Вы находитесь на странице: 1из 7

Journal of Stored Products Research xxx (2010) 1e7

Contents lists available at ScienceDirect

Journal of Stored Products Research


journal homepage: www.elsevier.com/locate/jspr

Development and validation of on-farm sampling methods for the collection


of marketing (quality) samples at harvest
J.D. Knight*, D.R. Wilkin
Centre for Environmental Policy, Imperial College London, Silwood Park Campus, Ascot, Berkshire SL5 7PY, UK

a r t i c l e i n f o

a b s t r a c t

Article history:
Accepted 20 May 2010

This paper examines the accuracy of different grain sampling methods by sampling on farms at harvest
during 2002 and 2003. A number of different methods of sampling are compared along with the
consequences of using composite samples. Results showed that there were no statistically signicant
differences between the sampling methods used for any of the parameters measured (moisture content,
dry matter, protein, nitrogen, hardness, specic weight, and nes) except for specic weight where use of
a grain spear resulted in a signicantly higher reading than other methods. Signicant variation occurred
between farms and elds but generally not within elds, i.e. trailers sampled coming from the same eld.
Using these results the effects of sampling intensity are discussed and proposals made on the number of
samples required to obtain a reliable estimate of the quality, whilst at the same time ensuring that the
sampling protocol is manageable within current farming systems.
2010 Elsevier Ltd. All rights reserved.

Keywords:
Sampling
Grain
Moisture content
Protein
Specic weight
Hardness
Wheat
Barley

1. Introduction
The cereals and oilseeds industry is an important part of the UK
rural and business economy. Production of grain e primarily wheat,
barley, and oats e and oilseed rape is in the order of 25 million
tonnes per annum and the grain-based food chain involves
approximately 70,000 growers, 650 traders, 350 processors and an
extensive retail/consumer base.
Grain is harvested in the UK during JulyeSeptember with most
of the crop being stored on-farm or in co-operative stores to ensure
a supply throughout the year. It is vital that growers, buyers,
processors and the consumers have immediate and accurate
information about the characteristics of the grain crops both at
harvest and whilst in store. This information is not only required to
ensure that the grain is stored safely to prevent spoilage through
the development of moulds and pests, but accurate information
is also needed about the quality of the wheat or barley so that it is
delivered for the right end-use, to the right location and at the
correct time. Failure to do so can have serious cost implications
since contaminated or out-of-specication grain will be rejected or,
at the very least, attract a penalty. However, it is not just the cost of
having poor information, but the cost of repeated sampling that is
an issue. The Cereal Liaison Group, convened by the Home Grown
Cereals Authority, has estimated the cost to industry of poor or

* Corresponding author. Tel.: 44 (0) 207 594 2496.


E-mail address: j.d.knight@imperial.ac.uk (J.D. Knight).

repeated sampling as about 2.5 million pounds per year (Hook,


2004), broken down as in Table 1.
Given that knowledge of the condition and quality of the grain is
fundamental to many of the decisions that have to be made relating
to storage and marketing, it is perhaps surprising that there is so
little information on the variability of the grain itself and the
implications of this for the design and implementation of sampling
methods. Work in precision agriculture has shown that there is
spatial variation in the quality of barley in a eld (Jensen et al.,
2001) but the importance of this has not really been investigated
with regard to sampling. Previous practical investigations of
sampling grain have suggested that the methods and equipment
used may inuence the sample that is collected (Wilkin, 1991, 1993)
whilst another study collected data concerning sampling equipment and methods currently in use on farms and at commercial
stores (Wilkin and Knight, 1995) which showed that there was little
standardisation in these methods.
Grain sampling is done for various reasons and occurs at
a number of different points along the supply chain. Sampling starts
with the farmer who is primarily interested in the moisture content
of the grain at harvest to determine whether or not it requires
drying to ensure safe storage. Typically, small samples of grain have
been taken from trailers bringing grain from the combine as they
enter the store. In many cases only a few samples were taken
throughout the harvest day. Very rarely were these samples tested
for quality parameters other than moisture content, or retained.
If grain is dried, it will be sampled during, or subsequent to,
drying to ensure that it has achieved the desired moisture content.

