by Burkhardt et al. (2009), Personne et al., (2009) and Sutton et al. (2009a).
As expected, emissions were also enhanced following fertilization, with a mean flux of 398 g N ha −1 day −1 .These fluxes are somewhat larger in magnitude to those observed earlier (Sutton et al., 1997;Milford et al., 2001).Expressed as the % emission of fertilizer N applied, in the present case 3.2% of the N (applied as calcium ammonium nitrate) was lost as ammonia within 10 days of application.The emission factor for ammonium nitrate fertilizer applied to grasslands in the UK is 1.6% of the applied N (van der Weerden and Jarvis, 1997).By comparison using AMANDA flux measurement, Milford et al. (2001) also estimated that 1.6% of fertilizer N (as ammonium nitrate) was lost as ammonia from a Scottish grassland.Strictly, the value from Milford et al. (2001) is not comparable with the present study, since their estimate referred to total emissions (including from grass cutting).Hence if the present losses following cutting are included, this gives an equivalent loss of 3.6% of the nitrogen, further emphasizing the larger values of this study.The larger emissions measured here are possibly a result of the high temperatures which were observed, particularly in the latter period of the campaign.The mean air temperature at 1 m was 17.6 • C during 5-15 June 2000, although daily maximums reached 39 • C (Nemitz et al., 2009b).The potential for NH 3 emission increases with increasing temperature due to the dependence of solubility of NH 3 on temperature, which affects the stomatal compensation point (Farquhar et al., 1980) as well as the NH 3 gaseous concentration at the soil and litter surface (Nemitz et al., 2001b(Nemitz et al., , 2004)).
The diurnal pattern of ammonia emission following fertilization is quite typical for the changing patterns of environmental conditions, with largest fluxes occurring during the day.Substantial emission also occurred at night immediately after fertilizer application (5-6 June), demonstrating the importance of surface emissions from the soil and litter surface.However, nocturnal emissions ceased after this date, and therefore process models and other measurements (David et al., 2009) are needed to interpret the contribution of different sources and sinks.

Conclusions
An inter-comparison of four NH 3 gradient measurement systems was conducted over intensively managed grassland during a four week period in May-June 2000 as part of the GRAMINAE Integrated Experiment.The inter-comparison was encouraging, with three of the flux measurements being within 11% of the mean and the other within 32%.The inter-comparison highlighted the need for regular calibration of flux gradient systems and regular checks against quality standards.Operating four systems resulted in exceptionally high data coverage of measured ammonia fluxes (98%) and provided a continuous record of the uncertainty of the data.The measurements supported previous findings of enhanced emissions from grass cutting with a mean daily flux of 98 g N ha −1 day −1 after cutting compared to measurements of predominantly deposition before cutting.Emissions were also enhanced following fertilization, with a mean flux of 398 g N ha −1 day −1 .The results of this intercomparison provide a robust dataset for the evaluation of the processes controlling ammonia exchange, while indicating the need for caution in interpreting unreplicated ammonia flux measurements.It is clear that measurement of ambient ammonia concentration and flux still remains a challenge and future research should focus on continuing to improve ammonia measurement techniques and the reliability of flux measurements.

Figure 1 .
Figure 1.Examples of χ(1m) for NH3 from the four different systems for pre-cutting, postcutting and post-fertilizing periods.Centre for Ecology and Hydrology (CEH), Hungarian Forest Research Institute (FRI), German Federal Agricultural Research Institute (FAL-D) and Swiss Federal Research Station for Agroecology and Agriculture (FAL-CH).FAL-CH gradient data comences on 30 May 2000.Tick labels mark 00:00 GMT.

Fig. 1 .Figure 2 .
Fig. 1.Examples of χ (1 m) for NH 3 from the four different systems for pre-cutting, post-cutting and post-fertilizing periods.Centre for Ecology and Hydrology (CEH), Hungarian Forest Research Institute (FRI), German Federal Agricultural Research Institute (FAL-D) and Swiss Federal Research Station for Agroecology and Agriculture (FAL-CH).FAL-CH gradient data comences on 30 May 2000.Tick labels mark 00:00 GMT.

