Prediction of grain protein in spring malting barley grown in northern Europe

Abstract

The ability to predict grain protein concentration at harvest (CP) in Swedish malting barley (Hordeum distichum) from observations of soil conditions, sowing day, fertilisation rate, remote sensing at early stem elongation and the temperature sum during grain filling, was tested for two cultivars; Astoria (Secobra, France) and Wikingett (Svalöf-Weibulls, Sweden) in 16 fertilisation trials in southern Sweden, encompassing 3 years (2001–2003). Fertilisation was applied either as a single dose at sowing, or as both a starting application and an application at early stem elongation. The highest total application rate was 160kgNha−1y−1. The soil was analysed for phosphorus, potassium, magnesium and calcium in the layer 0–0.3m, and mineral N down to 0.6m. Canopy reflectance observations at BBCH 32 was used to calculate a vegetation index (TCARI(32)) to reflect the canopy (leaf and straw) chlorophyll concentration. Harvested grain CP correlated only marginally with the observed soil variables. It was, however, for a specific cultivar, possible to make a prediction of grain CP based on day of sowing and TCARI(32) with Radj2=0.78. Part of the sowing day effect might be due to thermal stress during grain filling, as the risk for high temperatures during this phase was higher when sowing was late. This might also explain why the introduction of accumulated temperature during grain filling, into a model already including sowing day and TCARI(32) as independent variables, did not improve the predictability of grain CP.

Full text

Europ. J. Agronomy 27 (2007) 205–214
Prediction of grain protein in spring malting
barley grown in northern Europe
C.G. Pettersson ∗, H. Eckersten
Department of Crop Production Ecology, Box 7043, Swedish University of Agricultural Sciences (SLU),
SE-750 07 Uppsala, Sweden
Received 2 February 2007; received in revised form 6 April 2007; accepted 12 April 2007
Abstract
The ability to predict grain protein concentration at harvest (CP) in Swedish malting barley (Hordeum distichum) from observations of soil
conditions, sowing day, fertilisation rate, remote sensing at early stem elongation and the temperature sum during grain filling, was tested for two
cultivars; Astoria (Secobra, France) and Wikingett (Sval¨
of-Weibulls, Sweden) in 16 fertilisation trials in southern Sweden, encompassing 3 years
(2001–2003). Fertilisation was applied either as a single dose at sowing, or as both a starting application and an application at early stem elongation.
The highest total application rate was 160 kg N ha−1y−1. The soil was analysed for phosphorus, potassium, magnesium and calcium in the layer
0–0.3 m, and mineral N down to 0.6 m. Canopy reflectance observations at BBCH 32 was used to calculate a vegetation index (TCARI(32)) to
reflect the canopy (leaf and straw) chlorophyll concentration. Harvested grain CP correlated only marginally with the observed soil variables. It
was, however, for a specific cultivar, possible to make a prediction of grain CP based on day of sowing and TCARI(32) with R2
adj =0.78. Part of the
sowing day effect might be due to thermal stress during grain filling, as the risk for high temperatures during this phase was higher when sowing
was late. This might also explain why the introduction of accumulated temperature during grain filling, into a model already including sowing day
and TCARI(32) as independent variables, did not improve the predictability of grain CP.
© 2007 Elsevier B.V. All rights reserved.
Keywords: Canopy reflection; Cultivar; Grain filling; Grain protein; Malting barley; Sowing day number; Temperature sum; Vegetation index
1. Introduction
The ideal grain crude protein concentration (CP) of malting
barley for the production of European lager beer is 10.7% of dry
matter (DM), with a permitted range of 9.5–11.5%. At higher
levels, lower starch content results in less alcohol and excess
CP results in cloudy beer, whereas, at lower grain CP levels,
yeast activity is limited by nitrogen (N) shortage. Protein con-
centration also influences germination rate, and this can cause
problems when there is variation within the lot, even though its
mean protein content is acceptable (Palmer, 2000).
1.1. Soil, season and fertiliser
Much work has been done on the effects of application meth-
ods, rates and timing of N to malting barley. Unifying for most
∗Corresponding author. Tel.: +46 18 671428; fax: +46 18 672890.
E-mail address: cg.pettersson@vpe.slu.se (C.G. Pettersson).
studies is that predictions of grain CP are hard to do only using
information available at sowing. Framework for nitrogen appli-
cation has been developed using soil mineral N down to 1 m
depth sampled at sowing, and an assumption of possible grain
yield and its N needs (17 kg of N per 1000kg grain; Birch et al.,
1997). In contrast McTaggart and Smith (1992) found that crop
soil N uptake mainly originated from N mineralisation from soil
organic matter during the growing period, and to a less extent
from soil mineral N content at sowing. However, as mineralisa-
tion during the season was difficult to predict the soil mineral
N content still was the best soil predictor for crop N uptake.
The crude protein concentration of harvested grain (grain CP)
has been found to correlate with N application rate but with
large seasonal and site differences (Conry, 1995, 1997), and
in line with this McTaggart and Smith (1992) found grain CP
more affected by site and season than by fertiliser treatment.
Also, the availability of fertiliser N depends on an interaction
between the fertiliser N form and water conditions, where cal-
cium nitrate worked better than ammonium-nitrate under dry
conditions (McTaggart and Smith, 1992). Grain CP not only
1161-0301/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.eja.2007.04.002

206 C.G. Pettersson, H. Eckersten / Europ. J. Agronomy 27 (2007) 205–214
depends on N availability but also on factors influencing carbon
assimilation. Grain CP and grain yield have been found to inter-
act with sowing dates, with higher CP and lower yield after late
sowing (Conry, 1995, 1997), explained by sowing date influenc-
ing the timing between crop development stage and climate. The
knowledge around the crop-nitrogen relation of malting barley is
still expanding, and sowing date has been used (e.g., McKenzie
et al., 2005) as a way of understanding the system better.
