ArticlePDF Available

Hanekamp et al. (2017) A volatile discourse – reviewing aspects of ammonia emissions, models and atmospheric concentrations in The Netherlands

June 2017
Soil Use and Management 33:276–287

DOI:10.1111/sum.12354

Authors:

J.C. Hanekamp

University College Roosevelt

Marcel Crok

Clintel Foundation

In the Netherlands, there is a vigorous debate on ammonia emissions, atmospheric concentrations and deposition between stakeholders and research institutions. In this article, we scrutinise some aspects of the ammonia discourse. In particular, we want to improve the understanding of the methodology for handling experimentally determined ammonia emissions. We show that uncertainty in published results is substantial. This uncertainty is under- or even unreported, and as a result, data in national emission inventories are overconfident by a wide margin. Next, we demonstrate that the statistical handling of data on atmospheric ammonia concentrations to produce national yearly atmospheric averages is oversimplified and consequently atmospheric concentrations are substantially overestimated. Finally, we show that the much-discussed ‘ammonia gap’ – either the discrepancy between calculated and measured atmospheric ammonia concentrations or the difference observed between estimated NH3 emission levels and those indicated by atmospheric measurements – is an expression of the widespread overconfidence placed in atmospheric modelling.

Content uploaded by J.C. Hanekamp

Content may be subject to copyright.

Avolatilediscourse–reviewing aspects of ammonia

emissions, models and atmospheric concentrations in

The Netherlands

J. C. HANEKAMP

1,2

,W.M.BRIGGS &M.CROK

University College Roosevelt, Middelburg, the Netherlands, and

Environmental Health Sciences, University of Massachusetts,

Amherst, MA, USA

Abstract

In the Netherlands, there is a vigorous debate on ammonia emissions, atmospheric concentrations and

deposition between stakeholders and research institutions. In this article, we scrutinise some aspects of

the ammonia discourse. In particular, we want to improve the understanding of the methodology for

handling experimentally determined ammonia emissions. We show that uncertainty in published

results is substantial. This uncertainty is under- or even unreported, and as a result, data in national

emission inventories are overconﬁdent by a wide margin. Next, we demonstrate that the statistical

handling of data on atmospheric ammonia concentrations to produce national yearly atmospheric

averages is oversimpliﬁed and consequently atmospheric concentrations are substantially

overestimated. Finally, we show that the much-discussed ‘ammonia gap’ –either the discrepancy

between calculated and measured atmospheric ammonia concentrations or the difference observed

between estimated NH

emission levels and those indicated by atmospheric measurements –is an

expression of the widespread overconﬁdence placed in atmospheric modelling.

Keywords: Ammonia emission, manure application, ammonia emission modelling, Dutch Air Quality

Monitoring Network, atmospheric ammonia concentrations, statistics

Introduction

Atmospheric ammonia (NH

) concentrations in the

Netherlands are reported to be amongst the highest in the

world and are regarded as a hazard to biodiversity in natural

ecosystems. Livestock are the largest contributor to ammonia

emissions (PBL, 2016), and since 1993, major efforts have

been made to reduce emissions. As a practical approach, the

reduction of ammonia volatilization after manure application

to farmland, regarded as the largest single emission source,

has received much attention (Van Bruggen et al., 2011). In the

1990s, broadcast surface spreading made way for methods

such as shallow and narrow band injection on grassland and

deep placement on arable land (fallow). However, an

evaluation of the scientiﬁc underpinning of the calculation of

ammonia emission and deposition in the Netherlands stated

that ammonia concentrations in the air ‘have not decreased as

much as expected since the introduction of mitigation

measures. This has led stakeholders to question the

effectiveness of the Dutch ammonia policy.’ (Sutton et al.,

2015) This is signiﬁcant, as the Dutch agricultural community

has invested much in these strategies, and is regarded

internationally as environmentally innovative.

In this article, we analyse some parts of the scientiﬁc

discourse on Dutch ammonia emissions. We take as our

primary cue the article published by Huijsmans et al. (2016).

Therein the focus is put on ammonia emissions from the

application of cattle slurries to grassland. One of our goals

was the reproduction of the presented results using the

underlying data. We were motivated by both scientiﬁc

curiosity and the desire to try to resolve a continuous

dispute over the published results, which has implications for

agricultural policies in the Netherlands.

We broaden our scope with a discussion on the much-used

Ryden and McNeill model for ﬁtting measured ammonia

concentrations to emissions after manure application

experiments (Ryden & McNeill, 1984). We also analyse the

Dutch national data set of atmospheric ammonia concen-

trations as produced by the LML network (Landelijk Meetnet

Luchtkwaliteit –Dutch Air Quality Monitoring Network).

Correspondence: J.C. Hanekamp. E-mails: j.hanekamp@ucr.nl;

hjaap@xs4all.nl

Received September 2016; accepted after revision May 2017

Soil Use and Management, June 2017, 33, 276–287 doi: 10.1111/sum.12354

SoilUse

and Management

Reproducing ammonia emissions from manure

application

In 2009, an overview of experimentally assessed emissions of

ammonia (so-called emission factors) related to manure

application was published (Huijsmans & Schils, 2009). The

authors state that the emission factors, deﬁned as the

average total emission for each method as a percentage of

total ammoniacal nitrogen (TAN) applied with the manure,

are based on all available data, including the ranges in total

emissions for each method. This amounts to a total number

of observations of 199 on grassland and 58 on arable land,

relating to various application techniques.

