Environmental Pollution 118 (2002) 225–238

www.elsevier.com/locate/envpol

Limits to nitrogen use on grassland

H.F.M. ten Bergea,*, H.G. van der Meera, L. Carlierb, T. Baan Hofmana, J.J. Neetesona
a
Plant Research International, PO Box 16, 6700 AA Wageningen, The Netherlands
b
Centrum voor Landbouwkundig Onderzoek, Burg. Van Gansberghelaan 109, B-9820 Merelbeke, Belgium

Received 10 January 2001; accepted 31 August 2001

‘‘Capsule’’: A descriptive crop N response model was used to identify permissible N doses that do not result in
accumulation of residual mineral soil N.

Abstract
Data from nitrogen (N) response experiments on grassland in Belgium and the Netherlands were analysed with the help of a
descriptive crop N response model, to identify permissible doses below which no accumulation occurs of residual mineral soil N in
autumn, Nmin. Using different years as separate sets, a total of 29 data sets were obtained from eight locations on various soil types.
A large variation was found in Nmin base levels (unfertilised) between locations and between years at a given location. For doses low
enough not to affect crop N recovery, every 100 kg N applied was associated with 3–4 kg residual Nmin in autumn. This is con-
sidered very low compared to Nmin base levels, but values differed significantly from zero. After normalising N-doses from different
sources (mineral fertiliser and cattle slurry) with the help of a coefficient expressing effectiveness based on crop N uptake, no dif-
ference was found between fertiliser and slurry in terms of their effect on residual Nmin. The above also holds for nitrate leaching as
measured. The sources do differ, however, with respect to long-term effects and these are quantified with a first-order approxima-
tion. It it shown that, also after incorporation of long-term effects, much higher N-doses on grassland are justified than the 170 kg
N per ha per year in animal manures currently proposed by the European Commission. On normal productive cut grassland as in
the analysed experiments, total N doses in cattle slurry up to 400 kg per ha per year have very little effect on residual Nmin, if not
accompanied by high fertiliser doses. Introducing limits to the use of animal manures on grassland without limiting the input of
mineral fertiliser-N lacks any scientific ground. # 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Grassland; Nitrogen; Residual mineral soil nitrogen; Nitrate leaching; Critical dose; Long term effect; Cattle slurry; Fertiliser

1. Introduction to grassland in the past decades. In recent years, legis-

In parts of western Europe where agriculture is
intensive, there is general concern that nitrogen (N)
derived from animal manures and fertilisers applied to
crop land represents a threat to the quality of shallow
groundwater and surface waters. Intensive dairy prod-
uction is one of the sectors known for its high N
inputs. The very large capacity of grass—compared to
other crops—to continue absorbing N and responding
with additional growth up to very high N-input levels, is
associated with approximately three times higher pro-
tein contents in grass than in most arable crops. This
capacity has induced farmers to apply excessive N-doses
* Corresponding author. Tel.: +31-317-475951; fax: +31-317-
423110.
E-mail addresses: h.f.m.tenberge@plant.wag-ur.nl (H.F.M. ten
Berge).
0269-7491/02/$ - see front matter # 2002 Elsevier Science Ltd. All rights reserved.
PII: S0269-7491(01)00315-3
lation in several EU countries has forced farmers to
reduce input levels to comply with the Nitrates Directive
of the European Commission. The Directive states that
the NO-3-concentration in groundwater that is intended
to be used for drinking water may not exceed 50 mg/l
(Nitrates Directive 91/676 EEC; Commission of the
European Communities 1991). Different measures are
already in place or are being imposed by the various
national governments to meet this directive. Measures
vary from maximum-dose limits to detailed instructions
on when, where and how to apply N in various forms.
The Netherlands takes a whole-farm approach by
keeping farm mineral balances and taxing N and P sur-
pluses when they exceed an allowed limit. Alternatively,
some countries introduced a permissible amount of
residual mineral N in the soil profile after the last crop
harvest in autumn (Nmin) as an instrument to monitor
and control farmers’ N management practices.
226 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238
For any of these approaches to be just, thresholds
introduced must be founded on quantified effects of
applied N-doses on nitrate loss. Several experiments
have been conducted in the past to establish such rela-
tions, including for grassland. Since it is difficult to
measure nitrate leaching under field conditions, residual
nitrogen Nmin has been observed in most cases as an
indicator of potential nitrate leaching and we follow this
in our analysis. In this study we compile the results from
grassland experiments in Belgium and the Netherlands
where the response of Nmin to applied N-doses was
measured. Whereas the Nmin responses to applied doses
were documented in the various source reports cited, we
take the analysis one step further by using a crop N-
response model (QUADMOD) as a mechanism or frame
to assess permissible doses in function of local conditions.
The purpose of introducing this frame is five-fold: (1)
it enables incorporation of local conditions, notably
soil fertility and attainable biomass production, in
defining permissible N-doses; (2) once a robust relation
between model parameters and Nmin is established, it
can be used to analyse available data from the numer-
ous experiments where crop response (biomass and N-
yield) was observed but where Nmin observations are
missing; this allows to explore effects of spatial and
temporal variation and of management practices on
critical doses; (3) a crop response model enables to
assess yield loss resulting from constraints on N-input
and thus to evaluate the economic impact of imposed
limits; (4) N-surplus can be directly assessed as the dif-
ference between N-dose and crop N-offtake for its use in
surplus-based legislation; and (5) such frame permits to
interpret permissible doses in a long-term perspective
and for different types of N-manures, as shown in this
paper.
The QUADMOD-based analysis identifies the range
of N-doses where the apparent recovery of applied N by
the crop remains constant, that is, independent of the
dose. The upper end of this domain is the so-called criti-
cal point, and it is investigated whether this point can
serve as a threshold to define the safe permissible N-
dose with respect to residual Nmin. The hypotheses tested
here are (1) that doses below this critical point do not
augment Nmin beyond the base level found in unferti-
lised fields, and (2) that the effects of N-dose on Nmin
are independent of the source type (mineral fertiliser;
cattle slurry), once doses are normalised based on crop
N-uptake.
2. Materials and methods
2.1. Experiment locations
We analysed data sets from cut grass experiments
where harvested annual biomass, annual N yield, and
residual mineral soil nitrogen in late autumn (hereafter
referred to as Nmin) were measured in response to
applied nitrogen doses. All experiments included at least
four levels of applied N, including a zero N-input treat-
ment. The experiments were conducted at Ruurlo,
Droevendaal (near Wageningen), and Heino in the
Netherlands; and at Merelbeke, Tienen, Middelkerke,
Geel, and Bree in Belgium. The trials in the Netherlands
were three independent experiments. The five locations
in Belgium, however, all participated in a coordinated
scheme executed simultaneously at all sites. General
features of the sites are given in Table 1. At each location,
measurements were continued during three, four or five
consecutive years (Table 2).
The experimental field at Ruurlo was well drained in
summer, with groundwater at 1.20–1.50 m below the
soil surface, but wet in winter, with groundwater at
0.40–0.50 m depth, periodically rising to 0.20 m depth.
At Heino, the experiment was conducted on a typical
‘plaggen-soil’, a profile enriched during centuries with
farmyard manure and hence now characterised by a
0.80–1.0 m thick ‘plaggen epipedon’. The field was well
drained with water tables never shallower than 2.0 m
depth, and sensitive to drought in summer. The Droe-
vendaal site was situated on a fine sandy soil, homo-
genized by deep plowing down to 1 m some 20 years
before the experiment was conducted. Groundwater
fluctuated between 1.20 m depth in dry spells and 0.20
m depth during winter season. Upward water flow at
about 1 m depth prohibited drainwater collection and
nitrate leaching could therefore not be determined. The
soils at Merelbeke, Tienen and Geel were well-drained
permeable profiles; the experiment site at Middelkerke
was situated on a poorly drained heavy clay soil with
groundwater near the surface during winter periods. At
Bree, spells of drought occurred during summer due
to the coarse soil texture and absence of groundwater
near the root zone.
Table 1
General features of the experiment locations (further details are given
in the text)
Soil properties Hydrological position
Ruurlo (Netherlands) Sand topsoil, Wet (winter) to
loamy sand well drained
subsoil (summer)
Heino (Netherlands) Sand Well drained to dry
Droevendaal (Netherlands) Sand Wet (winter) to
well drained
(summer)
Merelbeke (Belgium) Sandy loam Well-drained
Tienen (Belgium) Loam Well-drained
Middelkerke (Belgium) Clay Poorly drained
Geel (Belgium) Sandy loam Well-drained
Bree (Belgium) Loamy sand Well drained to dry
 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238 227