There is considerable variation of malting barley grain yield
and grain CP also within fields. For example, in studies span-
ning 6 years and using systematic grids of machine harvested
(10 m ×2.4 m) plots in clay-loam production fields (Thyl´
en et
al., 1999; Pettersson et al., 2006) we found variation of ±1.25%
CP within fields of roughly 10 hectares (ha). Grain yield was
negatively correlated with grain CP, but the spatial distributions
of soil mineral N (Nmin) at sowing, soil clay content, and SOM,
alone or in combination, were not related to the spatial distri-
bution of CP. However, we found a covariation between grain
CP and apparent soil electrical conductivity (ECa), which, in
turn, depends upon factors like soil porosity, SOM, clay con-
tent, soil salinity or soil water content (Ehsani and Sullivan,
2002; Jabro et al., 2005). The role of ECain the study indicated
that canopy development and nitrogen status were mostly related
to soil porosity. No unifying way to use ECawas proposed, as
ECacould correlate to other factors in other fields.
This poses difficulties to the farmer in producing malting
barley grain of an acceptable and uniform quality. Precision agri-
culture promises solutions to the production of malting barley
grain by variable rate fertilisation but, to date, little precision
agriculture research has been done on malting barley.
1.2. Canopy reflectance and temperatures
Controlled fertilisation during crop growth depends upon
enough information on the performance of the crop after estab-
lishment. Canopy reflection data measured early in the season
can provide potentially valuable information of this kind (Lark
and Wheeler, 2003). The relationship between the spectral distri-
bution of incoming and reflected light is determined by how the
canopy is utilising the received solar energy. A common method
of expressing canopy reflectance measurements is to calculate
a vegetation index (VI), based on relationships between sep-
arate wavelengths or wavelength bands (Thiam and Eastman,
2001). Canopy reflectance depends on chlorophyll and plant
DM, and as nitrogen is an important component of chlorophyll
and the amount of N depends on plant size, observations are
often interpreted as a measure of plant N content. A VI calcu-
lated from reflection at growth stage 32 (BBCH 32, Lancashire
et al., 1991) provides a measure of the potential for growth
during stem elongation, the period when the upper limits for
grain population density, and thus for yield, are set. Later N
application can cause systematically higher grain CP than with
a single N application at sowing, as demonstrated in several,
non-published, studies of Swedish conditions. Based on these
observations, the current Swedish recommendation is to use
early stem elongation as the latest stage for planned N appli-
cations to barley.
In a previous study (Pettersson et al., 2006) we found that
canopy reflection at BBCH 32 can be a valuable measurement
for estimating patterns of grain CP, especially in combination
with information on soil electrical conductivity and temperature
conditions during grain filling. This study, which included many
(81) sites but only three different temperature regimes, provided
promising results but there is a need for more thorough test-
ing using a dataset representing a larger variation of climatic
conditions.
Temperature is a key factor controlling crop development,
and temperature sums are used as the main driving variables in
all major crop models (e.g., Porter et al., 1993). Temperatures
during anthesis and grain filling can have an additional impact on
grain yield and grain CP, especially when daily maximum tem-
perature (TMAX) reaches the heat shock range, (TMAX >32◦C;
Wardlaw and Wrigley, 1994). Also TMAX values in the range
20–32 ◦C might limit grain filling in wheat (Tribo¨
ı and Tribo¨
ı-
Blondel, 2002) and barley (Passarella et al., 2002) possibly due
to a shortened grain filling period. This is one of the reasons for
early sowing of spring established cereals in temperate regions,
which are generally sown as soon as the soil dries beyond field
capacity. Delayed sowing is expected to result in lower grain
yield and higher grain CP under climatic conditions with increas-
ing temperatures from sowing to grain filling. A more rapid
crop development causes a lower accumulated growth before
the start of grain filling and also shortens the grain filling period.
Relations between sowing dates, grain yield and CP has been
described in detail in the 1980s by Kirby et al. (1982, 1985) but
active research in the area is still performed (e.g., Juskiw and
Helm, 2003).
1.3. Objectives
The objective of the present study was to develop a method
for predicting grain CP of malting barley at harvest, using
observations of the crop, and site conditions during the grow-
ing period. Different combinations of predictors (sowing day,
canopy reflectance at early stem elongation, temperature sums
during grain filling, soil conditions and fertilisation regimes)
were evaluated, to become a base for developing a practical
methodology for variable fertilisation of malting barley in North-
ern Europe.
2. Materials and methods
2.1. Model
The day of sowing influences both the light climate that the
emerging seedling is exposed to and the risk of exposing the crop
to super-optimal temperatures during grain filling. Both factors
have an impact on the crop, and sowing day is of importance
for the development of a spring established, temperate climate
crop.
Remote sensing at BBCH 32, can give a measure of canopy
development and the nitrogen status of the canopy. In the present
work, the remote sensing observations were the radiation emis-
sion, as a fraction of incoming radiation, and they were used to

C.G. Pettersson, H. Eckersten / Europ. J. Agronomy 27 (2007) 205–214 207
calculate a vegetation index based on the absorption spectrum of
chlorophyll. The index was calculated from the fractional emit-
ted light at chlorophyll absorption maximum at 670 nm (R670)
combined with the absorption minimum at 700 nm (R700), and
also using the local chlorophyll absorption minimum at 550 nm
(R550) to form the “Transformed Chlorophyll in Absorption
Index” (TCARI, Haboudane et al., 2002).
TCARI =3(R700 −R670)−0.2(R700 −R550 )R700
R670  (1)
This VI has proven useful for barley grain CP predictions at
harvest in a related study (Pettersson et al., 2006). Finally, the
temperature sum (day-degrees (◦C d)) during grain filling was
calculated, as it is expected to influence grain CP at high levels,
by limiting the carbon (C) filling of the grain. The daily temper-
ature above a threshold (Tb), optimised by statistical methods,
was accumulated for a 3-week period starting at BBCH 45 (Late
boot stage: flag leaf sheath swollen) to cover anthesis and grain
filling. The start day of the accumulation of temperature was cho-
sen the same as the sampling day of canopy reflectance (data not
shown) and the apparent development stage was observed. The
“Stress Temperature Sum” (STS) (Pettersson et al., 2006)was
accumulated using the daily maximum temperature (TMAX)to
achieve a high sensitivity to temperatures in the medium-high
range (20–32 ◦C):
STS(Tb)=
BBCH 45+3 weeks

t=BBCH 45
TMAX(t)−Tb,T
MAX >T
b(2)
2.2. Field experiments
2.2.1. Field trial design
Two spring-sown, two row (Hordeum distichum), malt-
ing barley cultivars, Astoria (Secobra, France) and Wikingett
(Sval¨
of-Weibulls, Sweden) were established in 16 fertilisation
trials during the years 2001–2003, in a long-day environment
ranging from Lat. 58◦19to 59◦44N in southern Sweden
(Table 1).