Unfortunately, the data sets used in Huijsmans & Schils

(2009) are no longer available. Consequently, the results,

republished by Huijsmans et al. (2016), cannot be

reproduced. However, some experimental data sets gathered

since 2010 have been made available, although these data do

not contribute to those shown in Table 1. The results from

one of these data sets are reproduced and discussed below

(all other data sets received are mathematically treated in the

same way). The results reproduced here can act as a

template for other results derived from the unavailable data

sets, as the same experimental techniques and mathematical

model were used throughout the years (see e.g. Huijsmans,

2003). Consequently, we are partially able to explain the

derivation of the emission factors and the uncertainties

surrounding them, which are not or only obliquely reported.

Estimating ammonia emissions from manure application

In much of the Dutch policy-relevant work on the

volatilization of NH

following manure application, the

micrometeorological mass balance method was used

(Denmead, 1983; Ryden & McNeill, 1984; Huijsmans, 2003).

Although other methods are available, these will not be

reviewed here because of our focus on Dutch emissions.

Ryden & McNeill (1984) offered a quasi-physical model to

represent NH

ﬂux over a manured plot. The following

equation is proposed:

F¼1

xZzp

uc dz ð1Þ

where xis the fetch of the plot; z

and z

are two heights

where measurements take place; uis the instantaneous wind

speed; cthe instantaneous value of NH

; and uc the time-

averaged ﬂux at height z.

This simple model states that, at a certain height, as either

wind speed or NH

concentration increases, ﬂux increases.

This is a gross approximation at the boundary layer, where

volatilization involves many more factors other than

just wind speed including temperature, soil chemistry (pH),

soil moisture content, precipitation and TAN (e.g. Behera

et al., 2013). The solution to Equation (1) requires knowing

the functional relationship between uc and z. Instead of a

physical argument, Ryden and McNeill ﬁrst assumed uand c

to be (causally and probabilistically) independent. They next

created two empirical functional relationships between !

uand

cand height z. These relationships are themselves not

derived from chemical or physical principles but were

claimed to be observationally valid. The two equations are

uðzÞ¼Dln zþEð2aÞ

cðzÞ¼%Aln zþBð2bÞ

Equations (2a) and (2b) are substituted into Equation (1) for

uc, and the integral is then solved. The coefﬁcients A,B,D, and

Eare unknown but estimated by ordinary linear regression

between the natural log of height (z) and the observed wind

speed and observed NH

, the end result of which is

F¼1

xh%AD ðzðln zÞ2%2zln zþ2zÞþðBD %AEÞðzðln z%1ÞÞ

þEBz %!c1Dðzðlnz%1ÞÞ % !c1Exi!

ð3Þ

where !

c1is the ambient concentration of NH

on the

windward side of the site, and z

and z

are the height

Table 1 Emission factors (% total ammoniacal nitrogen [TAN] applied); mean and range for each slurry application method (Huijsmans & Schils, 2009)

Method Experiments (number)

Total emissions (average based

on all available data), % Minimum, % Maximum, %

Grassland

Surface spreading 81 74 28 100

Narrow band 29 26 9 52

Shallow injection 89 16 1 63

Arable land

Surface spreading 26 69 30 100

Surface incorporation 25 22 3 45

Deep placement 7 2 1 3

The Dutch ammonia-discourse critiqued 277

differences. Equation (3) shows the explicit dependence on

the regression coefﬁcients A,B,D, and E. The point

estimates of these coefﬁcients are input into the equation to

produce F. However, as the coefﬁcients are not known with

certainty, Fcannot be known with certainty either.

Just as there are sources of uncertainty in the relationships

of windspeed and NH

concentration with height, there are

methods to account for uncertainty in F. A crucial one is the

physical model itself and in the posited empirical regression

relationships, which we will not attempt to fathom here.

Other physical–chemical models that describe NH

concentrations after manure application are in use, as are

other statistical relationships and experimental set-ups. We

have not compared those with the model of Ryden & McNeill

(1984). What we can do is to incorporate parameter

uncertainty in Equation (3). To do this, the conﬁdence

intervals and the central estimates for the parameters A,B,D

and Eare input into Equation (3) to produce a 95%

conﬁdence interval and central estimate of F. Below we will

focus on this parameter uncertainty as a means of determining

the quality of the ﬁt of the assumed statistical relationships

shown in Equation 2a and 2b. With the original authors, we

assume statistical independence of the four parameters and of

uand c, which of course is a ﬁrst approximation.

Incorporating uncertainty in NH

ﬂux estimates

With this knowledge in hand, we reproduced the published

ammonia emissions by Huijsmans & Hol (2012). The data

sets we obtained did not include an estimate of measurement

uncertainty, and hence, the results of our analyses are

overconﬁdent.

Table 2 reproduces Table 4 in Huijsmans & Hol (2012),

which includes data from four different ﬁelds from two

separate weeks, which we were able to reproduce without

issue. However, estimates of uncertainty are now included.

For example, the central estimate of Fas found in line 4 is

18.7 kg NH

-N/ha. After incorporating parameter

uncertainty, the 95% conﬁdence interval is 14.8–42.0 kg NH

N/ha. It is clear from examining Table 2 that the uncertainty

in the regression coefﬁcients produces sizeable uncertainties in

F. We could not discover uncertainty measures used by

authors employing the model of Ryden & McNeill (1984).