Table 2
Treatment details of experiments at locations in The Netherlands (NL) and Belgium (B)

Location Setsa Year Fertiliser dose (kg N/ha) Cattle slurry dose (1000 kg/ha) Nmin Nitrate leaching
measured measured

Ruurlo (NL) 1–5 1980–1984 0, 200, 400, 600 0,10,20,40 SAb + +

0,10,20,40,80 injected + +
Heinoc (NL) 6–8 1982–1984 0, 200, 400, 600 0 + +
Heinod (NL) 9–11 1982–1984 0, 220, 440, 660 0 + +
Droevendaal (NL) 12–15 1987–1988 0, 200, 400 0, 60 + –
600 0
1989–1990 0, 200, 400 0, 60
600 0
Merelbeke (B) 16–18 1996–1998 0, 100, 200, 300, 400 0, 20, 40 + –
Tienen (B) 19–21 1996,1998 0, 80, 160, 240, 320 0, 20, 40 + –
1997 0, 100, 200, 300, 400 0, 20, 40
Middelkerke (B) 22–24 1996–1998 0, 100, 200, 300, 400 0, 20, 40 + –
Geel (B) 25–26 1996–1997 0, 100, 200, 300, 400 0, 20, 40 + –
Bree (B) 27–29 1996–1998 0, 100, 200, 300, 400 0, 20, 40 + –
a
Sets numbered in chronological order.
b
SA, surface application of slurry.
c
Without supplemental irrigation.
d
With supplemental irrigation.