0022-474X/$ e see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jspr.2010.05.004

Please cite this article in press as: Knight, J.D., Wilkin, D.R., Development and validation of on-farm sampling methods for the collection of
marketing..., Journal of Stored Products Research (2010), doi:10.1016/j.jspr.2010.05.004

J.D. Knight, D.R. Wilkin / Journal of Stored Products Research xxx (2010) 1e7

Table 1
Key user benets of improved sampling methodologies in the grain industry (from
Hook, 2004).
Benets

Beneciaries

Quantiable
amounts

Less duplication of sampling


and analysis
Less rejection costs and less
haulage
Administration costs reduced

Growers, traders
and processors
Industry and
environment
Food chain as
a whole

1.5 million/year
520 k/year
530 k/year

A number of farmers will continue to sample their grain during


storage to ensure that it remains within safe limits of moisture and
temperature to prevent loss of quality.
The second reason for sampling is to assess the quality of the
grain and shortly after harvest grain merchants visit farms and take
a small number of spear samples of the stored grain. Once the grain
is in store the ability to obtain a representative sample is generally
much reduced since it is difcult to sample much deeper than
about 2 m and many stores contain grain bulks much deeper than
this. It is not uncommon for 5 or more merchants to take and
analyse samples from the same bulk. The farmer is not charged
directly for this service and the indicative results are usually given
to the farmer. Grain is then sold on the basis of these indicative
results when the farmer wishes to sell. As a number of samples are
taken and analysed the farmer will be presented with a range of
test results and has to make a decision on which set of results to
base his marketing. Often those who sell to the most optimistic
values are disappointed when the grain is tested at the point of
delivery and nd it either rejected or subject to claims (discounted).
When the grain is sold it is usually loaded into lorries using
bucket loaders (typically w1.5 to 2 tonnes per bucket load) and
then taken to the point of delivery. A few farmers will take loading
samples from each bucket, but this is time consuming, potentially
dangerous and such samples are not accepted by buyers as representative of the load. On arrival at the processor (miller, maltster,
feed compounder) or export terminal the grain is sampled again
whilst in the lorry and tested before authorisation is given to tip the
load. Merchants and co-operatives taking grain into store will
also sample the lorry to determine where to store the grain. The
lorry sampling is generally performed using manual or vacuum
spear samplers. This is one of the very few points at which there is
an international standard (ISO 13690, 1999) for sampling grain in
bulk and this includes a section on lorries. It is included in most
contracts but seldom rigorously observed.
With the exception of the nal lorry sampling procedure there
is no widely accepted protocol for sampling whilst the grain is on
farm and as a result the samples that are taken are often not
representative and can lead to poor decision-making since the
characteristics of the grain are poorly understood. Obtaining
a representative sample is essential in getting an accurate determination of the quality parameters of the grain and the minimisation of errors, which can arise at a variety of points
throughout the entire sampling and analysis process, is a key part
of this. Since grain is heterogeneous the need to correctly extract
material from a lot, i.e. a sample, is essential. The reliability or
representativeness of a sample can be determined by applying the
Theory of Sampling as presented by Petersen et al. (2005) which
denes the total error associated with the taking and analysis of
any sample as the Global Estimation Error (Gy, 2004) and is
composed of Total Sampling Error (TSE) and Total Analytical
Error (TAE). In general the TAE is relatively minor, often 100x
smaller than the TSE (Petersen et al., 2005), and is not addressed
in this paper. The TSE is in turn made up of the errors associated