The uncertainties we present assume the validity of that

model, which itself is only a rough approximation of the

actual physics and chemistry. We only provide the

uncertainty inherent in the output of the model of Ryden &

McNeill (1984), which thus is still an underestimation of the

total uncertainty. The uncertainty in Fobviously is bounded

at the lower end by 0, and the empirical functional

relationship is logarithmic, which explains the asymmetry in

the conﬁdence intervals. The ﬂux Fitself is also limited to the

total amount of TAN applied and that from the ambient air.

In every case, the upper bounds of the conﬁdence intervals

are about two-times the central estimate, while the lower

bounds are about 10–20% of the central value.

Week 16 data of Table 2 for line 4 are shown in Figure 1.

For early periods (after manuring), Equation 2b does a poor

job at modelling the data. The data are below the model for

most of the range. In later periods, the approximation is

better, but as most of the NH

contribution comes from

earlier periods; overall the model of Ryden & McNeill (1984)

will overestimate values. Our analysis reveals this to be typical

for all the data we examined. Similar plots were made (not

shown) for windspeed and height, showing better ﬁts between

Equation 2a and measured data, but also showing the need

for uncertainty intervals. An interesting problem with the

model regression is seen in Periods 1 and 2, which are the

most important. At certain values of the log height, there are

predictions of negative NH

values, which of course are

physically impossible (Figure 1; red arrows). The assumption

of independence of the parameters for calculating the intervals

in Table 2 is open to question. As we are interested in the

predictive uncertainty of Fand given the shortcomings of the

regression models shown as Equation 2a and 2b detailed in

Figure 1, this is not unreasonable. Of course, the only way to

Table 2 Reproduction of Table 4 of Huijsmans & Hol (2012) including conﬁdence intervals around central estimates of F

Week NApplication

Ammonia emission kg NH

-N/ha Ammonia emission % NH

-N of total N applied

Mean [95% CI] Mean [95% CI]

16 1 On potato ridges +incorporation 1 28.7 [22.2, 50.9] 13.9 [10.8, 24.8]

2 On potato ridges +incorporation 2 23.9 [18.9, 47.1] 14.9 [11.7, 29.3]

3 In slits 1 29.7 [23.2, 69.1] 18.7 [14.5, 43.4]

4 In slits 2 18.7 [14.8, 42.0] 11.8 [9.32, 26.5]

17 5 On potato ridges +incorporation 1 15.6 [13.1, 23.1] 10.0 [8.44, 14.9]

6 On potato ridges +incorporation 2 25.5 [19.3, 151.0] 16.2 [12.3, 96.1]

7 In slits 1 37.1 [28.6, 88.2] 24.2 [18.7, 57.5]

8 In slits 2 25.6 [19.6, 60.9] 17.3 [13.2, 41.2]

The experiment consisted of two series of measurements executed with two different manuring techniques. The experiment was carried out twice

in weeks 16 and 17.

278 J. C. Hanekamp et al.

conﬁrm these or any assumptions is to verify (probabilistic)

predictions of Fwith actual measurement. The above has

implications for the emission calculations that are made on a

national scale and the connection with actual atmospheric

concentrations as measured by the LML network.

Ammonia in the Dutch atmosphere

Measurement of atmospheric concentrations of ammonia

helps to link the emission of ammonia with its deposition.

We agree with Erisman et al. (1998) that in ‘the causal

relation emission-concentration deposition effects, the

concentration and deposition observations might serve as a

test of the effectiveness of measures and the temporal

development of emissions.’ Sutton et al. (2003) noted that

‘[c]orrect interpretation of adequate atmospheric

measurements is essential, since monitoring data provide the

only means to evaluate trends in regional NH

emissions.’

The latter quote underscores the fundamental and (more

or less) exclusive empirical quality of atmospheric

concentrations above and beyond computed estimates of

ammonia emissions and deposition.

Atmospheric ammonia concentrations have been measured

since 1993 with the LML network. Ammonia is measured at

stations Vredepeel (S131), Huijbergen (S235), De Zilk

(S444), Wieringerwerf (S538), Leiduin (S540), Zegveld

(S633), Eibergen (S722), Lunteren (S734), Wekerom (S738),

Witteveen (S928) and Valthermond (S929); however, not all

stations have complete data sets either from 1993 or through

to 2014, the ﬁnal year for which we received data.

The individual data sets are aggregated to national

atmospheric averages to produce a mean of all the data from

all stations. The importance of these atmospheric

concentrations lies in the fact that ammonia deposition

calculations are calibrated with the data gathered through

the LML stations (see e.g. PBL, 2010). So, these empirical

data connect emission estimates and deposition calculations

and are thus highly important. The national averages seem

to indicate a decline between 1994 and 2004 from 10 to 8 lg/

, and thereafter, concentrations remain at roughly the

●

●●

–1.5 0.0 0.5 1.0

0200400600800

Perio d 1

Log height (m)

−N (µg/m

)

Data

Model

Bounds

●

Period 2

Log height (m)

●

Perio d 3

Log height (m)

●

●●●

Perio d 4

Log height (m)

●

●●

Period 5

Log height (m)

●

●●●●

Perio d 6

Log height (m)

●

●●●●

Perio d 7

Log height (m)

–1.0 –0.5

–1.5 0.0 0.5 1.0

–1.0 –0.5

–1.5 0.0 0.5 1.0–1.0 –0.5

–1.5 0.0 0.5 1.0

–1.0 –0.5

–1.5 0.0 0.5 1.0–1.0 –0.5

–1.5 0.0 0.5 1.0

–1.0 –0.5

–1.5 0.0 0.5 1.0

–1.0 –0.5

0200400600800

0 200 400 600 800

0200400600800

0 200 400 600 800

02004006008000 200 400 600 800

−N (µg/m

)NH

−N (µg/m

)

−N (µg/m

)NH

−N (µg/m

)

−N (µg/m

)

−N (µg/m

)

Figure 1 Log-linear empirical relationship between height and mean NH

concentration. (The small black circles indicate the actual data; data

were measured at seven periods (there were data missing in period 5 at height 3.16 m). Note the log scale for height. The dashed red line

represents Equation 2b, and the dotted black lines are the 95% uncertainty bounds to 2b.)

