356 M.A. Sutton et al. Agricultural and Forest Meteorology 105 2000 351–369
tion would apply if fluxes are not conserved due to gas–particle reactions above the canopy. This issue is
addressed separately by Nemitz et al. 2000c, where it is shown that above-canopy reactions would have
had little effect on the calculated NH
3
fluxes during the North Berwick experiment. For the analysis of the
measured fluxes, three levels of data filtration and cor- rection were considered:
1. A general filtering of data to remove major prob-
lems, such as fetch interruptions and denuder mal- functioning.
2. A more rigorous filtering to exclude conditions where the flux is likely to be estimated with less
certainty, such as very stable conditions or where the contribution of the rape field to the flux at the
top height was less than a defined value, according to a foot-print analysis.
3. Application of correction procedures to the remain- ing data to account for storage errors and density
corrections. For the foot-print analysis Schuepp et al., 1990
data were accepted if the field contributed to the top NH
3
sampling height by 65 or more, equivalent to almost 100 at the middle and bottom sampling
heights.
Fig. 1. Comparison of friction velocity u
∗
as measured with three ultrasonic anemometers logged digitally and an anemometer profile during the main campaign 20–21 June after filtering. For clarity, the mean of estimates is not shown. Heights are shown above ground.
4. Results
4.1. Micrometeorological exchange parameters The comparison of several different measurement
methods provided a powerful technique to establish reliable estimates of the micrometeorological parame-
ters. This is shown in Figs. 1 and 2, which illustrate the comparisons for u
∗
and H, respectively. Good agree- ment was obtained between the different determina-
tions of u
∗
, with the weighted mean of the relative standard deviation of the filtered values being 14 for
the overall dataset. Close agreement was also found for sensible heat fluxes using the different systems
Fig. 2, with the weighted mean of relative standard deviation being 26. In both these cases, best esti-
mates of the parameters were established by means of the available estimates, which were used in the subse-
quent analysis. The canopy was found to have a zero plane displacement of around 1.11 m, with typical val-
ues of z
in the range 65–125 mm. Less encouraging agreement was found for the es-
timation of λE Fig. 3, which was caused by a poor time-response of the Campbell dewpoint meter on the
profile system coupled to the switching of this device
M.A. Sutton et al. Agricultural and Forest Meteorology 105 2000 351–369 357
Fig. 2. Comparison of sensible heat flux H measured with three ultrasonic anemometers and stability corrected anemometertemperature profile 20–21 June. For clarity the mean of estimates is not shown. Heights are shown above ground.
between the two sampling heights, so the stabiliza- tion time was longer than the switching time of 2 min.
This resulted in underestimation of vertical vapour pressure gradients and hence of λE by this method.
These data were therefore not used in further analy- sis. Despite this limitation, good agreement Fig. 3
Fig. 3. Comparison of the latent heat flux λE measured with the Krypton hygrometer by eddy correlation, the profile system and dew point meter by aerodynamic gradient method, and an estimate required to close the energy balance of the surface 18–19 June.
was found between the eddy correlation estimate us- ing the KH
2
O and the estimate based on closure of the energy balance. These values were used to provide
best estimates of λE for calculation of bulk canopy stomatal resistance for water vapour applied in later
analysis.
358 M.A. Sutton et al. Agricultural and Forest Meteorology 105 2000 351–369
4.2. Determination of ammonia concentration gradients and fluxes by continuous denuders and
filter packs
As expected, the determination of concentration gradients and hence χ
∗
, Eq. 1 represented the largest uncertainty in calculating fluxes. The avail-
ability of the two AMANDA systems allowed for im- proved estimates where both systems were function-
ing acceptably well, although denuder malfunctioning occurred frequently. The quality of the measurement
estimates for both the concentration profiles and the estimation of χ 1 m is shown in Fig. 4, together with
estimates of the flux using the best estimates of u
∗
and H. This therefore shows the uncertainties due to the AMANDA gradient measurements.
