Content uploaded by Alexander F Bouwman
Author content
All content in this area was uploaded by Alexander F Bouwman on Feb 12, 2015
Content may be subject to copyright.
N
2
O and NO emission from agricultural fields and soils under natural
vegetation: summarizing available measurement data and modeling
of global annual emissions
Elke Stehfest
1,
* and Lex Bouwman
2
1
Center for Environmental Systems Research, University of Kassel, Kurt-Wolters-Strasse 3, 34109, Kassel,
Germany;
2
Netherlands Environmental Assessment Agency, 303, 3720 AH, Bilthoven, The Netherlands;
*Author for correspondence (e-mail: elke.stehfest@mnp.nl)
Received 20 May 2005; accepted in revised form 23 January 2006
Key words: Agriculture, Animal manure, Crop, Emission, Fertilizer, Model, Nitrogen, Nitric oxide,
Nitrous oxide, Vegetation
Abstract
The number of published N
2
O and NO emissions measurements is increasing steadily, providing additional
information about driving factors of these emissions and allowing an improvement of statistical N-emission
models. We summarized information from 1008 N
2
O and 189 NO emission measurements for agricultural
fields, and 207 N
2
O and 210 NO measurements for soils under natural vegetation. The factors that
significantly influence agricultural N
2
O emissions were N application rate, crop type, fertilizer type, soil
organic C content, soil pH and texture, and those for NO emissions include N application rate, soil N
content and climate. Compared to an earlier analysis the 20% increase in the number of N
2
O measurements
for agriculture did not yield more insight or reduced uncertainty, because the representation of environ-
mental and management conditions in agro-ecosystems did not improve, while for NO emissions the
additional measurements in agricultural systems did yield a considerable improvement. N
2
O emissions
from soils under natural vegetation are significantly influenced by vegetation type, soil organic C content,
soil pH, bulk density and drainage, while vegetation type and soil C content are major factors for NO
emissions. Statistical models of these factors were used to calculate global annual emissions from fertilized
cropland (3.3 Tg N
2
O-N and 1.4 Tg NO-N) and grassland (0.8 Tg N
2
O-N and 0.4 Tg NO-N). Global
emissions were not calculated for soils under natural vegetation due to lack of data for many vegetation
types.
Introduction
Human activities like fertilizer application, and
fossil fuel combustion have caused a major in-
crease in both nitrous oxide (N
2
O) and nitric
oxide (NO) emissions. The atmospheric increase
of N
2
O by 0.7 ppbv per year causes 6% of the
anthropogenic greenhouse effect and also con-
tributes to stratospheric ozone depletion (IPCC
2001). NO is involved in the regional balance of
oxidants of the atmosphere and its re-deposition
causes eutrophication and acidification of eco-
systems. Natural sources of N
2
O are soils and
oceans, and the anthropogenic increase is mainly
caused by accelerated soil emissions through the
application of N fertilizers and animal manure in
Nutrient Cycling in Agroecosystems (2006) 74: 207 –228 Springer 2006
DOI 10.1007/s10705-006-9000-7
agriculture. NO emissions mainly stem from fossil
fuel combustion, while soil emissions (both natu-
ral and accelerated by fertilizer addition) are
dominant in remote areas. Despite more than
three decades of research yielding numerous
publications on N
2
O and NO flux measurements,
the contribution of individual sources is still
uncertain (IPCC 2001).
In soils N
2
O and NO are intermediate products
of nitrification and denitrification, while denitrifi-
cation is also a sink for N
2
O (Tiedje 1988). The
controlling factors of N
2
O and NO emissions have
been reviewed elsewhere (for example, Firestone
and Davidson 1989; Robertson 1989; Bouwman
et al. 1993; Mosier et al. 1996; Bremner 1997;
Freney 1997; Bouwman et al. 2002a).
One way to learn about the processes and fac-
tors responsible for N
2
O and NO emissions is by
developing process-based ecosystem models such
as Daycent (Parton et al. 1996) and DNDC (Li
and Aber 2000). Parallel to these process-based
models, it is useful to summarize the emission
measurement data with statistical techniques to
identify correlations between controlling factors
and emissions. These approaches can be used to
develop emission factors such as those used by
IPCC (Bouwman 1996; Mosier et al. 1998) or
simple statistical models that describe the variation
of N
2
O and NO fluxes at larger scales which can
be used to assess management or mitigation op-
tions (Bouwman et al. 2002b; Freibauer and Kal-
tschmitt 2003). In addition, statistical approaches
are useful to validate process-based models and
to point to problems of biases and under-
representation in the data for specific climate or
land use conditions.
For N
2
O and NO emissions from agricultural
soils measurement data have recently been sum-
marized (Bouwman et al. 2002a) and an empirical
model has been developed (Bouwman et al.
2002b). For N
2
O and NO emissions from soils
under natural vegetation no regional or global
statistical emission model is known to the authors.
The aim of this study is to identify the factors with
significant influence on N
2
O and NO emissions
from agricultural fields and soils under natural
vegetation, and to develop statistical models for
estimating annual emissions. For N
2
O and NO
emissions from soils under natural ecosystems this
study is the first comprehensive statistical analysis
of published measurement data, and for
agricultural systems the data set used by Bouwman
et al. (2002a) was extended for this study.
Data and methods
Data set
We extended the N
2
O and NO emission data set
presented by Bouwman et al. (2002a) with data for
N
2
O and NO emissions from soils under natural
vegetation and added more measurements for
agricultural fields. The data set contains results
from field studies that were published in the peer-
reviewed literature, including various parameters
related to climate, soil, management and mea-
surement technique. The full data set can be ob-
tained from http://www.mnp.nl/en/publications/
2006.
The emissions are given as the sum of emissions
over the reported measurement period, and for
measurements covering more than one year the
values were converted to one year. Unless indi-
cated otherwise, hereafter the term ‘emission’ al-
ways refers to the emission during the time period
covered by the experiment.
The data set is unbalanced as the combinations
of classes are not represented by equal numbers in
all experiments, it is biased as some categories are
not represented at all, and it has many missing
values as most studies do not report all parameters
included in the data set. Similar to Bouwman et al.
(2002a), classes were designed with – as far as
possible – both similar ranges and balanced
numbers of measurements. Only in two cases we
had to use different classifications for N
2
O and
NO emissions from soils under natural vegetation
than for agricultural measurements.
For agricultural fields the factor climate was
classified according to de Pauw et al. (1996) and
grouped into temperate continental, temperate
oceanic, subtropical and tropical. The data for
soils under natural vegetation have an unbalanced
representation of these four climate classes, and
the factor climate was therefore further aggregated
to temperate and tropical. Arid, polar and boreal
climates were not part of the final data set due to
lack or complete absence of measurement data.
Annual precipitation and mean annual tempera-
ture from the literature reference itself was in-
cluded as a factor as well as climate data for the
208
geographical location of the measurement sites
from the 0.5 0.5 degree data provided by New
et al. (1999).
The data set contains 1125 measurements for
N
2
O and 199 for NO from agricultural fields,
which is a considerable improvement compared to
the 846 (N
2
O) and 99 (NO) measurements used by
Bouwman et al. (2002a). Some variables and
classes were excluded from the analysis before
summarizing and analyzing the data: (i) Organic
soils, as they are known to have very high N-
emission and because they strongly influenced the
predicted emissions for mineral soils. (ii) Experi-
ments with chemicals or additives like nitrification
inhibitors were excluded, because their use is still
very limited at the global scale (Trenkel 1997). (iii)
Grazing systems, because the N from animal
excretion is often not provided. The reduced data
set has 1008 measurements for agricultural N
2
O
from 204 references and 189 measurements for
agricultural NO from 58 references.
For soils under natural vegetation the data set
includes 247 N
2
O emission measurements and 231
NO emission measurements. Some variables and
classes were excluded before summarizing and
analyzing the data: (i) Organic soils (for the same
reason as above for agricultural fields). (ii) The
classes deciduous-legume (Alder) forest, marsh,
mixed forest and ‘other types’ which include only
two or three measurements. (iii) Two measure-
ments with N
2
O uptake >0.4 kg ha
)1
of N, which
caused the predicted emissions to be mostly neg-
ative. The reduced data set for soils under natural
vegetation has 207 measurements for N
2
O from 72
references and 210 measurements for NO from 52
references.
N deposition is not considered because of lim-
ited information provided in the references.
However, most N
2
O and NO emission measure-
ments from soils under deciduous and coniferous
forests stem from regions with N inputs from
atmospheric deposition >10 kg ha
)1
yr
)1
, which
is a threshold above which changes to sensitive
ecosystems may occur (Bobbink et al. 1998).
Therefore the data represent only N-affected
coniferous and deciduous forests. The measure-
ments from all other vegetation classes are not or
only slightly affected by N deposition and are
therefore represent the natural state of those
systems.
Data analysis
A statistical analysis of the data set based on the
above classification was done to identify factors
with a significant influence on N
2
O and NO
emissions and to develop empirical models for
N
2
O and NO emissions. We used the REML
directive of Genstat (Payne et al. 2000) which is
more appropriate for analyzing unbalanced data
sets with missing values than regression analysis.
Emissions are balanced by assuming all factor
classes to have an equal number of observations.
Balanced emission rates should therefore be used
as relative values for comparing the different
controlling factors or factor classes.