2.2. Treatments Sluismans (1992). Further details for the Ruurlo

Treatment details are listed in Table 2. At all loca-
tions except Heino, graded doses of mineral fertiliser-N
were combined with different doses of cattle slurry,
usually injected but in one case (Ruurlo) also surface-
applied. All possible combinations of the slurry and
fertiliser doses listed in the table for the respective loca-
tions were included as treatments in the experiments.
For example, at Droevendaal-1988 a total of
(3
2)+1=7 combinations of N-inputs were imposed as
treatments, excluding additional treatments omitted in
this study.
The Ruurlo experiment was conducted to determine
the effects of injected and surface applied cattle slurry
on N uptake, biomass yield, Nmin, and nitrate leaching.
It was laid out as split plot design, with slurry treat-
ments and Controls as main plots, and mineral fertiliser
treatments as sub-plots. The scheme was replicated
three times, with subplots measuring 15 m by 2.5 m.
Nmin was measured in samples collected from four bor-
ings per plot, so 12 per treatment, down to 1.0 m depth,
which were separated into 0.20 m sections and then
mixed per depth interval before analysis. Samples were
collected in early spring and then after every harvest,
immediately stored cool in the field, and analysed the
subsequent day. Only the data from the last sampling
per season (after last harvest) are used here. Nitrate
leaching was estimated in selected plots by sampling dur-
ing winter via suction cups inserted at 1 m depth. More
details on the assessment of nitrate leaching in this
experiment were given by Fonck (1982a, b, 1986a–c); the
Nmin measurements were reported by Wadman and
experiment are given in Van der Meer et al. (1987).
The experiment at Heino was a split plot setup with
three levels of water supply (main plots) and four levels
of fertiliser-N as subplots, to assess the effects of water
supply on biomass production and N-use in cut grass.
The scheme was executed in fourfold. The water supply
levels were: (1) natural rainfall; (2) irrigation when soil
moisture tension at 0.25 depth reached 50 kPa; and (3)
irrigation when soil moisture pressure at 0.25 depth
reached 20 kPa. Only the results of water supply levels
(1) and (3) are used here. The subplots measured 6
5 m.
No cattle slurry was applied at Heino. The amount of
mineral nitrogen present in the soil profile was deter-
mined in early spring and in autumn after the last har-
vest, down to 1.0 m with depth intervals of 0.20 m.
Twelve borings were collected (three per replicate sub-
plot) from objects with low N-doses (0 and 200 kg/ha),
and 48 borings (12 per replicate subplot) in the high-N
treatments (400 and 600 kg N/ha). Samples were mixed
per object and depth interval (low N-doses) and per
replicate subplot and depth interval (high N-doses)
before analysis. To determine nitrate leaching, soil
samples were collected between 0.8 and 3.0 m depth by
auger (0.25–0.50 m sections) and nitrate concentration
was determined in the extracted water. The depth of the
leaching front corresponding to the preceeding season
was determined by water balance calculation, and the
amount of nitrate leached per year was assessed by
integration over depth, from 0.80 m down to the leach-
ing front.
Data at Droevendaal were collected in an experiment to
assess the utilisation of (injected) slurry-N by cut grass as a
228 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238
function of application time and presence or absence of
nitrification inhibitor DCD. Slurry was applied either in
October, December, or early spring (between 25 Feb-
ruary and 2 April). We used for this analysis only the
data from spring application treatments without DCD.
All treatments were replicated six times, except for the
Control (no fertiliser nor slurry) which was laid out on
12 plots. Plot size was 11 m
3 m. Fertiliser-N was applied
at three levels in combination with cattle slurry, while a
fourth level of fertiliser-N dose (600 kg N /ha) was only
given in absence of cattle slurry (Table 2). In contrast to
experiments at the other sites, the timing of split fertili-
ser applications at Droevendaal was optimised with
respect to the expected time pattern of N-release from
the applied slurry. Slurry contained here on average 2.5
kg N in mineral form and 5.1 kg total N per 1000 kg.
Grass was harvested in 6–7 cuts per season. Nmin was
sampled after the last harvest, in intervals of 0.0–0.20,
0.20–0.40 and 0.40–0.60 m depth. Six borings were
taken per plot, samples were mixed per two replicate
plots and per depth interval, immediately cooled in the
field upon collection, and analysed on the following day.
Purpose of the experiment in Belgium was to assess
effects of fertiliser doses and doses of injected cattle
slurry on residual Nmin in the soil profile after the last
harvest, under cut grass. Setup and procedures were
identical across all five sites. The layout was a rando-
mized block design with four replications. The size of
subplots was 25
10 m. Samples for determination of
Nmin were composed by mixing samples from 25 borings
per plot, split into depth intervals of 0.30 m down to
0.90 m depth, except at Bree in 1996 and 1997 where
Nmin was sampled at depths 0.0–0.30 and 0.30–0.60 m
only. Slurry injection techniques and depths varied,
however, between sites and years. Details of the experi-
ments at the sites in Belgium were reported by Van
Bockstaele et al. (1996, 1997, 1998).