The estimates of air concentration by the two sys- tems were close, while periods of good agreement of
Fig. 4. Ammonia concentration as measured at six heights above the canopy by two 3-point AMANDA systems operated by the Centre for Ecology and Hydrology CEH and the Technical University of Madrid UPM 22–23 June. Sampling heights noted are referenced
above ground level. χ 1 m refers to NH
3
concentration at 1 m above the zero plane; F, net NH
3
flux emission is positive. Canopy height was 1.38 m. Sampling heights are shown above ground.
the flux are also apparent. However, there were also pe- riods of significant disagreement for the flux, as illus-
trated for the afternoon of 22 June in Fig. 4. Where the agreement was close, the average of the two systems
was used for further analysis. Where the differences were substantial as on 22 June, checks were made
to consider: a possible malfunctioning of one of the AMANDAs, such as indicated by unequal liquid flow
rates between the three denuder inlets, b compari- son with independent estimates from the filter packs,
where available, c agreement between the denuders when set to sample at the same height ‘co-sampling’
and d the consistency of the apparent fluxes with other measurement periods established as reliable. For
the example of 22 June it was shown that the UPM AMANDA sometimes suffered from unequal liquid
flow rates and that the daytime deposition fluxes of the UPM system were inconsistent with the emissions
M.A. Sutton et al. Agricultural and Forest Meteorology 105 2000 351–369 359
Fig. 5. Comparison of the CEH AMANDA system against the UPM AMANDA system and filter packs for periods where filter pack runs were made, showing: a NH
3
concentrations z − d = 1 m; b NH
3
fluxes. Open symbols show values filtered for either UPM AMANDA malfunctioning or where AMANDA data was available for 60 of a filter pack run. Units are mg m
− 3
for concentrations and ng m
− 2
s
− 1
for fluxes.
established for similar conditions on other days e.g. 23 June. In addition, although no filter packs runs
were made on this day, comparisons on other days see below, supported by co-sampling tests, indicated
that the UPM system repeated this behaviour in some other runs. A detailed filtering of the AMANDA data
was therefore made in order to establish best estimates for subsequent analysis.
A total of around 25 concentration profiles were made with the filter packs each of around 2 h du-
ration, providing an independent estimate of the NH
3
fluxes. In many cases the scatter from these runs was very high S.E. of the flux estimates typi-
cally ±20 ng m
− 2
s
− 1
. The comparison is shown in Fig. 5, and indicates a typical agreement between
NH
3
concentration estimates of ±25 and between flux estimates of ±50, although it is not possible to
say whether the errors lie mainly with the denuders or filter packs.
4.3. Filtering of measured ammonia fluxes and best estimates from continuous denuder sampling
A summary of the filtering and correction proce- dures applied to the data and the effect on the data re-
duction is shown in Table 1. Out of the first measure- ment period, which was studied most intensively, the
general data treatment provided 2880 flux estimates, equivalent to flux continuous measurements for around
2 3
of the time. Micrometeorological corrections were made depending on data availability 2270 data points
and the second stage filtering reduced the dataset to around 1500.
The effect of applying treatments b and c on the calculated fluxes is shown in Table 2, for the first
measurement campaign. Table 2 also shows means and variability of night-time and daytime fluxes for
both the pre- and post-cutting measurement periods. A comparison of treatments a and b shows a slightly
larger mean flux for the filtered data, which is largely a consequence of removing more data from stable
night-time conditions than from the daytime. How- ever, the effect on mean fluxes is rather small. The ef-
fect of the data correction procedures is even smaller as shown by a comparison of a and c. Here the
change on the mean flux between the corrected and uncorrected datasets is only 3, which is trivial com-
pared with other sources of error in the measurements. Even for the extreme values of minimum and maxi-
mum flux, the corrections are less than 10. The rea- son for this is that these corrections, although strictly
appropriate, are most relevant for slowly exchanging
360 M.A. Sutton et al. Agricultural and Forest Meteorology 105 2000 351–369
Table 1 Filter criteria and corrections applied to the NH
3
fluxes points are 10 min mean estimates
a
Correction Correctionsfilter criteria
No. of valid data points remaining
a General u
∗
, H filtered for obstructed wind sectors 2880
H from profile discarded as systematically different possible radiative heating
u
∗
from profile discarded for u1.68 m 0.5 m s
− 1
anemometer stalling NH
3
concentration data filtered for periods of malfunctioning systems b Filtering
Discard data for 0 L 5 m
2240 CNFtop NH
3
concentration 65 1710
u 1 m 0.8 m s
− 1
1598 Obstructed wind sector for NH
3
sampler 1528
c Corrections Correction for
2270
b
Storage errors Temperature gradients
Humidity gradients
a
u
∗
: friction velocity; H: sensible heat flux; uz − d: mean horizontal windspeed at height above the canopy displacement height z − d
; L: Monin–Obukhov stability length; CNF: the cumulative normalized contribution to the flux measurement denotes the fraction to which the flux measured at the top height is influenced by the fetch Section 3.4.
b
Indicates the number of data points for which sufficient data were available to apply the correction procedures.