The emissions were first log-transformed as this
resulted in a distribution that is closer to a normal
one than the untransformed data. Log transfor-
mation requires manipulation of negative and zero
fluxes. We used values for the minimum detectable
fluxes of 1.67 ng m
)2
s
)1
for N
2
O-N measure-
ments (Verchot et al. 1999) and 0.44 ng m
)2
s
)1
for NO-N (Meixner et al. 1997) for closed cham-
ber measurements (N
2
O) and open chambers with
forced flow-through (NO) (the most common
types for these gases in our data set). For all
measured fluxes smaller than this detectable
weekly flux of 0.01 kg N
2
O-N ha
)1
and 0.003 kg
NO-N ha
)1
, we set the emissions to these values.
Initially, all factors in the data set were included
in the REML analysis. They were treated as fixed
terms, i.e. the REML directive assigns a value to
each class of each factor, so that the resulting
model with the best possible fit is:
logðNemissionÞ¼AþX
n
i¼1
Eið1Þ
where N
emission
is the emission of N
2
O or NO ex-
pressed in kg ha
)1
of N over the time period
covered by the measurements, Ais a constant and
E
i
is the effect value for factor i.
The factor ‘reference’ was handled as a random
effect. Random effects are used in REML when
subgroups may have specific effects on the results,
but where group membership cannot be surveyed
(i.e. new measurements cannot be assigned to
existing groups). Including random effects may
increase the uncertainty of a prediction but de-
crease the deviance of the model.
209
We analyzed the significance of factors in two
ways. First, significant variables were identified by
creating a model that contained all factors. Sec-
ondly – in order to exclude interaction effects –
factors were added one by one to a core model,
only keeping the significant ones in the model be-
fore adding the next one. The reason for this
stepwise procedure is that in the model with all
factors some may not be significant if there are too
many other non-significant factors included.
REML tests the significance by (i) adding the
factors one after the other to the model, whereby
the results depend on the order of the factors, and
by (ii) dropping one variable at a time from the full
model. The Wald statistics tool is used to calculate
the change in deviance for a full model and a re-
duced model excluding one factor. The significance
is tested by comparing the change in deviance with
the chi-Square probability (see e.g. Snedecor and
Cochran 1980), indicating the chance that the full
model is significantly different from the reduced
one (P£0.05). Thus a model only containing the
significant factors is obtained.
A data summary for these significant variables
was compiled by calculating means (MEA) and
medians (MED) in order to investigate the skew-
ness of the data set. In addition, balanced medians
(BMED) and balanced means (BMEA) were cal-
culated for all classes from the statistical (bal-
anced) REML model with the significant factors.
As log transformation conserves the median, the
model described above can be used to calculate the
balanced median (BMED) by back-transforma-
tion of the REML results. For balanced means
(BMEA) a model was fitted with the same fixed
terms, but without prior log transformation of
emissions. A comparison between these balanced
values and the mean and median values can be
used to analyze the unbalanced features of the
data set. The values in the summary tables are
mean and median emissions calculated by aver-
aging reported emission values each having a
specific length of experiment. Therefore mean and
median emissions represent an average measure-
ment period for the factor class considered.
For the significant factors we assessed the sig-
nificance of differences between factor classes.
Predicted means (not back-transformed) and
standard errors of differences were calculated for
all factor classes, assuming average values for all
other classes. The difference between two factor
classes is significant if it is larger than the standard
error of the difference times the excentricity (l).
For classes that are expected to have different
emissions than another factor class, a one-tailed
test with l=1.64 is used. If there is no expectation,
the test is two-tailed with l=1.96.
The 95% confidence interval was calculated as
the prediction plus or minus 1.96 times the stan-
dard error. Back-transformation of the prediction
and its upper and lower bound yield the emission
and the confidence interval, whereby upper and
lower range of the confidence interval differ after
back-transformation. Since the confidence interval
is different for each combination of factor classes,
we present the average for all factor class combi-
nations that are covered in the data set.
Estimating global annual emissions
We used global 0.5 0.5 degree resolution data for
soil properties (Batjes 2002), climate (de Pauw
et al. 1996), land use and vegetation for the year
1995 (IMAGE-team 2001) and fertilizer and
manure application (Bouwman et al. 2005). The
categories of the global land use and vegetation
maps were grouped into the classes used in the
statistical model. We used country data on har-
vested areas and fertilizer use for 1998 (1997 –1999
average) obtained from Bruinsma (2003) and FAO
(2004) to correct the land use maps, and allocated
fertilizer use by crop on the basis of IFA/IFDC/
FAO (2003). By using harvested areas the crop-
ping intensity may exceed 100% in countries with
multiple cropping, such as China and India. To
identify areas affected by high N deposition we
used estimates of atmospheric N deposition from
the STOCHEM model (Collins et al. 1997) as de-
scribed by Bouwman et al. (2002c).
More spatial detail was considered not realistic
since data on agricultural management are avail-
able at the scale of countries at best. For example,
no statistical information is available for fertilizer
application mode, while fertilizer application rates
from IFA/IFDC/FAO (2003) are based on expert
knowledge for about 90 countries and animal
manure application rates are based on information
for world regions.
Fertilizer-induced emissions (FIE) were calcu-
lated for each grid cell as the emission with N
application minus the emissions for the same area
210
under zero N application (all other factors being
equal) and expressed as a percentage of the N
applied as fertilizer or animal manure. FIE ex-
presses the anthropogenic N
2
O emission from
fertilizer, animal manure and other N inputs
according to IPCC (1997). The exponential nature
of the model causes FIE rates to be positively
correlated to background emissions.
Results and discussion
Agricultural fields
Controlling factors for N
2
O
Crop type, fertilizer type and N application rate
are significant management-related factors for
N
2
O emissions. MEA and MED values for N
application rate increase along with N inputs, ex-
cept for the classes <100 kg ha
)1
(Table 1). This is
caused by the unbalanced features of the data,
because values for BMEA and BMED increase
almost linearly along with N application rate.
Differences between most classes are significant
(Table 2). These results show that the N applica-
tion is a major control of N
2
O emissions from
agricultural soils and confirms earlier work based
on smaller data sets (Bouwman 1996; Bouwman
et al. 2002a).
For fertilizer type, only ANP (lowest BMED
value) and CAN (highest BMED) are significantly
different from most other fertilizer types (Table 2).
Except for CAN and ANP the pronounced dif-
ferences for MEA and MED between fertilizer
types almost disappear after balancing.
For crop type, some differences between classes
in MEA values can be explained by outliers and
the unbalanced features of the data set, as MED
and BMED values are similar (for example le-
gumes compared to none, and grassland compared
to cereals, Table 1). A consistent picture for MED
and BMED is found for wetland rice with lowest,
and cereals and grass with somewhat lower values
compared to legumes. Wetland rice, cereals and
grass significantly differ from all other crop types
and among each other (Table 2), only the differ-
ence between cereals and grass is not significant.
The factor crop type reflects various differences
between crops. Legumes are N fixing crops and
generally receive no or small amounts of N fertil-
izer but the input from N fixation gives rise to high
N
2
O emissions. Inundation in wetland rice pro-
mote anaerobic conditions, whereby N
2
O is more
likely to be re-consumed before being emitted from
the soil (Davidson 1991). High N
2
O emissions
after drainage of the paddy field (Bronson et al.
1997) are not included in most measurements in
our data set. Low N
2
O emissions from grassland
may be related to the long growing season of grass
and higher N uptake than in crops with shorter
growing periods.
From the factors related to soil conditions, soil
organic C content, soil pH, and soil texture were
found to have a significant influence on N
2
O
emissions (Table 1). For soil organic C content
MEA, MED and BMED show continuously
increasing emissions with increasing C content
(Table 1). The class with C content >3% is sig-
nificantly different from both other classes
(Table 2). The increase of emissions along with
soil organic C reflects the positive correlation be-
tween soil organic C content and rates of nitrifi-
cation and denitrification (Tiedje 1988).
For soil pH the MEA, MED, BMEA and
BMED all clearly show lowest emissions for pH
values >7.3 (Table 2). The two classes with lower
pH show similar values within unbalanced and
balanced means and medians, whereby the medi-
ans are lower than the means. The pH class >7.3 is
also significantly different from the two classes
with lower pH (Table 2). The N
2
O emissions from
acidic soils exceed those from alkaline soils and
reflect the reported higher N
2
O emission from
nitrification (Martikainen and Boer 1993) or
higher N
2
O:N
2
fraction for denitrification (Alex-
ander 1977) at low pH compared to high pH.
The data for soil texture seem to be unbalanced
as MEA and MED values are lowest for fine tex-
tured soils, BMEA values are similar in all classes.
However, the balancing of logarithmic emissions
leads to significantly higher BMED values for fine
textured soils than for coarse and medium textures
(Tables 1 and 2). Fine-textured soils have more
capillary pores within aggregates than do sandy
soils, thereby holding soil water more tightly.
Anaerobic conditions may be more easily reached
and maintained for longer periods within aggre-
gates in fine-textured soils than in coarse-textured
soils (Bouwman et al. 1993).
Climate type is a significant factor, and MEA,
BMEA and BMED values for N
2
O emissions from
agricultural fields are highest for subtropical
211
Table 1. Number of observations (N), minimum (Min), maximum (Max), mean (MEA), median (MED), balanced mean (BMEA) and
balanced median (BMED, back-transformed after log-transformation) emissions
a
for those factors with a significant influence on N
2
O
emissions from agricultural fields.