The trials were of randomised split-plot design, assigning
the cultivars to the main plots and the fertiliser treatments to
sub-plots with the net size of 2.4 m ×10.0 m. Each combina-
tion of fertilisation and cultivar appeared twice in each trial. The
fertiliser treatments were: No fertilisation; single dose NPKS
applications banded 40 mm below between every second seed
row (combi-drill; Huhtapalo, 1982) at amounts of 70, 100 and
130 kg N ha−1; and split doses: first application NPKS at sow-
ing using combi-drill (70 and 100 kg N ha−1, respectively), and
second at BBCH 32 as calcium nitrate (30 or 60 kg N ha−1,
respectively). The treatments were performed with commer-
cially available products (Table 2), and the aim was to hold the
ratio between all applied macronutrients constant, but still use a
low number of fertilisers.
All trials were protected against fungal attack with a mix-
ture of 0.6 litre (l) ha−1Amistar (Azoxystrobin, 229 g l−1) and
0.5lha
−1Forbel (Fenpropimorph, 750 g l−1) at BBCH 37–39.
2.2.2. Measurements
Soil samples were taken at 0–0.3 and 0.3–0.6 m depth from
the trial sites before sowing and were analysed for NO3–N and
NH4–N, summed for the layer 0–0.6 m (Nmin,kgNha
−1). Plant
macro nutrients were analysed using Swedish standard soil anal-
Table 1
Locations, sowing and harvest days (day number from January 1) and soil properties of the 16 studied sites
Latitude
North
Longitude
East
Day number mg 100 g−1DM soil % DM soil kg N ha−1
Sowing Harvesting pH P–AL K–AL Mg–AL Ca–AL K–HCl Clay Silt Sand SOM NO3aNH4a
2001
158
◦2112◦42135 261 7.0 7.2 11.5 13.8 237 135 24.1 34.9 41.0 3.5 16.2 15.9
258
◦2013◦12133 249 6.8 1.1 11.0 39.7 323 240 49.5 35.3 15.2 3.5 113.9 70.1
359
◦4417◦00134 249 6.4 3.3 14.0 45.1 313 450 53.7 40.5 5.8 2.9 37.3 3.8
459
◦3116◦10137 249 6.4 1.4 14.5 65.2 290 400 47.7 46.0 6.4 4.9 23.3 14.1
558
◦2915◦30135 246 6.9 4.6 18.3 6.0 501 NA 35.0 45.0 20.0 3.5 27.7 10.6
659
◦0615◦05131 243 6.1 3.0 13.0 17.7 151 165 33.7 16.8 16.8 2.9 9.7 14.9
2002
759
◦4417◦00114 231 6.2 3.1 15.0 55.6 437 445 67.7 28.9 3.4 4.1 40.5 8.4
858
◦2815◦2995 222 7.4 12.2 13.5 11.7 1570 330 39.4 31.4 29.1 3.3 22.3 18.2
958
◦2012◦23100 223 6.2 4.5 8.5 12.2 232 145 28.1 44.7 27.1 4.3 7.2 11.8
10 58◦3113◦3492 219 6.5 3.2 9.5 14.2 239 230 22.6 63.1 14.4 3.2 17.5 10.1
11 59◦0915◦16115 221 6.1 9.0 17.0 4.4 133 80 8.4 32.2 59.4 3.1 NA NA
12 59◦3616◦38129 239 6.4 4.5 13.0 75.2 355 255 58.1 39.5 2.4 4.1 91.4 45.7
2003
13 59◦4416◦59140 241 6.5 2.7 13.5 91.9 447 330 74.9 16.3 9.2 4.7 32.3 7.7
14 58◦2715◦30131 241 7.0 8.8 14.5 15.9 528 235 49.3 10.5 40.2 4.3 31.2 9.6
15 58◦1912◦34107 230 7.0 6.6 16.5 34.6 284 330 42.6 30.1 27.3 2.4 10.8 5.2
16 59◦3616◦39142 250 6.4 6.2 26.5 50.0 355 556 56.5 33.7 9.8 3.9 23.0 10.1
aNmin 0–0.6 m.

208 C.G. Pettersson, H. Eckersten / Europ. J. Agronomy 27 (2007) 205–214
Table 2
Fertiliser treatments, using commercially available products
Fertiliser product at BBCH 00 N at BBCH 00
(kg ha−1)
N at BBCH 32
(kg ha−1)
Total N
(kg ha−1)
Total P
(kg ha−1)
Total K
(kg ha−1)
Total S
(kg ha−1)
A No fertilisation 0 0 0 0 0 0
B NPKS 21-4-7-2 70 0 70 12.7 24.3 7.3
C NPKS 21-4-7-2 100 0 100 18.1 34.7 10.5
D NPKS 21-4-7-2 130 0 130 23.5 45.2 13.6
E NPKS 20-6-8-2 70 30 100 21.0 28.0 10.0
F NPKS 20-6-8-2 70 60 130 21.0 28.0 13.0
G NPKS 20-6-8-2 100 30 130 30.0 40.0 13.0
H NPKS 20-6-8-2 100 60 160 30.0 40.0 16.0
The application at BBCH 00 (sowing) was banded 40 mm below between every second seed row, using products with an NH4:NO3ratio of 1:1. The applications at
BBCH 32 were top-dressed using sulphurised calcium nitrate (KsS, Yara Norway). The total N:S ratio was near 10:1 for all treatments.
yses, involving the extraction of plant-available nutrients with an
ammonium lactate solution (AL), plant unavailable soil reserves
with hydrochloric acid (HCl) solution, and soil pH in distilled
water (Anon., 1993). The plant nutrients were analysed from
soil samples 0–0.3 m (Table 1).