The Dutch ammonia-discourse critiqued 279

same level (Figure 2). To better understand the apparent

trend, we conducted a number of analyses. First, it is noted

in Figure 3 that the hourly LML data are highly variable,

both within and across stations. This persistent variability

makes summarizing the data, such as with a yearly national

average, problematical. Indeed, all one-number summaries of

data as complex as this necessarily omit crucial details and

this cautions against drawing overconﬁdent countrywide

conclusions.

As is evident from Figure 3, levels of NH

are high at

some stations (S131, S633 and S738), moderate at others

(S538, S722 and S734) and low at the rest (S235, S444, S540,

S928 and S929). Records are incomplete at four stations

(S540, S734, S928 and to some extent S929).

Figure 4 illustrates the difﬁculties of ascertaining

(national) trends. A popular method to deﬁne a trend is to

ﬁt a routine linear regression on the value of interest (here

) at the starting point of a time series to the endpoint,

1994

1996

1998

2000

2002

2004

2006

2008

2012

2010

(µg/m

)

Figure 2 Annual national average ammonia concentrations from

1994 to 2012 (RIVM, 2013). [Colour ﬁgure can be viewed at

wileyonlinelibrary.com]

100

200

300

400

500

1995

2000

2005

2010

2015

S131

100

200

300

400

500

S235

100

200

300

400

500

S444

100

200

300

400

500

S538

100

200

300

400

500

S540

100

200

300

400

500

S633

100

200

300

400

500

S722

100

200

300

400

500

S734

100

200

300

400

500

S738

100

200

300

400

500

S928

100

200

300

400

500

S929

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

NH3 (µg/m

)

NH3 (µg/m

)

NH3 (µg/m

)

NH3 (µg/m

)

NH3 (µg/m

)

NH3 (µg/m

)

NH3 (µg/m

)

NH3 (µg/m

)

NH3 (µg/m

)

NH3 (µg/m

)

NH3 (µg/m

)

Figure 3 Times series plots of atmospheric ammonia concentration for each LML station.[Colour ﬁgure can be viewed at wileyonlinelibrary.com]

280 J. C. Hanekamp et al.

using the date as the regressor. We performed a full trend

analysis of all the data of all stations.

Each point is the result of a different regression with a

different start date, but the same end date (the last date of

data for each station). The start date is stepped along the

series, starting with the ﬁrst date available, then the second

(excluding the ﬁrst), then the third (excluding the ﬁrst two

dates) and so on. This approach harbours a bias that is

frequently overlooked. The start and end dates of the data

series are arbitrary, and merely by choosing different dates,

trends oscillate between positive and negative, and become

signiﬁcant or non-signiﬁcant by the simple change of a date.

Some of this is due to the spiked nature of the series and

some due to the inherent variability of ammonia

concentrations. It is clear that no countrywide trend signal is

evinced, and neither is there any clear indication of a

consistent trend at any station (in contrast to ﬁndings by

Van Zanten et al., 2017).

Furthermore, every station records highly transient spikes

in NH

concentrations (Figure 5). Each subplot shows a

histogram of NH

concentrations at each individual LML

station. The data are grouped into ‘bins’ of NH

concentrations, and the frequency of observation for each

‘bin’ (count) is reported. Those points that occur with a

frequency <1 in a 1000 (0.1%) are shown with ‘+’ signs, to

highlight the large transient values. Clearly, none of the plots

have a normal distribution, and as a result, summary

measures like the mean and standard deviation can be

misleading (Galton, 1907).

With data that is symmetrical or even roughly ‘bell-

shaped’, a mean can be a useful one-number summary as an

expression of the average behaviour of the system. In cases

of high skew, which is evinced here, the median is preferred,

because it gives a more accurate summary of data tendency

and system behaviour. When data are symmetric, as when

using normal distributions to characterize uncertainty (which

is not defensible here), the mean and median coincide or are

very close. Thus, there is often little reason to prefer the

mean, unless there is some physical or causal reason. That

causal reason does not exist here. For instance, some

statistical models that represent ammonia concentrations

using normal distributions inappropriately characterize

measurements as a ‘global’ mean plus departures from that

mean as if some physical process ‘wants’ to ‘return’

ammonia levels to that mean. This makes no physical sense,

as the many causes for actual NH

concentrations are not

−1e−04

−5e−05

0e+00

5e−05

1e−04

1995 2000 2005 2010 2015

∆ NH3 (µg/m3) yr

S131

–1e−04

–5e−05

0e+00

5e−05

1e−04

S235

–1e−04

–5e−05

0e+00

5e−05

1e−04

S444

–1e−04

–5e−05

0e+00

5e−05

1e−04

S538

–0.010

–0.005

0.000

0.005

0.010

1995 2000 2005 2010 2015

S540

–1e−04

–5e−05

0e+00

5e−05

1e−04

S633

–1e−04

–5e−05

0e+00

5e−05

1e−04

S722

–0.010

–0.005

0.000

0.005

0.010

S734

–1e−04

–5e−05

0e+00

5e−05

1e−04

1995 2000 2005 2010 2015

S738

–1e−04

–5e−05

0e+00

5e−05

1e−04

S928

–1e−04

–5e−05

0e+00

5e−05

1e−04

S929

1995 2000 2005 2010 2015 1995 2000 2005 2010 2015 1995 2000 2005 2010 2015

1995 2000 2005 2010 2015 1995 2000 2005 2010 2015

∆ NH3 (µg/m3) yr

∆ NH3 (µg/m3) yr ∆ NH3 (µg/m3) yr ∆ NH3 (µg/m3) yr ∆ NH3 (µg/m3) yr

∆ NH3 (µg/m3) yr

Figure 4 Trend analyses for all LML stations. (Points which are ‘statistically signiﬁcant’ at 95% are black; those which are not ‘signiﬁcant’ are