The Nmin values reported in this study refer to total
depth intervals of 0.0–1.0 m (Ruurlo, Heino), 0.0–0.90
m (Merelbeke, Tienen, Middelkerke, Geel; and Bree-
1998), and 0.0–0.60 m (Droevendaal; Bree-1996, and
Bree-1997). The values represent the sum of measured
amounts of nitrate-N and ammonium-N, except at Bree
(1996, 1997) where only nitrate-N was determined. Nmin
was always determined after the last harvest, usually
late October or November.
2.3. Data sets
The results obtained at a specific location in a parti-
cular year were regarded for the current purpose as one
data set. The Heino data were considered as two sepa-
rate sets per year, because of the two contrasting water
regimes.Thus, the above experiments yielded in total 30
data sets. One of these was regarded invalid (Geel, 1997)
and was omitted, leaving 29 sets for further analysis. All
the individual data points in the sets as used in this study
represent treatment means, based on observations on
three or four replicate plots. This is so for biomass, N-
yield, Nmin, N-leaching (where measured) and,
obviously, applied doses. All symbols presented in fig-
ures refer to treatment means, and all statistics are based
on treatment means as primary information, thus ignor-
ing underlying information at the level of replicate plots.
2.4. Effective N in cattle slurry
We expressed N-doses in slurry as quantities of effec-
tive N (kg N/ha), to enable a comparison between N
inputs from mineral fertilisers and from cattle slurries.
The effectiveness or ‘working coefficient’ w of N in
slurry was defined as the equivalent dose of fertiliser (kg
N/ha) required to induce the same increment of crop N-
yield (with reference to the zero-N input treatment) as 1
kg of total slurry-N per ha. So if, for example, the
application of 100 kg fertiliser-N per ha results in
the same N-yield increment (e.g. 60 kg N per ha) as a
dose of 250 kg of total slurry-N per ha, the working
coefficient w for slurry-N equals 0.4 under the prevailing
set of conditions, that is, slurry quality, soil and weather
conditions, and application technique. The concept of
‘working coefficient’ was first introduced by Kolen-
brander and de la Lande Cremer (1967). This coefficient
w was determined for every location, year and slurry
dose. Its value differed by dose within the same year–
location, because the number of split applications and
the time of application differed with the total dose,
and hence different slurry batches (qualities) were
involved and environmental conditions during applica-
tion varied. While each slurry dose was combined with
four or five doses of mineral fertiliser (Table 2), only
treatments with up to 200 kg fertiliser-N were used in the
calculation of w, to avoid underestimation resulting from
reduced crop N demand. The effectiveness of slurry-N
was fairly stable for the experiments at Ruurlo and
Droevendaal, and an average value of 0.58 was used in
all calculations for injected slurry at these sites, and 0.24
for surface-applied slurry. However, w differed markedly
between locations, years and doses in the Belgian
experiments and hence case-specific w-values were used.
The doses of slurry-N applied were thus converted for
all available data sets into doses of effective N, i.e. fertili-
ser equivalents, based on the corresponding w-values.
Note that the above definition of effectiveness is entirely
based upon incremental crop N-uptake and that no
information on Nmin was used in the assessment of w,
which is a prerequisite for our interpretation of the results.
2.5. QUADMOD model
QUADMOD is a descriptive model that was fitted to
observed data of crop response to doses A of effective N
 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238
applied (kg N/ha). Crop response is expressed in terms
of harvestable biomass production Y (kg/ha) and N-
yield in harvested biomass, U (kg N/ha). The two rela-
tions, Y(U) and U(A), were quantified simultaneously
by numerical optimisation of the seven model param-
eters discussed later.
Both Y(U) and U(A) are divided in a lower and upper
segment, denoted as Domains I and II, respectively. The
transition between the two domains is marked by
the ‘critical point’, where U(A) ceases to be linear. The
model implies that for doses higher than Acrit (Domain
II), the fraction of N taken up by the crop from an
incremental dose
A decreases linearly with A, to reach a
value of zero for very high doses when the crop becomes
‘saturated’ with N at A=Amax,U=Umax andY=Ymax.
The values of A, U and Y at the critical point are denoted
as Acrit,Ucrit andYcrit, respectively. The constant slope dU/
dA for A < Acrit is referred to as the initial N-recovery,
ini. This parameter is identical with the well-known
‘apparent nitrogen recovery’ (ANR) for doses lower than
Acrit (Domain I). (‘Apparent’ refers to the common
notion that the uptake by the crop of N derived from
soil—and atmospheric deposition—is unaffected by fer-
tiliser application, and that the fraction of applied N that
is recovered by the crop can, therefore, be assessed by
subtraction of the N-yield observed in zero-N plots, U0.)
For higher doses this similarity with ANR vanishes
because ANR is the overall uptake efficiency (UU0)/A.
The N-yield in absence of applied nitrogen is expressed
by the model parameter S (kg N/ha), which is equivalent
with U0. So, the N-yield is given for the respective
domains by
U ¼ S þ ini A
ðDomain IÞ
ini
U ¼Ucrit þ ini ðA  Acrit Þ 
2ðAmax  Acrit Þ