with the collection of the sample from the lot and the material
properties (heterogeneity).
There is a need to develop a sampling methodology that is easy
and cheap to do but at the same time provides a truly representative
sample. These two factors are often in conict. For example, the use
of an easily taken and cheap grab sample is obviously awed since
it will only ever select from the top level of material and therefore not
every fragment of the lot has an equal probability of being sampled.
The sampling method employed therefore has to be able to access
any part of the lot (with equal ease and likelihood) to ensure an equal
probability of selecting any fragment of that lot. Grain can be sampled
at a number of points as it comes into storage and some are better
than others in terms of complying with the Theory of Sampling.
Flows of grain can be sampled relatively easily to cover the two
transverse dimensions i.e. both the width and depth of the ow, with
repeat increments taken over time. In contrast, three-dimensional
lots, such as heaps of grain, cannot easily be fully sampled across any
of the dimensions (width, length or depth) using conventional
methods such as spears. The heterogeneity of grain consists of two
elements, rstly the fact that individual grains differ from each other
in their properties (more or less protein, nitrogen etc.) known as
Constitution Heterogeneity and secondly, Distribution Heterogeneity, which is the physical, spatial distribution of the grains within
the lot. For example, grains from a particular part of a eld may have
a higher moisture content than those from another; this error can be
minimised by taking multiple increments to make up the sample and
ensuring thorough mixing of the primary sample before any further,
secondary sampling is undertaken.
This paper describes research into sampling naturally occurring
variable grain lots (wheat or barley), the selection of sampling
equipment and the implications for the design and implementation
of a reliable sampling scheme.

2. Materials and methods


2.1. Materials and rationale
All sampling was conducted on farm under true working
conditions so as to ensure that the performance of the sampling
equipment recorded during this work truly reected the accuracy
obtainable by farmers and merchants. The objective was to determine whether there was signicant variation in the grain at the
farm, eld or trailer level and whether the sampling method used
produced signicantly different results. The sampling tools tested
consisted of a pelican sampler and a 1 l capacity plastic jug, the
scoop, to sample a ow of grain; or a compartmental grain spear
to sample a heap of grain. Fig. 1 shows the various sampling
equipment and method of use.
The basic sampling unit was a trailer of grain. The trailer size
was a variable both within and between farms but typically was
around 12e16 tonnes. The sampling occurred over two harvests
with both wheat and barley being sampled during 2002 and
2003. In an attempt to include any differences due to geographic
variation, samples were taken from farms from a variety of
locations within England and Scotland. The number of samples
taken from each trailer varied due to the way in which it was
tipped or the sampling method being used. The maximum
number of samples taken from a single trailer was 20 (in order
to determine the within-trailer variation) whilst sampling
designed to assess different sampling methods was often
limited to single samples from individual trailers. Table 2 shows
the breakdown of the sampling methods used and the number
of samples taken.

Please cite this article in press as: Knight, J.D., Wilkin, D.R., Development and validation of on-farm sampling methods for the collection of
marketing..., Journal of Stored Products Research (2010), doi:10.1016/j.jspr.2010.05.004

J.D. Knight, D.R. Wilkin / Journal of Stored Products Research xxx (2010) 1e7

Fig. 1. The three different sampling methods used on the grain. (A) grain spear or trier. (B) Pelican sampler, shown in use across a stream of grain and (C) plastic jug, in use in
a controlled ow of grain. The spear, in the open position, is plunged into the heap of grain down to the full depth, closed, withdrawn and emptied into the sample container. In
effect this withdraws 10 samples over a range of depths into the heap.

Table 2
Table showing type and number of samples taken at each farm.
Location

Date
sampled

Number of
trailers sampled

Pelican sampler
(number of samples
per trailer)

Scoop (number
of samples
per trailer)

1 Barley
1 Wheat
2 Barley

N. Lincs
N. Lincs
N. Lincs

August 02
August 02
August 02

2
9
10

10

2 Wheat
3 Wheat

Kent

August 02

4
6

4 Wheat
5 Wheat
6 Wheat
7 Wheat
8 Wheat
8A Wheat
9 Wheat
10 Wheat
11 Barley
12 Barley
13 Barley
14 Wheat
16 Wheat
17 Barley
18 Barley
19 Wheat
20 Wheat
21 Wheat
22 Barley

N. Lincs
N. Lincs
Yorkshire
Yorkshire
Yorkshire
Perthshire
Perthshire
Perthshire
N. Lincs
N. Lincs
N. Lincs
N. Lincs
N. Lincs
N. Lincs
N. Lincs
Yorks.
Yorks.
Yorks.
N. Lincs

August 02
August 02
August 02
August 02
August 02
Sept 02
Sept 02
Sept 02
July 03
July 03
July 03
August 03
August 03
July 03
July 03
August 03
August 03
August 03
August 03