Table 2 Means and variability of 10 min AMANDA NH
3
flux estimates for the pre- and post-cutting measurement periods, noting the influence of filtering and correction procedures daytime is defined as 04:30–20:00 GMT
a
Period Mean
S.D. Median
Minimum Maximum
Sample size Pre-cutting
χ 1 m mg m
− 3
All data 1.03
0.90 0.90
0.00 20.97
2268 Day
1.17 0.71
1.04 0.09
9.18 1405
Night 0.79
1.11 0.71
0.00 20.97
864 a F ng m
− 2
s
− 1
general treatment All data
16.9 26.0
10.0 −
159.9 161.8
2370 Day
25.2 27.5
18.8 −
159.9 161.8
1467 Night
3.4 15.9
− 0.1
− 58.2
77.2 903
b F ng m
− 2
s
− 1
after filtering All data
19.0 28.1
12.2 −
159.9 161.8
1523 Day
27.1 28.6
21.0 −
159.9 161.8
1049 Night
− 1.6
11.6 −
0.9 −
58.2 39.9
429 c F ng m
− 2
s
− 1
corrections applied All data
16.4 25.4
9.8 −
148.2 178.5
2270 Day
24.6 26.4
18.3 −
148.2 178.5
1406 Night
3.2 16.4
− 0.05
− 58.6
77.1 864
Post-cutting χ
1 m mg m
− 3
All data 2.47
1.52 2.21
0.38 12.64
1247 Day
2.62 1.57
2.35 0.38
12.64 785
Night 2.21
1.40 1.94
0.50 7.21
462 F
ng m
− 2
s
− 1
All data 56.8
81.8 34.7
− 197.0
622.2 1244
Day 82.8
89.9 59.1
− 197.0
622.2 783
Night 12.7
35.5 0.2
− 96.6
131.1 461
a
χ 1 m: NH
3
concentration at 1 m above the zero plane displacement height; F: net NH
3
flux; S.D.: standard deviation.
M.A. Sutton et al. Agricultural and Forest Meteorology 105 2000 351–369 361
trace gases. For ammonia, the fluxes are rather large in relation to concentrations, so that the corrections
become trivial Sutton et al., 1993a.
4.4. Average and ranges of fluxes for the different measurement periods
Mean fluxes were around 17 and 57 ng m
− 2
s
− 1
for the pre- and post-cutting measurements, respectively.
Table 2 shows that both the largest and smallest fluxes occurred during the day, with extreme values
of −150 and +180 ng m
− 2
s
− 1
for the pre-cutting measurements. Even larger fluxes were recorded
for the post-cutting campaign, with values up to 620 ng m
− 2
s
− 1
. The pattern of air concentrations between the day
and the night is also of interest. Although concentra- tions are often more variable during night, probably as
a consequence of increased impact of plumes from lo- cal point sources during stable conditions, overall air
concentrations are larger during the day than during the night. This may be a consequence of larger emis-
sions during the daytime from this crop and others in the vicinity affecting NH
3
air concentrations. This explanation is supported by the observation of larger
average NH
3
concentrations during the post-cutting period, when net emissions were also larger. A fur-
ther possibility is variations in partitioning between gaseous NH
3
and aerosol NH
4 +
, although analysis by Nemitz et al. 2000c suggests that this would be a
second order effect for the measurements here. 4.5. Temporal patterns in ammonia fluxes
The time course of the NH
3
fluxes over the two measurement periods is shown together with χ 1 m,
R
n
, H and λE in Figs. 6 and 7. The NH
3
fluxes have been filtered using the general screening criteria a,
but, in order to show the temporal trends, results fil- tered under b are also included.
Examination of Fig. 6 shows some days such as 13–19 June which follow the classical pattern of NH
3
exchange over croplands, with emission during the day, and much smaller emissions or deposition at
night. This is consistent with the cooler night-time conditions giving smaller compensation points for the
canopy sources, closure of stomata and the increased effectiveness of leaf surface sinks at high relative hu-
midities. However, not all the data show this pattern, and substantial periods of emission occur for several
night-time periods, such as 12 and 19–23 June. These fluxes are most probably related to emissions from
decomposing abscised leaves at the bottom of the canopy Nemitz et al., 2000a.