red.)[Colour ﬁgure can be viewed at wileyonlinelibrary.com]

The Dutch ammonia-discourse critiqued 281

‘restorative’ in this way. Instead, factors such as soil

chemistry, fertilizer application timing, wind entrainment,

precipitation, atmospheric chemistry and other mechanisms

cause levels to continuously change, and often to swing

wildly and to spike, as the plots in fact show. All these

aspects are good reasons to choose the median as the most

accurate representation of the system’s behaviour.

The mean is highly sensitive to spikes in data. Table 3

summarizes mean and median for each LML station. The

mean/median ratio shows that the mean in this skewed data

is always higher than the median, with the smallest ‘boost’ in

the average being 27% (at S722). When the data series is

short, as it is at S540, the disparity is even larger, with the

mean being 310% larger than the median. Use of the mean

is therefore very misleading (Figure 6).

The wide heterogeneity of the LML data, both in

measurement and topography (station location and local

land use and so on), ﬁrmly argues against the use of an

annual nationwide mean across the LML stations (see

RIVM, 2013). This erroneously gives equal weight to each

station in the mean. Because of this, for instance, a station

that is on average ‘up wind’ will have more inﬂuence over

national totals than another that is downwind or near the

border of the country and which therefore cannot contribute

much to national agricultural emissions (Sutton et al., 2015).

To drive the point home with respect to the difﬁculty of

national yearly averages, a representative scatterplot of

S131

NH3 (µg/m3)

Count

0100200300400

010 000 30 000

++++ ++++++++++++++++++++ +++++++++++++++ ++++++++++++++++++++++ ++++++++++++++++++ ++++++++++++++++++++++++++ ++++++++ ++++++++ +++++++++++ +++++++++++++++++++++++ + +

S235

Count

0102030405060

05000 15 000 25 000

++++++++++++++++++++++++++++++++++++++++++++++++++++ + +++++++++++++++++++++++++++++++++ ++++++++++++++++++ ++++++++++++ ++++++++++++++++++++++ + ++++++++++++

S444

Count

01020304050

0 10 000 30 000

++++ ++++++++++++++++++++++++++++ +++++++ ++++++++++++++ ++++++++++++++++++++++++++ ++++++++++ + +++++++++++++++++++++++++++++++++ +++++++++++++++ + ++

S538

Count

0 50 100 150 200

020 00040 000

++++++++++++++ ++++++++++++++++++++++++++++++++ +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++ ++++++++++++++ ++++

S540

Count

010203040

0500 1000 1500

+++++

S633

Count

0 100 200 300

0 20 000 40 000 60 000

++++++++++++++++++++++++++++++++++++++++++++ ++++++++++ ++++++ ++++++++++ +++++ + +++++++++++++++++ +++++ +++++++++++++ +++++++++++++++++++++++++++++++++++ +++++++

S722

Count

050100150200

0 10 000 20 000

+++ ++++++++++++++++++++++++ ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++++ ++++ ++++++++++++++++++++++++ +

S734

Count

0 50 100 150 200 250

0200 400 600 800

++++++++

S738

Count

0100200300400500

0 10 000 30 000

+++++++++++++++++++++++++++++++++++ ++++++ ++ ++++++++++++++++++++++++++ +++ +++ ++++++++++++++++++++++++++++ ++++++++++++++++++ ++++++++ ++++++++++ +++++ +++

S928

Count

01020304050

0 2000 6000

+++++++++ + +++++++++++++++ + +++++++ +++++++++++++ +

S929

Count

020406080

100

05000 15 000

+++++++++++++++ ++++++++++++ ++++++++++++++++++++++++++++++++++++++++++ + + ++++++++++++++ ++++++ ++++++++

++++

NH3 (µg/m3)NH3 (µg/m3)NH3 (µg/m3)

NH3 (µg/m3)NH3 (µg/m3)NH3 (µg/m3)NH3 (µg/m3)

NH3 (µg/m3)NH3 (µg/m3)NH3 (µg/m3)

Figure 5 Histogram of atmospheric NH

concentrations at each LML station.