ðA  Acrit Þ2
ðDomain IIÞ
Parameters defining Y(U) are three characteristic con-
centrations U/Y (kg N/kg dry matter) of N in har-
vested biomass. This ratio increases with increased N-
supply. The minimum concentration min is maintained
under low N-supply, i.e. small values of S and A. In the
critical point the concentration reaches the value crit,
and  attains its maximum value max under high N-
supply. Finally, the scale parameter Ymax defines the
biomass yield plateau attained under these conditions of
high N-supply. The parameter  is the biomass yield in
the critical point, Ycrit, relative to the maximum yield
Ymax. With these parameters, the relation between bio-
mass yield and N-yield is written as:
229
 
1 min  crit
Y¼ U þ U2
min amin a2crit Ycrit ð3Þ
ðDomain IÞ
Q
Y ¼Ycrit þ QðU  Ucrit Þ 
pþ1
 pþ1
U  Ucrit ð4Þ

 ðUmax  Ucrit Þ
Umax  Ucrit
ðDomain IIÞ
where Q is the slope dY/dU in Ucrit and p is an expres-
sion in min, crit, max, and  to assure continuity of
dY/dU in the critical point.
2.6. Quantification of the critical dose Acrit per data set
The above model was fitted for each data set to
observed Y and U at given doses A of total effective N.
Data from all treatments, that is, doses of fertiliser-
N, slurry-N and their combinations, were pooled and
one single set of parameter values (S, ini, min, crit,
max,  and Ymax) was determined on the basis of all
observations made per location per year. The optimisa-
tion of these parameters based on likelihood maximisa-
tion was done by simulated annealing (statistical
cooling), a random search method proposed by Kirk-
patrick et al. (1983). An advantage of this method is its
capacity to identify the global optimum in problems
with local optima. Although the model could be fitted
very well to all available data sets, some data sets were
not suitable for the identification of all seven para-
ð1Þ
meters. This is because the agronomic meaning of the
obtained parameter values vanishes when fitted on
observations that do not represent the full range of both
Domains I and II. We therefore introduced the criteria
that the ratio between the lowest and highest observed
ð2Þ
yields within a data set (location-year) should be less
than 0.75; and that the highest observed N-yield should
not differ by more than 50 kg N/ha from the maximum
uptake obtained from the fitted model for this set,
maxYmax. Whereas all data sets did meet the first
requirement, five sets failed the second requirement and
the corresponding parameters were considered invalid
for further analysis.
The valid parameter values thus obtained were then
used to assess the critical dose for a given location and
year as:
Acrit ¼ ðUcrit  SÞ=ini ¼ ð crit Ymax  SÞ=ini ð5Þ
This critical level of effective N-input was used to sepa-
rate all data points per set into a batch corresponding to
doses below Acrit, and a batch corresponding to higher
230 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238
doses. The latter were ignored in the regression analysis
outlined below.
2.7. Regression analysis
All data pertaining to sets with a valid estimate of
Acrit and to doses below this threshold were pooled, and
one ‘overall’ linear regression model was fitted to these
data. This was to assess the effects of effective N-dose A
(kg effective N/ha), N-source type j (j=1. . .3, for ferti-
liser, cattle slurry, and the two combined, respectively)
and local conditions i (i=1. . .24) on residual mineral
soil nitrogen in late autumn, Nmin (kg N/ha). The
regression was executed so as to accept a different
intercept bi for each location–year combination i and to
presume a different slope aj for each N-source type:
Nmin ¼ bi þ aj A ð6Þ
Year-effects within a location were then assessed by
pairwise comparison of all bi associated with a given
location. The same procedure was applied to the data
on nitrate leaching, for the limited cases where data were
available.
3. Results and discussion
All 29 data sets together included 426 treatments and
thus a total of 426 data points on biomass, N-yield,
and Nmin were available. Data points for nitrate leaching
numbered only 96 in total, because this variable was
measured in selected treatments at only two locations.
Fig. 1 serves to illustrate the pattern of crop responses
Y(U) and U(A) as generally observed for cut grass. This
pattern was found at all sites, though the N-doses were
not always high enough for biomass and N-yields to
reach a true plateau. Five data sets were therefore unfit
to assess the threshold Acrit, based on the criterion sta-
ted earlier. Typical responses of N-yield, Nmin and N-
leaching are given in Fig. 2 which corresponds to the
same set, Ruurlo-1981.
All observations on Nmin are pooled in Fig. 3 and
shown as a function of the effective N-dose. The values
are expressed as the difference with respect to Nmin
measured under the unfertilised control plots. Negative
values thus indicate cases where Nmin was lower in
treatments which received N-input than in the control.
The QUADMOD parameters derived in the process
to quantify Acrit are given in Table 3 and the resulting
Acrit values per set are listed in Table 4. After intro-
ducing the respective local thresholds Acrit as an upper
limit to the effective N-dose, 158 data points remained
in the 24 sets where Acrit could be calculated. These
results are shown in Fig. 4, again versus the effective
N-dose.
All Figs. 1–4 suggest that none of the response vari-
ables was affected by N-source type, and Fig. 4 indicates
that Acrit could indeed be used as a safe threshold
defining the effective-N dose below which no response
of Nmin was observed. To subtantiate this, we investi-
gated by linear regression the possibility of N-dose
effects on Nmin and N-leaching below Acrit within sets,
i.e. within year–location combinations. Such effects
could be masked by our pooling of data across sets in
the figures shown. Results are given below.
3.1. Effects of N-dose, N-source type and local
conditions on Nmin and N-leaching
The selection of all data corresponding to N-doses
below the local threshold Acrit resulted in 33 control
observations (no N applied), 36 observations corre-
sponding to treatments with pure mineral fertiliser, 46
to treatments with pure cattle slurry, and 43 to combi-
nations of fertiliser with slurry, totalling 158 data points
from 24 sets. Each observation represents a treatment
mean, as explained earlier.
The results of the regression analysis are given in
Tables 5 (slope aj) and 6 (intercept bi). It is concluded
that a very low but significant effect of dose A on Nmin
was present for mineral fertiliser, and also for combi-
nations of mineral-N and slurry-N inputs. No sig-
nificant effect of the dose on Nmin was observed for pure
cattle slurry, although the slope a was practically iden-
tical with that found for fertiliser with slurry combined
(Table 5). Pairwise comparison of aj for j=1. . .3
revealed that differences between slopes were not sig-
nificant. The slope of 0.03 kg Nmin per kg of N admi-
nistered to the field implies that a dose of 300 kg
effective N applied per ha is associated with an amount
of 9 kg residual mineral N in the soil profile in late
autumn, on top of the level found in the control (zero N).
With respect to the intercept b, representing the value
of Nmin in absence of external N-inputs, large differ-
ences were generally found between locations (Table 6).
Within locations, b varied significantly between at least
some of the years at the locations Ruurlo, Droevendaal,
Merelbeke, Middelkerke and Tienen. At Heino the
year-effects on b were not significant, nor were differ-
ences between irrigated and non-irrigated treatments.
This is contrasting with the differences in soil N-supply
to the unfertilised crop, expressed by parameter S,
found associated with irrigation: although supplemental
irrigation at Heino markedly increased S, this had no
effect on Nmin values (see later, Fig. 5).
Of the data on N-leaching, only observations on
treatments with pure mineral fertiliser (four) and pure
slurry (14) remained, in addition to the 14 control
values, after application of the earlier stated criteria.
This total of 28 observations was derived from nine data
sets (see also Table 2). The values found for slope a are
 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238 231

Fig. 1. ‘Quadrant representation’ of typical crop response to applied N-dose, illustrated here for the case of Ruurlo-1981. The upper graph shows
the relation between N-yield and biomass yield; the lower graph the relation between N-dose and N-yield. The intercept on the horizontal axis
represents the uptake of N from soil, as observed under zero N-input. F, mineral fertilizer; S, spring-applied cattle slurry; FS, mineral fertiliser and
cattle slurry combined. Doses are normalised to effective-N, see text for further explanation. Note that biomass and N-yield responses do not differ
between F, S and FS treatments.

very close to those for Nmin (Table 5), suggesting that
Nmin may be a valid indicator for N-leaching under
grass. The a-values do not differ significantly from zero
nor do they differ between mineral fertiliser and slurry,
but the intercepts b were different between years and
locations.
3.2. Long-term effects of slurry application explored
The relation between the crop N-uptake observed
under zero-N treatments (parameter S) and Nmin
registered in those same plots (referred to as Nmin base
value) is given in Fig. 5, which combines the data from
232 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238