Factor/factor class NMin Max MEA MED BMEA BMED
N Application rate (kg ha
)1
)
0 –1 255 )0.60 9.00 1.09 0.56 )0.47 0.29
1 –50 30 0.01 3.10 1.03 0.94 1.31 0.61
50 –100 160 )0.75 12.93 1.62 0.80 1.61 0.71
100 –150 183 )0.01 16.31 1.58 0.87 2.13 0.89
150 –200 113 0.00 16.78 2.52 1.14 2.52 1.11
200 –250 79 0.01 15.60 2.64 1.42 2.83 1.41
>250 188 0.00 56.00 7.50 3.88 5.59 2.26
Soil organic C content (%)
<1 82 0.01 5.20 1.07 0.59 2.22 0.71
1 –3 447 )0.75 31.73 2.11 0.95 1.69 0.71
>3 180 )0.60 30.40 2.93 1.51 2.74 1.34
Soil pH
<5.5 95 0.00 24.20 2.78 0.91 2.63 1.08
5.5 –7.3 465 )0.75 41.80 2.49 1.10 2.74 1.02
>7.3 144 0.00 26.90 1.87 0.65 1.28 0.61
Texture
Coarse 509 )0.60 46.44 3.21 1.20 2.48 0.80
Medium 219 0.00 56.00 2.56 1.25 2.04 0.68
Fine 158 )0.75 19.00 1.77 0.94 2.14 1.24
Climate
Temp_C 464 )0.07 56.00 2.17 1.11 1.86 0.77
Temp_O 268 )0.60 31.73 2.80 1.15 1.20 0.71
S-Trop. 144 0.00 41.80 4.27 1.16 3.72 1.72
Trop. 132 )0.75 46.44 2.93 0.98 2.09 0.63
Crop type
Cereals 184 0.00 56.00 2.09 0.92 2.09 0.77
Grass 282 )0.60 46.44 3.49 1.11 2.57 0.63
Legume 36 0.00 4.20 1.53 1.31 2.70 1.58
Other 289 0.00 41.80 3.39 1.60 2.62 1.16
W-Rice 79 )0.75 4.72 0.79 0.53 0.58 0.31
None 107 0.01 19.60 2.29 1.16 2.74 1.64
Fertilizer type
AA 38 0.05 19.60 4.07 2.59 3.42 1.04
OAF 74 0.01 36.54 0.97 0.35 1.75 0.82
AN 131 0.00 30.40 3.20 1.41 2.73 1.12
CAN 73 0.05 11.20 2.58 1.80 2.37 1.56
KN 58 0.00 41.80 5.62 1.09 3.41 0.79
Mix 45 0.00 16.78 4.05 3.06 2.09 1.13
OS 48 0.00 31.73 5.64 3.35 2.70 0.81
Organic 88 0.03 56.00 4.49 1.00 2.97 1.15
U 131 )0.01 46.44 2.22 0.69 2.30 0.96
UAN 40 0.03 16.03 3.15 2.70 2.40 0.78
ANP 6 0.06 7.00 1.48 0.36 )1.73 0.26
Length of experiment (days)
0 –50 175 )0.01 16.31 1.19 0.16 1.20 0.28
50 –100 111 0.00 15.00 1.15 0.36 1.09 0.61
100 –200 311 )0.06 19.60 1.86 0.91 1.76 0.92
200 –300 77 )0.10 41.80 5.85 2.10 3.73 1.61
>300 334 )0.75 56.00 4.18 2.10 3.31 2.08
a
Emissions in kg N
2
O-N ha
)1
during the experimental period.
212
Table 2. Significance of differences between classes for BMED for N
2
O emissions from agricultural fields for those factors with a
significant influence.
Factor/factor class NFactor class
N Application rate
(kg ha
)1
)
0 –1 1 –50 50 –100 100 –150 150 –200 200 –250
0 –1 255
1 –50 30 d
50 –100 160 ds
100 –150 183 ddd
150 –200 113 ddds
200 –250 79 dddds
>250 188 dddddd
Soil organic C content (%) <1 1 –3
<1 82
1 –3 447 s
>3 180 dd
Soil pH <5.5 5.5 –7.3
<5.5 95
5.5 –7.3 465 h
>7.3 144 jj
Texture Coarse Medium
Coarse 509
Medium 219 h
Fine 158 jj
Climate Temp_C Temp_O S-Trop.
Temp_C 464
Temp_O 268 h
S-Trop. 144 jj
Trop. 132 hh j
Crop type Cereals Grass Leg. Other W-Rice
Cereals 184
Grass 282 h
Legumes 36 jj
Other 289 jjh
W-Rice 79 jjjj
None 107 jjhhj
Fertilizer type AA OAF AN CAN KN Mix OS Org. U UAN
AA 38
OAF 74 h
AN 131 hh
CAN 73 hjh
KN 58 hh hj
Mix 45 hhhhh
OS 48 hhhjhh
Organic 88 hhhhjhh
U 131 hhhjhhhh
UAN 40 hh hjhhhhh
ANP 6 jj jjjjhjjh
Length of experiment (days) 0 –50 50 –100 100 –200 200 –300
0 –50 175
50 –100 111 d
100 –200 311 dd
200 –300 77 ddd
>300 334 ddds
Solid = significant; open=not significant; circle = one-tailed test with excentricity = 1.64; cube = two-tailed test with
excentricity = 1.96.
213
climates, while the differences between the other
classes are small (Table 1). Only the BMED for
subtropical climates is significantly different for
the other climate types (Table 2). Surprisingly the
BMED values for N
2
O for tropical climates are
similar to those in temperate and lower than in
subtropical climates. Although not significantly
different, the BMED for continental temperate
climates is higher than for oceanic temperate cli-
mates, reflecting the higher winter emissions due to
temporary accumulation of soil N due to freeze-
thaw cycles (Kaiser and Ruser 2000).
The length of the experiment is the only mea-
surement-related factor with a significant influence
on N
2
O emissions (Table 1). As expected, N
2
O
emissions increase with the length of the mea-
surement period and for BMED the observed in-
crease is almost linear. Emissions for short
measurement periods tend to be disproportion-
ately high as studies often cover the high emission
period directly after N application. Differences
between classes are significant in all cases but one
(Table 2).
Controlling factors for NO
N application rate, soil N content, climate and
length of the experiment were found to have a
significant influence on NO emissions from agri-
cultural fields (Table 3). The values of MEA,
MED, BMEA and BMED generally increase
along with the amount of N fertilizer applied,
though only MEA and MED values for N appli-
cation rates >200 kg ha
)1
are markedly higher
than the other classes. However, after balancing
only the lowest N application rate is significantly
different from the others (Table 4). This may
indicate that the number of measurements is too
small to describe the high variability of NO emis-
sions in the data set. However, NO emissions that
increase along with N application rate are in
agreement with the findings for N
2
O.
MEA, MED, BMEA and BMED for soil N
content >0.2% exceed values for the two classes
with lower soil N content. Emissions for soil N
content >0.2% differ significantly from the class
with 0.05 –0.2% N, and the difference between the
two lower classes is also significant (Table 4). The
influence of soil N content reflects the positive
effect of soil organic matter on nitrification and
denitrification, similar to soil organic C for N
2
O
emissions.
The patterns of MEA, MED, BMEA and
BMED for the factor climate type vary (Table 4).
MEA, MED and BMED are highest for NO
Table 3. Number of observations (N), maximum (Max), minimum (Min), mean (MEA), median (MED), balanced mean (BMEA) and
balanced median (BMED, back-transformed after log transformation) emissions
a
for those factors with a significant influence on NO
emissions from agricultural fields.
Factor/factor class NMin Max MEA MED BMEA BMED
N Application rate (kg ha
)1
)
0–1 56 )0.18 2.62 0.35 0.09 1.61 0.10
1 –100 46 0.00 4.48 0.61 0.15 2.20 0.41
100 –200 56 0.00 3.00 0.38 0.14 2.15 0.53
>200 31 0.00 32.00 3.43 0.97 3.37 0.74
Soil N content (%)
<0.05 12 0.01 1.05 0.24 0.15 1.65 0.37
0.05 –0.2 18 0.00 0.47 0.08 0.03 1.93 0.14
>0.2 11 0.00 32.00 4.07 1.21 3.41 0.85
Climate
Temp_C 71 )0.18 4.48 0.37 0.11 1.57 0.17
Temp_O 22 0.00 32.00 1.69 0.19 4.21 0.30
S-Trop. 53 0.00 8.00 0.70 0.31 1.56 0.40
Trop. 43 0.00 10.70 1.74 0.54 2.00 0.78
Length of experiment (days)
0 –50 107 )0.03 2.62 0.22 0.05 2.04 0.08
50 –100 19 0.03 2.54 0.73 0.34 1.71 0.53
100 –200 33 )0.18 6.29 1.00 0.42 1.75 0.30
200 –300 7 0.28 1.13 0.55 0.39 2.00 0.36
>300 22 0.24 32.00 4.13 1.90 4.17 1.21
a
Emissions in kg NO-N ha
)1
during the experimental period.
214
emissions from tropical systems, and temperate
oceanic and subtropical climates show intermedi-
ate values of BMED. The two tropical climate
types are significantly different from the temperate
continental climate type, while the other differ-
ences are not significant (Table 4). Higher emis-
sions for tropical than for temperate climates
reflect the observed positive relationship between
temperature and NO emission (Williams and
Fehsenfeld 1991; Saad and Conrad 1993), and this
finding is consistent with for example Yienger and
Levy (1995) and Davidson and Kingerlee (1997).