Canopy reflectance was recorded in each plot at BBCH
32, with a hand-held passive sensor equipped with two ETA-
G 380/1050 (256 pixel) spectrometers (Steag ETA-Optik,
Heinsberg, Germany). One spectrometer was pointed vertically
upwards, and one sensor located 1.5 m above the canopy pointed
forwards the crop at a 60◦angle from the vertical line. The reflec-
tion spectrum from 400 to 1000 nanometres (nm) was recorded
at intervals of 10 nm, thus covering both the photosynthetically
active and the near-infrared (NIR) parts of the spectrum. Mea-
surements were made from each corner of the plots, and mean
values were used in the analyses. The whole dataset was used for
most analyses, but due to abundance of weeds (Euphorbia helio-
scopia) in trial 10 and 12 (Table 1), 14 out of 16 trials were used
for analyses using canopy reflection data. Air temperatures were
recorded at 1.8 m height at individual meteorological stations
(Swedish national meteorological network stations (SMHI) or
Swedish agricultural weather network (LantMet, 2006)). All
weather stations were situated within 30 km distance from the
field trial, and within the same agricultural landscape. The land-
scape is flat and small topographic effects, or effects from water
bodies, on temperature variability were expected on daily max-
imum temperature.
Canopy (leaf and straw) samples of 0.25 m2were cut at
BBCH 77 (late milk) within the borders but outside the net
harvesting area of the sub-plots. At this stage, the canopy
was expected to contain its highest amount of N (McTaggart
and Smith, 1995). The samples were dried at 105 ◦C for 24 h,
weighed for DM content, and milled. The N-content was mea-
sured with a LECO CNS-2000 (LECO Corporation, MI, USA).
The trials were machine harvested at BBCH 89 (Fully ripe:
grain hard, difficult to divide with thumbnail); using net har-
vest plots of 24 m2. Grain samples of 1000 g were placed in
cotton bags, dried at a maximum of 38 ◦C, and weighed to allow
calculation of moisture content at harvest. The dried grain sam-
ples were cleaned, but not screened, before analysing grain N
concentration and volume weight using a FOSS Infratec 1241
Grain Analyzer (FOSS, Denmark). The results were expressed
as crude protein concentration assumed to be 6.25 times the N
concentration.
2.2.3. Statistics
Data were analysed using the statistical software R-2.4.0. (R
Development Core Team, 2006). The R function for linear mod-
els (lm) was used for regressions and the R function for linear
mixed-effects models (lme; Pinheiro et al., 2006) for analysis
of variance. There were 4 replicates in each trial and 16 trials
(site-years) in the series. To estimate treatment means, an anal-
ysis of variance with a fixed and a random part was performed.
In the fixed part the observations for each plot were assigned
to fertiliser treatment or cultivar. In the random part the replicat
were nested in field-trials. Linear regressions for grain CP were
performed using environmental measurements as independent
variables in multiple linear regressions. R2statistics were used
for goodness of fit. The result from such an analysis could be
presented as Eq. (3), where ‘a’ is the intercept, and ‘b’ and ‘c’
are the change in grain CP for one unit change in variable 1
(Var1) and variable 2 (Var2), respectively.
Grain CP =a+b×Var1 +c×Var2 (3)
3. Results
The variation in the weather among years and sites resulted
in sowing dates ranging from day number (from January 1) 92
(April 2) to 142 (May 22), and growing periods (sowing to har-
vest) from 101 to 127 days long. Late sowing was associated
with a shorter total growing period, with a loss of 0.33 growing
days (R2= 0.78) for each day of delay.
3.1. Soil
Grain CP and grain yield were not strongly correlated with
soil chemical and physical variables (Table 3). Grain CP, but
not grain yield, was positively correlated to SOM, soil mineral
N at sowing in soil layer 0–0.6 m (Nmin) did not improve the
correlation. Grain CP was positively correlated with clay content
and clay-related nutrients (K–HCl and Mg–AL), and grain yield
was negatively correlated with the same soil variables. As the
ability of soil properties to explain the variations in grain CP was

C.G. Pettersson, H. Eckersten / Europ. J. Agronomy 27 (2007) 205–214 209
Table 3
Probability levels of linear regressions with grain yield and grain CP as depen-
dent variables and recorded soil properties as independent
Variable P-levels
Grain yield Grain CP
pH ns ns
P–AL ns (*)
K–AL – ns
Mg–AL ** (*)
K–HCl *** ns
Cu–HCl ** *
Clay ** *
Silt ns ns
Sand *(*)
SOM ns (*)
Nmin ns ns
P-levels: ns >0.1, (*) <0.1, *<0.05, **<0.01, ***<0.001.
low even if there were significant correlations at the 5% level, no
further analyses of soil impact on nitrogen relations were done.
3.2. Fertilisation
Grain CP and grain yield increased with the total amount of
applied N and the effects were neutral to the strategy of appli-
cation (Table 4). However, in treatments applying 60 kg N ha−1
at BBCH 32 (treatments F and H), grain CP and number of ears
m−2was significantly higher, and the proportion of large grains
(>2.5 mm) was significantly lower than for the corresponding
single dose application. These effects were not found in the low
rate treatments (30 kg N ha−1applied at BBCH 32; E and G).
Grain CP was linearly related to total applied N and, on average,
the cultivar Wikingett had a grain CP 0.8% higher than Asto-
ria. Increasing fertilisation by 1 kg Nha−1resulted in a 0.028%
increase of grain CP for both cultivars.
The choice of cultivar as well as the amount of applied N
affected grain CP, but the high significances in Table 4 rests on
the use of site-year (trial) in the random part in the analysis of
variance. The spread of the observations (Fig. 1) indicate that
other factors than those varied in the field trials dominated as
causes of variation in grain CP.
3.3. Day number at sowing
Day of sowing was related (R2
adj =0.63) to grain CP (Fig. 2),
whereas grain yield was not (data not shown). Grain CP
increased by 0.07% for each day of sowing delay within the
range of day numbers, 92–142.
The mean grain CP differed significantly between the two
cultivars, but the reaction to sowing day number was the same.
The intercepts for all significant regressions with grain CP as
dependent variable were cultivar specific, whereas the slopes
were the same for both cultivars. For this reason, the remaining
regressions are presented for Astoria only, and using treatment
letters (Table 4) as plotting symbols.