Fig. 2 .
Fig. 2. Example of NH 3 fluxes (F z ) from the 4 different systems for pre-cutting, post-cutting and post-fertilizing periods.Centre for Ecology and Hydrology (CEH), Hungarian Forest Research Institute (FRI), German Federal Agricultural Research Institute (FAL-D) and Swiss Federal Research Station for Agroecology and Agriculture (FAL-CH).FAL-CH gradient data comences on 30 May 2000.Tick labels mark 00:00 GMT.

Fig. 3 .
Fig. 3. Comparison of (a) χ (1 m) and (b) NH 3 flux at z o (corrected for advection) measurements.German Federal Agricultural Research Institute (FAL-D), Centre for Ecology and Hydrology (CEH) and Swiss Federal Research Station for Agroecology and Agriculture (FAL-CH).

Fig. 5 .
Fig. 5. Regression of NH 3 flux from each system against mean gradient flux for (a) German Federal Agricultural Research Institute (FAL-D), (b) Centre for Ecology and Hydrology (CEH), (c) Hungarian Forest Research Institute (FRI) and (d) Swiss Federal Research Station for Agroecology and Agriculture (FAL-CH).Data from 3, 8, 9 and 10 June 2000 are not included.

Figure 6 .
Figure 6.Mean gradient estimate of a) χ(1m)mg and b) net NH3 flux (Fz(zo)mg) showing response to management activities.The alternative gradient estimate is also shown (χ(1m)ag and Fz(zo)ag ) on 3, 8, 9 and 10 June 2000, these are days with high instrument uncertainty (see text).Vertical lines indicate cutting (dashed line), removal of the grass from the field (dotted line) and NH4NO3 fertilization (solid line).

Fig. 6 .
Fig. 6.Mean gradient estimate of (a) χ (1 m) mg and (b) net NH 3 flux (F z (z o ) mg ) showing response to management activities.The alternative gradient estimate is also shown (χ (1 m) ag and F z (z o ) ag on 3, 8, 9 and 10 June 2000, these are days with high instrument uncertainty (see text).Vertical lines indicate cutting (dashed line), removal of the grass from the field (dotted line) and NH 4 NO 3 fertilization (solid line).

3. 6
Time-course of uncertainties in the mean estimates of χ(1 m) and F z (z o ) The availability of up to four parallel measurements of the ammonia concentrations and fluxes enabled an assessment of the uncertainty in the mean estimates of F z (z o ) and χ(1 m).

Fig. 7 .
Fig. 7. Time-course of the uncertainty in mean F z (z o ) and χ (1 m) presenting 1 hourly running medians of the 15 min estimates, shown together with the magnitude of advection and storage errors: (a) Percentage and absolute Standard Error (SE) in the measured ammonia flux, (b) % SE in χ (1 m), (c) absolute storage error (F z,sto ) and F z,sto as a % of F z (1 m), (d) absolute advection error (F z,adv ), sum of the modelled advection errors at z=1 m due to the experimental field and farm and F z,adv as a % of F z (1 m).Vertical lines indicate cutting (dashed line), removal of the grass from the field (dotted line) and NH 4 NO 3 fertilization (solid line).

Table 1 .
Summary of the different continuous ammonia concentration profile sampling systems used to calculate ammonia fluxes in the present study.

Table 3 .
Results of the blind testing of aqueous ammonium Quality Control standards.The blind standards were prepared by a separate laboratory (ECN, Netherlands).On each occasion two standards were distributed and the difference in concentration measured by each analyser was compared with the actual concentration difference to give an indication of the performance across the full measurement range.The % difference shown was calculated as (100-(100 [NH + 4 ] lab / [NH + 4 ] QCstd )).This test was conducted on 22 May 2000 for FRI and a contaminated stripping solution container was found to be the cause; n/a: not available. *

Table 4 .
Summary of regression results of χ (1 m) of the individual systems versus the value for the mean gradient (χ (1 m) mg ) expressed in µg m −3 .The regression is given first as a simple linear function (as shown in

Table 5 .
Summary of linear regression results of F z (z o ) of the individual systems versus F z (z o ) mg expressed in ng m −2 s 1 , given as F z (z o ) individ =c F z (z o ) mg +b.The estimates are derived from the data shown in Fig.5.

Table 6 .
Variations in 15 min measurements of NH 3 concentration at 1 m from the mean gradient, χ (1 m) mg , throughout different measurement periods, before and after micromet filtering.

Table 8 .
Standard errors (SE) and % standard errors in the concentration and flux estimates between the 4 ammonia sampling systems according to the different management periods (for the full dataset).