MEA, MED and BMED increase along with the
length of the measurement period, but this trend is
not continuous because of the small number of
measurements in most classes. The shortest mea-
surement period (0 –50 days) has the largest
number of measurements, and is significantly dif-
ferent from the other classes. Furthermore, only
emissions for measurements covering >300 days
are significantly different from those for 100 –
200 days (Table 4).
Estimation of global annual N
2
O emissions
The summary model excludes the factor fertilizer
type, because there is no information about crop-
specific use of different fertilizer types at the global
scale, and because differences between most fer-
tilizer types are not significant. Furthermore, the
summary model handles N application rate as a
continuous variable, while this factor was classified
in the data summary for presentation purposes.
The effect values for the factors of the summary
model are listed in Table 5. For length of experi-
ment we use the class >300 days to calculate an-
nual emissions.
The N
2
O emissions calculated with the summary
model for agriculture and grassland show that the
broad patterns are mainly governed by N appli-
cation rate, while at smaller scales spatial vari-
ability is determined by differences in soil
parameters (Figure 1). Differences between crop
types mainly follow the effect values (Table 5),
Table 4. Significance of differences between classes for BMED
of NO emissions from agricultural fields for factors with a
significant influence.
Factor/factor class NFactor class
N Application rate
(kg ha
)
1)
0 –1 1 –100 100 –200
0–1 56
1 –100 46 d
100 –200 56 ds
>200 31 dss
Soil N content (%) <0.05 0.05 –0.2
<0.05 12
0.05 –0.2 18 d
>0.2 11 sd
Climate Temp_C Temp_OS-Trop.
Temp_C 71
Temp_O 22 h
S-Trop. 53 jh
Trop. 43 jhh
Length of experiment
(days)
0 –50 50–100 100 –200200 –300
0 –50 107
50 –100 19 d
100 –200 33 d
200 –300 7 dss
>300 22 dsds
Solid=significant; open=not significant; circle=one-tailed test
with excentricity=1.64; cube=two-tailed test with excentrici-
ty=1.96.
Table 5. Effect values and constant for the N
2
O and NO
summary model used for global emissions from agricultural
fields.
Factor/factor class N
2
O
model
NO
model
Constant )1.5160 )2.9950
N Application
rate per kg N ha
)1
0.0038 0.0061
Soil organic C content (for N
2
O)/soil N content (for NO) (%)
<1 0 <0.05 0
1 –3 0.0526 0.05 –0.2 )1.0211
>3 0.6334 >0.2 0.7892
Soil pH
<5.5 0
5.5 –7.3 )0.0693
>7.3 )0.4836
Texture
Coarse 0
Medium )0.1528
Fine 0.4312
Climate
Temp_C 0 0
Temp_O 0.0226 0.3511
S-Trop. 0.6117 0.5189
Trop. )0.3022 1.1167
Crop type
Cereals 0
Grass )0.3502
Legume 0.3783
Other 0.4420
W-Rice )0.8850
None 0.5870
Length of experiment
Per year (>300 days) 1.9910 2.5440
215
though lower crop-specific effects for grassland are
compensated for by higher fertilizer application
rates in some regions, and higher crop-specific ef-
fects for legumes are compensated for by lower
fertilizer application rates (data not shown).
Highest emission rates are calculated for other
crops, cereals and legumes in Europe and China.
High synthetic fertilizer inputs in crop systems
in East Asia, South Asia, North America and
Europe are reflected in the high emission sums for
these regions (Table 6). Even though more than
one crop is grown each year in large parts of China
and India, the aggregated emission rates are sup-
pressed by wide-spread rice cultivation. For
grassland the highest input of synthetic fertilizer is
in Europe, and animal manure is important in
Europe and North America, leading to high
emissions in these regions. Some regions with low
N application rates exhibit high N
2
O emission
sums because of the vast areas of grassland (e.g.
the former USSR).
The global annual N
2
O-N emission from fertil-
ized crops is 3.3 Tg with 0.1 Tg from rice crops,
0.4 Tg from legumes, 1 Tg from cereals and 1.9 Tg
from others crops (Table 6). Global annual emis-
sions from grassland amount to 0.8 Tg N
2
O-N.
The mean global FIE (see Section ‘Estimating
global annual emissions’) is 0.91% of the N ap-
plied in cropland (excluding legumes) and
grassland.
The average 95% confidence interval for calcu-
lated N
2
O emissions is )51% to +107%. This
uncertainty is comparable to that obtained by
Bouwman et al. (2002b) and that used as an
uncertainty range by IPCC (1997).
Estimation of global annual NO emissions
In the summary model the factor N application
rate was – as for N
2
O – included as a continuous
variable, and for length of experiment we used the
class >300 days to calculate annual emissions. As
no global map of soil N content was available we
used the map of soil organic C content assuming a
C/N ratio of 10 globally based on Brady (1990) as
a proxy for soil N content.
Similar to N
2
O, broad patterns of NO emissions
from agricultural soils are mainly governed by N
application rate, while at smaller scales spatial
variability is mainly determined by differences in
soil organic N content (Figure 2). As crop type is
not a model factor, differences between crops are
due to crop-specific fertilizer application rates
(data not shown). Highest emission rates are cal-
culated for all non-rice crops in Europe, and
intermediate values are observed in Northern
America and China (Figure 2). In spite of the low
kg N2O-N ha-1 yr-1
0- 2
2- 4
4- 6
6- 8
8-10
>10
Figure 1. Simulated annual N
2
O emission rates for agriculture and grassland. Values are weighted averages over the crop and
grassland areas within one grid cell and refer to land use in 1998.
216
fertilizer input, the NO emissions in many tropical
countries are rather high due to the high effect
value for tropical climate (Figure 2 and Table 5).
High NO emissions from South Korea (also ob-
served for N
2
O emissions) reflect high fertilizer
input rates in this country (IFA/IFDC/FAO
2003).
Tropical and subtropical climates promote high
NO emissions while fertilizer application rates are
generally low in these regions. Therefore, the cor-
relation between N fertilizer application and the
emission sum is not as strong as observed for N
2
O
(Table 6). Among the highest emissions come from
South and East Asia, where fertilizer application
rates are comparable to those in industrialized
countries (Table 6). However, high emissions are
also calculated for Latin America, Africa, and
Oceania, where fertilizer input is rather low. For
grasslands, which generally have lower fertilizer
and manure application rates, the same can be
seen, with high emissions from Latin America,
Africa and Oceania (Table 6).
The global annual NO-N emission from fertil-
ized crops is 1.4 Tg (Table 6), with 0.1 Tg from
rice crops, 0.1 Tg from legumes, 0.4 Tg from
cereals and 0.7 Tg from others crops, and total
emissions from grassland amount to 0.4 Tg NO-N
(Table 6). Our estimated global annual emission
from agricultural systems therefore is 1.8 Tg NO-
N. The calculated FIE for NO from agriculture
and grassland excluding legumes is 0.55%.
The relative 95%-confidence interval is )80%
and +406% for NO emissions from agricultural
fields. NO emission estimates are more uncertain
than those for N
2
O because of the smaller number
of available measurements in our data set. There is
no uncertainty estimate from the literature to
compare with, because Bouwman et al. (2002b)
did not assess the uncertainty (due to the limited
number of available measurements), while Veldk-
amp and Keller (1997) obtained an R
2
value,
which is not a true estimate of the uncertainty.
Soils under natural vegetation
Controlling factors for N
2
O
Soil organic C content, soil pH, bulk density,
drainage, vegetation type, length of the measure-
ment period and frequency of the measurements
have a significant influence on N
2
O emissions from
Table 6. Total fertilized area, N fertilizer and animal manure application, N
2
O and NO emissions for arable land and grassland for nine world regions
a
.
Region Cropland Grassland
b
Area
(Mha)
N fertilizer
(Gg y
)1
)
N manure
(Gg y
)1
)
N
2
O-N emission
(Gg y
)1
)
NO-N emission
(Gg y
)1
)
Area
(Mha)
N fertilizer
(Gg y
)1
)
N manure
(Gg y
)1
)
N
2
O -N emission
(Gg y
)1
)
NO-N emission
(Gg y
)1
)
North America 134 13,545 2532 459 116 173 0 1577 240 86
Latin America 116 5699 3373 363 177 73 55 145 79 58
North Africa and Middle East 61 4163 1271 150 50 30 23 50 32 12
West, East and Southern Africa 164 1202 1523 294 179 61 31 51 56 58
Europe 98 9231 3581 330 144 71 2418 1935 99 57
Former USSR 104 2132 2355 177 64 75 393 493 69 41
South Asia 219 15,686 6715 617 265 20 0 229 22 14
East Asia 216 25,323 6986 677 173 75 0 193 79 22
Southeast Asia, Oceania and Japan 118 7082 2631 278 220 97 138 144 134 70
World 1229 84,063 30,968 3345 1388 677 3058 4816 809 417
a
Totals for the world may differ from the sum of regional values due to rounding.
b
This includes grassland where manure or fertilizer is applied; grazing land only receiving N input from animal excretion during grazing is excluded.
217
soils under natural vegetation (Table 7). MEA,
MED, BMEA and BMED show continuously
increasing emissions with increasing soil organic C
content, indicating that the data set is well-bal-
anced. The classes with C content >1% are sig-
nificantly different from the class <1% (Table 8).