Table 4
Mean grain yield and grain quality characteristics for fertiliser treatments and cultivars
N strategy
(kgNha
−1)
Grain yield
(kg DM ha−1)
Grain protein
(% DM)
Grain N-yield
(kg ha−1)
Canopy N uptake at
BBCH 77 (kg ha−1)
Stand density at
BBCH 87 (year m−2)
Standing power at
BBCH 89 (%)
Grains
>2.5 mm (%)
Grain moisture
at harvest (%)
Grain volume
weight (g dm−3)
Grain weight
(g 1000−1)
A 0 2092 10.0 33.7 32.9 408 99 88.9 20.4 660 44.2
B 70 4079 10.1 65.8 71.8 638 94 90.6 18.2 671 47.0
C 100 4599 10.7 78.1 90.0 706 90 89.5 18.3 676 47.1
D 130 4960 11.4 89.3 110.0 737 87 88.0 18.5 680 47.1
E 70 + 30 4708 10.9 81.6 93.6 744 88 86.1 18.6 677 46.3
F 70 + 60 5048 12.0 95.6 111.9 835 81 81.5 19.3 676 45.1
G 100 + 30 5090 11.7 94.2 113.8 802 86 85.7 18.8 680 46.6
H 100 + 60 5225 12.6 103.8 127.3 846 75 81.7 19.2 678 45.6
L.S.D. 5% 247 0.3 4.1 8.7 61 8 0.9 0.8 5 0.9
Astoria 4594 10.8 79.4 93.7 738 89 85.9 18.8 677 46.8
Wikingett 4366 11.6 81.3 94.1 691 87 87.1 19.0 672 45.5
L.S.D. 5% 247 0.3 4.1 8.7 61 8 0.9 0.8 5 0.9
Least significant difference (L.S.D.) for pairwise comparisons were adjusted according to Tukey.

210 C.G. Pettersson, H. Eckersten / Europ. J. Agronomy 27 (2007) 205–214
Fig. 1. The relationship between malting barley grain CP at harvest and total
applied N. The dataset from the present study was used, and the N levels were
jittered to make the figure read clearer.The plotting symbols indicate the cultivars
Astoria (A) and Wikingett (W).
3.4. Canopy reflection
For fertilisation treatments B–D (70–130 kg N ha−1at sow-
ing) the vegetation index TCARI from BBCH 32 (TCARI(32))
was used to predict grain CP at harvest. The unfertilised treat-
ment A was not used, as no linear models could be fitted for
grain CP with reflection data at BBCH 32 as, despite dissimi-
lar TCARI(32) scores indicating different crop status, the final
grain CP for treatments A and B were the same (Fig. 1 and
Table 4). A linear regression for grain yield could be fitted from
Fig. 2. The relationship between malting barley grain CP at harvest and day
number at sowing, from January 1. Barley fertilisation was 70–130kgNha−1.
The plotting symbols and regression line types indicate the cultivars Astoria (A;
—) and Wikingett (W; -- -). R2
adj =0.63.
Fig. 3. The relationship between grain CP of Astoria malting barley at harvest,
and the vegetation index TCARI estimated at BBCH 32. Fertilisations treatments
B, C and D (70–130 kg N ha−1) were not explicitly part of the regression model.
R2= 0.41.
TCARI(32), but that was not the main objective. The correla-
tion between TCARI(32) and grain CP was negative (Fig. 3)
with R2= 0.41.
3.5. Temperature sum during grain filling
The temperature sum during grain filling (STS; Eq. (2))was
assessed as a predictor of CP, using seven base temperatures for
TMAX (Tb= 17, 19, 21, 23, 25, 27 and 29 ◦C, respectively). For Tb
between 17 and 23 ◦C, all P-values were < 0.001 but strongest for
Tb=21◦C. Higher scores for STS(21) resulted in higher grain
CP (R2
adj =0.52), with the cultivars showing different reactions
to STS(21); Wikingett being slightly more temperature sensitive
than Astoria. Grain yield and grain weight were not significantly
correlated with STS(21).
Growing degree days (GDD) were accumulated for the same
three-week period as STS, to compare accumulated daily mean
temperatures (GDD = TMEAN −Tb) with accumulated daily
maximum temperatures (Eq. (2)) in regressions for grain CP.
Four different base temperatures for TMEAN were used (Tb=1,5,
9 and 15 ◦C). The two highest Tbvalues (9 and 15 ◦C), produced
GDD that correlated with grain CP but did not reach the same
significance levels as STS(21). Accumulated TMAX values were
more useful than accumulated TMEAN values, as descriptors of
thermal influence on grain filling.
3.6. Combinations of predictors
Combining day number from January 1 at sowing, and canopy
reflection measurement at BBCH 32 (TCARI(32)) improved the
relationship with grain CP considerably compared with using
only one predictor (Fig. 4).
Combining TCARI(32) and the best accumulated sum of
daily maximum temperature, STS(21), also improved the rela-

C.G. Pettersson, H. Eckersten / Europ. J. Agronomy 27 (2007) 205–214 211
Fig. 4. The relationship between grain CP of Astoria malting barley at harvest,
and a prediction using a linear function of day number, from January 1, at sowing
and the vegetation index TCARI estimated at BBCH 32. Fertilisation treatments
B, C and D (70–130 kg N ha−1) were not explicitly part of the regression model.
The line represents the 1:1 relation between predicted and observed grain CP.
R2
adj =0.73.
tionship with grain CP (R2
adj =0.63), but not as much as when
sowing day was used. A final attempt to use all three vari-
ables (day number at sowing, TCARI(32) and STS(21)) for
predictions of grain CP was not successful, as STS(21) was
not significantly related to grain CP when introduced into a
model already using sowing day number and TCARI(32) as
independent variables.
As a planned, sensor controlled, two-step fertilisation of malt-
ing barley would not likely start with an application rate of
130 kg N ha−1at sowing a further regression was performed
using the low rate treatments B and C only (70–100 kg Nha−1).
This regression revealed a slightly better fit (R2
adj =0.78). In
conclusion, the best prediction of Astoria grain CP at harvest
was achieved with the following equation:
Grain CP =9.75 +0.064(Sowing day −120)
−15.4(TCARI(32) −0.150) (4)
Both sowing day and TCARI(32) were centered to their, respec-
tive, observation means. In this way the offset (a) of Eq. (3) was
the expected grain CP with both variables at their observation
means. Eq. (4) can be used at BBCH 32 to predict the grain CP at
harvest if no further fertiliser is applied. A decision concerning
a second application would need this information.
4. Discussion
4.1. Soil
In this study, soil chemical and mechanical properties were
of little value in predicting grain CP. Soil organic matter (SOM)
was positively correlated with grain CP, but at a low probability
level, and introducing SOM in multiple regressions for grain
CP was not successful. This might be due to a limited range
of SOM in the present study (2.4–4.9%; Table 1), compared to
studies reporting strong influence from SOM on plant N uptake
(e g., McTaggart and Smith, 1993). This strengthens the view
that crop development was related more to soil structure than
to chemical properties. In a previous study (Pettersson et al.,
2006) we suggested soil structure as a reason for the role that
soil apparent electrical conductivity (ECa) played for spatially
distributed protein patterns, but did not see any way of using
ECaas a general descriptor of soil structure.