Fig. 2. Crop N-yield in harvested grass biomass (left-hand axis); and residual mineral soil nitrogen in autumn (Nmin) (right-hand axis), and N-
leaching observed at Ruurlo-1981 for different effective N-doses. These graphs demonstrate the typical pattern found at all locations: accumulation
of Nmin at doses where N-yield approaches the maximum absorption capacity of the grass crop. The critical dose (see text) calculated for this case by
QUADMOD optimisation was 188 kg effective-N per ha per year.

all locations and years. No clear pattern between the two
variables emerges from this comparison. The absence of
a correlation between Nmin base values and S, over the
wide range in S covered here, supports the notion that
Nmin accumulates only under conditions where N-sup-
ply is high, relative to the crop’s absorption capacity.
The Nmin base values at Middelkerke deviated markedly
from those at other sites; these are regarded as excep-
tionally high and cannot be explained.
Although the above analysis involving the data from
eight locations and several years provides no indication
that N derived from animal slurries would contribute
differently to Nmin nor to N-leaching than fertiliser-N,
it may be argued that the long term effect of animal
 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238 233

Fig. 3. Residual mineral soil nitrogen in autumn (Nmin) for effective N-doses applied as mineral fertiliser, cattle slurry and the two combined. Nmin is
expressed here as the difference with respect to the value observed in unfertilised plots in the corresponding location–year, in text referred to as the
Nmin base level. All observations (treatment means) of all 29 data sets are included.

Table 3
Mean values and standard deviation (S.D.) of QUADMOD parameters across n available data sets suitable for identification of the respective
parameters

S (kg N/ha) ini min crit max g Ymax (kg/ha) Acrit (kg N/ha)

Mean 111 0.79 0.020 0.029 0.037 0.87 13201 270

S.D. 41.9 0.165 0.0034 0.0034 0.0058 .149 2357 127
N 29 29 29 24 24 24 24 24

manures on these two variables remains uncertain. For
afterall, the above observations are based on within-
year effects only. As the working coefficient w of cattle
slurry is always well below unity, the fate of the fraction
(1w) remained outside the scope of our analysis. Part
of this fraction is lost by ammonia volatilisation during
and shortly after application of slurry to the field. The
remaining part has been coined Nr, for organic nitrogen
‘resistant’ to decomposition, which is—by definition—N
that is not available to the crop within the year of slurry
application. Continued application of large amounts of
Nr will, if maintained during decades, increase soil fer-
tility S and thereby reduce the threshold dose Acrit
which can be interpreted as the safely permissible N-
dose. Below follows an attempt to estimate these long-
term effects.
We presume that the accumulation of Nr-inputs in
soil affects no other parameters than S, which is a
measure of soil fertility with respect to nitrogen. For a
first approximation of how S changes as a result of
annual Nr-inputs, we use the simplest possible model of
organic matter decay. Since the early nineteenth century
(Thaer, 1809; von Wulffen, 1823, 1830, 1847, summar-
ized by de Wit, 1974) it is known that the buildup—and
breakdown—of soil fertility is the net result of two
opposing fluxes: the annual inputs brought into the soil
system, and the liberation and subsequent uptake of
nutrients from the stock contained in the soil. Von
Wulffen expressed this release as a fixed fraction,
annually, of the total stock, and introduced the concept
that an equilibrium state should finally be approached if
the annual input remained constant during many years.
234 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238

Table 4 (no N retained) to 1 (all N retained). (Note that Acrit

Critical N-dose Acrit (kg N/ha as fertiliser equivalents) determined by here always refers to the effective N-dose!)
numerical parameter optimisation via simultaneous fitting of the two The parameter S is not identical with the annual
QUADMOD functions Y(U) and U(A) to observed data
mineralisation but is obviously closely related. On the
Location Year one hand, S includes—in addition to N from mineral-
1980 1981 1982 1983 1984
isation of soil organic matter—nitrogen derived from
atmospheric deposition; on the other, not all nitrogen
Ruurlo 175 188 83 128 190 that becomes available from these two sources is actu-
Heinoa 230 205 249 ally captured by the crop and thus included in the value
Heinob 112 170 250
of S. Let this fraction captured, q, be 80% in temperate
1987 1988 1989 1990 grasslands, and let the annual deposition be noted as d
(approximately 50 kg N/ha/year in the Netherlands). It
Droevendaal 450 450 410 450
follows that the parameter S, once equilibrated to long-
1996 1997 1998 term constant input, is then given by

Merelbeke (353)c 373 399 S1 ¼ q ðI þ dÞ ð8Þ

Tienen 58 377 345
Middelkerke 275 405 88
Geel 278 (365) – After introducing Eqs. (7) and (8) in Eq. (5), the per-
Bree (384) (164) 317 missible dose in the equilibrium state follows as
a
Without supplemental irrigation. Ucrit  S1 crit Ymax  qd
b
With supplemental irrigation. Acrit;1 ¼ ¼ ð9Þ
c
Value in brackets: set does not meet crit 2.
ini ini þ q

where  represents the term in brackets in Eq. (7). With

This state was referred to as the ‘Beharrungspunkt’ and
it was also recognized that the corresponding fertility
level would be high in systems where the annual fraction
released (year1) was small, and low where this coeffi-
cient was large. Von Wulffen’s analysis was based on
elaboration of numerous empirical long-term records of
crop yields, and his first order approach still stands, be
it of old age and approximative, as a robust model of
organic matter-related soil fertility development. The
implication for our analysis is that, under any given
level of annual Nr input, both the total pool of organic
soil nitrogen and the annual N-release from this pool by
mineralisation should in the long run approach constant
values, such that the amount of N released per year is
equal to the net annual N input into the soil pool, that
is, after subtraction of crop offtake and N-losses not
captured into organic matter. The net N-input by ferti-
liser and slurry equals, if the N-supply is maintained at
Acrit:
 