10
5
5
6
5
6
1
1
10
10
3
10
10
9
6
12
12
12
4

10
1
5 (rst trailer)
1 (from 2 to 9)
1
5 (rst 4 trailers)
1 (from 5 and 6)
5
5
1
1
1
1
20
25
5
5
5
1
1
3
3
1
1
1
5

Sampling
method
Farm no./cereal

Spear (number of
samples per heap)

5 (from 3 trailers)
5 from heap

5 (rst 4 heaps)
1 (from 5 and 6)

1
1

1
1

5
5
5

Please cite this article in press as: Knight, J.D., Wilkin, D.R., Development and validation of on-farm sampling methods for the collection of
marketing..., Journal of Stored Products Research (2010), doi:10.1016/j.jspr.2010.05.004

J.D. Knight, D.R. Wilkin / Journal of Stored Products Research xxx (2010) 1e7

2.2. Assessment of samples


All samples were checked for moisture content and temperature
using a Protimeter GrainMaster i electrical moisture meter immediately on collection.
More detailed assessments of the properties of each sample
were made using a Foss Infratec grain analyser 1241 GA-TWM, and
data from this method were used in the analysis. This machine
measured moisture content, specic weight, protein (in the case of
wheat) or nitrogen (in the case of barley) and made an assessment
of hardness for wheat. As samples had to be transported to the
location of the Infratec machine, delays up to a maximum of 24 h
occurred between the collection and assessment of some samples.
In these cases, samples were sealed in plastic bags and transported
in a cool box to reduce the inuence of temperature and moisture
changes on the properties of the grain.
The level of screenings was also assessed for each sample using
a single sieve with a mesh size of 2.5 mm. This was only
a comparative test and not done using standard slotted sieves due
to constraints of time. The results simply provide an indication of
the variability of the level of screenings or nes that occurs
between samples.
Whilst the intention was to obtain the same number of samples
for each sampling method and at each farm location this was not
always possible due to the sampling taking place on-farm during
harvest. This resulted in a variety of trailer types and tipping
methods being used but had the advantage that the sampling was
done under normal working conditions using the methods that
reected on-farm practice.
Sampling was done intensively when trailers were tted with
a grain hatch with many samples being taken from each trailer to
ascertain the within-trailer variation; samples taken from trailers using
only opening tailgates were sampled less intensively since the speed of
discharge was such that fewer samples could be taken in the time.
These latter samples helped to monitor between-trailer variations.
The experiment also examined the comparability of the quality
measurement resulting from individual samples taken from
a trailer or series of trailers with the quality measurement resulting
from a sub-sample taken from a composite of those individual
samples. Sub-samples from the composites were taken after thorough mixing of the samples had taken place. Multiple samples from
each composite were taken, each sample being discarded after
measurement. The aspects tested are listed below.
(a) Within-trailer variation was measured on 6 occasions using
between 5 and 25 samples per trailer depending on the rate at
which the trailer was emptied.
(b) Between-trailer variation was measured on 11 occasions using
both multiple and single samples from each trailer.
(c) Pelican and spear sampling results were compared on 5 occasions using both multiple and single samples from each trailer.
(d) Pelican and scoop sampling results were compared on 1 occasion,
using multiple samples taken with each sampler from each trailer.
(e) Single and composite sampling results, those derived from
single samples and those obtained from bulked samples, were
compared on 5 occasions.

2.3. Statistical analysis


Residual maximum likelihood estimation (REML) using the
nlme library in R (2002, The R Development Core Team) was used
to analyse the results to account for the unbalanced incomplete
nature of the data. All the data collected for samples of wheat were
combined to detect differences between farms, trailers, and