Table 3 Mean and median of all hourly measurement values per

LML station and ratios of mean to median

Name

Mean

lg/m

Median

lg/m

Mean/median ratio

S131 18.37 13.13 1.40

S235 2.84 1.76 1.61

S444 1.83 0.82 2.23

S538 4.73 2.84 1.67

S540 2.67 0.86 3.10

S633 9.01 6.10 1.48

S722 9.48 7.45 1.27

S734 21.65 15.99 1.35

S738 16.86 11.56 1.46

S928 2.46 1.63 1.51

S929 4.35 3.19 1.36

282 J. C. Hanekamp et al.

hourly NH

values for two stations, S444 and S538, is given

in Figure 7. This and the other plots we have produced (not

shown) reveal that hourly NH

concentrations at one station

are not well correlated with NH

concentrations at other

stations. This implies that the causes of NH

in the

atmosphere are localized and not countrywide. There is some

1995 2000 2005 2010 2015

10 15 20

S131

NH3 (µg/m3)

1.0 2.0 3.0

S235

0.5 1.0 1.5 2.0 2.5

S444

123456

S538

0.0 1.0 2.0 3.0

S540

4 6 8 10 12 14

S633

6 8 10 12

S722

12 14 16 18 20 22

S734

5 10 15 20

S738

1.5 2.0 2.5

S928

123456

S929

1995 2000 2005 2010 2015

1995 2000 2005 2010 2015 1995 2000 2005 2010 2015 1995 2000 2005 2010 2015

1995 2000 2005 2010 2015

NH3 (µg/m3)

Figure 6 Monthly mean (solid) and monthly median (dotted) atmospheric ammonia concentrations for all LML stations.

200

150

100

020 40 60

S444

Correlation between two LML stations

S538

Figure 7 A representative scatterplot of NH

concentrations for two stations, S444 and

S538.

The Dutch ammonia-discourse critiqued 283

unsurprising evidence of a seasonal cycle embedded in the

variability at each station. This localization argues against a

countrywide single-number annual summary. Obviously,

correlation necessarily increases between stations when data

are aggregated to, say, monthly averages. This is purely the

result of smoothing the data and does not in any way

indicate increases in causal relationships or ‘links’ between

sets of data (Briggs, 2016).

Unravelling emission and concentration estimates

In the years after the LML network became operational, it

was noticed that atmospheric ammonia concentrations did

not greatly change. That is unexpected because several

emission reduction measures had been put in place and

emissions in this period were estimated to have decreased by

35%. Later, this discrepancy between calculated emissions

and the LML concentration measurements was dubbed the

‘ammonia gap’. Erisman et al. (2001) deﬁned this gap as ‘the

difference observed between the estimated NH

emission

levels in the Netherlands and those indicated by atmospheric

measurements’. Sutton et al. (2003) take Erisman et al.’s

deﬁnition to its logical next step. They state that the ‘gap’ is

‘the lack of a detectable reduction in NH

concentrations

following the implementation of abatement measures in

1993.’ Here, a downward signal is expected in the

atmospheric ammonia concentrations as a result of

abatement measures, which nevertheless does not show. That

missing signal is referred to as the ‘gap’. Conversely, in 1995,

the RIVM issued a report in which an evaluation was

presented of calculations of atmospheric ammonia

concentrations in the Netherlands related to measurements

(RIVM, 1995). With the aid of annual averaged values,

modelled ammonia concentrations that are based on national

emission estimates were 27% lower than the measured

concentrations. This discrepancy persisted despite adding

16% to estimated national emissions. Although the report

does not mention the ‘ammonia gap’ as such, the ﬁrst

outlines thereof seem to emerge with the discrepancy deﬁned

as the difference between averages of measured and modelled

atmospheric ammonia concentrations. In line with this, De

Ruiter et al. (2006) deﬁned the ‘ammonia gap’ as the

difference between the modelled atmospheric ammonia

concentrations (with the operational priority substances

(OPS) model) and the measured concentrations, the former

being substantially lower (by some 25 to 30%) than the latter

(see Velders et al., 2010). So, the ﬁrst deﬁnition of the

‘ammonia gap’ focuses on the ostensible difference between

estimated ammonia emissions (which ostensibly change) and

the measured atmospheric ammonia concentrations (which

do not change), whereas the second deﬁnition revolves

around the discrepancy between modelled and empirically

measured atmospheric ammonia concentrations. It seems

that considering the ﬁrst deﬁnition, the ‘gap’ is now bigger

than ever.

There is a rich scientiﬁc literature on the Dutch ammonia

emissions. This gives the opportunity to see how the

emission calculations changed but also how views evolved.

In the 1980s, estimates of the total ammonia emissions in the

Netherlands were in the order of 200 kt/year (Erisman,

1989). Between 1990 and 2016 ammonia emissions estimates

roughly doubled with most of the increase occurring around

1990. The biggest change in historical emissions was due to

increases in emission factors for manure application (see Van

Bruggen et al., 2011). In the nineties, an emission factor of

50% of applied TAN for surface spreading was assumed

(Van Bruggen et al., 2011, p. 107). Following the work of

100

150

200

250

300

350

NH3 (kt)

Emission of ammonia from agricultural sources

2015, RIVM

2011, Van Bruggen et al.

2006, De Ruiter et al.

2000, Van Jaarsveld et al.

1989, Erisman

1980

1982

1984

1986

1988

1990

1992

1994

1996

2000

1998

2002

2004

2006

2008

2010

2012

2014

Figure 8 Different estimates (with references)

of ammonia emission trends.[Colour ﬁgure

can be viewed at wileyonlinelibrary.com]

284 J. C. Hanekamp et al.

Huijsmans (2003), the emission factor for 1990 was changed

to 74%, as surface spreading of manure was the

predominant method at the time. As low-emission techniques

for manure application were introduced from 1991 onwards

(see Table 1 for examples), the estimated emissions have

decreased considerably according to the updated calculations

(Figure 8).