These findings agree with those for N
2
O emissions
from agricultural soils and literature (Tiedje 1988).
For soil pH the values for MEA, BMEA and
BMED indicate decreasing N
2
O emissions with
increasing pH (Table 7), and the BMED for
pH>7.3 is significantly different from both other
classes (Table 8). This also is consistent with the
findings for agricultural emissions.
The factors bulk density and drainage have an
effect on soil hydrological conditions and gas ex-
change. N
2
O emissions decrease along with soil
bulk density as is apparent for MEA, MED,
BMEA and BMED. Classes with bulk density
>1gm
)3
are significantly different from those
with bulk density <1 g m
)3
(Table 8). MEA and
MED values for the factor soil drainage class show
lower emissions for poorly drained soils. However,
BMEA and BMED are significantly higher for
poorly drained than for well-drained soils
(Table 8), which may be attributed to the small
number of measurements from poorly drained
soils. In general, poor drainage and high bulk
density both limit gas diffusion. Under low gas
diffusivity N
2
O is more likely to be re-consumed
before being emitted from the soil (Davidson
1991).
For vegetation type the patterns of MEA,
MED, BMEA and MED are not consistent as the
data are highly unbalanced (Table 7). For BMED
the hierarchy of emissions is rainforest > conif-
erous/deciduous forest (N affected) > savannah/
tropical dry forest. Emissions of N
2
O from rain-
forest are significantly higher than from grassland,
savannah and tropical dry forest, and emissions
from grassland are significantly lower than those
from deciduous forest and rainforest (Table 8).
Climate is not a significant factor for soils under
natural vegetation. However, climate is partly
represented by the factor vegetation type in many
cases, since most grassland sites in our data set are
from temperate regions. High values for tropical
rainforest reflect that these forests generally cycle
2 –4 times more N between soil and vegetation
than do most temperate ecosystems (Vitousek
1984; Jordan 1985; Vitousek and Sanford 1986).
Part of this difference may be related to the pres-
ence and N fixation activity of leguminous species
which are generally more abundant in tropical
than in temperate ecosystems (Crews 1999). An
important anthropogenic N input to ‘natural’
kg NO-N ha-1 yr-1
0-1
1-2
2-3
3-4
4-5
>5
Figure 2. Simulated annual NO emission rates for agriculture and grassland. Values are weighted averages over the crop and grassland
areas within one grid cell and refer to land use in 1998.
218
ecosystems is atmospheric N deposition, causing
the high values we find for the N-affected tem-
perate forests as observed by many authors (for
example, Brumme and Beese 1992).
As expected, N
2
O emissions increase with the
length of the experiment, confirming our results
for agricultural fields. There is a continuous trend
for BMED, while for MEA, MED, and BMEA the
class 50 –100 days breaks the otherwise continuous
increase. The differences between classes are sig-
nificant in most cases (Table 8). The factor fre-
quency of the measurements also has a significant
influence on N
2
O emissions (Table 7), but only the
class with less than one measurement per week is
significantly lower than the other classes (Table 8).
Controlling factors for NO
Soil organic C content, vegetation type and length
of experiment have a significant influence on NO
emissions from soils under natural vegetation.
MEA, MED, BMEA, and BMED continuously
increase along with soil C content (Table 9),
whereby the class >3% C is significantly different
from both other classes (Table 10). This finding is
consistent with the results for N
2
O emissions from
agricultural fields and soils under natural vegeta-
tion. Soil organic C content is a general indicator
of soil fertility, similar to soil N content for NO
emissions from agricultural fields.
For the factor vegetation type MEA is highest
for NO emissions from coniferous, deciduous and
Table 7. Number of observations (N), minimum (Min), maximum (Max), mean (MEA), median (MED), balanced mean (BMEA) and
balanced median (BMED, back-transformed after log transformation) emissions for those factors with a significant influence on N
2
O
emissions
a
from soils under natural vegetation.
Factor class N Min Max MEA MED BMEA BMED
Soil organic C content (%)
<1 5 0.02 0.16 0.06 0.03 0.64 0.06
1 –3 38 0.00 2.43 0.36 0.06 0.89 0.12
>3 44 0.00 7.45 1.07 0.31 1.04 0.19
Soil pH
<5.5 109 )0.03 7.45 0.52 0.04 1.32 0.27
5.5 –7.3 29 0.00 1.28 0.24 0.11 0.94 0.21
>7.3 4 0.02 0.04 0.03 0.04 0.32 0.02
Bulk density (g cm
)3
)
0.5 –1 26 0.02 6.89 1.18 0.55 1.19 0.33
1 –1.25 58 0.00 7.45 0.43 0.05 0.79 0.08
>1.25 8 0.00 0.31 0.05 0.01 0.59 0.05
Drainage
P 14 0.00 1.08 0.25 0.08 1.13 0.19
W 121 )0.03 7.45 0.55 0.08 0.58 0.07
Vegetation type
Coniferous 51 )0.03 2.10 0.13 0.01 0.92 0.14
Deciduous 18 0.00 1.15 0.48 0.46 0.42 0.15
Grass 31 0.00 1.08 0.11 0.06 0.63 0.07
Rain forest 77 0.00 7.45 0.85 0.21 1.37 0.24
Savannah 17 0.00 0.09 0.02 0.02 0.93 0.07
Tropical dry forest 13 0.01 0.70 0.11 0.04 0.87 0.08
Length of experiment (days)
0 –50 122 0.00 1.08 0.06 0.02 )1.04 0.01
50 –100 10 0.13 3.19 0.90 0.35 )1.60 0.09
100 –200 21 )0.03 1.90 0.29 0.10 1.95 0.15
200 –300 11 0.00 2.72 0.81 0.35 2.36 0.27
>300 43 0.01 7.45 1.29 0.67 2.62 0.41
Frequency of measurements
>1 per day 75 0.00 7.45 0.40 0.03 1.92 0.17
Daily 54 0.00 1.08 0.09 0.02 1.69 0.18
Every 2 –3 days 6 0.03 0.31 0.14 0.09 2.40 0.24
Every 4 –7 days 14 0.08 2.20 0.74 0.30 )0.78 0.08
<1 per week 58 )0.03 5.86 0.69 0.26 )0.94 0.03
a
Emissions in kg N
2
O-N ha
)1
during the experimental period.
219
tropical dry forest, while MED values for these
classes are low, indicating skewness of the data
(Table 9). BMEA and BMED for tropical systems
differ from MEA and MED, indicating unbal-
anced features of the data set. Most tropical
emission measurements stem from soils with a C
content >3%. As emissions are positively corre-
lated with soil organic C content this causes the
observed reduction of balanced values. BMED
values are highest for coniferous forest, interme-
diate for savannah, grassland and deciduous for-
est, and lowest for tropical rainforest. Most classes
are significantly different from two or three other
classes (Table 10). Vegetation type is also a rep-
resentation of climate in many cases (tropical
rainforest, savanna and dry forests, temperate
coniferous and deciduous). Our results for savan-
nas confirm literature reporting high emissions of
NO for climates with wet-dry cycles (Davidson
and Kingerlee 1997), and high values for temper-
ate N-affected forests reflect the accelerated N
cycling due to atmospheric N deposition.
Finally, our results indicate that the length of
the experiment is a significant factor, similar to our
results for N
2
O and NO from agricultural fields
and N
2
O emissions from soils under natural veg-
etation. The NO experiments generally cover
shorter periods than N
2
O measurements (see
Table 9; the class 0 –50 days has by far the largest
number). BMEA and BMED increase along with
Table 8. Significance of differences between classes for BMED of N
2
O emissions from soils under natural vegetation for factors with a
significant influence.
Factor/factor class N Factor class
Soil organic C content (%) <1 1 –3
<1 5
1–3 38 d
>3 44 ds
Soil pH <5.5 5.5 –7.3
<5.5 109
5.5 –7.3 29 h
>7.3 4 jj
Bulk density (g cm
)3
)0.5 –1 1 –1.25
0.5 –1 26
1 –1.25 58 j
>1.25 8 jh
Drainage P
P14
W 121 j
Vegetation type Coniferous Deciduous Grass Rain forest Savannah
Coniferous 51
Deciduous 18 h
Grass 31 hj
Rain forest 77 hhj
Savannah 17 hhhj
Tropical dry forest 13 hhhj h
Length of experiment (days) 0 –50 50 –100 100 –200 200 –300
0 –50 122
50 –100 10 d
100 –200 21 ds
200 –300 11 dds
>300 43 ddds
Frequency of measurements 1234
>1 per day 75
Daily 54 h
Every 2 –3 days 6 hh
Every 4 –7 days 14 hhh
<1 per week 58 jjjj
Solid=significant; open=not significant; circle=one-tailed test with excentricity=1.64; cube=two-tailed test with excentricity=1.96.
220
the length of experiment, and for MEA and MED
the continuous increase is only disturbed by the
class 100 –200 days.
Estimation of global annual N
2
O and NO emissions
The effect values for the parameters in the sum-
mary models for N
2
O and NO emissions are listed
in Table 11. For the factor length of experiment
we used the effect values for the class >300 days
to calculate annual emissions, and for frequency of
experiment we used >1 measurement per day.