sampling methods. The same procedure was adopted for the


samples for barley. The REML was used to detect which of the
random elements of the model (farm, trailer, sampler type and
sample number) best described the variance that was seen in the
xed factor (moisture, protein, hardness, specic weight or nes).
The REML analysis was run with all random effects and then had
effects deleted one by one. The model (i.e. random effects structure)
that gave the lowest value for the Akaikes Information Criterion
(AIC), indicating the best t, was selected.
3. Results
3.1. Principal ndings
The analysis showed that for moisture content, protein (dry
matter) DM and hardness, farms and trailers accounted for the
signicant variance in the model but there was no variation
accounted for by samples within trailers or sampling methods. For
specic weight, signicant variance in the model was accounted for
by farms and sampling method which could probably be explained
by the fact that spear sampling can have a polishing effect on the
grain, changing its characteristics and therefore specic weight.
The results for nes only showed an effect for farms. However, the
residuals from the nes analysis were highly asymmetric and the
result should therefore be treated with great caution.
The overall analysis showed that there was no statistically
signicant variation within trailers. This is not to say that variation
did not occur or that such variation could make an important
difference to individuals relying upon a particular sample to make
a decision. To illustrate this point Table 3 shows the worst case
scenario with the maximum and minimum values obtained in
single trailers. The results were recorded from a number of different
trailers, i.e. moisture content from one trailer, protein DM from
another. However, more typical results can be seen in Table 4 which
shows the maximum and minimum values for 10 samples from
a single trailer of barley and single trailer of wheat.
The overall analysis showed the variation between trailers to be
statistically signicant and therefore determines the minimum
sample unit, i.e. each trailer has to be sampled to get a good estimate

Table 3
Greatest ranges of values recorded in single trailers. (Individual
factors were not necessarily drawn from the same trailer.)
Quality factor

Range of values

Protein (DM)
Nitrogen (DM)
Moisture content
Hardness
Specic weight
Fines

10.0e12.0%
1.7e1.8%
16.9e17.9%
30.3e52.7
60.8e69.5 kg/hl
0.02e0.52%

Table 4
Greatest ranges of values recorded in a single trailer.
Quality factor

Range of values (barley)

Range of values (wheat)

Protein (DM)
Nitrogen (DM)
Moisture content
Hardness
Specic weighta
Fines

e
2.1e2.2%
16.6e17.4%
e
60.5e62.7 kg/hl
0.05e0.50%

13.1e13.5%
e
21.0e21.2%
83.2e90.5
62.5e65.1 kg/hl
0.0e0.11%

a
The Infratec instrument had not been calibrated for specic weight and when
checked against a calibrated instrument was found to consistently under-read by
about 5 units. Thus the results reported here are underestimates of the specic
weight of the grain but are consistent and show a real difference.

Please cite this article in press as: Knight, J.D., Wilkin, D.R., Development and validation of on-farm sampling methods for the collection of
marketing..., Journal of Stored Products Research (2010), doi:10.1016/j.jspr.2010.05.004

J.D. Knight, D.R. Wilkin / Journal of Stored Products Research xxx (2010) 1e7
Table 5
Differences in quality parameter values recorded from two combines operating in
the same eld.
Trailer

Protein, DM

Moisture

Hardness

Specic weighta

Fines (%)

Combine 1
Combine 2

9.3
11.8

14.3
16.3

42.9
67.1

74.9
70.6

0.06
0.06

a
The Infratec instrument had not been calibrated for specic weight and when
checked against a calibrated instrument was found to consistently under-read by
about 5 units. Thus the results reported here are underestimates of the specic
weight of the grain but are consistent and show a real difference.

of the quality of the grain. This variation is evident in Table 5 where


two combines were operating in the same eld but in different parts.
3.2. Comparison of pelican and spear sampling
The only statistically signicant difference recorded between
these two sampling methods for both wheat and barley was for
specic weight with spear sampling giving a signicantly higher
value than the pelican sampler (Table 6).
3.3. Comparison of pelican and scoop
This comparison was only made using barley and revealed no
signicant differences for any of the qualities that were measured.
When a probability distribution was produced from the mean and
standard deviations of the samples it can be seen that the scoop
tended to provide a wider range of values, although not producing
a statistically signicant difference (Fig. 2).
3.4. Comparison of single and composite samples
All tests were done on wheat. In general there were no statistically signicant differences between the mean results obtained
Table 6
Results from the comparison of pelican sampling versus spear sampling.
Spear mean
( SE), specic
weighta
Barley 62.4  0.06
Wheat 67.7  0.11