The lack of success in using soil mineral N (Nmin) at 0–0.6 m
deep in grain CP regressions was unexpected. Nmin has been
used as a planning tool for the farmer and in cereal fertilisation
studies since the mid 1970s, and is still being developed (e g.,
Mengel et al., 2006) as a coarse but generally useful tool. As
there was no clear cause of the failure, so that possible erroneous
observations could be removed from the dataset, Nmin was not
used further in the analyses.
4.2. Fertilisation
A two-step fertilisation regime with an adjusting second fer-
tiliser application (Table 2) was neutral to grain CP when the
second application was limited to 30 kg Nha−1. When as much
as 60 kg N ha−1was applied at BBCH 32 (F and H in Table 4),
grain CP and ears m−2were consistently higher and the fraction
of large grains (>2.5 mm) was consistently lower than treatments
with similar total N application rate. This might be due to an
imbalance between available N and available assimilates during
grain filling, following the high fertilisation rate at BBCH 32.
More available N, than what chlorophyll amount and leaf area
could balance with CO2assimilation, would result in higher
grain CP. The same imbalance between N and leaf area might
also cause more small tillers to produce ears, giving smaller
grains than average compared with treatments with the same
total N application.
4.3. Day number at sowing
A long daylength at emergence was expected to lower the
accumulated temperature sum needed for the crop to fulfil each
development phase, resulting in a smaller canopy during grain
filling (Kirby, 1995; Miralles et al., 2003; Paynter et al., 2004).
This was confirmed in the present work where the crop lost 0.33
development days for each day of sowing delay. A late sowing
could also increase the risk of exposing the crop to high tem-
peratures during grain filling (Wardlaw and Wrigley, 1994). The
two phenomena (rapid development and risk of high temperature
exposure due to late sowing) are both known to be associated
with increased CP, providing the explanation of why such a sim-
ple variable as day number at sowing can have a major influence
in grain CP predictions. It might also be the explanation why
the introduction of STS(21), into a model already including
the sowing day number and TCARI(32) as independent vari-
ables, did not improve the predictability. Sowing day number
and STS(21) were not independent and sowing day number did

212 C.G. Pettersson, H. Eckersten / Europ. J. Agronomy 27 (2007) 205–214
already incorporate parts of the information in STS(21). For
instance, STS(21) was never higher than 20d ◦C for sowing day
numbers smaller than 120 (April 30th). When evaluating day
of sowing, it is important to hold latitude of the study in mind.
Sowing dates that are considered normal in the present study
performed between latitudes 58◦19and 59◦44N in Sweden,
would be extremely late appearing, for example, in the study of
Conry (1995) working with a dataset from Ireland.
A 40kgNha
−1increase in N application rate would, using
linear regression, result in a mean increase in grain CP of about
1% for both cultivars, but 15 days delay of sowing would also
result in a similar change.
4.4. Canopy reflectance
The objective of this study was to evaluate methodologies
to predict grain CP at harvest. Fertilisation level influences the
grain CP, and the idea was that canopy reflectance observation
might give information on the influence of fertilisation as well
as interacting weather and soil factors on canopy development
prior to BBCH 32, and that the same model could be applied to
all treatments.
The ratio TCARI/OSAVI (TC/OS), proposed by Haboudane
et al. (2002), was not successful in this investigation. The
ratio was developed in maize, and has been used for spatial
chlorophyll regressions in various crops including winter wheat
(Huang et al., 2005) vineyards (Zarco-Tejada et al., 2005), and
for grain CP in malting barley (Pettersson et al., 2006). Since
OSAVI(32) gave no satisfactory correlations with grain CP in
the present study, whereas TCARI(32) did, it was not surpris-
ing that TC/OS(32) gave no better regressions than TCARI(32)
itself. The problem, that very low TCARI scores can correspond
to both high and low chlorophyll concentrations, was described
for TCARI < 0.10 (Haboudane et al., 2002). As this range of
the VI was not apparent in the data from barley fertilised with
70–130 kg N ha−1(Fig. 3), the problem would not be necessary
to address in this context.
4.5. Temperature sum during grain filling
High temperatures are known to shorten the grain filling dura-
tion and speed up the grain-filling rate. As these two reactions
are opposed, in terms of the amount of assimilate allocated to
the grain, an optimal temperature can be expected. Tashiro and
Wardlaw (1989) found that grain weights of wheat were reduced
at daily mean temperatures (TMEAN) above 18 ◦C, while grain
weights of rice were not reduced until TMEAN exceeded 27 ◦C.
As the loading of N into the grains is not as sensitive to high
temperatures as the loading of C, high temperatures during grain
filling tend to result in higher protein concentrations (Grashoff
and d’Antuono, 1997; Boonchoo et al., 1998). In line with this,
daily maximum temperatures (TMAX) in the range 20–30 ◦C
have been found to limit grain C filling in both wheat (Tribo¨
ı and
Tribo¨
ı-Blondel, 2002) and in malting barley (Passarella et al.,
2002), resulting in increased protein concentration. Using the
definition from Wardlaw and Wrigley (1994),TMAX >32◦C,
only one out of 16 trials (no. 5, Table 1) was exposed to heat
shock temperatures during grain filling, the remaining fifteen
trials recorded TMAX temperatures in the range 20–31 ◦C.
In the present study, accumulated TMAX >21◦C during a
three-week period covering grain filling (STS(21)) was a suc-
cessful independent regression variable for grain CP, in line with
recent studies in malting barley (Wang et al., 2001; Pettersson
et al., 2006). Grain weight, however, did not correlate well with
STS(21). One possible explanation is that grain weight, to a
large extent, depends on canopy development before grain fill-
ing. A good relationship with grain weight can, however, occur in
experiments such as those of Passarella et al. (2002) where tem-
peratures were manipulated within the site. Compared canopies
then have similar leaf area and potential grain weight at the start
of grain filling, and effects of thermal stress during grain filling
can be explored, which otherwise would be statistically invisible
in field studies under natural conditions.