f
I ¼ Acrit  þ rð1  ini Þ ð7Þ
w values of Acrit and S. The time coefficient does affect,
where f is the fraction of Acrit supplied in the form of
cattle slurry,  is the fraction of slurry-N which is in the
not-directly available form Nr, here estimated to be 0.3; w
was defined in Section 2.4. The first term within brackets
thus represents the Nr-input, the second represents all
effective nitrogen from fertiliser and slurry that is not
recovered by the crop but yet retained in the organic N-
pool. Since little is known about the fate of non-recovered
N, we introduce the factor r which may range from 0
the help of Eqs. (7)–(9) we can now quantify pairs of
Acrit and S values that are consistent in a steady state
situation. We use the mean values of ini, crit,  and
Ymax obtained earlier (Table 3), and a value of 0.5 for w.
Now it follows that the permissible dose ranges between
200 and 225 kg effective N per ha per year if all N were
given as slurry, with the lower value for r=1 and the
higher for r=0. This is equivalent with 400–450 kg total
N in slurry. If 50% of the effective N dose is given as
fertiliser-N and 50% as effective slurry-N, the permis-
sible dose would be 240–280 kg effective N per ha. This
is equivalent with the same amounts in total slurry N
(w=0.5 offsets the fact that half the dose is given as
fertiliser). If all N is applied in mineral fertiliser, the
doses are between 300 and 360 kg N/ha. The corre-
sponding S-values are 150 (r=0) to 170 (r=1) kg N per
ha for the case with 100% slurry-N; 105–135 kg N per
ha for 50% fertiliser-N and 50% slurry-N; and 40–90 kg
N per ha if all N is given as mineral fertiliser.
Note that no assumptions on the value of the above
time coefficient are needed to estimate the equilibrium
however, the rate at which the steady state is approa-
ched, and the final size of the organic soil nitrogen pool,
as noted already by von Wulffen. The latter depends, in
addition, on the net annual input.
4. Conclusions
Based on the analysis of 29 data sets from eight loca-
tions in the Netherlands and Belgium, each representing
 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238 235

Fig. 4. Residual mineral soil nitrogen in autumn (Nmin) for effective N-doses applied as mineral fertiliser, cattle slurry and the two combined. Nmin is
expressed here as the difference with respect to the value observed in unfertilised plots in the corresponding location–year. Only data below the
critical dose, assessed for 24 sets separately, are included.

Table 5
Effect of (effective) N-dose on residual mineral nitrogen in the soil profile in autumn, Nmin, and on nitrogen leaching

Nmin N-leaching

a (kg Nmin per kg effective N) t-Probability a (kg Nmin per kg effective N) t-Probability

Mineral fertilizer 0.0387 <0.001a 0.0494 0.29 n.s.b

Mineral fertiliser and cattle slurry combined 0.0287 0.004a –
Cattle slurry 0.0281 0.285 n.s. 0.0247 0.707 n.s.
a
Significantly different from 0.
b
n.s., Non-significantly different from 0.