Pelican mean
(SE), specic
weighta

t-statistic

Probability Degrees of
freedom

61.0  0.29
66.4  0.43

5.07
2.99

<0.0001
0.002

17
39

a
The Infratec instrument had not been calibrated for specic weight and when
checked against a calibrated instrument was found to consistently under-read by
about 5 units. Thus the results reported here are underestimates of the specic
weight of the grain but are consistent and show a real difference.

from single samples and composite samples created from those


individual ones (Table 7). Of the 5 comparisons examined 2 showed
differences for nes, 2 showed differences for hardness and a single
sample showed differences for specic weight.
As has already been mentioned the measurement of the nes
was problematic. In one case the composite sample showed
a higher level and in the other a lower level of ne material. Fine
material is, of course, not xed within a grain as is protein or
moisture so that the act of sampling itself may disrupt the
distribution. It is also very likely that ne material is distributed in
an extremely uneven manner thus increasing the error in
sampling. One of the differences in the hardness values was highly
signicant although this appears to be more due to small variability within the sample than a large difference in the actual
values (Tables 7 and 8).

3.5. Effect of changing sampling intensity


To demonstrate the effect of sampling intensity the variability of
a batch of grain (wheat) was used to show the margin of error
associated with taking a single sample as opposed to taking two
samples from a trailer of grain. Figs. 3 and 4 show the sort of level of
variation that was obtained and the effects of taking one or two
samples to get a measure of the moisture content for each load. The
principles are the same for the other factors measured (protein,
specic weight, hardness, nes).
Fig. 3 shows the range of values that are likely to be obtained
from a series of samples of grain coming from a single eld, based
on data obtained from a total of 40 samples. The mean and standard
deviation were calculated and used to produce a probability
distribution curve of the range of moisture contents that may be
found. The black area is the region that contains samples that are
within the value of the mean plus or minus 0.5%, i.e. 18.45  0.5%.
This region contains 83.7% of the possible results, that is to say
there is a probability of 83.7% that any result will lie within this area
or a 14.3% chance (1 in 7 chance) that it is outside. If the tolerance is
extended to 18.45  0.7% then 95% of the values will fall within this
range, i.e. a 5% chance (1 in 20 chance) that a value for moisture
content is 0.7% more or less than the mean, i.e. it lies outside the
range 19.15e17.75%.
Fig. 4 shows the impact of taking two samples. The probability of
getting a value that is greater than 0.5% (1 in 20) of the mean is only
3.4% and if 0.7% is used then the probability of getting a value
outside this range is reduced to 0.05% or one chance in 200.

Fig. 2. Comparison of range and frequency of values obtained for moisture content comparing pelican and scoop sampling.

Please cite this article in press as: Knight, J.D., Wilkin, D.R., Development and validation of on-farm sampling methods for the collection of
marketing..., Journal of Stored Products Research (2010), doi:10.1016/j.jspr.2010.05.004

J.D. Knight, D.R. Wilkin / Journal of Stored Products Research xxx (2010) 1e7

Table 7
Comparison of quality measurement means (SE) calculated from single samples against means from composite samples for 5 farms. All samples were wheat.
Farm
1
2
7
8
8a

Single mean
Composite mean
Single mean
Composite mean
Single mean
Composite mean
Single mean
Composite mean
Single mean
Composite mean

Protein, DM

Moisture

Hardness

Specic weighta

Fines (%)

10.5  0.05
10.5  0.02
10.4  0.60
11.2  0.04
10.1  0.03
10.1  0.02
12.2  0.04
12.2  0.04
13.2  0.06
13.2  0.02

13.4  0.08
13.4  0.02
14.9  0.37
15.1  0.01
12.4  0.17
12.8  0.03
19.6  0.16
19.5  0.05
18.0  0.07
17.9  0.01

37.1  1.09
38.6  0.40
54.9  5.88
63.5  0.60
55.2  0.46
60.1  0.68*
68.1  0.87
66.8  0.55
64.3  0.55
62.7  0.36*

74.3  0.21
74.5  0.12
73.3  1.08
71.7  0.41
76.0  0.15
75.9  0.09
70.5  0.43
70.7  0.21
66.3  0.19
66.7  0.08*

0.08  0.010
0.13  0.015*
0.04  0.007
0.06  0.015
0.12  0.014
0.08  0.007
0.03  0.004
0.02  0.002*
0.04  0.005
0.03  0.004

* indicates a statistically signicant difference; see Table 8


a
The Infratec instrument had not been calibrated for specic weight and when checked against a calibrated instrument was found to consistently under-read by about 5
units. Thus the results reported here are underestimates of the specic weight of the grain but are consistent and show a real difference.