4.6. Combinations of predictors
To use day number at sowing to predict grain CP is of prac-
tical advantage as it, apart from Nmin and possibly SOM, is the
only available information at sowing for adjusting the fertilisa-
tion rate. The predictability using sowing day (Fig. 2) was fairly
high (R2
adj =0.63) and may be used to adjust the fertilisation
at sowing. But the predictability was considerably improved by
extending the model using the TCARI(32) (R2
adj =0.73; Fig. 4).
Where a second, adjusting, fertilisation is planned it is more
realistic to assume that the fertilisation at sowing is limited to
below 100 kg N ha−1. In this case the remote sensing data at
BBCH 32 contributed even more to the predictions of grain
CP at harvest (R2
adj =0.78; Eq. (4)). The type of effect day
of sowing has on predicted grain CP as expressed in Eq. (4) is
expected to be general, but the exact use is only valid within
the study. Expanding the use of sowing day south of Lat. 58◦
N would call for field trial series with a broader north-south
range, and an evaluation that incorporates the latitude in the
model.
4.7. Precision agriculture implications
Systems for sensor controlled variable rate fertilisation of
small grains have, so far, varied the fertiliser application accord-
ing to expected correlations between a vegetation index (VI)
and predicted crop parameters. Some systems use the sensors to
control the relative variation in application rate inside the field,
from a mean field fertilisation rate estimated with other meth-
ods (Link et al., 2002), whereas other systems make use of the
absolute values of a VI to control the fertilisation rates (Schwab
et al., 2005). The proposed model for VI of our study (Eq. (4))
would be applicable to both systems to estimate the grain CP at
harvest if no further fertilisation is made, and would be useful
information for decisions on a second fertilisation rate. The pre-
dictions depend on the date of sowing and in this way represent
a new thought in precision agriculture. The crop water supply
was imbedded in our predictions but might be valuable to express
explicit to further improve specific grain CP predictions.

C.G. Pettersson, H. Eckersten / Europ. J. Agronomy 27 (2007) 205–214 213
5. Conclusions
It was possible, for specific cultivars, to make predictions of
malting barley grain CP at harvest, from day number at sowing
and the vegetation index TCARI sampled at BBCH 32. That
means that all needed data was available at a time when an
adjusting fertilisation was still possible. Some of the sowing
day effect, but not all, might be due to thermal stress during
grain filling (STS(21)), as the risk for high STS(21) observations
was higher when sowing was late. It might also explain why
the introduction of STS(21) into a model already including the
sowing day number and TCARI(32) as independent variables,
did not improve the predictability of grain CP.
Acknowledgements
We thank Anders Anderson and Stefan Reusch, Yara, who
helped us with the N-sensor datasets and solved many problems
during the project, and Christian Ritz, University of Copen-
hagen, who advised the statistical analyses. We also thank
Prof. Robert Hay, SLU, who helped us incorporate crop phys-
iology thinking into the project and improved the language
of the manuscript. The project has been possible due to eco-
nomic support from Stiftelsen Svensk V¨
axtn¨
aringsforskning and
Lantm¨
annen.
References
Anon., 1993. SS 028310:1993. Soil analysis—Extraction and determination of
phosphorus, potassium, calcium, magnesium and sodium from soils with
ammonium lactate/acetic acid solution. Svensk Standard. SIS F¨
orlag AB,
Stockholm, p. 7 (in Swedish).
Birch, C.J., Fukai, S., Broad, I.J., 1997. Estimation of responses of yield and
grain protein concentration of malting barley to nitrogen fertiliser using plant
nitrogen uptake. Aust. J. Agric. Res. 48, 635–648.
Boonchoo, S., Fukai, S., Hetherington, S.E., 1998. Barley yield and grain protein
concentration as affected by assimilate and nitrogen availability. Aust. J.
Agric. Res. 49, 695–706.
Conry, M.J., 1995. Comparison of early, normal and late sowing at three rates
of nitrogen on the yield, grain nitrogen and screenings content of Blenheim
spring malting barley in Ireland. J. Agricult. Sci. 125, 183–188.
Conry, M.J., 1997. Effect of fertiliser N on the grain yield and quality of spring
malting barley grown on five contrasting soils in Ireland. Proc. R. Irish Acad.
97, 185–196.
Ehsani, R., Sullivan, M., 2002. Soil Electrical Conductivity (EC) Sensors. Ohio
state university. Extension factsheet. Food, Agricultural and Biological Engi-
neering, AEX(565-02), p. 3.
Grashoff, C., d’Antuono, L.F., 1997. Effect of shading and nitrogen application
on yield, grain size distribution and concentrations of nitrogen and water
soluble carbohydrates in malting spring barley (Hordeum vulgare L.). Eur.
J. Agron. 6, 275–293.
Haboudane, D., Miller, J.R., Tremblay, N., Zarco-Tejada, P.J., Dextraze, L.,
2002. Integrated narrow-band vegetation indices for prediction of crop
chlorophyll content for application to precision agriculture. Rem. Sens.
Environ. 81, 416–426.
Huang, W., Huang, M., Liu, L.Y., Wang, J., Zhao, C., Wang, J., 2005. Inversion
of the severity of winter wheat yellow rust using proper spectral index.Trans.
Chin. Soc. Agr. Eng. 21, 97–103.
Huhtapalo, A., 1982. Scandinavian principles for fertilizer placement, utiliza-
tion of fertilizer-N nitrogen. In: Proceedings of the 9th Conference of the
International Soil Tillage Research Organization,ISTRO, Osijek Yugoslavia.
ISTRO, Wageningen, pp. 669–674.
Jabro, J.D., Evans, R.G., Kim, Y., Stevens, B.W., Iversen, W., 2005. ASABE
technical library; www.asabe.org. American Society of Agricultural and
Biological Engineers, St. Joseph, MI.
Juskiw,P.E., Helm, J.H., 2003. Barley response to seeding date in central Alberta.
Can. J. Plant Sci. 83, 275–281.
Kirby,E.J.M., 1995. Factors affecting rate of leaf emergence in barley and wheat.
Crop Sci. 35, 11–19.
Kirby, E.J.M., Appleyard, M., Fellowes, G., 1982. Effect of sowing date on the
temperature response of leaf emergence and leaf size in barley. Plant Cell
Environ. 5, 477–484.
Kirby, E.J.M., Appleyard, M., Fellowes, G., 1985. Variation in development of
wheat and barley in response to sowing date and variety. J. Agricult. Sci.