a location–year combination, it is concluded that 3–4 kg
of residual mineral nitrogen (Nmin) remained by late
autumn in the soil profile per each 100 kg of effective
nitrogen applied, under permanent cut grassland, as
long as doses remained under a threshold referred to
as Acrit. This threshold was assessed by parameter opti-
misation and marks the maximum N-dose where the
apparent nitrogen recovery by the crop still maintains
its initial value, i.e. the value at low N-doses. The mean
of this constant recovery (over all data sets) was 0.8 and
the biomass yield level corresponding to Acrit was on
average 87% of the maximum biomass yield attainable
under unlimited N-supply. The mean of the threshold
Acrit was 270 kg effective N applied per ha per year. Acrit
can be regarded as a safe permissible dose, with respect
to residual mineral N accumulation in soil.
The response of Nmin to effective N-dose was not dif-
ferent between the N-sources: mineral fertiliser, cattle
slurry, and the two combined. This implies that the
‘effectiveness’ of N contained in slurry as expressed in
the ‘working coefficient’ w—which is based upon crop N-
uptake—applies equally well to normalise N-doses from
different sources when the purpose is to quantify Nmin in
late autumn. In other words: nitrogen that is ‘effective
for uptake’ is also effective in terms of residual Nmin.
From the limited cases where N-leaching was meas-
ured it appears that the same conclusions as mentioned
for Nmin hold for N-leaching: there are no differences
 236
H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238
Fig. 5. Residual mineral soil nitrogen in autumn (Nmin) observed in absence of N-input, plotted versus crop N-uptake from unfertilised soil. All data sets included. For Heino data, N refers to
treatments with natural water supply, R to supplemental irrigation.
 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238
Table 6
N_min base level, defined as the amount of residual mineral nitrogen
in the soil profile in autumn, under zero N-inputa
Location b (kg N per ha)
Ruurlo 23–49
Heino-N 25–36
Heino-R 22–30
Droevendaal 11–38
Merelbeke 27–48
Tienen 24–40
Middelkerke 140–176
Geel 77
Bree 14
a
The presented values are the intercept values b, as found by linear
regression on all data below the local critical dose from 24 sets. Ran-
ges refer to values across years within locations. Single values refer to
1 year at locations where no critical dose could be assessed in remain-
ing years. For further details see text.
between N-sources if doses are expressed as effective-N;
the response is 2–5 kg N-leached per 100 kg effective N
applied for doses below Acrit; this was, however, not
significantly different from zero due to the relatively
high variance in N-leaching data. The numerical simi-
larity of the response (2–5 kg N per ha leached versus 3–4
kg residual Nmin per ha, per 100 kg N applied) suggests
that Nmin in late autumn is a good indicator for N-
leaching from cut grassland.
The responses of several kg Nmin and N-leached per
100 kg N applied are small compared to the observed
Nmin base values (zero N-input) of usually 20–40 kg per
ha.
No relation was found between the Nmin base value
and S, the crop N-yield under zero N-input. Values of the
latter ranged between 30 and 200 kg N per ha per year.
Long-term application of organic, not readily avail-
able nitrogen (Nr) through cattle slurry will lead to net
accumulation or net breakdown of organic-N in soil,
depending on the annually applied dose in relation to
the initial soil status and the time coefficient of organic-
N turnover. Whether accumulation or breakdown, a
steady state will be approached where the annual release
of N by mineralisation equals the annual net N-input,
where the size of the organic N pool is constant
and where S attains a constant value in equilibrium with
these variables. One particular permissible dose corre-
sponds to this state because Acrit and S are com-
plementary sources to meet the crop N demand. In such
steady state, Acrit would amount to 200 kg effective N
per ha on grassland for Dutch and Belgian growth condi-
tions, if all N were supplied in slurry. This corresponds to
roughly 400 kg total N per ha in slurry, and the associated
S value would reach 150–170 kg N per ha. Supplying all of
Acrit as fertiliser would lower the steady state S-value and
thereby increase Acrit to exceed 300 kg effective N per ha,
with S-values of 40–90 kg N per ha.
237
All these conclusions refer to cut grass only, and have
to be adjusted for grazing. Yet, this broad set of
experimental data demonstrates that the European
Commission’s viewpoint that N application in animal
manures should in general be limited to 170 kg N per ha
lacks a scientific basis. For grassland, values up to 400
kg total N in slurry and more are justified, depending on
concurrent mineral fertiliser use. Establishing a limit on
N supply in animal manures while ignoring mineral N-
inputs is not justified, as one substitutes the other and
not only so with respect to crop nutrition, but also to N-
losses to the environment.
Acknowledgements
We greatly appreciate the support provided by bio-
metrists Michiel Jansen, Saskia Burgers, and Jacques
Withagen at the Department of Biometry, Plant
Research International.
References
Fonck, H., 1982a. Stikstofconcentraties in bodemvocht en grondwater
onder grasland op zandgrond in afhankelijkheid van runder-
drijfmest- en kunstmeststikstofdosering. 1e onderzoeksjaar, Nota
1337. ICW, Wageningen.
Fonck, H., 1982b. Stikstofconcentraties in bodemvocht en grondwater
onder grasland op zandgrond in afhankelijkheid van runder-
drijfmest- en kunstmeststikstofdosering. 2e onderzoeksjaar, Nota
1407. ICW, Wageningen.
Fonck, H., 1986a. Stikstofconcentraties in bodemvocht en grondwater
onder grasland op zandgrond in afhankelijkheid van runder-
drijfmest- en kunstmeststikstofdosering. 3e onderzoeksjaar, Nota
1707. ICW, Wageningen.
Fonck, H., 1986b. Stikstofconcentraties in bodemvocht en grondwater
onder grasland op zandgrond in afhankelijkheid van runder-
drijfmest- en kunstmeststikstofdosering. 4e onderzoeksjaar, Nota
1685. ICW, Wageningen.
Fonck, H., 1986c. Stikstofconcentraties in bodemvocht en grondwater
onder grasland op zandgrond in afhankelijkheid van runder-
drijfmest- en kunstmeststikstofdosering. 5e onderzoeksjaar, Nota
1690. ICW, Wageningen.
Kirkpatrick, S., Gelatt Jr., C.D., Vecchi, M.P., 1983. Optimization by
simulated annealing. Science 220 May, 671–680.
Kolenbrander, G.J., de la Lande Cremer, L.C.N., 1967. Stalmest en
gier. Waarde en mogelijkheden. Veenman, Wageningen.
van der Meer, H.G., Thompson, R.B., Snijders, P.J.M., Geurink, J.H.,
1987. Utilization of nitrogen from injected and surface-spread cattle
slurry applied to grassland. In: van der Meer, H.G., Unwin, R.J.,
van Dijk, T.A., Ennik, G.C. (Eds.), Animal Manure on Grassland
and Fodder Crops. Fertilizer or Waste? Developments in Plant and
Soil Sciences. Martinus Nijhoff Publishers, Dordrecht, The Nether-
lands, pp. 47–71.
Thaer, A., 1809. Grundsätze der rationellen Landwirtschaft.
Van Bockstaele, E., Carlier, L., Verbruggen, I., Michiels, J., 1996.
Annual Report EG 2078/92: 1996 Grasland, to the European
Commission.
Van Bockstaele, E., Carlier, L., Verbruggen, I., Michiels, J., 1997.
Annual Report EG 2078/92: 1997 Grasland, to the European
Commission.
238 H.F.M. ten Berge et al. / Environmental Pollution 118 (2002) 225–238

Van Bockstaele, E., Carlier, L., Verbruggen, I., Michiels, J., 1998.
Annual Report EG 2078/92: 1998 Grasland, to the European
Commission.
Wadman, W.P., Sluijsmans, C.M.J. (Eds.), 1992. Mestinjectie op
grasland. De betekenis voor de bodemvruchtbaarheid en risico’s
voor nitraatuitspoeling; Ruurlo 1980-1984. DLO-Staring Centrum
en DLO-Instituut voor Bodemvruchtbaarheid (Rapport DLO-
Instituut voor Bodemvruchtbaarheid).
de Wit, C.T., 1974. Early theoretical concepts in soil fertility. Neth. J.
Agric. Sci. 22, 319–324.
von Wulffen, C., 1823. Ideeen zur Grundlage einer Statik des Land-
baues. Mogliner Annalen, Band XI.
von Wulffen, C., 1830. Die Vorschule der Statik des Landbaues.
Magdenburg.
von Wulffen, C. 1847. Entwurf einer Methodik zur Berechnung der
Feldsysteme. Berlin.