Table 8
Details of statistically signicant differences between single sample values and those
obtained from composite samples.
Single
samples
mean value
Farm 1
e Fines
Farm 7
e Hardness
Farm 8
e Fines
Farm 8A
e Hardness
Farm 8A
e Specic
Weighta

Composite
sample
mean value

T statistic Probability Degrees of


freedom

0.08  0.010 0.13  0.015 2.60


55.2  0.46

60.1  0.68

6.07

0.01

12

<0.0001

13

0.03  0.004 0.02  0.002

2.21

0.02

20

64.3  0.55

62.7  0.55

2.27

0.02

20

66.3  0.19

66.7  0.08

1.83

0.04

20

a
The Infratec instrument had not been calibrated for specic weight and when
checked against a calibrated instrument was found to consistently under-read by
about 5 units. Thus the results reported here are underestimates of the specic
weight of the grain but are consistent and show a real difference.

4. Discussion
All the results gathered indicate that there is inherent variation
in the grain that comes from a single eld and differences are
signicant at farm and eld level and also between trailers. The
variation in the quality of grain within a single trailer was not
shown to be statistically different using the methods tested here.
However, there was variability and whilst not statistically signicant it could be large enough to make the difference between the
grain being rejected or accepted if sampling was inadequate. The
problem facing the grower, merchant or end-user is to get

Fig. 3. The probability distribution curve of moisture content results likely for a single
grain sample from a trailer.

a reasonable indication of the quality of a batch of grain without


expending excess time, and therefore money, in detecting these
differences. The experiments here show that a single sample per
trailer gives a reasonably reliable estimate of the quality of the grain
within a trailer. Taking more than a single sample would often be
impractical and may or may not improve the accuracy of the result.
The key point to come out of this work is that grain is not homogeneous at a scale greater than a trailer and therefore should not be
sampled as if it is so, since this would more than likely give rise to
misleading or inaccurate information.
Given the error associated with measuring moisture content
(and the other parameters) it was concluded from the results that
taking a single sample was a simple and reasonably reliable method
of estimating the qualities of the grain. It is obviously possible to get
greater accuracy with more samples but the extra time, effort and
storage space for samples, not to mention cost, would make this
unacceptable to the majority of growers.
The different sampling methods tested did not appear to show
any statistically signicant differences for the important quality
parameters although spear sampling did serve to increase the
specic weight of a sample by the polishing effect that the extra
handling had on the grain. The level of nes detected by different
sampling methods varied but the nes in the samples were
a continual problem and no rm conclusion can be drawn from the
data although the extra handling and opportunity for ne material
to fall out of the grain may be part of the explanation. Further work
is required to clarify this situation. The use of a pelican sampler,
a scoop or spear sampling appears to be equally effective and does
not give signicantly different results for nitrogen, protein, hardness or moisture content values. Thus it would appear possible to
sample grain safely and reliably on intake from the combine
whatever method is being used to tip the grain.

Fig. 4. The probability distribution curve of moisture content results likely for the
mixture of two grain samples from a trailer.