Given the high uncertainty of the summary
models and the limited representation of different
ecosystems and climatic zones in the data set we
Table 9. Number of observations (N), minimum (Min), maximum (Max), mean (MEA), median (MED), balanced mean (BMEA),
balanced median (BMED, back-transformed after log transformation) for those factors with a significant influence on NO emissions
a
from soils under natural vegetation.
Factor class N7 Min Max MEA MED BMEA BMED
Soil organic C content (%)
<1 31 0.00 0.20 0.01 0.00 1.01 0.13
1 –3 52 0.00 3.38 0.19 0.01 1.02 0.14
>3 25 0.00 10.85 1.09 0.10 1.31 0.48
Vegetation type
Coniferous 53 0.00 8.04 0.47 0.01 2.01 0.45
Deciduous 10 0.00 2.49 0.40 0.01 0.88 0.17
Grass 43 0.00 0.69 0.08 0.00 1.02 0.29
Rain forest 33 0.00 2.38 0.39 0.04 0.39 0.11
Savannah 60 0.00 3.38 0.11 0.00 1.11 0.29
Tropical dry forest 11 0.00 10.85 1.30 0.02 1.26 0.10
Length of experiment (days)
0 –50 168 0.00 0.47 0.02 0.00 0.09 0.01
50 –100 8 0.08 2.82 0.69 0.45 0.59 0.26
100 –200 5 0.16 1.31 0.62 0.43 1.32 0.33
200 –300 6 0.18 1.09 0.66 0.58 1.38 0.47
>300 23 0.00 10.85 2.15 0.82 2.19 0.60
a
Emissions in kg NO-N ha
)1
during the experimental period.
Table 10. Significance of differences between classes for BMED of NO emissions from soils under natural vegetation for factors with a
significant influence.
Factor/factor class N Factor class
Soil organic C content (%) <1 1 –3
<1 31
1–3 52 s
>3 25 dd
Vegetation type Coniferous Deciduous Grass Rain forest Savannah
Coniferous 36
Deciduous 0 j
Grass 21 hh
Rain forest 59 jhj
Savannah 31 hhhj
Tropical dry forest 10 jhjh j
Length of experiment (days) 0 –50 50 –100 100 –200 200 –300
0 –50 168
50 –100 8 d
100 –200 5 ds
200 –300 6 dss
>300 23 dsss
Solid=significant; open=not significant; circle=one-tailed test with excentricity=1.64; cube=two-tailed test with excentricity=1.96.
221
regard the global emission maps (Figures 3 and 4)
as an illustration of the interacting effect of sig-
nificant factors at the global scale and not as
reliable estimates of natural N
2
O emission rates.
As most measurements for coniferous and
deciduous forests stem from areas with high
atmospheric N deposition the estimation of global
emissions excludes all temperate forests where N
deposition is less than 10 kg N ha
)1
y
)1
. Hence,
large areas in northern latitudes are not consid-
ered. The high effect value for tropical rainforest
(Table 11) leads to high N
2
O emissions from
tropical regions (Figure 3). The N
2
O emission
rates calculated for tundra and grasslands in
northern latitudes are comparable to those from
tropical savanna and dry forest systems due to the
combined effect of poorly drained soils and low
soil bulk density. Low pH values, which are
mainly found in tropical regions and high lati-
tudes, are further causes of high emission rates in
these two regions (Figure 3). The effect of the soil
organic C content on simulated global emission
patterns is not as strong as could be expected from
the summary model, as soil organic C content
exceeds 3% only in few regions according to the
global input dataset.
Regarding NO emissions, variation in emissions
can directly be attributed to the distribution of
vegetation types and their effect values (Table 11
and Figure 4). Lowest emissions are calculated for
rainforest and tropical dry forest. Higher NO-N
emissions of about 0.6 kg ha
)1
y
)1
are estimated
for temperate grasslands and savannah, which
together cover the largest area included in the
estimation. The soil organic C content in the data
set often exceeds 3%, while in the global soil map
it is lower than 3% in most areas. Therefore the
statistical model produces lower estimates of NO
(and also N
2
O) emissions from soils under natural
vegetation than one would conclude from the
measurement data per se. The average 95%-con-
fidence interval is )84% and +621% for N
2
O,
and )73% and +274% for NO emissions from
soils under natural vegetation.
Comparison with other studies
Agricultural fields
Compared to the data of Bouwman et al. (2002a)
(which we will refer to as subset), the data set of
N
2
O measurements for agricultural fields was ex-
tended with 162 measurements, and the results are
similar to those found with the data subset. The
20% increase in the number of measurements and
re-classification of climate types cause small dif-
ferences, i.e. climate is now a significant factor,
while soil drainage is not significant as it was for
the subset of data.
For N
2
O there is only little reduction of the
uncertainty due to the addition of new data, pos-
sibly because the subset had already a large num-
ber of measurements in primarily temperate
climates, and additional measurements in the same
climate types do not add much information.
Unfortunately, the representation of tropical cli-
mates did not increase substantially (relative con-
tribution of subtropical and tropical systems is 13
and 11% in the subset, and is now 14 and 13%,
respectively), so the representation of global envi-
ronmental conditions in agricultural systems has
not really improved.
Table 11. Effect values for the N
2
O and the NO model for soils
under natural vegetation.
N
2
O model NO model
Constant )2.8900 )3.952
Soil organic C content (%)
<1 0 0
1 –3 0.6683 0.0569
>3 1.0918 1.3265
Soil pH
<5.5 0
5.5 –7.3 )0.2750
>7.3 )2.4179
Bulk density (g cm
)3
)
0 –1 0.9941
1 –1.25 )0.3786
>1.25 )0.8597
Drainage
P0
W)1.0462
Vegetation type
Coniferous 0 0
Deciduous 0.0115 )0.9540
Grass )0.7941 )0.4335
Rain forest 0.4995 )1.4246
Savannah )0.6881 )0.4238
Tropical dry forest )0.5811 )1.5296
Length of experiment
Per year (>300 days) 3.6120 3.771
Frequency of experiment
>daily 0
222
The global estimate for annual N
2
O-N emis-
sions from cropland (3.3 Tg) we obtain here ex-
ceeds that based on the subset. This difference is
related to the summary model and the handling of
the data. Although the models are quite similar,
four different climate classes are now included,
which results in more variation and higher emis-
sions in sub-tropical and tropical climates. In
kg N2O-N ha-1 yr-1
0 - 0.25
0.25 - 0.5
0.5 - 0.75
0.75 - 1.0
1.0 - 1.25
> 1.25
Figure 3. Simulated annual N
2
O emission rates for natural ecosystems for 1998 land cover. Agricultural area, regrowth forest, arid
climate and polar climate are excluded.
kg NO-N ha-1 yr-1
0 - 0.25
0.25 - 0.5
0.5 - 0.75
0.75 - 1.0
1.0 - 1.25
> 1.25
Figure 4. Simulated annual NO emission rates for natural ecosystems for 1998 land cover. Agricultural area, regrowth forest, arid
climate and polar climate are excluded.
223
addition, in this study we have a more detailed
classification of crop types, which may lead to
higher emission estimates in some regions.
Our estimate for annual global N
2
O emissions
from fertilized grassland differs from the results
based on the data subset. In this study the grass-
land area of about 700 Mha includes primarily
managed grassland in mixed agricultural systems,
only excluding pastoral grazing land (Bouwman
et al. 2005). In contrast, (Bouwman et al. 2002b)
considered only those grassland areas receiving
fertilizer N inputs and therefore obtained lower
N
2
O (and NO) emissions.
Freibauer and Kaltschmitt (2003) used stepwise
multivariate linear regression to analyze N
2
O
emissions from Europe. Results were based on
measurements for arable sites in temperate oceanic
(61) and temperate continental (46) climates, and
for grassland sites (72). In our study we used
available data from all over the world, with 464
measurements for temperate oceanic and 268
measurements for temperate continental climates.
It is therefore difficult to compare our results in
terms of uncertainty with those of Freibauer and
Kaltschmitt (2003).
As Freibauer and Kaltschmitt (2003) did not
calculate total European emissions, we can only
compare the FIE estimates. Our estimate for FIE
for N
2
O is 0.91%. Based on their regression,
Freibauer and Kaltschmitt (2003) calculate FIE
values for N
2
O for arable soils in temperate oce-
anic climates of 0.2, 0.8% in temperate continental
climates, and 0.3% for grassland. This contradicts
their mean FIE obtained directly from the litera-
ture (1.3, 2.2 and 1.2% for arable soils in tem-
perate oceanic and temperate continental climates,
and grassland, respectively). Our FIE and their
direct mean values are consistent with the 1.25%
currently used as default FIE by the IPCC meth-
odology for national greenhouse gas inventories
(Bouwman 1996; IPCC 1997), and with the 0.9%
obtained by Bouwman et al. (2002b) based on the
subset.
For NO emissions from agricultural soils 90
measurements were added compared to the subset
(Table 12). This 91% increase resulted in soil
drainage as a significant control of NO emissions,
while for the subset this was soil organic C con-
tent. Furthermore, climate is a significant factor
additional to the N application rate (significant for
both the subset and extended data set). The fre-
quency of measurements is no longer significant,
while it was a major factor for the subset. We
believe that results are less uncertain than those
based on the subset. This does not mean that the
data now represent the full variability of world
agricultural systems, but temperate continental
(36%), subtropical (28%) and tropical (23%) are
better represented than in the subset.