(Cambridge) 104, 383–396.
Lancashire, P.D., Bleiholder, H., van den Boom, T., Langeluddeke, P., Stauss,
R., 1991. A uniform decimal code for growth stages of crops and weeds.
Ann. Appl. Biol. 119, 561–601.
LantMet., 2006. The Swedish agricultural weather station network. URL:
http://www.dacom.nl/lantmet new/ (accessed 2006-01-15).
Lark, R.M., Wheeler, H.C., 2003. Experimental and analythical methods for
studying within-field variation of crop responses to inputs. In: Stafford,
J., Werner, A. (Eds.), Precision Agriculture. Proceedings of the Fourth
European Conference on Precision Agriculture. WageningenAcademic Pub-
lishers, Wageningen, pp. 341–346.
Link, A., Panitzki, M., Reusch, S., 2002. Hydro N-Sensor: tractor-mounted
remote sensing for variable nitrogen fertilization. In: Robert, P.C. (Ed.), Pro-
ceedings of Sixth International Conference on Precision Agriculture (CD).
ASA/CSSA/SSSA, Madison, USA, pp. 1012–1018.
McKenzie, R.H., Middleton, A.B., Bremer, E., 2005. Fertilization, seeding date,
and seeding rate for malting barley yield and quality in southern Alberta.
Can. J. Plant Sci. 85, 603–614.
McTaggart, I.P., Smith, K.A., 1992. The effect of fertiliser and soil nitrogen on
the overall uptake of nitrogen in the plant, and the grain nitrogen content of
spring-sown malting barley. HGCA Project Report 46, pp. 126.
McTaggart, I.P., Smith, K.A., 1993. Estimation of potentially mineralisable
nitrogen in soil by KCl extraction. II: Comparison with soil N uptake in
the field. Plant Soil 157, 175–184.
McTaggart, I.P., Smith, K.A., 1995. The effect of rate, form and timing of fer-
tilizer N on nitrogen uptake and grain N content in spring malting barley. J.
Agric. Sci. 125, 341–353.
Mengel, K., H¨
utsch, B., Kane, Y., 2006. Nitrogen fertilizer application rates on
cereal crops according to available mineral and organic soil nitrogen. Eur. J.
Agron. 24, 343–348.
Miralles, D.J., Slafer, G.A., Richards, R.A., Rawson, H.M., 2003. Quantitative
development response to the length of exposure to long photoperiod in wheat
and barley. J. Agricult. Sci. 141, 159–167.
Palmer, G.H., 2000. Malt performance is more related to inhomogeneity of pro-
tein and b-glucan breakdown than to standard malt analyses. J. Inst. Brewing
106, 189–192.
Passarella, V.S., Savin, R., Slafer, G.A., 2002. Grain weight and malting quality
in barley as affected by brief periods of increased spike temperature under
field conditions. Aust. J. Agric. Res. 53, 1219–1227.
Paynter, B.H., Juskiw, P.E., Helm, J.H., 2004. Leaf development in two-row
spring barley under long-day and short-day field conditions. Can. J. Plant
Sci. 84, 477–486.
Pettersson, C.G., S¨
oderstr¨
om, M., Eckersten, H., 2006. Canopy reflectance,
thermal stress, and apparent soil electrical conductivity as predictors for
within-field variability in grain yield and grain protein of malting barley.
Prec. Agric. 7, 343–359.
Pinheiro, J., Bates, D., DebRoy, S., Sarkar, D., 2006. nlme: linear and nonlinear
mixed effects models. R package version 3, pp. 1–77.
Porter, J.R., Jamieson, P.D., Wilson, D.R., 1993. Comparison of the wheat simu-
lation models AFRCWHEAT2, Ceres-Wheat and SWHEAT for non-limiting
conditions of crop growth. Field Crops Res. 33, 131–157.
R Development Core Team, 2006. R: A Language and Environment for Statis-
tical Computing. R Foundation for Statistical Computing, Vienna, Austria,
ISBN 3-900051-07-0, URL: http://www.R-project.org.
Schwab, G.J., Pena-Yewtukhiw, E.M., Wendroth, O., Murdock, L.W.,
Stombaugh, T., 2005. Wheat and yield population response to variable rate

214 C.G. Pettersson, H. Eckersten / Europ. J. Agronomy 27 (2007) 205–214
N fertilization strategies using active NDVI sensors. In: Stafford, J. (Ed.),
Precision Agriculture. Proceedings of the 5th European Conference on Pre-
cision Agriculture. Uppsala, Sweden. Wageningen Academic Publishers,
Wageningen, The Netherlands, pp. 235–242.
Tashiro, T., Wardlaw, I.F., 1989. A comparison of the effect of high temperature
on grain development in wheat and rice. Ann. Bot. (Lond.) 64, 59–65.
Thiam, A., Eastman, J.R., 2001. Vegetation indices. In: Eastman, J.R. (Ed.),
idrisi32 Release 2. Clark Labs, Worcester, MA, pp. 89–101 (Chapter 4).
Thyl´
en, L., Algerbo, P.A., Pettersson, C.G., 1999. Grain quality variations within
fields of malting barley. In: Stafford, J.V. (Ed.), Precision Agriculture 1999.
Proceedings of the Second European Conference on Precision Agriculture.
Sheffield Academic Press Ltd., Sheffield, UK, pp. 287–296.
Tribo¨
ı, E., Tribo¨
ı-Blondel, A.-M., 2002. Productivity and grain or seed compo-
sition: a new approach to an old problem—invited paper. Eur. J. Agron. 16,
163–186.
Wang, J.M., Zhang, G.P., Chen, J.X., 2001. Cultivar and environmental effects
on protein content and grain weight of malting barley. J. Zhej. Uni. (Agr.
Life Sci.) 27, 503–507.
Wardlaw, I.F., Wrigley, C.W., 1994. Heat tolerance in temperate cereals: an
overview. Aust. J. Plant Physiol. 21, 695–703.
Zarco-Tejada, P.J., Berj´
on, A., L´
opez-Lozano, R., Miller, J.R., Martin, P.,
Cachorro, V., Gonz ´
alez, M.R., de Frutos, A., 2005. Assessing vineyard con-
dition with hyperspectral indices: leaf canopy reflectance simulation in a
row-structured discontinous canopy. Rem. Sens. Environ. 99, 271–287.