Please cite this article in press as: Knight, J.D., Wilkin, D.R., Development and validation of on-farm sampling methods for the collection of
marketing..., Journal of Stored Products Research (2010), doi:10.1016/j.jspr.2010.05.004

J.D. Knight, D.R. Wilkin / Journal of Stored Products Research xxx (2010) 1e7

The use of composite samples can save time since the


measurement of the sample need not be done immediately for each
trailer of grain but the samples can be taken from each trailer,
placed in a container, mixed thoroughly and then sampled and the
measurements recorded. This reduces the number of sample bags
that have to be handled and stored and the number of samples for
analysis. The results indicate that for most of the measurements the
results from composite samples are essentially the same as the
average of individual samples. Again there may be some variation
in the value that is obtained for nes between the single samples
and the bulk. This can probably be explained by the extra handling
of the grain giving greater opportunity for the nes to work their
way to the bottom of the sample during handling and mixing if not
done very thoroughly. The difference in the specic weight detected was probably the result of the extra handling that took place
resulting in a polishing of the grain and therefore a higher specic
weight. The records of differences in hardness are not easily
explicable but may have been occasioned by an outlier result being
accompanied by small standard errors.
All persons involved in the grain trade should be made aware of
the fact that there is a relatively large variation in grain quality and
therefore sampling can only ever be indicative of the quality of the
grain and never a denitive value. By following industry supported
protocols (Knight et al., 2003) it should be possible to obtain reliable samples on which to base marketing decisions. However,
specications set in contracts used for the trading of grain need to
acknowledge that this variation exists and criteria should be set to
include this variation.
Acknowledgements
The work in this project was funded by The Home Grown
Cereals Authority project number 2761. The authors of this paper

wish to acknowledge the valuable assistance given by many sectors


of the grain industry with the development of the protocols. We
also wish to thank the farmers who allowed us to work on their
farms at an extremely busy time of the year.
We also wish to thank Foss Instruments for the loan of the
Infratec and the training given in its use and Protimeter for the loan
of a moisture meter, moisture probe and a temperature probe.

References
Gy, P., 2004. Sampling of discrete materials e a new introduction to the theory of
sampling. I. Qualitative approach. Chemometrics and Intelligent Laboratory
Systems 74, 7e24.
Hook, S.C.W., 2004. A National Grain Sampling and Analysis System for Improved
Food Marketing and Safety. Home-Grown Cereals Authority Report 349. HGCA,
Stoneleigh Park, Kenilworth, Warwickshire CV8 2TL, UK.
Jensen, T., Zeller, L., Kelly, R., 2001. Mechanising on-the-run grain sampling. In:
Rowe, B. (Ed.), Proceedings of the Tenth Australian Agronomy Conference,
Hobart, Australia. Australian Society of Agronomy Inc. Available at: http://www.
regional.org.au/au/asa/2001/1/a/jensen.htm (accessed 20.01.10).
Knight, J.D., Wilkin, R., Rivett, J., 2003. Development and Validation of On-Farm
Sampling Protocols for Collection of Marketing (Quality) Samples at Harvest.
Home-Grown Cereals Authority Report 301. HGCA, Stoneleigh Park, Kenilworth,
Warwickshire CV8 2TL, UK.
ISO 13690, 1999. Cereals, Pulses and Milled Products. Sampling of Static Batches.
ISO, ISBN 0 580 34624 2, 26 pp.
Petersen, L., Minkkinen, P., Esbensen, K., 2005. Representative sampling for reliable
data analysis: theory of Sampling. Chemometrics and Intelligent Laboratory
Systems 77, 261e277.
Wilkin, D.R., 1991. An Assessment of Methods of Sampling Bulk Grain. Home-Grown
Cereals Authority Report 34. HGCA, Stoneleigh Park, Kenilworth, Warwickshire
CV8 2TL, UK.
Wilkin, D.R., 1993. An Assessment of Practical Methods for Collecting Samples from
Lorry-loads of Grain. Home-Grown Cereals Authority Report 79. HGCA, Stoneleigh Park, Kenilworth, Warwickshire CV8 2TL, UK.
Wilkin, D.R., Knight, J.D., 1995. The Collection of Samples of Grain: an Assessment of
Current Methods and Problems. Home-Grown Cereals Authority Report 118.
HGCA, Stoneleigh Park, Kenilworth, Warwickshire CV8 2TL, UK.

Please cite this article in press as: Knight, J.D., Wilkin, D.R., Development and validation of on-farm sampling methods for the collection of
marketing..., Journal of Stored Products Research (2010), doi:10.1016/j.jspr.2010.05.004