Given the substantial uncertainties relating to (nationwide)

emissions, expressed in our discussion of the use of the

model of Ryden & McNeill (1984), and all the complex

processes that take place in the atmosphere, it is not

surprising that trends in emissions and atmospheric

concentrations differ or that model calculations are at odds

with measured concentrations. Nevertheless, in terms of

emission factors, overconﬁdence is inbuilt in the national

emission inventories. The National Emission Model for

Agriculture (NEMA) model is used to calculate agricultural

emissions to the atmosphere in the Netherlands, with

ammonia emission factors are simply speciﬁed as one

decimal numerals (Table 4).

Coining the term ‘ammonia gap’ suggests that discrepancies

between either the calculated atmospheric ammonia

concentrations and actual measurements or changing emissions

and measured (roughly) unchanging atmospheric concen-

trations are somehow physically real: this evidently is a false

notion. More precisely, the gap has been ‘reiﬁed into reality’.

Reiﬁcation is a widespread and classical fallacy dubbed by

Alfred North Whitehead (1925) as the ‘fallacy of misplaced

concreteness’. Reiﬁcation is making something which is

hypothetical or abstract physically real. Abstractions in science

are quite common and necessary; however, trouble arises when

we start to think of abstractions as if they were concrete

realities themselves –thereby ‘reifying’ them. This

predicament is intensiﬁed when we think of the abstractions as

somehow more real than the concrete realities from which they

have been abstracted (see Briggs, 2016). Using the mean

averages for the LML data and even aggregating the data to a

national scale exacerbates this reiﬁcation. Reducing the

temporally and spatially highly variable atmospheric

concentrations to one annual number produces over-certainty,

or at least masks the substantial uncertainty present. The

median as the correct and thereby more accurate expression of

the complex data lowers local atmospheric ammonia

concentrations substantially (see Table 3 and Figure 6).

Concluding discussion

We are aware that we have put large question marks over

seemingly settled aspects of the ammonia discourse in the

Table 4 Manure application techniques and emission factors as

speciﬁed in the NEMA model

Manure application, techniques and

emission factors

Share,

Emission

Factor (EF), %

Grassland –slurry

Shallow injection 61 19.0

Slit coulter 13 22.5

Narrow band 25 26.0

Surface spreading 1 74.0

Cropland (uncropped) –slurry

Deep placement 71 2.0

Shallow injection 9 24.0

Slit coulter 9 30.0

Narrow band 7 36.0

Direct incorporation 4 22.0

2-pass incorporation 0 46.0

Surface spreading 0 69.0

Manuring before crops were planted.

1990 1995 2000 2005 2010 2015

0 10 20 30 40

concentrations

Date

LML NH3 (µg/m3)

S738

S444

150 200 250 300 350

NH3 Emissions (Gg)

LML

Emissions

Figure 9 Estimated ammonia emissions (Gg)

since 1990 (RIVM, 2015) and ammonia

concentrations at two LML stations, one

high and one low measurement; median

values between 1993 and 2014.[Colour ﬁgure

can be viewed at wileyonlinelibrary.com]

The Dutch ammonia-discourse critiqued 285

Netherlands. Firstly, the non-availability of the experimental

data that underpin current agricultural policies and form a

key part of the paper of Huijsmans et al. (2016) is worrying.

Reproducibility is a crucial epistemic value that nowadays is

at the forefront of scientiﬁc discussion (e.g. Ioannidis, 2005;

Horton, 2015; Peng, 2015; Browman, 2016; Munaf"

oet al.,

2017).

Secondly, we have shown that the uncertainty in ﬁtting

experimental emission data is substantial, and yet, model-

generated conﬁdence intervals for the central emission

estimates are not reported (Table 2). Emission factors

derived from experiment and modelling (Ryden & McNeill,

1984) are published and used with an amount of

overconﬁdence and accuracy that is not appropriate for the

actual data gathering and subsequent statistical work (see

e.g. Vonk et al., 2016).

Changing manure application from broadcast spreading to

shallow injection, for instance, is likely to have resulted in

emission reduction in the Netherlands, but by how much

remains unknown. Our analysis of one data set gives enough

insight into the model uncertainty to seriously question the

NEMA model in which emission factors from experimental

research are applied to national emission inventories with a

precision of one decimal place. Even our own analysis is

likely to be fraught with overconﬁdence, as for instance,

measurement uncertainty is unknown. As a result, emission

factors in the NEMA model carry sizeable uncertainties that

are not made explicit and should carry over to the emission

totals produced by the model. Indeed, Sutton et al. (2015)

remark that no overall synthesis in the uncertainty in the

trend of all contributions to Dutch ammonia emissions has

been conducted.

We would assert that the uncertainty in the output of the

equation given in Ryden & McNeill (1984) at least would result

in emission uncertainties that would perhaps result in overlap

between the different manure application techniques (see

Table 1). Thus, the widely reported dividing lines between

emissions from different manure application techniques are

likely blurred. In some ways, this is expressed in Figure 9 where

the ammonia emission estimates since 1990 are overlaid with

LML-median value series for two stations, one with high and

the other with low atmospheric ammonia concentrations. The

1990 value for estimated ammonia emissions is 373 kt (RIVM,

2015) or 372 kt (RIVM, 2016). Clearly, there is no correlation

between estimated emissions and atmospheric median

concentrations. The picture is similar for all LML stations:

Thirdly, current ofﬁcial output from the LML database

leaves much to be desired. Two things are quite clear from

our analyses: (i) median values should be the default when

summarizing the highly skewed LML data set as it best

describes the average system behaviour; (ii) no countrywide

trend signal is evinced from the hourly LML data, and

neither is there any consistent trend at any station. Use of

the mean gives a misleading picture and suggests ‘average’

atmospheric ammonia values that are simply too high

(Galton, 1907). Ammonia concentrations in the Dutch

atmosphere are substantially lower than reported in the

ofﬁcial outputs, resulting, we venture, in lower subsequent

deposition. Actual ammonia emissions could be lower as

well, although this was explicitly not researched here. With

the reduction of data to a one-number annual summary,

much useful information is being discarded.