Our estimated global annual NO emission sum
from agricultural systems (1.8 Tg) is much lower
than the 5 Tg estimate in the inventory of David-
son and Kingerlee (1997), also lower than the
2.6 Tg reported in a recent summary at the global
N cycle (Galloway et al. 2004) and similar to the
1.6 Tg reported by Bouwman et al. (2002b).
However, a proper comparison is difficult because
of differences in the types and areas of grassland in
the various studies.
The calculated FIE of 0.55% for NO from agri-
culture and grassland excluding legumes agrees
with the estimate of 0.5% by (Veldkamp and Keller
1997) and is somewhat lower than the 0.7% of
Bouwman et al. (2002b) based on a smaller data set.
Soils under natural vegetation
For global N
2
O and NO emissions from soils un-
der natural vegetation no purely statistical emis-
sion model has been developed so far, but
empirical approaches have been applied both for
global emissions of N
2
O (Bouwman et al. 1993;
Kreileman and Bouwman 1994) and NO (Yienger
and Levy 1995). Additionally, process based
models have been used to estimate N
2
O emissions
(Nevison and Holland 1997; Potter et al. 1997).
Table 12. Comparison of number of measurements of N
2
O and
NO for agricultural fields in this study and Bouwman et al.
(2002a).
Crop type Number of N
2
O
measurements
Number of NO
measurements
2002 This study 2002 This study
Grass 193 282 23 55
Legumes 36 36 16 14
Wetland rice 61 79 2 2
All other
(incl. ‘not known’)
556 611 58 118
Total 846 1008 99 189
224
The global N
2
O emission rates calculated with
our summary model differ from the pattern sug-
gested by the above cited references. The main
reason for this discrepancy is that both the empir-
ical model (Bouwman et al. 1993) and the process
models (Nevison and Holland 1997; Potter et al.
1997) strongly link N
2
O emission rates to one or
more of the parameters Normalized Difference
Vegetation Index (NDVI), Net Primary Produc-
tion (NPP), decomposition rate or temperature. All
these parameters peak in tropical systems. In
addition, high emissions are calculated by the sta-
tistical model for high latitudes due to the the im-
pact of drainage class and bulk density. Therefore
the emission sums and average emission rates for
broad vegetation classes differ between Bouwman
et al. (1993) and this study (Table 13). While
Bouwman et al. (1993) covers the entire area of
temperate forests and assumes no N deposition,
here only N affected temperate forests are included
with higher emission rates. For the three other
vegetation classes the areas are not directly com-
parable because of different classifications (Ta-
ble 13). The emission estimate for closed tropical
rainforest is similar, while the emissions we calcu-
late for open tropical forest and grassland/steppe
are lower compared to Bouwman et al. (1993).
The patterns of global NO emissions from soil
under natural vegetation calculated with the sta-
tistical model differ from both the empirical model
(Yienger and Levy 1995) and the process-based
approach (Potter et al. 1997). Analogous to N
2
O,
the NO emissions according to Potter et al. (1997)
are strongly linked to NPP, decomposition rates
and temperature, thus predicting highest emissions
in tropical systems. In contrast, Yienger and Levy
(1995) basically derived a biome-specific NO
emission potential from a compilation of mea-
surement data (which is highest for tropical sys-
tems), and superimposed a temperature response
function. Though they additionally accounted for
other effects like pulsing, this basic mechanism
also causes their emission estimates to roughly
increase with decreasing latitude.
A more recent biome stratification of NO
emissions based on mean values and expert judg-
ment covers a larger variety of systems though not
deriving an empirical model (Davidson and
Kingerlee 1997). Our calculated NO emissions are
systematically lower than these values (Table 13),
which can be attributed to the effect of soil organic
C content and the log transformation that we used
to reduce effects of outliers. However, the relative
emission rates for vegetation classes are similar in
both cases, with lowest emissions calculated for
tropical rainforest. Given the high uncertainty
range of the statistical model and the problematic
interaction of the two parameters C content and
vegetation class, we recognize that the estimation
of global NO emissions from soils under natural
vegetation presented is highly uncertain due to
scarcity of data.
Table 13. Comparison of N
2
O and NO emission estimates from soils under natural vegetation.
Vegetation classes Area (Mha) Emission (N
2
O-N or NO-N) Area (Mha) Emission (N
2
O-N or NO-N)
Gg y
)1
kg ha
)1
y
)1
Gg y
)1
kg ha
)1
y
)1
This study Bouwman et al. (1993)
A. N
2
O emission estimates
Temperate forest
a
230 147 0.64 2246 500 0.22
Open tropical forest
b
1598 333 0.21 1028 1000 0.97
Closed tropical forest
c
854 1170 1.37 1682 2300 1.37
Grassland/steppe 2765 403 0.15 3147 1500 0.48
This study Davidson and Kingerlee (1997)
B. NO emission estimates
Temperate forest
a
230 105 0.46 100 300 3.00
Open tropical forest
b
1598 670 0.42 2400 7400 3.08
Closed tropical forest
c
854 186 0.22 1600 1320 0.83
Grassland/steppe 2765 1559 0.56 900 1100 1.22
a
N affected temperate forest, except for the estimate of Bouwman et al. (1993) which covers the entire temperate forest area.
b
Including shrubland, savanna and tropical woodland.
c
Including warm humid, deciduous and montane tropical forest and warm mixed forest.
225
Conclusions
The analysis based on an extended version of the
data set presented in Bouwman et al. (2002b)
does not yield a considerable improvement or
reduction of uncertainty in the N
2
O emissions
from agricultural fields compared to the reduced
data set. This is because the representation of
environmental and management conditions in
agricultural systems did not improve. For NO
emissions from agricultural fields the analysis is
based on a much larger number of measurements
(200%) compared to Bouwman et al. (2002b),
now covering a higher diversity of environmental
conditions. The uncertainty of NO emission esti-
mates was considerably reduced compared to the
previous analysis.
For agricultural N
2
O a better understanding of
important processes and better emission estimates
can be expected by improving the representation
of tropical and subtropical agricultural systems,
and by including more measurements for legumes
and for wetland rice covering also the post-drain-
age period. In contrast, agricultural NO measure-
ments in the database already cover temperate and
tropical systems likewise, but the number of mea-
surements is substantially lower than for N
2
O,
which is reflected in a lower number of significant
factors and a much higher uncertainty range.
For emissions from soils under natural vegeta-
tion this analysis is the first comprehensive statis-
tical analysis of published measurement data
(about 200 for both N
2
O and NO). Given the
incomplete coverage of global vegetation zones
and the high uncertainty of the developed statis-
tical models, global annual emissions cannot be
calculated with this approach. Far more mea-
surement data, and particularly a better represen-
tation of the vegetation types tropical dry forest,
savanna, tundra and temperate ecosystems that
are not affected by N deposition, preferably cov-
ering prolonged periods, are needed to understand
the complexity of interactions.
The statistical models presented in this study are
useful to estimate seasonal or annual N
2
O and NO
emissions based on site-specific environmental and
management parameters, and as they include
estimates of uncertainty, they can serve as a
benchmark to process-based models applied at
larger spatial scales.
Acknowledgments
Financial support was granted by the Interna-
tional Max Planck School for Earth Systems
Modeling (Hamburg, Germany). The work of
AFB is part of the project Integrated Terrestrial
Modeling (S/550005/01/DD) of the Netherlands
Environmental Assessment Agency. We thank Leo
Boumans for his advice on the statistical analysis.
References
Alexander M. 1977. Introduction to Soil Microbiology, 2nd ed.
Wiley and Sons, New York, 467 pp.
Batjes N.H. 2002. Revised soil parameter estimates for the soil
types of the world. Soil Use Manage 18: 232 –235.
Bobbink R., Hornung M. and Roelofs J.G.M. 1998. The effects
of air-borne nitrogen pollutants on species diversity in nat-
ural and semi-natural European vegetation. J. Ecol. 86: 717–
738.
Bouwman A.F. 1996. Direct emission of nitrous oxide from
agricultural soils. Nutr. Cycl. Agroecosyst. 46: 53 –70.
Bouwman A.F., Boumans L.J.M. and Batjes N.H. 2002a.
Emissions of N
2
O and NO from fertilized fields. Summary of
available measurement data. Global Biogeochem. Cycles
16(4): 1058. doi: 10.1029/2001GB001811.
Bouwman A.F., Boumans L.J.M. and Batjes N.H. 2002b.
Modeling global annual N
2
O and NO emissions from fertil-
ized fields. Global Biogeochem. Cycles 16(4): 1080.
doi:10.1029/2001GB001812.
Bouwman A.F., Fung I., Matthews E. and John J. 1993. Global
analysis of the potential for N
2
O production in natural soils.
Global Biogeochem. Cycles 7: 557 –597.
Bouwman A.F., Van Drecht G. and Van der Hoek K.W. 2005.
Nitrogen surface balances in intensive agricultural produc-
tion systems in different world regions for the period 1970 –
2030. Pedosphere 15(2): 137 –155.
Bouwman A.F., Van Vuuren D.P., Derwent R.G. and Posch
M. 2002c. A global analysis of acidification and eutrophica-
tion of terrestrial ecosystems. Water Air Soil Pollut. 141:
349 –382.
Brady N.C. 1990. The Nature and Properties of Soils, 10th ed.
Macmillan Publishing Company, New York.
Bremner J.M. 1997. Sources of nitrous oxide in soils. Nutr.
Cycl. Agroecosyst. 49: 7 –16.