Diurnal variation in NH
3
fluxes is also seen in the post-cutting measurement period, although
here the larger daytime emissions typically up to
200–300 ng m
− 2
s
− 1
alternate with
smaller night-time emissions, with only a few periods of
nocturnal NH
3
deposition. 4.6. Average daily fluxes
A summary of the fluxes for each day of the two measurement periods is shown in Fig. 8, alongside
the percent data coverage of fluxes for each day. Data capture was 50 or more for all except two of the
measurement days, and in these cases no average flux for the day is stated. The typical daily flux for
the measurement period over the ripening crop was 28 ng NH
3
m
− 2
s
− 1
, which is equivalent to around 20 g N ha
− 1
per day. For the post-cutting period the emissions were more
variable. However, the overall emissions were even larger than the pre-cutting period, and showed a gen-
eral tendency to increase with time. Typical emissions immediately after cutting were around 30 g N ha
− 1
per day and increased to around 80 g ha
− 1
per day 110 ng NH
3
m
− 2
s
− 1
at the end of the measurements. 4.7. Comparison of the continuous and passive flux
sampling estimates A summary of the comparison between the
AMANDA flux estimates with those from the passive denuder and shuttle samplers is shown in Table 3.
Although several runs were made with the RVAU passive denuders, there was substantial scatter in the
results partly related to the short run periods of only a few days, and the result have therefore been averaged
here. The passive estimates of the mean flux shown in Table 3 are very close to the AMANDA reference,
being within 25. While this level of agreement is encouraging, this may be fortuitous, since the scatter
in the measurements is larger than the means. This is especially the case for the ADAS shuttles, where only
362 M.A. Sutton et al. Agricultural and Forest Meteorology 105 2000 351–369
Fig. 6. Fluxes of NH
3
and heat as measured during the main campaign at late flowering of the rape canopy. Fluxes of NH
3
F
NH
3
are shown as dashed lines where these pass the measurement criteria parts a and b and result from the mean of both the CEH and UPM
AMANDA systems where both functioned adequately Table 1. Fluxes are shown as dotted lines were filtered out under part b, or gaps which have been filled with data from the UPM AMANDA only, filter packs or by linear interpolation for gaps of less than 3 h. Air
concentrations of NH
3
at 1 m above the displacement height χ
NH
3
1 m are shown as solid lines.
M.A. Sutton et al. Agricultural and Forest Meteorology 105 2000 351–369 363
Fig. 7. Fluxes of NH
3
F
NH
3
as measured during the post-cutting campaign. The field was cut on the 22 July. Air concentrations χ
NH
3
are the lower of the two traces in each graph.
Fig. 8. Daily average NH
3
fluxes shown as points during both experimental periods obtained by interpolation of measured fluxes. Percentage data coverage is shown as bars.
364 M.A. Sutton et al. Agricultural and Forest Meteorology 105 2000 351–369
Table 3 Comparison of mean NH
3
concentrations and fluxes from the classical aerodynamic gradient method AMANDA and the passive flux sampling approaches
a
Run periods compared
b
Measuring system χ
1 m ± S.D. mg m
− 3
F ± S.D. ng m
− 2
s
− 1
Runs: 1, 2, 3 AMANDA
1.1 ± 0.1 20.4 ± 11.8
Runs: 1, 2, 3 Passive denuders
2.0 ± 0.8
c
25.0 ± 45.6 Run 3
AMANDA 1.2
23.3 Run 3
Passive shuttles 1.3
c
22.4 ±145
d a
χ 1 m: NH
3
concentration at 1 m above the zero plane displacement height; F: net NH
3
flux; S.D.: standard deviation.
b
Timing of runs: Run 1: 8–1161995; Run 2: 11–1461995; Run 3: 14–1761995. Passive samplers changed at noon GMT.
c
Windspeed weighted concentration.
d
As only one run, the standard deviation refers to the scatter in the vertical concentration profile of eight samplers.
one run was made, and where the scatter was much larger than for the passive denuders, for which the av-
erage standard deviation of an individual profile was 35 ng m
− 2
s
− 1
. The NH
3
concentrations measured by the passive samplers are similar or larger than the
AMANDA reference, and this may partly reflect the fact that these represent windspeed weighted con-
centrations. However, if this were the case it would suggest the existence of larger NH
3
concentrations occurring during high windspeed conditions, mostly
during the day. In fact there is no clear relationship be- tween χ 1 m from the AMANDA measurements and
windspeed data not shown. This would suggest that the difference in χ measured by the two systems is
actually not due to the difference between time aver- aged and windspeed weighted average concentration.
5. Discussion