Finally, the extensive research efforts over the past thirty

years in the Netherlands have provided much valuable data on

the agricultural impact on the local and national environment

and beyond. However, lack of data transparency,

oversimpliﬁed statistical procedures, and the resulting

spurious accuracy of published and applied emission results

dilutes the usefulness of these research efforts. This has

created an environment in which results used for regulatory

purposes seemingly can be employed without the usual and

compulsory scientiﬁc provisos. This needs to change and we

have provided some tools to make that change happen.

Acknowledgements

We thank Addo van Pul of the RIVM for graciously making

available all LML data sets and further open discussions.

Equally, we want to thank Gerard Velthof of Wageningen

University and Research for sending us the NEMA model.

Wageningen University is also thanked for making available

some data sets of ammonia emission manure experiments. The

two anonymous referees are graciously acknowledged for their

constructive and insightful critique.

This research is made possible through a crowd-funding

effort resulting from the authors’ interest in this subject.

Geesje Rotgers of V-Focus is graciously acknowledged for

her tireless coordination efforts (http://www.v-focus.nl/

ammoniak2015/in Dutch). As a result, we do not have any

competing interests.

References

Behera, S.N., Sharma, M., Aneja, V.P. & Balasubramanian, R.

2013. Ammonia in the atmosphere: a review on emission sources,

atmospheric chemistry and deposition on terrestrial bodies.

Environmental Science and Pollution Research,20, 8092–8131.

Briggs, W. 2016. Uncertainty –The soul of modeling, probability &

statistics. Springer, Switzerland.

Browman, H.I. 2016. Applying organized scepticism to ocean

acidiﬁcation research. CES Journal of Marine Science,73, 529–536.

De Ruiter, J.F., van Pul, W.A.J., van Jaarsveld, J.A. & Buijsman, E.

2006. Zuur- en stikstofdepositie in Nederland in de periode 1981–

2002. Rapport 500037005/2006, Milieu- en Natuurplanbureau,

Bilthoven, The Netherlands. (In Dutch.)

Denmead, O.T. 1983. Micrometeorological methods for measuring

gaseous losses of nitrogen in the ﬁeld. In: Gaseous loss of nitrogen

from plant-soil systems (eds J.R. Freney & J.R. Simpson), pp. 133–

157. Kluwer Academic Publisher, Dordrecht.

286 J. C. Hanekamp et al.

Erisman, J.W. 1989. Ammonia emissions in the Netherlands in 1987

and 1988. Report nr. 228471006, Bilthoven, The Netherlands.

Erisman, J.W., Bleeker, A. & Van Jaarsveld, J.A. 1998. Evaluation

of ammonia emission abatement on the basis of measurements

and model calculations. Environmental Pollution,102, 269–274.

Erisman, J.W., Mosquera, J. & Hensen, A. 2001. Two options to

explain the ammonia gap in the Netherlands. Environmental

Science & Policy,4, 97–105.

Galton, F. 1907. One vote, one value. Nature,75, 414.

Horton, R. 2015. What is medicine’s 5 sigma? Lancet,385, 1380.

Huijsmans, J.F.M. 2003. Manure application and ammonia

volatilization. PhD thesis, Wageningen University.

Huijsmans, J.F.M. & Hol, J.M.G. 2012. Ammoniakemissie bij

mesttoediening en inwerken in aardappelruggen en bij mesttoediening

in sleuven op niet beteeld geploegd kleibouwland. Plant Research

International, onderdeel van Wageningen UR Business Unit

Agrosysteemkunde, rapport 445. (In Dutch.)

Huijsmans, J.F.M. & Schils, R.L.M. 2009. Ammonia and nitrous

oxide emissions following ﬁeld-application of manure: state of the

art measurement in the Netherlands. International Fertiliser

Society Proceedings,655,1–36.

Huijsmans, J.F.M., Schr€

oder, J.J., Mosquera, J., Vermeulen, G.D.,

Ten Berge, H.F.M. & Neeteson, J.J. 2016. Ammonia emissions from

cattle slurries applied to grassland: should application techniques be

reconsidered? Soil Use and Management,32(Suppl. 1), 109–116.

Ioannidis, J.P.A. 2005. Why most published research ﬁndings are

false. PLoS Medicine,2, e124.

Munaf"

o, M.R., et al. 2017. A manifesto for reproducible science.

Nature Human Behaviour,1,1–8 https://doi.org/10.1038/s41562-

016-0021.

Peng, R. 2015. The reproducibility crisis in science: a statistical

counterattack. Signiﬁcance,12, 30–32.

Planbureau voor de Leefomgeving (PBL –Netherlands Environmental

Assessment Agency). 2010. Grootschalige stikstofdepositie in

Nederland. Herkomst en ontwikkeling in de tijd. Rapport

500088007/2010, Den Haag, The Netherlands. (In Dutch.)

Planbureau voor de Leefomgeving (PBL –Netherlands

Environmental Assessment Agency). 2016. Balans van de

Leefomgeving 2016. Richting geven –Ruimte maken. PBL-

publicatienummer 1838, Den Haag, The Netherlands. (In Dutch.)