Bronson K.F., Neue H.U., Singh U. and Abao E.B. 1997.
Automated chamber measurements of methane and nitrous
oxide flux in a flooded rice soil: 2. Fallow period emissions.
Soil Sci. Soc. Am. J. 61: 988 –993.
Bruinsma J.E. 2003. World Agriculture: Towards 2015/2030.
An FAO Perspective. Earthscan, London, 432 pp.
Brumme R. and Beese F. 1992. Effects of liming and nitrogen
fertilization on emissions of CO
2
and N
2
O from a temperate
forest. J. Geophys. Res. 97: 12851 –12858.
Collins W.J., Stevenson D.S., Johnson C.E. and Derwent R.G.
1997. Tropospheric ozone in a global-scale three-dimensional
226
Lagrangian model and its response to NO
x
emission controls.
J. Atmos. Chem. 26: 223 –274.
Crews T. 1999. The presence of nitrogen fixing legumes in
terrestrial communities: Evolutionary vs ecological consid-
erations. Biogeochemistry 46: 233 –246.
Davidson E.A. 1991. Fluxes of nitrous oxide and nitric oxide
from terrestrial ecosystems. In: Rogers J.E. and Whitman
W.B. (eds), Microbial Production and Consumption of
Greenhouse Gases: Methane, Nitrogen Oxides and Halome-
thanes, American Society of Microbiology, Washington DC,
pp. 219 –235.
Davidson E.A. and Kingerlee W. 1997. A global inventory of
nitric oxide emissions from soils. Nutr. Cycl. Agroecosyst. 48:
37 –50.
de Pauw E., Nachtergaele F.O., Antoine J., Fisher G. and Van
Velthuizen H.T. 1996. A provisional world climatic resource
inventory based on the length-of-growing-period concept. In:
Batjes N.H., Kauffman J.H. and Spaargaren O.C. (eds),
National Soil Reference Collections and Databases (NAS-
REC), International Soil Reference and Information Centre
(ISRIC), Wageningen, pp. 30 –43.
FAO 2004. FAOSTAT Database Collections (http://
www.apps.fao.org). Food and Agriculture Organization of
the United Nations, Rome.
Firestone M.K. and Davidson E.A. 1989. Microbiological
basis for NO and N
2
O production and consumption in soils.
In: Andreae M.O. and Schimel D.S. (eds), Exchange of Trace
Gases between Terrestrial Ecosystems and the Atmosphere,
Wiley and Sons, Chichester, pp. 7 –21.
Freibauer A. and Kaltschmitt M. 2003. Controls and models
for estimating direct nitrous oxide emissions from temperate
and sub-boreal agricultural mineral soils in Europe. Biogeo-
chemistry 63: 93 –115.
Freney J.R. 1997. Emission of nitrous oxide from soils used for
agriculture. Nutr. Cycl. Agroecosyst. 49: 1–6 (DOI: 10.1023/
A:1009702832489).
Galloway J.N., Dentener F.J., Capone D.G., Boyer E.W.,
Howarth R.W., Seitzinger S.P., Asner G.P., Cleveland C.C.,
Green P.A., Holland E.A., Karl D.M., Michaels A.F., Porter
J.H., Townsend A.R. and Vo
¨ro
¨smarty C.J. 2004. Nitrogen
cycles: past, present, and future. Biogeochemistry 70: 153 –
226.
IFA/IFDC/FAO 2003. Fertilizer Use by Crop 5th ed. Food and
Agriculture Organization of the United Nations, Rome.
IMAGE-team 2001. The IMAGE 2.2 implementation of the
SRES scenarios. A comprehensive analysis of emissions, cli-
mate change and impacts in the 21st century. CD-ROM
publication 481508018, National Institute for Public Health
and the Environment, Bilthoven.
IPCC 1997. Guidelines for National Greenhouse Gas Inven-
tories. Intergovernmental panel on Climate Change/Organi-
zation for Economic Cooperation and Development, Paris.
IPCC 2001. Third assessment report. Working Group I.
Cambridge University Press, Cambridge.
Jordan C.F. 1985. Nutrient Cycling in Tropical Forest Eco-
systems: Principles and their Application in Management and
Conservation. Wiley, Chichester, 179 pp.
Kaiser E.A. and Ruser R. 2000. Nitrous oxide emissions from
arable soils in Germany. An evaluation of six long-term field
experiments. J. Plant Nutr. Soil Sci. (Zeitschrift fu
¨r Plan-
zenerna
¨hrung and Bodenkunde 163: 249 –259.
Kreileman G.J.J. and Bouwman A.F. 1994. Computing land
use emissions of greenhouse gases. Water Air Soil Pollut. 76:
231 –258.
Li C. and Aber J. 2000. A process-oriented model of N
2
O and
NO emissions from forest soils: I Model development. J.
Geophys. Res. 105: 4369 –4384.
Martikainen P.J. and de Boer W. 1993. Nitrous oxide pro-
duction and nitrification in acidic soil from a Dutch conif-
erous forest. Soil Biol. Biochem. 25: 343 –347.
Meixner F.X., Fickinger T., Marufu L., Serca D., Nathaus F.J.,
Makina E., Mukurumbira L. and Andreae M.O. 1997. Pre-
liminary results on nitric oxide emission from a southern Afri-
can savanna ecosystem. Nutr. Cycl. Agroecosyst. 48: 123 –138.
Mosier A.R., Duxbury J.M., Freney J.R., Heinemeier O. and
Minami K. 1996. Nitrous oxide emissions from agricultural
fields: assessment, measurement and mitigation. Plant Soil
181: 95 –108.
Mosier A.R., Kroeze C., Nevison C., Oenema O., Seitzinger S.
and Van Cleemput O. 1998. Closing the global atmospheric
N
2
O budget: nitrous oxide emissions through the agricultural
nitrogen cycle. Nutr. Cycl. Agroecosyst. 52: 225 –248.
Nevison C. and Holland E. 1997. A reexamination of the impact
of anthropogenically fixed nitrogen on atmopheric N
2
O and
the stratospheric O
3
layer. J. Geophys. Res. 102: 25519 –25536.
New M., Hulme M. and Jones P. 1999. Representing twentieth-
century space-time climate variability. Part I: Development
of a 1961 –90 mean monthly terrestrial climatology mean
monthly terrestrial climatology. J. Climate 12: 829 –856.
Parton W.J., Mosier A.R., Ojima D.S., Valente D.W., Weier K.
and Kulmala A.E. 1996. Generalized model for N
2
and N
2
O
production from nitrification and denitrification. Global
Biogeochem. Cycles 10: 401 –412.
Payne R.W., Baird D.B., Gilmore A.R., Harding S.A., Lane
P.W., Murray D.A., Soutar D.M., Thompson R., Todd
A.D., Wilson G.T., Webster R. and Welham S.J. 2000.
Genstat Release 4.2. Reference manual. Lawes Agricultural
Trust (Rothamsted Experimental Station), Harpenden,
Hertfordshire, UK.
Potter C.S., Riley R.H. and Klooster S.A. 1997. Simulation
modeling of nitrogen trace gas emissions along an age gra-
dient of tropical forest soils. Ecol. Model. 97(3): 179 –196.
Robertson G.P. 1989. Nitrification and denitrification in hu-
mid tropical ecosystems: potential controls on nitrogen
retention. In: Proctor J. (ed.), Mineral Nutrients in Tropical
Forest and Savanna Ecosystems, Blackwell Scientific Publi-
cations, Oxford, pp. 55 –69.
Saad A.L.O. and Conrad R. 1993. Temperature dependence of
nitrification, denitrification and turnover of nitric oxide in
different soils. Biol. Fertil. Soils 15: 21 –27.
Snedecor G.W. and Cochran W.G. 1980. Statistical Methods,
7th ed. The Iowa State University Press, Ames, Iowa, 507 pp.
Tiedje J.M. 1988. Ecology of denitrification and dissimilatory
nitrate reduction to ammonium. In: Zehnder A.J.B. (ed.),
Biology of Anaerobic Microorganisms, Wiley and Sons, New
York, pp. 179 –244.
Trenkel M.E. 1997. Improving Fertilizer Use Efficiency. Con-
trolled-release and Stabilized Fertilizers in Agriculture.
International Fertilizer Industry Association, Paris.
Veldkamp E. and Keller M. 1997. Fertilizer-induced nitric
oxide emissions from agricultural soils. Nutr. Cycl. Agro-
ecosyst. 48: 69 –77.
227
Verchot L.V., Davidson E.A., Cattanio J.H., Ackerman I.L.,
Erikson H.E. and Keller M. 1999. Land use change and
biogeochemical controls of nitrogen oxide emissions from
soils in eastern Amazonia. Global Biogeochem. Cycles 13:
31 –46.
Vitousek P.M. 1984. Litterfall, nutrient cycling and nutrient
limitation in tropical forests. Ecology 65: 285 –298.
Vitousek P.M. and Sanford R.L. 1986. Nutrient cycling in moist
tropical forest. Annu. Rev. Ecol. Systemat. 17: 137 –167.
Williams E.J. and Fehsenfeld F.C. 1991. Measurement of soil
nitrogen oxide emissions at three North American ecosys-
tems. J. Geophys. Res. 96: 1033 –1042.
Yienger J.J. and Levy H. II 1995. Empirical model of global
soil-biogenic NO
x
emissions. Global Biogeochem. Cycles
100: 11447 –11464.
228
